Eric Lee / smarc-ti-linux-kernel

Download as
Browse Code ยป

Commit eed31172a351856ad18081f501946e1670b6a1f6

Authored by Liu Bo
Committed by Greg Kroah-Hartman
1 parent 6d2a63f3d1

Btrfs: fix scrub_print_warning to handle skinny metadata extents 

commit 6eda71d0c030af0fc2f68aaa676e6d445600855b upstream.

The skinny extents are intepreted incorrectly in scrub_print_warning(),
and end up hitting the BUG() in btrfs_extent_inline_ref_size.

Reported-by: Konstantinos Skarlatos <k.skarlatos@gmail.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Showing 3 changed files with 24 additions and 15 deletions Inline Diff

1 /* 1 /*
2 * Copyright (C) 2011 STRATO. All rights reserved. 2 * Copyright (C) 2011 STRATO. All rights reserved.
3 * 3 *
4 * This program is free software; you can redistribute it and/or 4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public 5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation. 6 * License v2 as published by the Free Software Foundation.
7 * 7 *
8 * This program is distributed in the hope that it will be useful, 8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details. 11 * General Public License for more details.
12 * 12 *
13 * You should have received a copy of the GNU General Public 13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the 14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA. 16 * Boston, MA 021110-1307, USA.
17 */ 17 */
18 18
19 #include <linux/vmalloc.h> 19 #include <linux/vmalloc.h>
20 #include "ctree.h" 20 #include "ctree.h"
21 #include "disk-io.h" 21 #include "disk-io.h"
22 #include "backref.h" 22 #include "backref.h"
23 #include "ulist.h" 23 #include "ulist.h"
24 #include "transaction.h" 24 #include "transaction.h"
25 #include "delayed-ref.h" 25 #include "delayed-ref.h"
26 #include "locking.h" 26 #include "locking.h"
27 27
28 struct extent_inode_elem { 28 struct extent_inode_elem {
29 u64 inum; 29 u64 inum;
30 u64 offset; 30 u64 offset;
31 struct extent_inode_elem *next; 31 struct extent_inode_elem *next;
32 }; 32 };
33 33
34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb, 34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
35 struct btrfs_file_extent_item *fi, 35 struct btrfs_file_extent_item *fi,
36 u64 extent_item_pos, 36 u64 extent_item_pos,
37 struct extent_inode_elem **eie) 37 struct extent_inode_elem **eie)
38 { 38 {
39 u64 offset = 0; 39 u64 offset = 0;
40 struct extent_inode_elem *e; 40 struct extent_inode_elem *e;
41 41
42 if (!btrfs_file_extent_compression(eb, fi) && 42 if (!btrfs_file_extent_compression(eb, fi) &&
43 !btrfs_file_extent_encryption(eb, fi) && 43 !btrfs_file_extent_encryption(eb, fi) &&
44 !btrfs_file_extent_other_encoding(eb, fi)) { 44 !btrfs_file_extent_other_encoding(eb, fi)) {
45 u64 data_offset; 45 u64 data_offset;
46 u64 data_len; 46 u64 data_len;
47 47
48 data_offset = btrfs_file_extent_offset(eb, fi); 48 data_offset = btrfs_file_extent_offset(eb, fi);
49 data_len = btrfs_file_extent_num_bytes(eb, fi); 49 data_len = btrfs_file_extent_num_bytes(eb, fi);
50 50
51 if (extent_item_pos < data_offset || 51 if (extent_item_pos < data_offset ||
52 extent_item_pos >= data_offset + data_len) 52 extent_item_pos >= data_offset + data_len)
53 return 1; 53 return 1;
54 offset = extent_item_pos - data_offset; 54 offset = extent_item_pos - data_offset;
55 } 55 }
56 56
57 e = kmalloc(sizeof(*e), GFP_NOFS); 57 e = kmalloc(sizeof(*e), GFP_NOFS);
58 if (!e) 58 if (!e)
59 return -ENOMEM; 59 return -ENOMEM;
60 60
61 e->next = *eie; 61 e->next = *eie;
62 e->inum = key->objectid; 62 e->inum = key->objectid;
63 e->offset = key->offset + offset; 63 e->offset = key->offset + offset;
64 *eie = e; 64 *eie = e;
65 65
66 return 0; 66 return 0;
67 } 67 }
68 68
69 static void free_inode_elem_list(struct extent_inode_elem *eie) 69 static void free_inode_elem_list(struct extent_inode_elem *eie)
70 { 70 {
71 struct extent_inode_elem *eie_next; 71 struct extent_inode_elem *eie_next;
72 72
73 for (; eie; eie = eie_next) { 73 for (; eie; eie = eie_next) {
74 eie_next = eie->next; 74 eie_next = eie->next;
75 kfree(eie); 75 kfree(eie);
76 } 76 }
77 } 77 }
78 78
79 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte, 79 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
80 u64 extent_item_pos, 80 u64 extent_item_pos,
81 struct extent_inode_elem **eie) 81 struct extent_inode_elem **eie)
82 { 82 {
83 u64 disk_byte; 83 u64 disk_byte;
84 struct btrfs_key key; 84 struct btrfs_key key;
85 struct btrfs_file_extent_item *fi; 85 struct btrfs_file_extent_item *fi;
86 int slot; 86 int slot;
87 int nritems; 87 int nritems;
88 int extent_type; 88 int extent_type;
89 int ret; 89 int ret;
90 90
91 /* 91 /*
92 * from the shared data ref, we only have the leaf but we need 92 * from the shared data ref, we only have the leaf but we need
93 * the key. thus, we must look into all items and see that we 93 * the key. thus, we must look into all items and see that we
94 * find one (some) with a reference to our extent item. 94 * find one (some) with a reference to our extent item.
95 */ 95 */
96 nritems = btrfs_header_nritems(eb); 96 nritems = btrfs_header_nritems(eb);
97 for (slot = 0; slot < nritems; ++slot) { 97 for (slot = 0; slot < nritems; ++slot) {
98 btrfs_item_key_to_cpu(eb, &key, slot); 98 btrfs_item_key_to_cpu(eb, &key, slot);
99 if (key.type != BTRFS_EXTENT_DATA_KEY) 99 if (key.type != BTRFS_EXTENT_DATA_KEY)
100 continue; 100 continue;
101 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); 101 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
102 extent_type = btrfs_file_extent_type(eb, fi); 102 extent_type = btrfs_file_extent_type(eb, fi);
103 if (extent_type == BTRFS_FILE_EXTENT_INLINE) 103 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
104 continue; 104 continue;
105 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ 105 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
106 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); 106 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
107 if (disk_byte != wanted_disk_byte) 107 if (disk_byte != wanted_disk_byte)
108 continue; 108 continue;
109 109
110 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie); 110 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
111 if (ret < 0) 111 if (ret < 0)
112 return ret; 112 return ret;
113 } 113 }
114 114
115 return 0; 115 return 0;
116 } 116 }
117 117
118 /* 118 /*
119 * this structure records all encountered refs on the way up to the root 119 * this structure records all encountered refs on the way up to the root
120 */ 120 */
121 struct __prelim_ref { 121 struct __prelim_ref {
122 struct list_head list; 122 struct list_head list;
123 u64 root_id; 123 u64 root_id;
124 struct btrfs_key key_for_search; 124 struct btrfs_key key_for_search;
125 int level; 125 int level;
126 int count; 126 int count;
127 struct extent_inode_elem *inode_list; 127 struct extent_inode_elem *inode_list;
128 u64 parent; 128 u64 parent;
129 u64 wanted_disk_byte; 129 u64 wanted_disk_byte;
130 }; 130 };
131 131
132 static struct kmem_cache *btrfs_prelim_ref_cache; 132 static struct kmem_cache *btrfs_prelim_ref_cache;
133 133
134 int __init btrfs_prelim_ref_init(void) 134 int __init btrfs_prelim_ref_init(void)
135 { 135 {
136 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", 136 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
137 sizeof(struct __prelim_ref), 137 sizeof(struct __prelim_ref),
138 0, 138 0,
139 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, 139 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
140 NULL); 140 NULL);
141 if (!btrfs_prelim_ref_cache) 141 if (!btrfs_prelim_ref_cache)
142 return -ENOMEM; 142 return -ENOMEM;
143 return 0; 143 return 0;
144 } 144 }
145 145
146 void btrfs_prelim_ref_exit(void) 146 void btrfs_prelim_ref_exit(void)
147 { 147 {
148 if (btrfs_prelim_ref_cache) 148 if (btrfs_prelim_ref_cache)
149 kmem_cache_destroy(btrfs_prelim_ref_cache); 149 kmem_cache_destroy(btrfs_prelim_ref_cache);
150 } 150 }
151 151
152 /* 152 /*
153 * the rules for all callers of this function are: 153 * the rules for all callers of this function are:
154 * - obtaining the parent is the goal 154 * - obtaining the parent is the goal
155 * - if you add a key, you must know that it is a correct key 155 * - if you add a key, you must know that it is a correct key
156 * - if you cannot add the parent or a correct key, then we will look into the 156 * - if you cannot add the parent or a correct key, then we will look into the
157 * block later to set a correct key 157 * block later to set a correct key
158 * 158 *
159 * delayed refs 159 * delayed refs
160 * ============ 160 * ============
161 * backref type | shared | indirect | shared | indirect 161 * backref type | shared | indirect | shared | indirect
162 * information | tree | tree | data | data 162 * information | tree | tree | data | data
163 * --------------------+--------+----------+--------+---------- 163 * --------------------+--------+----------+--------+----------
164 * parent logical | y | - | - | - 164 * parent logical | y | - | - | -
165 * key to resolve | - | y | y | y 165 * key to resolve | - | y | y | y
166 * tree block logical | - | - | - | - 166 * tree block logical | - | - | - | -
167 * root for resolving | y | y | y | y 167 * root for resolving | y | y | y | y
168 * 168 *
169 * - column 1: we've the parent -> done 169 * - column 1: we've the parent -> done
170 * - column 2, 3, 4: we use the key to find the parent 170 * - column 2, 3, 4: we use the key to find the parent
171 * 171 *
172 * on disk refs (inline or keyed) 172 * on disk refs (inline or keyed)
173 * ============================== 173 * ==============================
174 * backref type | shared | indirect | shared | indirect 174 * backref type | shared | indirect | shared | indirect
175 * information | tree | tree | data | data 175 * information | tree | tree | data | data
176 * --------------------+--------+----------+--------+---------- 176 * --------------------+--------+----------+--------+----------
177 * parent logical | y | - | y | - 177 * parent logical | y | - | y | -
178 * key to resolve | - | - | - | y 178 * key to resolve | - | - | - | y
179 * tree block logical | y | y | y | y 179 * tree block logical | y | y | y | y
180 * root for resolving | - | y | y | y 180 * root for resolving | - | y | y | y
181 * 181 *
182 * - column 1, 3: we've the parent -> done 182 * - column 1, 3: we've the parent -> done
183 * - column 2: we take the first key from the block to find the parent 183 * - column 2: we take the first key from the block to find the parent
184 * (see __add_missing_keys) 184 * (see __add_missing_keys)
185 * - column 4: we use the key to find the parent 185 * - column 4: we use the key to find the parent
186 * 186 *
187 * additional information that's available but not required to find the parent 187 * additional information that's available but not required to find the parent
188 * block might help in merging entries to gain some speed. 188 * block might help in merging entries to gain some speed.
189 */ 189 */
190 190
191 static int __add_prelim_ref(struct list_head *head, u64 root_id, 191 static int __add_prelim_ref(struct list_head *head, u64 root_id,
192 struct btrfs_key *key, int level, 192 struct btrfs_key *key, int level,
193 u64 parent, u64 wanted_disk_byte, int count, 193 u64 parent, u64 wanted_disk_byte, int count,
194 gfp_t gfp_mask) 194 gfp_t gfp_mask)
195 { 195 {
196 struct __prelim_ref *ref; 196 struct __prelim_ref *ref;
197 197
198 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) 198 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
199 return 0; 199 return 0;
200 200
201 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask); 201 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
202 if (!ref) 202 if (!ref)
203 return -ENOMEM; 203 return -ENOMEM;
204 204
205 ref->root_id = root_id; 205 ref->root_id = root_id;
206 if (key) 206 if (key)
207 ref->key_for_search = *key; 207 ref->key_for_search = *key;
208 else 208 else
209 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); 209 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
210 210
211 ref->inode_list = NULL; 211 ref->inode_list = NULL;
212 ref->level = level; 212 ref->level = level;
213 ref->count = count; 213 ref->count = count;
214 ref->parent = parent; 214 ref->parent = parent;
215 ref->wanted_disk_byte = wanted_disk_byte; 215 ref->wanted_disk_byte = wanted_disk_byte;
216 list_add_tail(&ref->list, head); 216 list_add_tail(&ref->list, head);
217 217
218 return 0; 218 return 0;
219 } 219 }
220 220
221 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, 221 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
222 struct ulist *parents, struct __prelim_ref *ref, 222 struct ulist *parents, struct __prelim_ref *ref,
223 int level, u64 time_seq, const u64 *extent_item_pos, 223 int level, u64 time_seq, const u64 *extent_item_pos,
224 u64 total_refs) 224 u64 total_refs)
225 { 225 {
226 int ret = 0; 226 int ret = 0;
227 int slot; 227 int slot;
228 struct extent_buffer *eb; 228 struct extent_buffer *eb;
229 struct btrfs_key key; 229 struct btrfs_key key;
230 struct btrfs_key *key_for_search = &ref->key_for_search; 230 struct btrfs_key *key_for_search = &ref->key_for_search;
231 struct btrfs_file_extent_item *fi; 231 struct btrfs_file_extent_item *fi;
232 struct extent_inode_elem *eie = NULL, *old = NULL; 232 struct extent_inode_elem *eie = NULL, *old = NULL;
233 u64 disk_byte; 233 u64 disk_byte;
234 u64 wanted_disk_byte = ref->wanted_disk_byte; 234 u64 wanted_disk_byte = ref->wanted_disk_byte;
235 u64 count = 0; 235 u64 count = 0;
236 236
237 if (level != 0) { 237 if (level != 0) {
238 eb = path->nodes[level]; 238 eb = path->nodes[level];
239 ret = ulist_add(parents, eb->start, 0, GFP_NOFS); 239 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
240 if (ret < 0) 240 if (ret < 0)
241 return ret; 241 return ret;
242 return 0; 242 return 0;
243 } 243 }
244 244
245 /* 245 /*
246 * We normally enter this function with the path already pointing to 246 * We normally enter this function with the path already pointing to
247 * the first item to check. But sometimes, we may enter it with 247 * the first item to check. But sometimes, we may enter it with
248 * slot==nritems. In that case, go to the next leaf before we continue. 248 * slot==nritems. In that case, go to the next leaf before we continue.
249 */ 249 */
250 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) 250 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
251 ret = btrfs_next_old_leaf(root, path, time_seq); 251 ret = btrfs_next_old_leaf(root, path, time_seq);
252 252
253 while (!ret && count < total_refs) { 253 while (!ret && count < total_refs) {
254 eb = path->nodes[0]; 254 eb = path->nodes[0];
255 slot = path->slots[0]; 255 slot = path->slots[0];
256 256
257 btrfs_item_key_to_cpu(eb, &key, slot); 257 btrfs_item_key_to_cpu(eb, &key, slot);
258 258
259 if (key.objectid != key_for_search->objectid || 259 if (key.objectid != key_for_search->objectid ||
260 key.type != BTRFS_EXTENT_DATA_KEY) 260 key.type != BTRFS_EXTENT_DATA_KEY)
261 break; 261 break;
262 262
263 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); 263 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
264 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); 264 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
265 265
266 if (disk_byte == wanted_disk_byte) { 266 if (disk_byte == wanted_disk_byte) {
267 eie = NULL; 267 eie = NULL;
268 old = NULL; 268 old = NULL;
269 count++; 269 count++;
270 if (extent_item_pos) { 270 if (extent_item_pos) {
271 ret = check_extent_in_eb(&key, eb, fi, 271 ret = check_extent_in_eb(&key, eb, fi,
272 *extent_item_pos, 272 *extent_item_pos,
273 &eie); 273 &eie);
274 if (ret < 0) 274 if (ret < 0)
275 break; 275 break;
276 } 276 }
277 if (ret > 0) 277 if (ret > 0)
278 goto next; 278 goto next;
279 ret = ulist_add_merge(parents, eb->start, 279 ret = ulist_add_merge(parents, eb->start,
280 (uintptr_t)eie, 280 (uintptr_t)eie,
281 (u64 *)&old, GFP_NOFS); 281 (u64 *)&old, GFP_NOFS);
282 if (ret < 0) 282 if (ret < 0)
283 break; 283 break;
284 if (!ret && extent_item_pos) { 284 if (!ret && extent_item_pos) {
285 while (old->next) 285 while (old->next)
286 old = old->next; 286 old = old->next;
287 old->next = eie; 287 old->next = eie;
288 } 288 }
289 eie = NULL; 289 eie = NULL;
290 } 290 }
291 next: 291 next:
292 ret = btrfs_next_old_item(root, path, time_seq); 292 ret = btrfs_next_old_item(root, path, time_seq);
293 } 293 }
294 294
295 if (ret > 0) 295 if (ret > 0)
296 ret = 0; 296 ret = 0;
297 else if (ret < 0) 297 else if (ret < 0)
298 free_inode_elem_list(eie); 298 free_inode_elem_list(eie);
299 return ret; 299 return ret;
300 } 300 }
301 301
302 /* 302 /*
303 * resolve an indirect backref in the form (root_id, key, level) 303 * resolve an indirect backref in the form (root_id, key, level)
304 * to a logical address 304 * to a logical address
305 */ 305 */
306 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info, 306 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
307 struct btrfs_path *path, u64 time_seq, 307 struct btrfs_path *path, u64 time_seq,
308 struct __prelim_ref *ref, 308 struct __prelim_ref *ref,
309 struct ulist *parents, 309 struct ulist *parents,
310 const u64 *extent_item_pos, u64 total_refs) 310 const u64 *extent_item_pos, u64 total_refs)
311 { 311 {
312 struct btrfs_root *root; 312 struct btrfs_root *root;
313 struct btrfs_key root_key; 313 struct btrfs_key root_key;
314 struct extent_buffer *eb; 314 struct extent_buffer *eb;
315 int ret = 0; 315 int ret = 0;
316 int root_level; 316 int root_level;
317 int level = ref->level; 317 int level = ref->level;
318 int index; 318 int index;
319 319
320 root_key.objectid = ref->root_id; 320 root_key.objectid = ref->root_id;
321 root_key.type = BTRFS_ROOT_ITEM_KEY; 321 root_key.type = BTRFS_ROOT_ITEM_KEY;
322 root_key.offset = (u64)-1; 322 root_key.offset = (u64)-1;
323 323
324 index = srcu_read_lock(&fs_info->subvol_srcu); 324 index = srcu_read_lock(&fs_info->subvol_srcu);
325 325
326 root = btrfs_read_fs_root_no_name(fs_info, &root_key); 326 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
327 if (IS_ERR(root)) { 327 if (IS_ERR(root)) {
328 srcu_read_unlock(&fs_info->subvol_srcu, index); 328 srcu_read_unlock(&fs_info->subvol_srcu, index);
329 ret = PTR_ERR(root); 329 ret = PTR_ERR(root);
330 goto out; 330 goto out;
331 } 331 }
332 332
333 if (path->search_commit_root) 333 if (path->search_commit_root)
334 root_level = btrfs_header_level(root->commit_root); 334 root_level = btrfs_header_level(root->commit_root);
335 else 335 else
336 root_level = btrfs_old_root_level(root, time_seq); 336 root_level = btrfs_old_root_level(root, time_seq);
337 337
338 if (root_level + 1 == level) { 338 if (root_level + 1 == level) {
339 srcu_read_unlock(&fs_info->subvol_srcu, index); 339 srcu_read_unlock(&fs_info->subvol_srcu, index);
340 goto out; 340 goto out;
341 } 341 }
342 342
343 path->lowest_level = level; 343 path->lowest_level = level;
344 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq); 344 ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
345 345
346 /* root node has been locked, we can release @subvol_srcu safely here */ 346 /* root node has been locked, we can release @subvol_srcu safely here */
347 srcu_read_unlock(&fs_info->subvol_srcu, index); 347 srcu_read_unlock(&fs_info->subvol_srcu, index);
348 348
349 pr_debug("search slot in root %llu (level %d, ref count %d) returned " 349 pr_debug("search slot in root %llu (level %d, ref count %d) returned "
350 "%d for key (%llu %u %llu)\n", 350 "%d for key (%llu %u %llu)\n",
351 ref->root_id, level, ref->count, ret, 351 ref->root_id, level, ref->count, ret,
352 ref->key_for_search.objectid, ref->key_for_search.type, 352 ref->key_for_search.objectid, ref->key_for_search.type,
353 ref->key_for_search.offset); 353 ref->key_for_search.offset);
354 if (ret < 0) 354 if (ret < 0)
355 goto out; 355 goto out;
356 356
357 eb = path->nodes[level]; 357 eb = path->nodes[level];
358 while (!eb) { 358 while (!eb) {
359 if (WARN_ON(!level)) { 359 if (WARN_ON(!level)) {
360 ret = 1; 360 ret = 1;
361 goto out; 361 goto out;
362 } 362 }
363 level--; 363 level--;
364 eb = path->nodes[level]; 364 eb = path->nodes[level];
365 } 365 }
366 366
367 ret = add_all_parents(root, path, parents, ref, level, time_seq, 367 ret = add_all_parents(root, path, parents, ref, level, time_seq,
368 extent_item_pos, total_refs); 368 extent_item_pos, total_refs);
369 out: 369 out:
370 path->lowest_level = 0; 370 path->lowest_level = 0;
371 btrfs_release_path(path); 371 btrfs_release_path(path);
372 return ret; 372 return ret;
373 } 373 }
374 374
375 /* 375 /*
376 * resolve all indirect backrefs from the list 376 * resolve all indirect backrefs from the list
377 */ 377 */
378 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info, 378 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
379 struct btrfs_path *path, u64 time_seq, 379 struct btrfs_path *path, u64 time_seq,
380 struct list_head *head, 380 struct list_head *head,
381 const u64 *extent_item_pos, u64 total_refs) 381 const u64 *extent_item_pos, u64 total_refs)
382 { 382 {
383 int err; 383 int err;
384 int ret = 0; 384 int ret = 0;
385 struct __prelim_ref *ref; 385 struct __prelim_ref *ref;
386 struct __prelim_ref *ref_safe; 386 struct __prelim_ref *ref_safe;
387 struct __prelim_ref *new_ref; 387 struct __prelim_ref *new_ref;
388 struct ulist *parents; 388 struct ulist *parents;
389 struct ulist_node *node; 389 struct ulist_node *node;
390 struct ulist_iterator uiter; 390 struct ulist_iterator uiter;
391 391
392 parents = ulist_alloc(GFP_NOFS); 392 parents = ulist_alloc(GFP_NOFS);
393 if (!parents) 393 if (!parents)
394 return -ENOMEM; 394 return -ENOMEM;
395 395
396 /* 396 /*
397 * _safe allows us to insert directly after the current item without 397 * _safe allows us to insert directly after the current item without
398 * iterating over the newly inserted items. 398 * iterating over the newly inserted items.
399 * we're also allowed to re-assign ref during iteration. 399 * we're also allowed to re-assign ref during iteration.
400 */ 400 */
401 list_for_each_entry_safe(ref, ref_safe, head, list) { 401 list_for_each_entry_safe(ref, ref_safe, head, list) {
402 if (ref->parent) /* already direct */ 402 if (ref->parent) /* already direct */
403 continue; 403 continue;
404 if (ref->count == 0) 404 if (ref->count == 0)
405 continue; 405 continue;
406 err = __resolve_indirect_ref(fs_info, path, time_seq, ref, 406 err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
407 parents, extent_item_pos, 407 parents, extent_item_pos,
408 total_refs); 408 total_refs);
409 /* 409 /*
410 * we can only tolerate ENOENT,otherwise,we should catch error 410 * we can only tolerate ENOENT,otherwise,we should catch error
411 * and return directly. 411 * and return directly.
412 */ 412 */
413 if (err == -ENOENT) { 413 if (err == -ENOENT) {
414 continue; 414 continue;
415 } else if (err) { 415 } else if (err) {
416 ret = err; 416 ret = err;
417 goto out; 417 goto out;
418 } 418 }
419 419
420 /* we put the first parent into the ref at hand */ 420 /* we put the first parent into the ref at hand */
421 ULIST_ITER_INIT(&uiter); 421 ULIST_ITER_INIT(&uiter);
422 node = ulist_next(parents, &uiter); 422 node = ulist_next(parents, &uiter);
423 ref->parent = node ? node->val : 0; 423 ref->parent = node ? node->val : 0;
424 ref->inode_list = node ? 424 ref->inode_list = node ?
425 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL; 425 (struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
426 426
427 /* additional parents require new refs being added here */ 427 /* additional parents require new refs being added here */
428 while ((node = ulist_next(parents, &uiter))) { 428 while ((node = ulist_next(parents, &uiter))) {
429 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, 429 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
430 GFP_NOFS); 430 GFP_NOFS);
431 if (!new_ref) { 431 if (!new_ref) {
432 ret = -ENOMEM; 432 ret = -ENOMEM;
433 goto out; 433 goto out;
434 } 434 }
435 memcpy(new_ref, ref, sizeof(*ref)); 435 memcpy(new_ref, ref, sizeof(*ref));
436 new_ref->parent = node->val; 436 new_ref->parent = node->val;
437 new_ref->inode_list = (struct extent_inode_elem *) 437 new_ref->inode_list = (struct extent_inode_elem *)
438 (uintptr_t)node->aux; 438 (uintptr_t)node->aux;
439 list_add(&new_ref->list, &ref->list); 439 list_add(&new_ref->list, &ref->list);
440 } 440 }
441 ulist_reinit(parents); 441 ulist_reinit(parents);
442 } 442 }
443 out: 443 out:
444 ulist_free(parents); 444 ulist_free(parents);
445 return ret; 445 return ret;
446 } 446 }
447 447
448 static inline int ref_for_same_block(struct __prelim_ref *ref1, 448 static inline int ref_for_same_block(struct __prelim_ref *ref1,
449 struct __prelim_ref *ref2) 449 struct __prelim_ref *ref2)
450 { 450 {
451 if (ref1->level != ref2->level) 451 if (ref1->level != ref2->level)
452 return 0; 452 return 0;
453 if (ref1->root_id != ref2->root_id) 453 if (ref1->root_id != ref2->root_id)
454 return 0; 454 return 0;
455 if (ref1->key_for_search.type != ref2->key_for_search.type) 455 if (ref1->key_for_search.type != ref2->key_for_search.type)
456 return 0; 456 return 0;
457 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid) 457 if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
458 return 0; 458 return 0;
459 if (ref1->key_for_search.offset != ref2->key_for_search.offset) 459 if (ref1->key_for_search.offset != ref2->key_for_search.offset)
460 return 0; 460 return 0;
461 if (ref1->parent != ref2->parent) 461 if (ref1->parent != ref2->parent)
462 return 0; 462 return 0;
463 463
464 return 1; 464 return 1;
465 } 465 }
466 466
467 /* 467 /*
468 * read tree blocks and add keys where required. 468 * read tree blocks and add keys where required.
469 */ 469 */
470 static int __add_missing_keys(struct btrfs_fs_info *fs_info, 470 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
471 struct list_head *head) 471 struct list_head *head)
472 { 472 {
473 struct list_head *pos; 473 struct list_head *pos;
474 struct extent_buffer *eb; 474 struct extent_buffer *eb;
475 475
476 list_for_each(pos, head) { 476 list_for_each(pos, head) {
477 struct __prelim_ref *ref; 477 struct __prelim_ref *ref;
478 ref = list_entry(pos, struct __prelim_ref, list); 478 ref = list_entry(pos, struct __prelim_ref, list);
479 479
480 if (ref->parent) 480 if (ref->parent)
481 continue; 481 continue;
482 if (ref->key_for_search.type) 482 if (ref->key_for_search.type)
483 continue; 483 continue;
484 BUG_ON(!ref->wanted_disk_byte); 484 BUG_ON(!ref->wanted_disk_byte);
485 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte, 485 eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
486 fs_info->tree_root->leafsize, 0); 486 fs_info->tree_root->leafsize, 0);
487 if (!eb || !extent_buffer_uptodate(eb)) { 487 if (!eb || !extent_buffer_uptodate(eb)) {
488 free_extent_buffer(eb); 488 free_extent_buffer(eb);
489 return -EIO; 489 return -EIO;
490 } 490 }
491 btrfs_tree_read_lock(eb); 491 btrfs_tree_read_lock(eb);
492 if (btrfs_header_level(eb) == 0) 492 if (btrfs_header_level(eb) == 0)
493 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); 493 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
494 else 494 else
495 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); 495 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
496 btrfs_tree_read_unlock(eb); 496 btrfs_tree_read_unlock(eb);
497 free_extent_buffer(eb); 497 free_extent_buffer(eb);
498 } 498 }
499 return 0; 499 return 0;
500 } 500 }
501 501
502 /* 502 /*
503 * merge two lists of backrefs and adjust counts accordingly 503 * merge two lists of backrefs and adjust counts accordingly
504 * 504 *
505 * mode = 1: merge identical keys, if key is set 505 * mode = 1: merge identical keys, if key is set
506 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here. 506 * FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
507 * additionally, we could even add a key range for the blocks we 507 * additionally, we could even add a key range for the blocks we
508 * looked into to merge even more (-> replace unresolved refs by those 508 * looked into to merge even more (-> replace unresolved refs by those
509 * having a parent). 509 * having a parent).
510 * mode = 2: merge identical parents 510 * mode = 2: merge identical parents
511 */ 511 */
512 static void __merge_refs(struct list_head *head, int mode) 512 static void __merge_refs(struct list_head *head, int mode)
513 { 513 {
514 struct list_head *pos1; 514 struct list_head *pos1;
515 515
516 list_for_each(pos1, head) { 516 list_for_each(pos1, head) {
517 struct list_head *n2; 517 struct list_head *n2;
518 struct list_head *pos2; 518 struct list_head *pos2;
519 struct __prelim_ref *ref1; 519 struct __prelim_ref *ref1;
520 520
521 ref1 = list_entry(pos1, struct __prelim_ref, list); 521 ref1 = list_entry(pos1, struct __prelim_ref, list);
522 522
523 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head; 523 for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
524 pos2 = n2, n2 = pos2->next) { 524 pos2 = n2, n2 = pos2->next) {
525 struct __prelim_ref *ref2; 525 struct __prelim_ref *ref2;
526 struct __prelim_ref *xchg; 526 struct __prelim_ref *xchg;
527 struct extent_inode_elem *eie; 527 struct extent_inode_elem *eie;
528 528
529 ref2 = list_entry(pos2, struct __prelim_ref, list); 529 ref2 = list_entry(pos2, struct __prelim_ref, list);
530 530
531 if (mode == 1) { 531 if (mode == 1) {
532 if (!ref_for_same_block(ref1, ref2)) 532 if (!ref_for_same_block(ref1, ref2))
533 continue; 533 continue;
534 if (!ref1->parent && ref2->parent) { 534 if (!ref1->parent && ref2->parent) {
535 xchg = ref1; 535 xchg = ref1;
536 ref1 = ref2; 536 ref1 = ref2;
537 ref2 = xchg; 537 ref2 = xchg;
538 } 538 }
539 } else { 539 } else {
540 if (ref1->parent != ref2->parent) 540 if (ref1->parent != ref2->parent)
541 continue; 541 continue;
542 } 542 }
543 543
544 eie = ref1->inode_list; 544 eie = ref1->inode_list;
545 while (eie && eie->next) 545 while (eie && eie->next)
546 eie = eie->next; 546 eie = eie->next;
547 if (eie) 547 if (eie)
548 eie->next = ref2->inode_list; 548 eie->next = ref2->inode_list;
549 else 549 else
550 ref1->inode_list = ref2->inode_list; 550 ref1->inode_list = ref2->inode_list;
551 ref1->count += ref2->count; 551 ref1->count += ref2->count;
552 552
553 list_del(&ref2->list); 553 list_del(&ref2->list);
554 kmem_cache_free(btrfs_prelim_ref_cache, ref2); 554 kmem_cache_free(btrfs_prelim_ref_cache, ref2);
555 } 555 }
556 556
557 } 557 }
558 } 558 }
559 559
560 /* 560 /*
561 * add all currently queued delayed refs from this head whose seq nr is 561 * add all currently queued delayed refs from this head whose seq nr is
562 * smaller or equal that seq to the list 562 * smaller or equal that seq to the list
563 */ 563 */
564 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq, 564 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
565 struct list_head *prefs, u64 *total_refs) 565 struct list_head *prefs, u64 *total_refs)
566 { 566 {
567 struct btrfs_delayed_extent_op *extent_op = head->extent_op; 567 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
568 struct rb_node *n = &head->node.rb_node; 568 struct rb_node *n = &head->node.rb_node;
569 struct btrfs_key key; 569 struct btrfs_key key;
570 struct btrfs_key op_key = {0}; 570 struct btrfs_key op_key = {0};
571 int sgn; 571 int sgn;
572 int ret = 0; 572 int ret = 0;
573 573
574 if (extent_op && extent_op->update_key) 574 if (extent_op && extent_op->update_key)
575 btrfs_disk_key_to_cpu(&op_key, &extent_op->key); 575 btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
576 576
577 spin_lock(&head->lock); 577 spin_lock(&head->lock);
578 n = rb_first(&head->ref_root); 578 n = rb_first(&head->ref_root);
579 while (n) { 579 while (n) {
580 struct btrfs_delayed_ref_node *node; 580 struct btrfs_delayed_ref_node *node;
581 node = rb_entry(n, struct btrfs_delayed_ref_node, 581 node = rb_entry(n, struct btrfs_delayed_ref_node,
582 rb_node); 582 rb_node);
583 n = rb_next(n); 583 n = rb_next(n);
584 if (node->seq > seq) 584 if (node->seq > seq)
585 continue; 585 continue;
586 586
587 switch (node->action) { 587 switch (node->action) {
588 case BTRFS_ADD_DELAYED_EXTENT: 588 case BTRFS_ADD_DELAYED_EXTENT:
589 case BTRFS_UPDATE_DELAYED_HEAD: 589 case BTRFS_UPDATE_DELAYED_HEAD:
590 WARN_ON(1); 590 WARN_ON(1);
591 continue; 591 continue;
592 case BTRFS_ADD_DELAYED_REF: 592 case BTRFS_ADD_DELAYED_REF:
593 sgn = 1; 593 sgn = 1;
594 break; 594 break;
595 case BTRFS_DROP_DELAYED_REF: 595 case BTRFS_DROP_DELAYED_REF:
596 sgn = -1; 596 sgn = -1;
597 break; 597 break;
598 default: 598 default:
599 BUG_ON(1); 599 BUG_ON(1);
600 } 600 }
601 *total_refs += (node->ref_mod * sgn); 601 *total_refs += (node->ref_mod * sgn);
602 switch (node->type) { 602 switch (node->type) {
603 case BTRFS_TREE_BLOCK_REF_KEY: { 603 case BTRFS_TREE_BLOCK_REF_KEY: {
604 struct btrfs_delayed_tree_ref *ref; 604 struct btrfs_delayed_tree_ref *ref;
605 605
606 ref = btrfs_delayed_node_to_tree_ref(node); 606 ref = btrfs_delayed_node_to_tree_ref(node);
607 ret = __add_prelim_ref(prefs, ref->root, &op_key, 607 ret = __add_prelim_ref(prefs, ref->root, &op_key,
608 ref->level + 1, 0, node->bytenr, 608 ref->level + 1, 0, node->bytenr,
609 node->ref_mod * sgn, GFP_ATOMIC); 609 node->ref_mod * sgn, GFP_ATOMIC);
610 break; 610 break;
611 } 611 }
612 case BTRFS_SHARED_BLOCK_REF_KEY: { 612 case BTRFS_SHARED_BLOCK_REF_KEY: {
613 struct btrfs_delayed_tree_ref *ref; 613 struct btrfs_delayed_tree_ref *ref;
614 614
615 ref = btrfs_delayed_node_to_tree_ref(node); 615 ref = btrfs_delayed_node_to_tree_ref(node);
616 ret = __add_prelim_ref(prefs, ref->root, NULL, 616 ret = __add_prelim_ref(prefs, ref->root, NULL,
617 ref->level + 1, ref->parent, 617 ref->level + 1, ref->parent,
618 node->bytenr, 618 node->bytenr,
619 node->ref_mod * sgn, GFP_ATOMIC); 619 node->ref_mod * sgn, GFP_ATOMIC);
620 break; 620 break;
621 } 621 }
622 case BTRFS_EXTENT_DATA_REF_KEY: { 622 case BTRFS_EXTENT_DATA_REF_KEY: {
623 struct btrfs_delayed_data_ref *ref; 623 struct btrfs_delayed_data_ref *ref;
624 ref = btrfs_delayed_node_to_data_ref(node); 624 ref = btrfs_delayed_node_to_data_ref(node);
625 625
626 key.objectid = ref->objectid; 626 key.objectid = ref->objectid;
627 key.type = BTRFS_EXTENT_DATA_KEY; 627 key.type = BTRFS_EXTENT_DATA_KEY;
628 key.offset = ref->offset; 628 key.offset = ref->offset;
629 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0, 629 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
630 node->bytenr, 630 node->bytenr,
631 node->ref_mod * sgn, GFP_ATOMIC); 631 node->ref_mod * sgn, GFP_ATOMIC);
632 break; 632 break;
633 } 633 }
634 case BTRFS_SHARED_DATA_REF_KEY: { 634 case BTRFS_SHARED_DATA_REF_KEY: {
635 struct btrfs_delayed_data_ref *ref; 635 struct btrfs_delayed_data_ref *ref;
636 636
637 ref = btrfs_delayed_node_to_data_ref(node); 637 ref = btrfs_delayed_node_to_data_ref(node);
638 638
639 key.objectid = ref->objectid; 639 key.objectid = ref->objectid;
640 key.type = BTRFS_EXTENT_DATA_KEY; 640 key.type = BTRFS_EXTENT_DATA_KEY;
641 key.offset = ref->offset; 641 key.offset = ref->offset;
642 ret = __add_prelim_ref(prefs, ref->root, &key, 0, 642 ret = __add_prelim_ref(prefs, ref->root, &key, 0,
643 ref->parent, node->bytenr, 643 ref->parent, node->bytenr,
644 node->ref_mod * sgn, GFP_ATOMIC); 644 node->ref_mod * sgn, GFP_ATOMIC);
645 break; 645 break;
646 } 646 }
647 default: 647 default:
648 WARN_ON(1); 648 WARN_ON(1);
649 } 649 }
650 if (ret) 650 if (ret)
651 break; 651 break;
652 } 652 }
653 spin_unlock(&head->lock); 653 spin_unlock(&head->lock);
654 return ret; 654 return ret;
655 } 655 }
656 656
657 /* 657 /*
658 * add all inline backrefs for bytenr to the list 658 * add all inline backrefs for bytenr to the list
659 */ 659 */
660 static int __add_inline_refs(struct btrfs_fs_info *fs_info, 660 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
661 struct btrfs_path *path, u64 bytenr, 661 struct btrfs_path *path, u64 bytenr,
662 int *info_level, struct list_head *prefs, 662 int *info_level, struct list_head *prefs,
663 u64 *total_refs) 663 u64 *total_refs)
664 { 664 {
665 int ret = 0; 665 int ret = 0;
666 int slot; 666 int slot;
667 struct extent_buffer *leaf; 667 struct extent_buffer *leaf;
668 struct btrfs_key key; 668 struct btrfs_key key;
669 struct btrfs_key found_key; 669 struct btrfs_key found_key;
670 unsigned long ptr; 670 unsigned long ptr;
671 unsigned long end; 671 unsigned long end;
672 struct btrfs_extent_item *ei; 672 struct btrfs_extent_item *ei;
673 u64 flags; 673 u64 flags;
674 u64 item_size; 674 u64 item_size;
675 675
676 /* 676 /*
677 * enumerate all inline refs 677 * enumerate all inline refs
678 */ 678 */
679 leaf = path->nodes[0]; 679 leaf = path->nodes[0];
680 slot = path->slots[0]; 680 slot = path->slots[0];
681 681
682 item_size = btrfs_item_size_nr(leaf, slot); 682 item_size = btrfs_item_size_nr(leaf, slot);
683 BUG_ON(item_size < sizeof(*ei)); 683 BUG_ON(item_size < sizeof(*ei));
684 684
685 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 685 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
686 flags = btrfs_extent_flags(leaf, ei); 686 flags = btrfs_extent_flags(leaf, ei);
687 *total_refs += btrfs_extent_refs(leaf, ei); 687 *total_refs += btrfs_extent_refs(leaf, ei);
688 btrfs_item_key_to_cpu(leaf, &found_key, slot); 688 btrfs_item_key_to_cpu(leaf, &found_key, slot);
689 689
690 ptr = (unsigned long)(ei + 1); 690 ptr = (unsigned long)(ei + 1);
691 end = (unsigned long)ei + item_size; 691 end = (unsigned long)ei + item_size;
692 692
693 if (found_key.type == BTRFS_EXTENT_ITEM_KEY && 693 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
694 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 694 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
695 struct btrfs_tree_block_info *info; 695 struct btrfs_tree_block_info *info;
696 696
697 info = (struct btrfs_tree_block_info *)ptr; 697 info = (struct btrfs_tree_block_info *)ptr;
698 *info_level = btrfs_tree_block_level(leaf, info); 698 *info_level = btrfs_tree_block_level(leaf, info);
699 ptr += sizeof(struct btrfs_tree_block_info); 699 ptr += sizeof(struct btrfs_tree_block_info);
700 BUG_ON(ptr > end); 700 BUG_ON(ptr > end);
701 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) { 701 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
702 *info_level = found_key.offset; 702 *info_level = found_key.offset;
703 } else { 703 } else {
704 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); 704 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
705 } 705 }
706 706
707 while (ptr < end) { 707 while (ptr < end) {
708 struct btrfs_extent_inline_ref *iref; 708 struct btrfs_extent_inline_ref *iref;
709 u64 offset; 709 u64 offset;
710 int type; 710 int type;
711 711
712 iref = (struct btrfs_extent_inline_ref *)ptr; 712 iref = (struct btrfs_extent_inline_ref *)ptr;
713 type = btrfs_extent_inline_ref_type(leaf, iref); 713 type = btrfs_extent_inline_ref_type(leaf, iref);
714 offset = btrfs_extent_inline_ref_offset(leaf, iref); 714 offset = btrfs_extent_inline_ref_offset(leaf, iref);
715 715
716 switch (type) { 716 switch (type) {
717 case BTRFS_SHARED_BLOCK_REF_KEY: 717 case BTRFS_SHARED_BLOCK_REF_KEY:
718 ret = __add_prelim_ref(prefs, 0, NULL, 718 ret = __add_prelim_ref(prefs, 0, NULL,
719 *info_level + 1, offset, 719 *info_level + 1, offset,
720 bytenr, 1, GFP_NOFS); 720 bytenr, 1, GFP_NOFS);
721 break; 721 break;
722 case BTRFS_SHARED_DATA_REF_KEY: { 722 case BTRFS_SHARED_DATA_REF_KEY: {
723 struct btrfs_shared_data_ref *sdref; 723 struct btrfs_shared_data_ref *sdref;
724 int count; 724 int count;
725 725
726 sdref = (struct btrfs_shared_data_ref *)(iref + 1); 726 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
727 count = btrfs_shared_data_ref_count(leaf, sdref); 727 count = btrfs_shared_data_ref_count(leaf, sdref);
728 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset, 728 ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
729 bytenr, count, GFP_NOFS); 729 bytenr, count, GFP_NOFS);
730 break; 730 break;
731 } 731 }
732 case BTRFS_TREE_BLOCK_REF_KEY: 732 case BTRFS_TREE_BLOCK_REF_KEY:
733 ret = __add_prelim_ref(prefs, offset, NULL, 733 ret = __add_prelim_ref(prefs, offset, NULL,
734 *info_level + 1, 0, 734 *info_level + 1, 0,
735 bytenr, 1, GFP_NOFS); 735 bytenr, 1, GFP_NOFS);
736 break; 736 break;
737 case BTRFS_EXTENT_DATA_REF_KEY: { 737 case BTRFS_EXTENT_DATA_REF_KEY: {
738 struct btrfs_extent_data_ref *dref; 738 struct btrfs_extent_data_ref *dref;
739 int count; 739 int count;
740 u64 root; 740 u64 root;
741 741
742 dref = (struct btrfs_extent_data_ref *)(&iref->offset); 742 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
743 count = btrfs_extent_data_ref_count(leaf, dref); 743 count = btrfs_extent_data_ref_count(leaf, dref);
744 key.objectid = btrfs_extent_data_ref_objectid(leaf, 744 key.objectid = btrfs_extent_data_ref_objectid(leaf,
745 dref); 745 dref);
746 key.type = BTRFS_EXTENT_DATA_KEY; 746 key.type = BTRFS_EXTENT_DATA_KEY;
747 key.offset = btrfs_extent_data_ref_offset(leaf, dref); 747 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
748 root = btrfs_extent_data_ref_root(leaf, dref); 748 root = btrfs_extent_data_ref_root(leaf, dref);
749 ret = __add_prelim_ref(prefs, root, &key, 0, 0, 749 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
750 bytenr, count, GFP_NOFS); 750 bytenr, count, GFP_NOFS);
751 break; 751 break;
752 } 752 }
753 default: 753 default:
754 WARN_ON(1); 754 WARN_ON(1);
755 } 755 }
756 if (ret) 756 if (ret)
757 return ret; 757 return ret;
758 ptr += btrfs_extent_inline_ref_size(type); 758 ptr += btrfs_extent_inline_ref_size(type);
759 } 759 }
760 760
761 return 0; 761 return 0;
762 } 762 }
763 763
764 /* 764 /*
765 * add all non-inline backrefs for bytenr to the list 765 * add all non-inline backrefs for bytenr to the list
766 */ 766 */
767 static int __add_keyed_refs(struct btrfs_fs_info *fs_info, 767 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
768 struct btrfs_path *path, u64 bytenr, 768 struct btrfs_path *path, u64 bytenr,
769 int info_level, struct list_head *prefs) 769 int info_level, struct list_head *prefs)
770 { 770 {
771 struct btrfs_root *extent_root = fs_info->extent_root; 771 struct btrfs_root *extent_root = fs_info->extent_root;
772 int ret; 772 int ret;
773 int slot; 773 int slot;
774 struct extent_buffer *leaf; 774 struct extent_buffer *leaf;
775 struct btrfs_key key; 775 struct btrfs_key key;
776 776
777 while (1) { 777 while (1) {
778 ret = btrfs_next_item(extent_root, path); 778 ret = btrfs_next_item(extent_root, path);
779 if (ret < 0) 779 if (ret < 0)
780 break; 780 break;
781 if (ret) { 781 if (ret) {
782 ret = 0; 782 ret = 0;
783 break; 783 break;
784 } 784 }
785 785
786 slot = path->slots[0]; 786 slot = path->slots[0];
787 leaf = path->nodes[0]; 787 leaf = path->nodes[0];
788 btrfs_item_key_to_cpu(leaf, &key, slot); 788 btrfs_item_key_to_cpu(leaf, &key, slot);
789 789
790 if (key.objectid != bytenr) 790 if (key.objectid != bytenr)
791 break; 791 break;
792 if (key.type < BTRFS_TREE_BLOCK_REF_KEY) 792 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
793 continue; 793 continue;
794 if (key.type > BTRFS_SHARED_DATA_REF_KEY) 794 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
795 break; 795 break;
796 796
797 switch (key.type) { 797 switch (key.type) {
798 case BTRFS_SHARED_BLOCK_REF_KEY: 798 case BTRFS_SHARED_BLOCK_REF_KEY:
799 ret = __add_prelim_ref(prefs, 0, NULL, 799 ret = __add_prelim_ref(prefs, 0, NULL,
800 info_level + 1, key.offset, 800 info_level + 1, key.offset,
801 bytenr, 1, GFP_NOFS); 801 bytenr, 1, GFP_NOFS);
802 break; 802 break;
803 case BTRFS_SHARED_DATA_REF_KEY: { 803 case BTRFS_SHARED_DATA_REF_KEY: {
804 struct btrfs_shared_data_ref *sdref; 804 struct btrfs_shared_data_ref *sdref;
805 int count; 805 int count;
806 806
807 sdref = btrfs_item_ptr(leaf, slot, 807 sdref = btrfs_item_ptr(leaf, slot,
808 struct btrfs_shared_data_ref); 808 struct btrfs_shared_data_ref);
809 count = btrfs_shared_data_ref_count(leaf, sdref); 809 count = btrfs_shared_data_ref_count(leaf, sdref);
810 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset, 810 ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
811 bytenr, count, GFP_NOFS); 811 bytenr, count, GFP_NOFS);
812 break; 812 break;
813 } 813 }
814 case BTRFS_TREE_BLOCK_REF_KEY: 814 case BTRFS_TREE_BLOCK_REF_KEY:
815 ret = __add_prelim_ref(prefs, key.offset, NULL, 815 ret = __add_prelim_ref(prefs, key.offset, NULL,
816 info_level + 1, 0, 816 info_level + 1, 0,
817 bytenr, 1, GFP_NOFS); 817 bytenr, 1, GFP_NOFS);
818 break; 818 break;
819 case BTRFS_EXTENT_DATA_REF_KEY: { 819 case BTRFS_EXTENT_DATA_REF_KEY: {
820 struct btrfs_extent_data_ref *dref; 820 struct btrfs_extent_data_ref *dref;
821 int count; 821 int count;
822 u64 root; 822 u64 root;
823 823
824 dref = btrfs_item_ptr(leaf, slot, 824 dref = btrfs_item_ptr(leaf, slot,
825 struct btrfs_extent_data_ref); 825 struct btrfs_extent_data_ref);
826 count = btrfs_extent_data_ref_count(leaf, dref); 826 count = btrfs_extent_data_ref_count(leaf, dref);
827 key.objectid = btrfs_extent_data_ref_objectid(leaf, 827 key.objectid = btrfs_extent_data_ref_objectid(leaf,
828 dref); 828 dref);
829 key.type = BTRFS_EXTENT_DATA_KEY; 829 key.type = BTRFS_EXTENT_DATA_KEY;
830 key.offset = btrfs_extent_data_ref_offset(leaf, dref); 830 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
831 root = btrfs_extent_data_ref_root(leaf, dref); 831 root = btrfs_extent_data_ref_root(leaf, dref);
832 ret = __add_prelim_ref(prefs, root, &key, 0, 0, 832 ret = __add_prelim_ref(prefs, root, &key, 0, 0,
833 bytenr, count, GFP_NOFS); 833 bytenr, count, GFP_NOFS);
834 break; 834 break;
835 } 835 }
836 default: 836 default:
837 WARN_ON(1); 837 WARN_ON(1);
838 } 838 }
839 if (ret) 839 if (ret)
840 return ret; 840 return ret;
841 841
842 } 842 }
843 843
844 return ret; 844 return ret;
845 } 845 }
846 846
847 /* 847 /*
848 * this adds all existing backrefs (inline backrefs, backrefs and delayed 848 * this adds all existing backrefs (inline backrefs, backrefs and delayed
849 * refs) for the given bytenr to the refs list, merges duplicates and resolves 849 * refs) for the given bytenr to the refs list, merges duplicates and resolves
850 * indirect refs to their parent bytenr. 850 * indirect refs to their parent bytenr.
851 * When roots are found, they're added to the roots list 851 * When roots are found, they're added to the roots list
852 * 852 *
853 * FIXME some caching might speed things up 853 * FIXME some caching might speed things up
854 */ 854 */
855 static int find_parent_nodes(struct btrfs_trans_handle *trans, 855 static int find_parent_nodes(struct btrfs_trans_handle *trans,
856 struct btrfs_fs_info *fs_info, u64 bytenr, 856 struct btrfs_fs_info *fs_info, u64 bytenr,
857 u64 time_seq, struct ulist *refs, 857 u64 time_seq, struct ulist *refs,
858 struct ulist *roots, const u64 *extent_item_pos) 858 struct ulist *roots, const u64 *extent_item_pos)
859 { 859 {
860 struct btrfs_key key; 860 struct btrfs_key key;
861 struct btrfs_path *path; 861 struct btrfs_path *path;
862 struct btrfs_delayed_ref_root *delayed_refs = NULL; 862 struct btrfs_delayed_ref_root *delayed_refs = NULL;
863 struct btrfs_delayed_ref_head *head; 863 struct btrfs_delayed_ref_head *head;
864 int info_level = 0; 864 int info_level = 0;
865 int ret; 865 int ret;
866 struct list_head prefs_delayed; 866 struct list_head prefs_delayed;
867 struct list_head prefs; 867 struct list_head prefs;
868 struct __prelim_ref *ref; 868 struct __prelim_ref *ref;
869 struct extent_inode_elem *eie = NULL; 869 struct extent_inode_elem *eie = NULL;
870 u64 total_refs = 0; 870 u64 total_refs = 0;
871 871
872 INIT_LIST_HEAD(&prefs); 872 INIT_LIST_HEAD(&prefs);
873 INIT_LIST_HEAD(&prefs_delayed); 873 INIT_LIST_HEAD(&prefs_delayed);
874 874
875 key.objectid = bytenr; 875 key.objectid = bytenr;
876 key.offset = (u64)-1; 876 key.offset = (u64)-1;
877 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) 877 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
878 key.type = BTRFS_METADATA_ITEM_KEY; 878 key.type = BTRFS_METADATA_ITEM_KEY;
879 else 879 else
880 key.type = BTRFS_EXTENT_ITEM_KEY; 880 key.type = BTRFS_EXTENT_ITEM_KEY;
881 881
882 path = btrfs_alloc_path(); 882 path = btrfs_alloc_path();
883 if (!path) 883 if (!path)
884 return -ENOMEM; 884 return -ENOMEM;
885 if (!trans) { 885 if (!trans) {
886 path->search_commit_root = 1; 886 path->search_commit_root = 1;
887 path->skip_locking = 1; 887 path->skip_locking = 1;
888 } 888 }
889 889
890 /* 890 /*
891 * grab both a lock on the path and a lock on the delayed ref head. 891 * grab both a lock on the path and a lock on the delayed ref head.
892 * We need both to get a consistent picture of how the refs look 892 * We need both to get a consistent picture of how the refs look
893 * at a specified point in time 893 * at a specified point in time
894 */ 894 */
895 again: 895 again:
896 head = NULL; 896 head = NULL;
897 897
898 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); 898 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
899 if (ret < 0) 899 if (ret < 0)
900 goto out; 900 goto out;
901 BUG_ON(ret == 0); 901 BUG_ON(ret == 0);
902 902
903 if (trans) { 903 if (trans) {
904 /* 904 /*
905 * look if there are updates for this ref queued and lock the 905 * look if there are updates for this ref queued and lock the
906 * head 906 * head
907 */ 907 */
908 delayed_refs = &trans->transaction->delayed_refs; 908 delayed_refs = &trans->transaction->delayed_refs;
909 spin_lock(&delayed_refs->lock); 909 spin_lock(&delayed_refs->lock);
910 head = btrfs_find_delayed_ref_head(trans, bytenr); 910 head = btrfs_find_delayed_ref_head(trans, bytenr);
911 if (head) { 911 if (head) {
912 if (!mutex_trylock(&head->mutex)) { 912 if (!mutex_trylock(&head->mutex)) {
913 atomic_inc(&head->node.refs); 913 atomic_inc(&head->node.refs);
914 spin_unlock(&delayed_refs->lock); 914 spin_unlock(&delayed_refs->lock);
915 915
916 btrfs_release_path(path); 916 btrfs_release_path(path);
917 917
918 /* 918 /*
919 * Mutex was contended, block until it's 919 * Mutex was contended, block until it's
920 * released and try again 920 * released and try again
921 */ 921 */
922 mutex_lock(&head->mutex); 922 mutex_lock(&head->mutex);
923 mutex_unlock(&head->mutex); 923 mutex_unlock(&head->mutex);
924 btrfs_put_delayed_ref(&head->node); 924 btrfs_put_delayed_ref(&head->node);
925 goto again; 925 goto again;
926 } 926 }
927 spin_unlock(&delayed_refs->lock); 927 spin_unlock(&delayed_refs->lock);
928 ret = __add_delayed_refs(head, time_seq, 928 ret = __add_delayed_refs(head, time_seq,
929 &prefs_delayed, &total_refs); 929 &prefs_delayed, &total_refs);
930 mutex_unlock(&head->mutex); 930 mutex_unlock(&head->mutex);
931 if (ret) 931 if (ret)
932 goto out; 932 goto out;
933 } else { 933 } else {
934 spin_unlock(&delayed_refs->lock); 934 spin_unlock(&delayed_refs->lock);
935 } 935 }
936 } 936 }
937 937
938 if (path->slots[0]) { 938 if (path->slots[0]) {
939 struct extent_buffer *leaf; 939 struct extent_buffer *leaf;
940 int slot; 940 int slot;
941 941
942 path->slots[0]--; 942 path->slots[0]--;
943 leaf = path->nodes[0]; 943 leaf = path->nodes[0];
944 slot = path->slots[0]; 944 slot = path->slots[0];
945 btrfs_item_key_to_cpu(leaf, &key, slot); 945 btrfs_item_key_to_cpu(leaf, &key, slot);
946 if (key.objectid == bytenr && 946 if (key.objectid == bytenr &&
947 (key.type == BTRFS_EXTENT_ITEM_KEY || 947 (key.type == BTRFS_EXTENT_ITEM_KEY ||
948 key.type == BTRFS_METADATA_ITEM_KEY)) { 948 key.type == BTRFS_METADATA_ITEM_KEY)) {
949 ret = __add_inline_refs(fs_info, path, bytenr, 949 ret = __add_inline_refs(fs_info, path, bytenr,
950 &info_level, &prefs, 950 &info_level, &prefs,
951 &total_refs); 951 &total_refs);
952 if (ret) 952 if (ret)
953 goto out; 953 goto out;
954 ret = __add_keyed_refs(fs_info, path, bytenr, 954 ret = __add_keyed_refs(fs_info, path, bytenr,
955 info_level, &prefs); 955 info_level, &prefs);
956 if (ret) 956 if (ret)
957 goto out; 957 goto out;
958 } 958 }
959 } 959 }
960 btrfs_release_path(path); 960 btrfs_release_path(path);
961 961
962 list_splice_init(&prefs_delayed, &prefs); 962 list_splice_init(&prefs_delayed, &prefs);
963 963
964 ret = __add_missing_keys(fs_info, &prefs); 964 ret = __add_missing_keys(fs_info, &prefs);
965 if (ret) 965 if (ret)
966 goto out; 966 goto out;
967 967
968 __merge_refs(&prefs, 1); 968 __merge_refs(&prefs, 1);
969 969
970 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs, 970 ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
971 extent_item_pos, total_refs); 971 extent_item_pos, total_refs);
972 if (ret) 972 if (ret)
973 goto out; 973 goto out;
974 974
975 __merge_refs(&prefs, 2); 975 __merge_refs(&prefs, 2);
976 976
977 while (!list_empty(&prefs)) { 977 while (!list_empty(&prefs)) {
978 ref = list_first_entry(&prefs, struct __prelim_ref, list); 978 ref = list_first_entry(&prefs, struct __prelim_ref, list);
979 WARN_ON(ref->count < 0); 979 WARN_ON(ref->count < 0);
980 if (roots && ref->count && ref->root_id && ref->parent == 0) { 980 if (roots && ref->count && ref->root_id && ref->parent == 0) {
981 /* no parent == root of tree */ 981 /* no parent == root of tree */
982 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); 982 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
983 if (ret < 0) 983 if (ret < 0)
984 goto out; 984 goto out;
985 } 985 }
986 if (ref->count && ref->parent) { 986 if (ref->count && ref->parent) {
987 if (extent_item_pos && !ref->inode_list && 987 if (extent_item_pos && !ref->inode_list &&
988 ref->level == 0) { 988 ref->level == 0) {
989 u32 bsz; 989 u32 bsz;
990 struct extent_buffer *eb; 990 struct extent_buffer *eb;
991 bsz = btrfs_level_size(fs_info->extent_root, 991 bsz = btrfs_level_size(fs_info->extent_root,
992 ref->level); 992 ref->level);
993 eb = read_tree_block(fs_info->extent_root, 993 eb = read_tree_block(fs_info->extent_root,
994 ref->parent, bsz, 0); 994 ref->parent, bsz, 0);
995 if (!eb || !extent_buffer_uptodate(eb)) { 995 if (!eb || !extent_buffer_uptodate(eb)) {
996 free_extent_buffer(eb); 996 free_extent_buffer(eb);
997 ret = -EIO; 997 ret = -EIO;
998 goto out; 998 goto out;
999 } 999 }
1000 ret = find_extent_in_eb(eb, bytenr, 1000 ret = find_extent_in_eb(eb, bytenr,
1001 *extent_item_pos, &eie); 1001 *extent_item_pos, &eie);
1002 free_extent_buffer(eb); 1002 free_extent_buffer(eb);
1003 if (ret < 0) 1003 if (ret < 0)
1004 goto out; 1004 goto out;
1005 ref->inode_list = eie; 1005 ref->inode_list = eie;
1006 } 1006 }
1007 ret = ulist_add_merge(refs, ref->parent, 1007 ret = ulist_add_merge(refs, ref->parent,
1008 (uintptr_t)ref->inode_list, 1008 (uintptr_t)ref->inode_list,
1009 (u64 *)&eie, GFP_NOFS); 1009 (u64 *)&eie, GFP_NOFS);
1010 if (ret < 0) 1010 if (ret < 0)
1011 goto out; 1011 goto out;
1012 if (!ret && extent_item_pos) { 1012 if (!ret && extent_item_pos) {
1013 /* 1013 /*
1014 * we've recorded that parent, so we must extend 1014 * we've recorded that parent, so we must extend
1015 * its inode list here 1015 * its inode list here
1016 */ 1016 */
1017 BUG_ON(!eie); 1017 BUG_ON(!eie);
1018 while (eie->next) 1018 while (eie->next)
1019 eie = eie->next; 1019 eie = eie->next;
1020 eie->next = ref->inode_list; 1020 eie->next = ref->inode_list;
1021 } 1021 }
1022 eie = NULL; 1022 eie = NULL;
1023 } 1023 }
1024 list_del(&ref->list); 1024 list_del(&ref->list);
1025 kmem_cache_free(btrfs_prelim_ref_cache, ref); 1025 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1026 } 1026 }
1027 1027
1028 out: 1028 out:
1029 btrfs_free_path(path); 1029 btrfs_free_path(path);
1030 while (!list_empty(&prefs)) { 1030 while (!list_empty(&prefs)) {
1031 ref = list_first_entry(&prefs, struct __prelim_ref, list); 1031 ref = list_first_entry(&prefs, struct __prelim_ref, list);
1032 list_del(&ref->list); 1032 list_del(&ref->list);
1033 kmem_cache_free(btrfs_prelim_ref_cache, ref); 1033 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1034 } 1034 }
1035 while (!list_empty(&prefs_delayed)) { 1035 while (!list_empty(&prefs_delayed)) {
1036 ref = list_first_entry(&prefs_delayed, struct __prelim_ref, 1036 ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1037 list); 1037 list);
1038 list_del(&ref->list); 1038 list_del(&ref->list);
1039 kmem_cache_free(btrfs_prelim_ref_cache, ref); 1039 kmem_cache_free(btrfs_prelim_ref_cache, ref);
1040 } 1040 }
1041 if (ret < 0) 1041 if (ret < 0)
1042 free_inode_elem_list(eie); 1042 free_inode_elem_list(eie);
1043 return ret; 1043 return ret;
1044 } 1044 }
1045 1045
1046 static void free_leaf_list(struct ulist *blocks) 1046 static void free_leaf_list(struct ulist *blocks)
1047 { 1047 {
1048 struct ulist_node *node = NULL; 1048 struct ulist_node *node = NULL;
1049 struct extent_inode_elem *eie; 1049 struct extent_inode_elem *eie;
1050 struct ulist_iterator uiter; 1050 struct ulist_iterator uiter;
1051 1051
1052 ULIST_ITER_INIT(&uiter); 1052 ULIST_ITER_INIT(&uiter);
1053 while ((node = ulist_next(blocks, &uiter))) { 1053 while ((node = ulist_next(blocks, &uiter))) {
1054 if (!node->aux) 1054 if (!node->aux)
1055 continue; 1055 continue;
1056 eie = (struct extent_inode_elem *)(uintptr_t)node->aux; 1056 eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1057 free_inode_elem_list(eie); 1057 free_inode_elem_list(eie);
1058 node->aux = 0; 1058 node->aux = 0;
1059 } 1059 }
1060 1060
1061 ulist_free(blocks); 1061 ulist_free(blocks);
1062 } 1062 }
1063 1063
1064 /* 1064 /*
1065 * Finds all leafs with a reference to the specified combination of bytenr and 1065 * Finds all leafs with a reference to the specified combination of bytenr and
1066 * offset. key_list_head will point to a list of corresponding keys (caller must 1066 * offset. key_list_head will point to a list of corresponding keys (caller must
1067 * free each list element). The leafs will be stored in the leafs ulist, which 1067 * free each list element). The leafs will be stored in the leafs ulist, which
1068 * must be freed with ulist_free. 1068 * must be freed with ulist_free.
1069 * 1069 *
1070 * returns 0 on success, <0 on error 1070 * returns 0 on success, <0 on error
1071 */ 1071 */
1072 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, 1072 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1073 struct btrfs_fs_info *fs_info, u64 bytenr, 1073 struct btrfs_fs_info *fs_info, u64 bytenr,
1074 u64 time_seq, struct ulist **leafs, 1074 u64 time_seq, struct ulist **leafs,
1075 const u64 *extent_item_pos) 1075 const u64 *extent_item_pos)
1076 { 1076 {
1077 int ret; 1077 int ret;
1078 1078
1079 *leafs = ulist_alloc(GFP_NOFS); 1079 *leafs = ulist_alloc(GFP_NOFS);
1080 if (!*leafs) 1080 if (!*leafs)
1081 return -ENOMEM; 1081 return -ENOMEM;
1082 1082
1083 ret = find_parent_nodes(trans, fs_info, bytenr, 1083 ret = find_parent_nodes(trans, fs_info, bytenr,
1084 time_seq, *leafs, NULL, extent_item_pos); 1084 time_seq, *leafs, NULL, extent_item_pos);
1085 if (ret < 0 && ret != -ENOENT) { 1085 if (ret < 0 && ret != -ENOENT) {
1086 free_leaf_list(*leafs); 1086 free_leaf_list(*leafs);
1087 return ret; 1087 return ret;
1088 } 1088 }
1089 1089
1090 return 0; 1090 return 0;
1091 } 1091 }
1092 1092
1093 /* 1093 /*
1094 * walk all backrefs for a given extent to find all roots that reference this 1094 * walk all backrefs for a given extent to find all roots that reference this
1095 * extent. Walking a backref means finding all extents that reference this 1095 * extent. Walking a backref means finding all extents that reference this
1096 * extent and in turn walk the backrefs of those, too. Naturally this is a 1096 * extent and in turn walk the backrefs of those, too. Naturally this is a
1097 * recursive process, but here it is implemented in an iterative fashion: We 1097 * recursive process, but here it is implemented in an iterative fashion: We
1098 * find all referencing extents for the extent in question and put them on a 1098 * find all referencing extents for the extent in question and put them on a
1099 * list. In turn, we find all referencing extents for those, further appending 1099 * list. In turn, we find all referencing extents for those, further appending
1100 * to the list. The way we iterate the list allows adding more elements after 1100 * to the list. The way we iterate the list allows adding more elements after
1101 * the current while iterating. The process stops when we reach the end of the 1101 * the current while iterating. The process stops when we reach the end of the
1102 * list. Found roots are added to the roots list. 1102 * list. Found roots are added to the roots list.
1103 * 1103 *
1104 * returns 0 on success, < 0 on error. 1104 * returns 0 on success, < 0 on error.
1105 */ 1105 */
1106 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans, 1106 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1107 struct btrfs_fs_info *fs_info, u64 bytenr, 1107 struct btrfs_fs_info *fs_info, u64 bytenr,
1108 u64 time_seq, struct ulist **roots) 1108 u64 time_seq, struct ulist **roots)
1109 { 1109 {
1110 struct ulist *tmp; 1110 struct ulist *tmp;
1111 struct ulist_node *node = NULL; 1111 struct ulist_node *node = NULL;
1112 struct ulist_iterator uiter; 1112 struct ulist_iterator uiter;
1113 int ret; 1113 int ret;
1114 1114
1115 tmp = ulist_alloc(GFP_NOFS); 1115 tmp = ulist_alloc(GFP_NOFS);
1116 if (!tmp) 1116 if (!tmp)
1117 return -ENOMEM; 1117 return -ENOMEM;
1118 *roots = ulist_alloc(GFP_NOFS); 1118 *roots = ulist_alloc(GFP_NOFS);
1119 if (!*roots) { 1119 if (!*roots) {
1120 ulist_free(tmp); 1120 ulist_free(tmp);
1121 return -ENOMEM; 1121 return -ENOMEM;
1122 } 1122 }
1123 1123
1124 ULIST_ITER_INIT(&uiter); 1124 ULIST_ITER_INIT(&uiter);
1125 while (1) { 1125 while (1) {
1126 ret = find_parent_nodes(trans, fs_info, bytenr, 1126 ret = find_parent_nodes(trans, fs_info, bytenr,
1127 time_seq, tmp, *roots, NULL); 1127 time_seq, tmp, *roots, NULL);
1128 if (ret < 0 && ret != -ENOENT) { 1128 if (ret < 0 && ret != -ENOENT) {
1129 ulist_free(tmp); 1129 ulist_free(tmp);
1130 ulist_free(*roots); 1130 ulist_free(*roots);
1131 return ret; 1131 return ret;
1132 } 1132 }
1133 node = ulist_next(tmp, &uiter); 1133 node = ulist_next(tmp, &uiter);
1134 if (!node) 1134 if (!node)
1135 break; 1135 break;
1136 bytenr = node->val; 1136 bytenr = node->val;
1137 cond_resched(); 1137 cond_resched();
1138 } 1138 }
1139 1139
1140 ulist_free(tmp); 1140 ulist_free(tmp);
1141 return 0; 1141 return 0;
1142 } 1142 }
1143 1143
1144 int btrfs_find_all_roots(struct btrfs_trans_handle *trans, 1144 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1145 struct btrfs_fs_info *fs_info, u64 bytenr, 1145 struct btrfs_fs_info *fs_info, u64 bytenr,
1146 u64 time_seq, struct ulist **roots) 1146 u64 time_seq, struct ulist **roots)
1147 { 1147 {
1148 int ret; 1148 int ret;
1149 1149
1150 if (!trans) 1150 if (!trans)
1151 down_read(&fs_info->commit_root_sem); 1151 down_read(&fs_info->commit_root_sem);
1152 ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots); 1152 ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1153 if (!trans) 1153 if (!trans)
1154 up_read(&fs_info->commit_root_sem); 1154 up_read(&fs_info->commit_root_sem);
1155 return ret; 1155 return ret;
1156 } 1156 }
1157 1157
1158 /* 1158 /*
1159 * this makes the path point to (inum INODE_ITEM ioff) 1159 * this makes the path point to (inum INODE_ITEM ioff)
1160 */ 1160 */
1161 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, 1161 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1162 struct btrfs_path *path) 1162 struct btrfs_path *path)
1163 { 1163 {
1164 struct btrfs_key key; 1164 struct btrfs_key key;
1165 return btrfs_find_item(fs_root, path, inum, ioff, 1165 return btrfs_find_item(fs_root, path, inum, ioff,
1166 BTRFS_INODE_ITEM_KEY, &key); 1166 BTRFS_INODE_ITEM_KEY, &key);
1167 } 1167 }
1168 1168
1169 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, 1169 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1170 struct btrfs_path *path, 1170 struct btrfs_path *path,
1171 struct btrfs_key *found_key) 1171 struct btrfs_key *found_key)
1172 { 1172 {
1173 return btrfs_find_item(fs_root, path, inum, ioff, 1173 return btrfs_find_item(fs_root, path, inum, ioff,
1174 BTRFS_INODE_REF_KEY, found_key); 1174 BTRFS_INODE_REF_KEY, found_key);
1175 } 1175 }
1176 1176
1177 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, 1177 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1178 u64 start_off, struct btrfs_path *path, 1178 u64 start_off, struct btrfs_path *path,
1179 struct btrfs_inode_extref **ret_extref, 1179 struct btrfs_inode_extref **ret_extref,
1180 u64 *found_off) 1180 u64 *found_off)
1181 { 1181 {
1182 int ret, slot; 1182 int ret, slot;
1183 struct btrfs_key key; 1183 struct btrfs_key key;
1184 struct btrfs_key found_key; 1184 struct btrfs_key found_key;
1185 struct btrfs_inode_extref *extref; 1185 struct btrfs_inode_extref *extref;
1186 struct extent_buffer *leaf; 1186 struct extent_buffer *leaf;
1187 unsigned long ptr; 1187 unsigned long ptr;
1188 1188
1189 key.objectid = inode_objectid; 1189 key.objectid = inode_objectid;
1190 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY); 1190 btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1191 key.offset = start_off; 1191 key.offset = start_off;
1192 1192
1193 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 1193 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1194 if (ret < 0) 1194 if (ret < 0)
1195 return ret; 1195 return ret;
1196 1196
1197 while (1) { 1197 while (1) {
1198 leaf = path->nodes[0]; 1198 leaf = path->nodes[0];
1199 slot = path->slots[0]; 1199 slot = path->slots[0];
1200 if (slot >= btrfs_header_nritems(leaf)) { 1200 if (slot >= btrfs_header_nritems(leaf)) {
1201 /* 1201 /*
1202 * If the item at offset is not found, 1202 * If the item at offset is not found,
1203 * btrfs_search_slot will point us to the slot 1203 * btrfs_search_slot will point us to the slot
1204 * where it should be inserted. In our case 1204 * where it should be inserted. In our case
1205 * that will be the slot directly before the 1205 * that will be the slot directly before the
1206 * next INODE_REF_KEY_V2 item. In the case 1206 * next INODE_REF_KEY_V2 item. In the case
1207 * that we're pointing to the last slot in a 1207 * that we're pointing to the last slot in a
1208 * leaf, we must move one leaf over. 1208 * leaf, we must move one leaf over.
1209 */ 1209 */
1210 ret = btrfs_next_leaf(root, path); 1210 ret = btrfs_next_leaf(root, path);
1211 if (ret) { 1211 if (ret) {
1212 if (ret >= 1) 1212 if (ret >= 1)
1213 ret = -ENOENT; 1213 ret = -ENOENT;
1214 break; 1214 break;
1215 } 1215 }
1216 continue; 1216 continue;
1217 } 1217 }
1218 1218
1219 btrfs_item_key_to_cpu(leaf, &found_key, slot); 1219 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1220 1220
1221 /* 1221 /*
1222 * Check that we're still looking at an extended ref key for 1222 * Check that we're still looking at an extended ref key for
1223 * this particular objectid. If we have different 1223 * this particular objectid. If we have different
1224 * objectid or type then there are no more to be found 1224 * objectid or type then there are no more to be found
1225 * in the tree and we can exit. 1225 * in the tree and we can exit.
1226 */ 1226 */
1227 ret = -ENOENT; 1227 ret = -ENOENT;
1228 if (found_key.objectid != inode_objectid) 1228 if (found_key.objectid != inode_objectid)
1229 break; 1229 break;
1230 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY) 1230 if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1231 break; 1231 break;
1232 1232
1233 ret = 0; 1233 ret = 0;
1234 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 1234 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1235 extref = (struct btrfs_inode_extref *)ptr; 1235 extref = (struct btrfs_inode_extref *)ptr;
1236 *ret_extref = extref; 1236 *ret_extref = extref;
1237 if (found_off) 1237 if (found_off)
1238 *found_off = found_key.offset; 1238 *found_off = found_key.offset;
1239 break; 1239 break;
1240 } 1240 }
1241 1241
1242 return ret; 1242 return ret;
1243 } 1243 }
1244 1244
1245 /* 1245 /*
1246 * this iterates to turn a name (from iref/extref) into a full filesystem path. 1246 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1247 * Elements of the path are separated by '/' and the path is guaranteed to be 1247 * Elements of the path are separated by '/' and the path is guaranteed to be
1248 * 0-terminated. the path is only given within the current file system. 1248 * 0-terminated. the path is only given within the current file system.
1249 * Therefore, it never starts with a '/'. the caller is responsible to provide 1249 * Therefore, it never starts with a '/'. the caller is responsible to provide
1250 * "size" bytes in "dest". the dest buffer will be filled backwards. finally, 1250 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1251 * the start point of the resulting string is returned. this pointer is within 1251 * the start point of the resulting string is returned. this pointer is within
1252 * dest, normally. 1252 * dest, normally.
1253 * in case the path buffer would overflow, the pointer is decremented further 1253 * in case the path buffer would overflow, the pointer is decremented further
1254 * as if output was written to the buffer, though no more output is actually 1254 * as if output was written to the buffer, though no more output is actually
1255 * generated. that way, the caller can determine how much space would be 1255 * generated. that way, the caller can determine how much space would be
1256 * required for the path to fit into the buffer. in that case, the returned 1256 * required for the path to fit into the buffer. in that case, the returned
1257 * value will be smaller than dest. callers must check this! 1257 * value will be smaller than dest. callers must check this!
1258 */ 1258 */
1259 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, 1259 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1260 u32 name_len, unsigned long name_off, 1260 u32 name_len, unsigned long name_off,
1261 struct extent_buffer *eb_in, u64 parent, 1261 struct extent_buffer *eb_in, u64 parent,
1262 char *dest, u32 size) 1262 char *dest, u32 size)
1263 { 1263 {
1264 int slot; 1264 int slot;
1265 u64 next_inum; 1265 u64 next_inum;
1266 int ret; 1266 int ret;
1267 s64 bytes_left = ((s64)size) - 1; 1267 s64 bytes_left = ((s64)size) - 1;
1268 struct extent_buffer *eb = eb_in; 1268 struct extent_buffer *eb = eb_in;
1269 struct btrfs_key found_key; 1269 struct btrfs_key found_key;
1270 int leave_spinning = path->leave_spinning; 1270 int leave_spinning = path->leave_spinning;
1271 struct btrfs_inode_ref *iref; 1271 struct btrfs_inode_ref *iref;
1272 1272
1273 if (bytes_left >= 0) 1273 if (bytes_left >= 0)
1274 dest[bytes_left] = '\0'; 1274 dest[bytes_left] = '\0';
1275 1275
1276 path->leave_spinning = 1; 1276 path->leave_spinning = 1;
1277 while (1) { 1277 while (1) {
1278 bytes_left -= name_len; 1278 bytes_left -= name_len;
1279 if (bytes_left >= 0) 1279 if (bytes_left >= 0)
1280 read_extent_buffer(eb, dest + bytes_left, 1280 read_extent_buffer(eb, dest + bytes_left,
1281 name_off, name_len); 1281 name_off, name_len);
1282 if (eb != eb_in) { 1282 if (eb != eb_in) {
1283 btrfs_tree_read_unlock_blocking(eb); 1283 btrfs_tree_read_unlock_blocking(eb);
1284 free_extent_buffer(eb); 1284 free_extent_buffer(eb);
1285 } 1285 }
1286 ret = inode_ref_info(parent, 0, fs_root, path, &found_key); 1286 ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1287 if (ret > 0) 1287 if (ret > 0)
1288 ret = -ENOENT; 1288 ret = -ENOENT;
1289 if (ret) 1289 if (ret)
1290 break; 1290 break;
1291 1291
1292 next_inum = found_key.offset; 1292 next_inum = found_key.offset;
1293 1293
1294 /* regular exit ahead */ 1294 /* regular exit ahead */
1295 if (parent == next_inum) 1295 if (parent == next_inum)
1296 break; 1296 break;
1297 1297
1298 slot = path->slots[0]; 1298 slot = path->slots[0];
1299 eb = path->nodes[0]; 1299 eb = path->nodes[0];
1300 /* make sure we can use eb after releasing the path */ 1300 /* make sure we can use eb after releasing the path */
1301 if (eb != eb_in) { 1301 if (eb != eb_in) {
1302 atomic_inc(&eb->refs); 1302 atomic_inc(&eb->refs);
1303 btrfs_tree_read_lock(eb); 1303 btrfs_tree_read_lock(eb);
1304 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 1304 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1305 } 1305 }
1306 btrfs_release_path(path); 1306 btrfs_release_path(path);
1307 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); 1307 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1308 1308
1309 name_len = btrfs_inode_ref_name_len(eb, iref); 1309 name_len = btrfs_inode_ref_name_len(eb, iref);
1310 name_off = (unsigned long)(iref + 1); 1310 name_off = (unsigned long)(iref + 1);
1311 1311
1312 parent = next_inum; 1312 parent = next_inum;
1313 --bytes_left; 1313 --bytes_left;
1314 if (bytes_left >= 0) 1314 if (bytes_left >= 0)
1315 dest[bytes_left] = '/'; 1315 dest[bytes_left] = '/';
1316 } 1316 }
1317 1317
1318 btrfs_release_path(path); 1318 btrfs_release_path(path);
1319 path->leave_spinning = leave_spinning; 1319 path->leave_spinning = leave_spinning;
1320 1320
1321 if (ret) 1321 if (ret)
1322 return ERR_PTR(ret); 1322 return ERR_PTR(ret);
1323 1323
1324 return dest + bytes_left; 1324 return dest + bytes_left;
1325 } 1325 }
1326 1326
1327 /* 1327 /*
1328 * this makes the path point to (logical EXTENT_ITEM *) 1328 * this makes the path point to (logical EXTENT_ITEM *)
1329 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for 1329 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1330 * tree blocks and <0 on error. 1330 * tree blocks and <0 on error.
1331 */ 1331 */
1332 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, 1332 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1333 struct btrfs_path *path, struct btrfs_key *found_key, 1333 struct btrfs_path *path, struct btrfs_key *found_key,
1334 u64 *flags_ret) 1334 u64 *flags_ret)
1335 { 1335 {
1336 int ret; 1336 int ret;
1337 u64 flags; 1337 u64 flags;
1338 u64 size = 0; 1338 u64 size = 0;
1339 u32 item_size; 1339 u32 item_size;
1340 struct extent_buffer *eb; 1340 struct extent_buffer *eb;
1341 struct btrfs_extent_item *ei; 1341 struct btrfs_extent_item *ei;
1342 struct btrfs_key key; 1342 struct btrfs_key key;
1343 1343
1344 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) 1344 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1345 key.type = BTRFS_METADATA_ITEM_KEY; 1345 key.type = BTRFS_METADATA_ITEM_KEY;
1346 else 1346 else
1347 key.type = BTRFS_EXTENT_ITEM_KEY; 1347 key.type = BTRFS_EXTENT_ITEM_KEY;
1348 key.objectid = logical; 1348 key.objectid = logical;
1349 key.offset = (u64)-1; 1349 key.offset = (u64)-1;
1350 1350
1351 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); 1351 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1352 if (ret < 0) 1352 if (ret < 0)
1353 return ret; 1353 return ret;
1354 1354
1355 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0); 1355 ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1356 if (ret) { 1356 if (ret) {
1357 if (ret > 0) 1357 if (ret > 0)
1358 ret = -ENOENT; 1358 ret = -ENOENT;
1359 return ret; 1359 return ret;
1360 } 1360 }
1361 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]); 1361 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1362 if (found_key->type == BTRFS_METADATA_ITEM_KEY) 1362 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1363 size = fs_info->extent_root->leafsize; 1363 size = fs_info->extent_root->leafsize;
1364 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY) 1364 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1365 size = found_key->offset; 1365 size = found_key->offset;
1366 1366
1367 if (found_key->objectid > logical || 1367 if (found_key->objectid > logical ||
1368 found_key->objectid + size <= logical) { 1368 found_key->objectid + size <= logical) {
1369 pr_debug("logical %llu is not within any extent\n", logical); 1369 pr_debug("logical %llu is not within any extent\n", logical);
1370 return -ENOENT; 1370 return -ENOENT;
1371 } 1371 }
1372 1372
1373 eb = path->nodes[0]; 1373 eb = path->nodes[0];
1374 item_size = btrfs_item_size_nr(eb, path->slots[0]); 1374 item_size = btrfs_item_size_nr(eb, path->slots[0]);
1375 BUG_ON(item_size < sizeof(*ei)); 1375 BUG_ON(item_size < sizeof(*ei));
1376 1376
1377 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); 1377 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1378 flags = btrfs_extent_flags(eb, ei); 1378 flags = btrfs_extent_flags(eb, ei);
1379 1379
1380 pr_debug("logical %llu is at position %llu within the extent (%llu " 1380 pr_debug("logical %llu is at position %llu within the extent (%llu "
1381 "EXTENT_ITEM %llu) flags %#llx size %u\n", 1381 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1382 logical, logical - found_key->objectid, found_key->objectid, 1382 logical, logical - found_key->objectid, found_key->objectid,
1383 found_key->offset, flags, item_size); 1383 found_key->offset, flags, item_size);
1384 1384
1385 WARN_ON(!flags_ret); 1385 WARN_ON(!flags_ret);
1386 if (flags_ret) { 1386 if (flags_ret) {
1387 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) 1387 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1388 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK; 1388 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1389 else if (flags & BTRFS_EXTENT_FLAG_DATA) 1389 else if (flags & BTRFS_EXTENT_FLAG_DATA)
1390 *flags_ret = BTRFS_EXTENT_FLAG_DATA; 1390 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
1391 else 1391 else
1392 BUG_ON(1); 1392 BUG_ON(1);
1393 return 0; 1393 return 0;
1394 } 1394 }
1395 1395
1396 return -EIO; 1396 return -EIO;
1397 } 1397 }
1398 1398
1399 /* 1399 /*
1400 * helper function to iterate extent inline refs. ptr must point to a 0 value 1400 * helper function to iterate extent inline refs. ptr must point to a 0 value
1401 * for the first call and may be modified. it is used to track state. 1401 * for the first call and may be modified. it is used to track state.
1402 * if more refs exist, 0 is returned and the next call to 1402 * if more refs exist, 0 is returned and the next call to
1403 * __get_extent_inline_ref must pass the modified ptr parameter to get the 1403 * __get_extent_inline_ref must pass the modified ptr parameter to get the
1404 * next ref. after the last ref was processed, 1 is returned. 1404 * next ref. after the last ref was processed, 1 is returned.
1405 * returns <0 on error 1405 * returns <0 on error
1406 */ 1406 */
1407 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb, 1407 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1408 struct btrfs_extent_item *ei, u32 item_size, 1408 struct btrfs_key *key,
1409 struct btrfs_extent_inline_ref **out_eiref, 1409 struct btrfs_extent_item *ei, u32 item_size,
1410 int *out_type) 1410 struct btrfs_extent_inline_ref **out_eiref,
1411 int *out_type)
1411 { 1412 {
1412 unsigned long end; 1413 unsigned long end;
1413 u64 flags; 1414 u64 flags;
1414 struct btrfs_tree_block_info *info; 1415 struct btrfs_tree_block_info *info;
1415 1416
1416 if (!*ptr) { 1417 if (!*ptr) {
1417 /* first call */ 1418 /* first call */
1418 flags = btrfs_extent_flags(eb, ei); 1419 flags = btrfs_extent_flags(eb, ei);
1419 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 1420 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1420 info = (struct btrfs_tree_block_info *)(ei + 1); 1421 if (key->type == BTRFS_METADATA_ITEM_KEY) {
1421 *out_eiref = 1422 /* a skinny metadata extent */
1422 (struct btrfs_extent_inline_ref *)(info + 1); 1423 *out_eiref =
1424 (struct btrfs_extent_inline_ref *)(ei + 1);
1425 } else {
1426 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1427 info = (struct btrfs_tree_block_info *)(ei + 1);
1428 *out_eiref =
1429 (struct btrfs_extent_inline_ref *)(info + 1);
1430 }
1423 } else { 1431 } else {
1424 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1); 1432 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1425 } 1433 }
1426 *ptr = (unsigned long)*out_eiref; 1434 *ptr = (unsigned long)*out_eiref;
1427 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size) 1435 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1428 return -ENOENT; 1436 return -ENOENT;
1429 } 1437 }
1430 1438
1431 end = (unsigned long)ei + item_size; 1439 end = (unsigned long)ei + item_size;
1432 *out_eiref = (struct btrfs_extent_inline_ref *)*ptr; 1440 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1433 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref); 1441 *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1434 1442
1435 *ptr += btrfs_extent_inline_ref_size(*out_type); 1443 *ptr += btrfs_extent_inline_ref_size(*out_type);
1436 WARN_ON(*ptr > end); 1444 WARN_ON(*ptr > end);
1437 if (*ptr == end) 1445 if (*ptr == end)
1438 return 1; /* last */ 1446 return 1; /* last */
1439 1447
1440 return 0; 1448 return 0;
1441 } 1449 }
1442 1450
1443 /* 1451 /*
1444 * reads the tree block backref for an extent. tree level and root are returned 1452 * reads the tree block backref for an extent. tree level and root are returned
1445 * through out_level and out_root. ptr must point to a 0 value for the first 1453 * through out_level and out_root. ptr must point to a 0 value for the first
1446 * call and may be modified (see __get_extent_inline_ref comment). 1454 * call and may be modified (see __get_extent_inline_ref comment).
1447 * returns 0 if data was provided, 1 if there was no more data to provide or 1455 * returns 0 if data was provided, 1 if there was no more data to provide or
1448 * <0 on error. 1456 * <0 on error.
1449 */ 1457 */
1450 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, 1458 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1451 struct btrfs_extent_item *ei, u32 item_size, 1459 struct btrfs_key *key, struct btrfs_extent_item *ei,
1452 u64 *out_root, u8 *out_level) 1460 u32 item_size, u64 *out_root, u8 *out_level)
1453 { 1461 {
1454 int ret; 1462 int ret;
1455 int type; 1463 int type;
1456 struct btrfs_tree_block_info *info; 1464 struct btrfs_tree_block_info *info;
1457 struct btrfs_extent_inline_ref *eiref; 1465 struct btrfs_extent_inline_ref *eiref;
1458 1466
1459 if (*ptr == (unsigned long)-1) 1467 if (*ptr == (unsigned long)-1)
1460 return 1; 1468 return 1;
1461 1469
1462 while (1) { 1470 while (1) {
1463 ret = __get_extent_inline_ref(ptr, eb, ei, item_size, 1471 ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1464 &eiref, &type); 1472 &eiref, &type);
1465 if (ret < 0) 1473 if (ret < 0)
1466 return ret; 1474 return ret;
1467 1475
1468 if (type == BTRFS_TREE_BLOCK_REF_KEY || 1476 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1469 type == BTRFS_SHARED_BLOCK_REF_KEY) 1477 type == BTRFS_SHARED_BLOCK_REF_KEY)
1470 break; 1478 break;
1471 1479
1472 if (ret == 1) 1480 if (ret == 1)
1473 return 1; 1481 return 1;
1474 } 1482 }
1475 1483
1476 /* we can treat both ref types equally here */ 1484 /* we can treat both ref types equally here */
1477 info = (struct btrfs_tree_block_info *)(ei + 1); 1485 info = (struct btrfs_tree_block_info *)(ei + 1);
1478 *out_root = btrfs_extent_inline_ref_offset(eb, eiref); 1486 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1479 *out_level = btrfs_tree_block_level(eb, info); 1487 *out_level = btrfs_tree_block_level(eb, info);
1480 1488
1481 if (ret == 1) 1489 if (ret == 1)
1482 *ptr = (unsigned long)-1; 1490 *ptr = (unsigned long)-1;
1483 1491
1484 return 0; 1492 return 0;
1485 } 1493 }
1486 1494
1487 static int iterate_leaf_refs(struct extent_inode_elem *inode_list, 1495 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1488 u64 root, u64 extent_item_objectid, 1496 u64 root, u64 extent_item_objectid,
1489 iterate_extent_inodes_t *iterate, void *ctx) 1497 iterate_extent_inodes_t *iterate, void *ctx)
1490 { 1498 {
1491 struct extent_inode_elem *eie; 1499 struct extent_inode_elem *eie;
1492 int ret = 0; 1500 int ret = 0;
1493 1501
1494 for (eie = inode_list; eie; eie = eie->next) { 1502 for (eie = inode_list; eie; eie = eie->next) {
1495 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), " 1503 pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1496 "root %llu\n", extent_item_objectid, 1504 "root %llu\n", extent_item_objectid,
1497 eie->inum, eie->offset, root); 1505 eie->inum, eie->offset, root);
1498 ret = iterate(eie->inum, eie->offset, root, ctx); 1506 ret = iterate(eie->inum, eie->offset, root, ctx);
1499 if (ret) { 1507 if (ret) {
1500 pr_debug("stopping iteration for %llu due to ret=%d\n", 1508 pr_debug("stopping iteration for %llu due to ret=%d\n",
1501 extent_item_objectid, ret); 1509 extent_item_objectid, ret);
1502 break; 1510 break;
1503 } 1511 }
1504 } 1512 }
1505 1513
1506 return ret; 1514 return ret;
1507 } 1515 }
1508 1516
1509 /* 1517 /*
1510 * calls iterate() for every inode that references the extent identified by 1518 * calls iterate() for every inode that references the extent identified by
1511 * the given parameters. 1519 * the given parameters.
1512 * when the iterator function returns a non-zero value, iteration stops. 1520 * when the iterator function returns a non-zero value, iteration stops.
1513 */ 1521 */
1514 int iterate_extent_inodes(struct btrfs_fs_info *fs_info, 1522 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1515 u64 extent_item_objectid, u64 extent_item_pos, 1523 u64 extent_item_objectid, u64 extent_item_pos,
1516 int search_commit_root, 1524 int search_commit_root,
1517 iterate_extent_inodes_t *iterate, void *ctx) 1525 iterate_extent_inodes_t *iterate, void *ctx)
1518 { 1526 {
1519 int ret; 1527 int ret;
1520 struct btrfs_trans_handle *trans = NULL; 1528 struct btrfs_trans_handle *trans = NULL;
1521 struct ulist *refs = NULL; 1529 struct ulist *refs = NULL;
1522 struct ulist *roots = NULL; 1530 struct ulist *roots = NULL;
1523 struct ulist_node *ref_node = NULL; 1531 struct ulist_node *ref_node = NULL;
1524 struct ulist_node *root_node = NULL; 1532 struct ulist_node *root_node = NULL;
1525 struct seq_list tree_mod_seq_elem = {}; 1533 struct seq_list tree_mod_seq_elem = {};
1526 struct ulist_iterator ref_uiter; 1534 struct ulist_iterator ref_uiter;
1527 struct ulist_iterator root_uiter; 1535 struct ulist_iterator root_uiter;
1528 1536
1529 pr_debug("resolving all inodes for extent %llu\n", 1537 pr_debug("resolving all inodes for extent %llu\n",
1530 extent_item_objectid); 1538 extent_item_objectid);
1531 1539
1532 if (!search_commit_root) { 1540 if (!search_commit_root) {
1533 trans = btrfs_join_transaction(fs_info->extent_root); 1541 trans = btrfs_join_transaction(fs_info->extent_root);
1534 if (IS_ERR(trans)) 1542 if (IS_ERR(trans))
1535 return PTR_ERR(trans); 1543 return PTR_ERR(trans);
1536 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); 1544 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1537 } else { 1545 } else {
1538 down_read(&fs_info->commit_root_sem); 1546 down_read(&fs_info->commit_root_sem);
1539 } 1547 }
1540 1548
1541 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, 1549 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1542 tree_mod_seq_elem.seq, &refs, 1550 tree_mod_seq_elem.seq, &refs,
1543 &extent_item_pos); 1551 &extent_item_pos);
1544 if (ret) 1552 if (ret)
1545 goto out; 1553 goto out;
1546 1554
1547 ULIST_ITER_INIT(&ref_uiter); 1555 ULIST_ITER_INIT(&ref_uiter);
1548 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { 1556 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1549 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val, 1557 ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1550 tree_mod_seq_elem.seq, &roots); 1558 tree_mod_seq_elem.seq, &roots);
1551 if (ret) 1559 if (ret)
1552 break; 1560 break;
1553 ULIST_ITER_INIT(&root_uiter); 1561 ULIST_ITER_INIT(&root_uiter);
1554 while (!ret && (root_node = ulist_next(roots, &root_uiter))) { 1562 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1555 pr_debug("root %llu references leaf %llu, data list " 1563 pr_debug("root %llu references leaf %llu, data list "
1556 "%#llx\n", root_node->val, ref_node->val, 1564 "%#llx\n", root_node->val, ref_node->val,
1557 ref_node->aux); 1565 ref_node->aux);
1558 ret = iterate_leaf_refs((struct extent_inode_elem *) 1566 ret = iterate_leaf_refs((struct extent_inode_elem *)
1559 (uintptr_t)ref_node->aux, 1567 (uintptr_t)ref_node->aux,
1560 root_node->val, 1568 root_node->val,
1561 extent_item_objectid, 1569 extent_item_objectid,
1562 iterate, ctx); 1570 iterate, ctx);
1563 } 1571 }
1564 ulist_free(roots); 1572 ulist_free(roots);
1565 } 1573 }
1566 1574
1567 free_leaf_list(refs); 1575 free_leaf_list(refs);
1568 out: 1576 out:
1569 if (!search_commit_root) { 1577 if (!search_commit_root) {
1570 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); 1578 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1571 btrfs_end_transaction(trans, fs_info->extent_root); 1579 btrfs_end_transaction(trans, fs_info->extent_root);
1572 } else { 1580 } else {
1573 up_read(&fs_info->commit_root_sem); 1581 up_read(&fs_info->commit_root_sem);
1574 } 1582 }
1575 1583
1576 return ret; 1584 return ret;
1577 } 1585 }
1578 1586
1579 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, 1587 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1580 struct btrfs_path *path, 1588 struct btrfs_path *path,
1581 iterate_extent_inodes_t *iterate, void *ctx) 1589 iterate_extent_inodes_t *iterate, void *ctx)
1582 { 1590 {
1583 int ret; 1591 int ret;
1584 u64 extent_item_pos; 1592 u64 extent_item_pos;
1585 u64 flags = 0; 1593 u64 flags = 0;
1586 struct btrfs_key found_key; 1594 struct btrfs_key found_key;
1587 int search_commit_root = path->search_commit_root; 1595 int search_commit_root = path->search_commit_root;
1588 1596
1589 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); 1597 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1590 btrfs_release_path(path); 1598 btrfs_release_path(path);
1591 if (ret < 0) 1599 if (ret < 0)
1592 return ret; 1600 return ret;
1593 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) 1601 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1594 return -EINVAL; 1602 return -EINVAL;
1595 1603
1596 extent_item_pos = logical - found_key.objectid; 1604 extent_item_pos = logical - found_key.objectid;
1597 ret = iterate_extent_inodes(fs_info, found_key.objectid, 1605 ret = iterate_extent_inodes(fs_info, found_key.objectid,
1598 extent_item_pos, search_commit_root, 1606 extent_item_pos, search_commit_root,
1599 iterate, ctx); 1607 iterate, ctx);
1600 1608
1601 return ret; 1609 return ret;
1602 } 1610 }
1603 1611
1604 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off, 1612 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1605 struct extent_buffer *eb, void *ctx); 1613 struct extent_buffer *eb, void *ctx);
1606 1614
1607 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, 1615 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1608 struct btrfs_path *path, 1616 struct btrfs_path *path,
1609 iterate_irefs_t *iterate, void *ctx) 1617 iterate_irefs_t *iterate, void *ctx)
1610 { 1618 {
1611 int ret = 0; 1619 int ret = 0;
1612 int slot; 1620 int slot;
1613 u32 cur; 1621 u32 cur;
1614 u32 len; 1622 u32 len;
1615 u32 name_len; 1623 u32 name_len;
1616 u64 parent = 0; 1624 u64 parent = 0;
1617 int found = 0; 1625 int found = 0;
1618 struct extent_buffer *eb; 1626 struct extent_buffer *eb;
1619 struct btrfs_item *item; 1627 struct btrfs_item *item;
1620 struct btrfs_inode_ref *iref; 1628 struct btrfs_inode_ref *iref;
1621 struct btrfs_key found_key; 1629 struct btrfs_key found_key;
1622 1630
1623 while (!ret) { 1631 while (!ret) {
1624 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path, 1632 ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1625 &found_key); 1633 &found_key);
1626 if (ret < 0) 1634 if (ret < 0)
1627 break; 1635 break;
1628 if (ret) { 1636 if (ret) {
1629 ret = found ? 0 : -ENOENT; 1637 ret = found ? 0 : -ENOENT;
1630 break; 1638 break;
1631 } 1639 }
1632 ++found; 1640 ++found;
1633 1641
1634 parent = found_key.offset; 1642 parent = found_key.offset;
1635 slot = path->slots[0]; 1643 slot = path->slots[0];
1636 eb = btrfs_clone_extent_buffer(path->nodes[0]); 1644 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1637 if (!eb) { 1645 if (!eb) {
1638 ret = -ENOMEM; 1646 ret = -ENOMEM;
1639 break; 1647 break;
1640 } 1648 }
1641 extent_buffer_get(eb); 1649 extent_buffer_get(eb);
1642 btrfs_tree_read_lock(eb); 1650 btrfs_tree_read_lock(eb);
1643 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 1651 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1644 btrfs_release_path(path); 1652 btrfs_release_path(path);
1645 1653
1646 item = btrfs_item_nr(slot); 1654 item = btrfs_item_nr(slot);
1647 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); 1655 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1648 1656
1649 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) { 1657 for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1650 name_len = btrfs_inode_ref_name_len(eb, iref); 1658 name_len = btrfs_inode_ref_name_len(eb, iref);
1651 /* path must be released before calling iterate()! */ 1659 /* path must be released before calling iterate()! */
1652 pr_debug("following ref at offset %u for inode %llu in " 1660 pr_debug("following ref at offset %u for inode %llu in "
1653 "tree %llu\n", cur, found_key.objectid, 1661 "tree %llu\n", cur, found_key.objectid,
1654 fs_root->objectid); 1662 fs_root->objectid);
1655 ret = iterate(parent, name_len, 1663 ret = iterate(parent, name_len,
1656 (unsigned long)(iref + 1), eb, ctx); 1664 (unsigned long)(iref + 1), eb, ctx);
1657 if (ret) 1665 if (ret)
1658 break; 1666 break;
1659 len = sizeof(*iref) + name_len; 1667 len = sizeof(*iref) + name_len;
1660 iref = (struct btrfs_inode_ref *)((char *)iref + len); 1668 iref = (struct btrfs_inode_ref *)((char *)iref + len);
1661 } 1669 }
1662 btrfs_tree_read_unlock_blocking(eb); 1670 btrfs_tree_read_unlock_blocking(eb);
1663 free_extent_buffer(eb); 1671 free_extent_buffer(eb);
1664 } 1672 }
1665 1673
1666 btrfs_release_path(path); 1674 btrfs_release_path(path);
1667 1675
1668 return ret; 1676 return ret;
1669 } 1677 }
1670 1678
1671 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, 1679 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1672 struct btrfs_path *path, 1680 struct btrfs_path *path,
1673 iterate_irefs_t *iterate, void *ctx) 1681 iterate_irefs_t *iterate, void *ctx)
1674 { 1682 {
1675 int ret; 1683 int ret;
1676 int slot; 1684 int slot;
1677 u64 offset = 0; 1685 u64 offset = 0;
1678 u64 parent; 1686 u64 parent;
1679 int found = 0; 1687 int found = 0;
1680 struct extent_buffer *eb; 1688 struct extent_buffer *eb;
1681 struct btrfs_inode_extref *extref; 1689 struct btrfs_inode_extref *extref;
1682 struct extent_buffer *leaf; 1690 struct extent_buffer *leaf;
1683 u32 item_size; 1691 u32 item_size;
1684 u32 cur_offset; 1692 u32 cur_offset;
1685 unsigned long ptr; 1693 unsigned long ptr;
1686 1694
1687 while (1) { 1695 while (1) {
1688 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref, 1696 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1689 &offset); 1697 &offset);
1690 if (ret < 0) 1698 if (ret < 0)
1691 break; 1699 break;
1692 if (ret) { 1700 if (ret) {
1693 ret = found ? 0 : -ENOENT; 1701 ret = found ? 0 : -ENOENT;
1694 break; 1702 break;
1695 } 1703 }
1696 ++found; 1704 ++found;
1697 1705
1698 slot = path->slots[0]; 1706 slot = path->slots[0];
1699 eb = btrfs_clone_extent_buffer(path->nodes[0]); 1707 eb = btrfs_clone_extent_buffer(path->nodes[0]);
1700 if (!eb) { 1708 if (!eb) {
1701 ret = -ENOMEM; 1709 ret = -ENOMEM;
1702 break; 1710 break;
1703 } 1711 }
1704 extent_buffer_get(eb); 1712 extent_buffer_get(eb);
1705 1713
1706 btrfs_tree_read_lock(eb); 1714 btrfs_tree_read_lock(eb);
1707 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 1715 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1708 btrfs_release_path(path); 1716 btrfs_release_path(path);
1709 1717
1710 leaf = path->nodes[0]; 1718 leaf = path->nodes[0];
1711 item_size = btrfs_item_size_nr(leaf, slot); 1719 item_size = btrfs_item_size_nr(leaf, slot);
1712 ptr = btrfs_item_ptr_offset(leaf, slot); 1720 ptr = btrfs_item_ptr_offset(leaf, slot);
1713 cur_offset = 0; 1721 cur_offset = 0;
1714 1722
1715 while (cur_offset < item_size) { 1723 while (cur_offset < item_size) {
1716 u32 name_len; 1724 u32 name_len;
1717 1725
1718 extref = (struct btrfs_inode_extref *)(ptr + cur_offset); 1726 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1719 parent = btrfs_inode_extref_parent(eb, extref); 1727 parent = btrfs_inode_extref_parent(eb, extref);
1720 name_len = btrfs_inode_extref_name_len(eb, extref); 1728 name_len = btrfs_inode_extref_name_len(eb, extref);
1721 ret = iterate(parent, name_len, 1729 ret = iterate(parent, name_len,
1722 (unsigned long)&extref->name, eb, ctx); 1730 (unsigned long)&extref->name, eb, ctx);
1723 if (ret) 1731 if (ret)
1724 break; 1732 break;
1725 1733
1726 cur_offset += btrfs_inode_extref_name_len(leaf, extref); 1734 cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1727 cur_offset += sizeof(*extref); 1735 cur_offset += sizeof(*extref);
1728 } 1736 }
1729 btrfs_tree_read_unlock_blocking(eb); 1737 btrfs_tree_read_unlock_blocking(eb);
1730 free_extent_buffer(eb); 1738 free_extent_buffer(eb);
1731 1739
1732 offset++; 1740 offset++;
1733 } 1741 }
1734 1742
1735 btrfs_release_path(path); 1743 btrfs_release_path(path);
1736 1744
1737 return ret; 1745 return ret;
1738 } 1746 }
1739 1747
1740 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root, 1748 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1741 struct btrfs_path *path, iterate_irefs_t *iterate, 1749 struct btrfs_path *path, iterate_irefs_t *iterate,
1742 void *ctx) 1750 void *ctx)
1743 { 1751 {
1744 int ret; 1752 int ret;
1745 int found_refs = 0; 1753 int found_refs = 0;
1746 1754
1747 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx); 1755 ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1748 if (!ret) 1756 if (!ret)
1749 ++found_refs; 1757 ++found_refs;
1750 else if (ret != -ENOENT) 1758 else if (ret != -ENOENT)
1751 return ret; 1759 return ret;
1752 1760
1753 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx); 1761 ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1754 if (ret == -ENOENT && found_refs) 1762 if (ret == -ENOENT && found_refs)
1755 return 0; 1763 return 0;
1756 1764
1757 return ret; 1765 return ret;
1758 } 1766 }
1759 1767
1760 /* 1768 /*
1761 * returns 0 if the path could be dumped (probably truncated) 1769 * returns 0 if the path could be dumped (probably truncated)
1762 * returns <0 in case of an error 1770 * returns <0 in case of an error
1763 */ 1771 */
1764 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, 1772 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1765 struct extent_buffer *eb, void *ctx) 1773 struct extent_buffer *eb, void *ctx)
1766 { 1774 {
1767 struct inode_fs_paths *ipath = ctx; 1775 struct inode_fs_paths *ipath = ctx;
1768 char *fspath; 1776 char *fspath;
1769 char *fspath_min; 1777 char *fspath_min;
1770 int i = ipath->fspath->elem_cnt; 1778 int i = ipath->fspath->elem_cnt;
1771 const int s_ptr = sizeof(char *); 1779 const int s_ptr = sizeof(char *);
1772 u32 bytes_left; 1780 u32 bytes_left;
1773 1781
1774 bytes_left = ipath->fspath->bytes_left > s_ptr ? 1782 bytes_left = ipath->fspath->bytes_left > s_ptr ?
1775 ipath->fspath->bytes_left - s_ptr : 0; 1783 ipath->fspath->bytes_left - s_ptr : 0;
1776 1784
1777 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr; 1785 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1778 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len, 1786 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1779 name_off, eb, inum, fspath_min, bytes_left); 1787 name_off, eb, inum, fspath_min, bytes_left);
1780 if (IS_ERR(fspath)) 1788 if (IS_ERR(fspath))
1781 return PTR_ERR(fspath); 1789 return PTR_ERR(fspath);
1782 1790
1783 if (fspath > fspath_min) { 1791 if (fspath > fspath_min) {
1784 ipath->fspath->val[i] = (u64)(unsigned long)fspath; 1792 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1785 ++ipath->fspath->elem_cnt; 1793 ++ipath->fspath->elem_cnt;
1786 ipath->fspath->bytes_left = fspath - fspath_min; 1794 ipath->fspath->bytes_left = fspath - fspath_min;
1787 } else { 1795 } else {
1788 ++ipath->fspath->elem_missed; 1796 ++ipath->fspath->elem_missed;
1789 ipath->fspath->bytes_missing += fspath_min - fspath; 1797 ipath->fspath->bytes_missing += fspath_min - fspath;
1790 ipath->fspath->bytes_left = 0; 1798 ipath->fspath->bytes_left = 0;
1791 } 1799 }
1792 1800
1793 return 0; 1801 return 0;
1794 } 1802 }
1795 1803
1796 /* 1804 /*
1797 * this dumps all file system paths to the inode into the ipath struct, provided 1805 * this dumps all file system paths to the inode into the ipath struct, provided
1798 * is has been created large enough. each path is zero-terminated and accessed 1806 * is has been created large enough. each path is zero-terminated and accessed
1799 * from ipath->fspath->val[i]. 1807 * from ipath->fspath->val[i].
1800 * when it returns, there are ipath->fspath->elem_cnt number of paths available 1808 * when it returns, there are ipath->fspath->elem_cnt number of paths available
1801 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the 1809 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1802 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise, 1810 * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1803 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would 1811 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1804 * have been needed to return all paths. 1812 * have been needed to return all paths.
1805 */ 1813 */
1806 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) 1814 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1807 { 1815 {
1808 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path, 1816 return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1809 inode_to_path, ipath); 1817 inode_to_path, ipath);
1810 } 1818 }
1811 1819
1812 struct btrfs_data_container *init_data_container(u32 total_bytes) 1820 struct btrfs_data_container *init_data_container(u32 total_bytes)
1813 { 1821 {
1814 struct btrfs_data_container *data; 1822 struct btrfs_data_container *data;
1815 size_t alloc_bytes; 1823 size_t alloc_bytes;
1816 1824
1817 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); 1825 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1818 data = vmalloc(alloc_bytes); 1826 data = vmalloc(alloc_bytes);
1819 if (!data) 1827 if (!data)
1820 return ERR_PTR(-ENOMEM); 1828 return ERR_PTR(-ENOMEM);
1821 1829
1822 if (total_bytes >= sizeof(*data)) { 1830 if (total_bytes >= sizeof(*data)) {
1823 data->bytes_left = total_bytes - sizeof(*data); 1831 data->bytes_left = total_bytes - sizeof(*data);
1824 data->bytes_missing = 0; 1832 data->bytes_missing = 0;
1825 } else { 1833 } else {
1826 data->bytes_missing = sizeof(*data) - total_bytes; 1834 data->bytes_missing = sizeof(*data) - total_bytes;
1827 data->bytes_left = 0; 1835 data->bytes_left = 0;
1828 } 1836 }
1829 1837
1830 data->elem_cnt = 0; 1838 data->elem_cnt = 0;
1831 data->elem_missed = 0; 1839 data->elem_missed = 0;
1832 1840
1833 return data; 1841 return data;
1834 } 1842 }
1835 1843
1836 /* 1844 /*
1837 * allocates space to return multiple file system paths for an inode. 1845 * allocates space to return multiple file system paths for an inode.
1838 * total_bytes to allocate are passed, note that space usable for actual path 1846 * total_bytes to allocate are passed, note that space usable for actual path
1839 * information will be total_bytes - sizeof(struct inode_fs_paths). 1847 * information will be total_bytes - sizeof(struct inode_fs_paths).
1840 * the returned pointer must be freed with free_ipath() in the end. 1848 * the returned pointer must be freed with free_ipath() in the end.
1841 */ 1849 */
1842 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, 1850 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1843 struct btrfs_path *path) 1851 struct btrfs_path *path)
1844 { 1852 {
1845 struct inode_fs_paths *ifp; 1853 struct inode_fs_paths *ifp;
1846 struct btrfs_data_container *fspath; 1854 struct btrfs_data_container *fspath;
1847 1855
1848 fspath = init_data_container(total_bytes); 1856 fspath = init_data_container(total_bytes);
1849 if (IS_ERR(fspath)) 1857 if (IS_ERR(fspath))
1850 return (void *)fspath; 1858 return (void *)fspath;
1851 1859
1852 ifp = kmalloc(sizeof(*ifp), GFP_NOFS); 1860 ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1853 if (!ifp) { 1861 if (!ifp) {
1854 kfree(fspath); 1862 kfree(fspath);
1855 return ERR_PTR(-ENOMEM); 1863 return ERR_PTR(-ENOMEM);
1856 } 1864 }
1857 1865
1858 ifp->btrfs_path = path; 1866 ifp->btrfs_path = path;
1859 ifp->fspath = fspath; 1867 ifp->fspath = fspath;
1860 ifp->fs_root = fs_root; 1868 ifp->fs_root = fs_root;
1861 1869
1862 return ifp; 1870 return ifp;
1863 } 1871 }
1864 1872
1865 void free_ipath(struct inode_fs_paths *ipath) 1873 void free_ipath(struct inode_fs_paths *ipath)
1866 { 1874 {
1867 if (!ipath) 1875 if (!ipath)
1868 return; 1876 return;
1869 vfree(ipath->fspath); 1877 vfree(ipath->fspath);
1870 kfree(ipath); 1878 kfree(ipath);
1871 } 1879 }
1872 1880
1 /* 1 /*
2 * Copyright (C) 2011 STRATO. All rights reserved. 2 * Copyright (C) 2011 STRATO. All rights reserved.
3 * 3 *
4 * This program is free software; you can redistribute it and/or 4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public 5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation. 6 * License v2 as published by the Free Software Foundation.
7 * 7 *
8 * This program is distributed in the hope that it will be useful, 8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details. 11 * General Public License for more details.
12 * 12 *
13 * You should have received a copy of the GNU General Public 13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the 14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA. 16 * Boston, MA 021110-1307, USA.
17 */ 17 */
18 18
19 #ifndef __BTRFS_BACKREF__ 19 #ifndef __BTRFS_BACKREF__
20 #define __BTRFS_BACKREF__ 20 #define __BTRFS_BACKREF__
21 21
22 #include <linux/btrfs.h> 22 #include <linux/btrfs.h>
23 #include "ulist.h" 23 #include "ulist.h"
24 #include "extent_io.h" 24 #include "extent_io.h"
25 25
26 struct inode_fs_paths { 26 struct inode_fs_paths {
27 struct btrfs_path *btrfs_path; 27 struct btrfs_path *btrfs_path;
28 struct btrfs_root *fs_root; 28 struct btrfs_root *fs_root;
29 struct btrfs_data_container *fspath; 29 struct btrfs_data_container *fspath;
30 }; 30 };
31 31
32 typedef int (iterate_extent_inodes_t)(u64 inum, u64 offset, u64 root, 32 typedef int (iterate_extent_inodes_t)(u64 inum, u64 offset, u64 root,
33 void *ctx); 33 void *ctx);
34 34
35 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root, 35 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
36 struct btrfs_path *path); 36 struct btrfs_path *path);
37 37
38 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, 38 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
39 struct btrfs_path *path, struct btrfs_key *found_key, 39 struct btrfs_path *path, struct btrfs_key *found_key,
40 u64 *flags); 40 u64 *flags);
41 41
42 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, 42 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
43 struct btrfs_extent_item *ei, u32 item_size, 43 struct btrfs_key *key, struct btrfs_extent_item *ei,
44 u64 *out_root, u8 *out_level); 44 u32 item_size, u64 *out_root, u8 *out_level);
45 45
46 int iterate_extent_inodes(struct btrfs_fs_info *fs_info, 46 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
47 u64 extent_item_objectid, 47 u64 extent_item_objectid,
48 u64 extent_offset, int search_commit_root, 48 u64 extent_offset, int search_commit_root,
49 iterate_extent_inodes_t *iterate, void *ctx); 49 iterate_extent_inodes_t *iterate, void *ctx);
50 50
51 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, 51 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
52 struct btrfs_path *path, 52 struct btrfs_path *path,
53 iterate_extent_inodes_t *iterate, void *ctx); 53 iterate_extent_inodes_t *iterate, void *ctx);
54 54
55 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath); 55 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath);
56 56
57 int btrfs_find_all_roots(struct btrfs_trans_handle *trans, 57 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
58 struct btrfs_fs_info *fs_info, u64 bytenr, 58 struct btrfs_fs_info *fs_info, u64 bytenr,
59 u64 time_seq, struct ulist **roots); 59 u64 time_seq, struct ulist **roots);
60 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, 60 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
61 u32 name_len, unsigned long name_off, 61 u32 name_len, unsigned long name_off,
62 struct extent_buffer *eb_in, u64 parent, 62 struct extent_buffer *eb_in, u64 parent,
63 char *dest, u32 size); 63 char *dest, u32 size);
64 64
65 struct btrfs_data_container *init_data_container(u32 total_bytes); 65 struct btrfs_data_container *init_data_container(u32 total_bytes);
66 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, 66 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
67 struct btrfs_path *path); 67 struct btrfs_path *path);
68 void free_ipath(struct inode_fs_paths *ipath); 68 void free_ipath(struct inode_fs_paths *ipath);
69 69
70 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, 70 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
71 u64 start_off, struct btrfs_path *path, 71 u64 start_off, struct btrfs_path *path,
72 struct btrfs_inode_extref **ret_extref, 72 struct btrfs_inode_extref **ret_extref,
73 u64 *found_off); 73 u64 *found_off);
74 74
75 int __init btrfs_prelim_ref_init(void); 75 int __init btrfs_prelim_ref_init(void);
76 void btrfs_prelim_ref_exit(void); 76 void btrfs_prelim_ref_exit(void);
77 #endif 77 #endif
78 78
1 /* 1 /*
2 * Copyright (C) 2011, 2012 STRATO. All rights reserved. 2 * Copyright (C) 2011, 2012 STRATO. All rights reserved.
3 * 3 *
4 * This program is free software; you can redistribute it and/or 4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public 5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation. 6 * License v2 as published by the Free Software Foundation.
7 * 7 *
8 * This program is distributed in the hope that it will be useful, 8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details. 11 * General Public License for more details.
12 * 12 *
13 * You should have received a copy of the GNU General Public 13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the 14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA. 16 * Boston, MA 021110-1307, USA.
17 */ 17 */
18 18
19 #include <linux/blkdev.h> 19 #include <linux/blkdev.h>
20 #include <linux/ratelimit.h> 20 #include <linux/ratelimit.h>
21 #include "ctree.h" 21 #include "ctree.h"
22 #include "volumes.h" 22 #include "volumes.h"
23 #include "disk-io.h" 23 #include "disk-io.h"
24 #include "ordered-data.h" 24 #include "ordered-data.h"
25 #include "transaction.h" 25 #include "transaction.h"
26 #include "backref.h" 26 #include "backref.h"
27 #include "extent_io.h" 27 #include "extent_io.h"
28 #include "dev-replace.h" 28 #include "dev-replace.h"
29 #include "check-integrity.h" 29 #include "check-integrity.h"
30 #include "rcu-string.h" 30 #include "rcu-string.h"
31 #include "raid56.h" 31 #include "raid56.h"
32 32
33 /* 33 /*
34 * This is only the first step towards a full-features scrub. It reads all 34 * This is only the first step towards a full-features scrub. It reads all
35 * extent and super block and verifies the checksums. In case a bad checksum 35 * extent and super block and verifies the checksums. In case a bad checksum
36 * is found or the extent cannot be read, good data will be written back if 36 * is found or the extent cannot be read, good data will be written back if
37 * any can be found. 37 * any can be found.
38 * 38 *
39 * Future enhancements: 39 * Future enhancements:
40 * - In case an unrepairable extent is encountered, track which files are 40 * - In case an unrepairable extent is encountered, track which files are
41 * affected and report them 41 * affected and report them
42 * - track and record media errors, throw out bad devices 42 * - track and record media errors, throw out bad devices
43 * - add a mode to also read unallocated space 43 * - add a mode to also read unallocated space
44 */ 44 */
45 45
46 struct scrub_block; 46 struct scrub_block;
47 struct scrub_ctx; 47 struct scrub_ctx;
48 48
49 /* 49 /*
50 * the following three values only influence the performance. 50 * the following three values only influence the performance.
51 * The last one configures the number of parallel and outstanding I/O 51 * The last one configures the number of parallel and outstanding I/O
52 * operations. The first two values configure an upper limit for the number 52 * operations. The first two values configure an upper limit for the number
53 * of (dynamically allocated) pages that are added to a bio. 53 * of (dynamically allocated) pages that are added to a bio.
54 */ 54 */
55 #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */ 55 #define SCRUB_PAGES_PER_RD_BIO 32 /* 128k per bio */
56 #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */ 56 #define SCRUB_PAGES_PER_WR_BIO 32 /* 128k per bio */
57 #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */ 57 #define SCRUB_BIOS_PER_SCTX 64 /* 8MB per device in flight */
58 58
59 /* 59 /*
60 * the following value times PAGE_SIZE needs to be large enough to match the 60 * the following value times PAGE_SIZE needs to be large enough to match the
61 * largest node/leaf/sector size that shall be supported. 61 * largest node/leaf/sector size that shall be supported.
62 * Values larger than BTRFS_STRIPE_LEN are not supported. 62 * Values larger than BTRFS_STRIPE_LEN are not supported.
63 */ 63 */
64 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */ 64 #define SCRUB_MAX_PAGES_PER_BLOCK 16 /* 64k per node/leaf/sector */
65 65
66 struct scrub_page { 66 struct scrub_page {
67 struct scrub_block *sblock; 67 struct scrub_block *sblock;
68 struct page *page; 68 struct page *page;
69 struct btrfs_device *dev; 69 struct btrfs_device *dev;
70 u64 flags; /* extent flags */ 70 u64 flags; /* extent flags */
71 u64 generation; 71 u64 generation;
72 u64 logical; 72 u64 logical;
73 u64 physical; 73 u64 physical;
74 u64 physical_for_dev_replace; 74 u64 physical_for_dev_replace;
75 atomic_t ref_count; 75 atomic_t ref_count;
76 struct { 76 struct {
77 unsigned int mirror_num:8; 77 unsigned int mirror_num:8;
78 unsigned int have_csum:1; 78 unsigned int have_csum:1;
79 unsigned int io_error:1; 79 unsigned int io_error:1;
80 }; 80 };
81 u8 csum[BTRFS_CSUM_SIZE]; 81 u8 csum[BTRFS_CSUM_SIZE];
82 }; 82 };
83 83
84 struct scrub_bio { 84 struct scrub_bio {
85 int index; 85 int index;
86 struct scrub_ctx *sctx; 86 struct scrub_ctx *sctx;
87 struct btrfs_device *dev; 87 struct btrfs_device *dev;
88 struct bio *bio; 88 struct bio *bio;
89 int err; 89 int err;
90 u64 logical; 90 u64 logical;
91 u64 physical; 91 u64 physical;
92 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO 92 #if SCRUB_PAGES_PER_WR_BIO >= SCRUB_PAGES_PER_RD_BIO
93 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO]; 93 struct scrub_page *pagev[SCRUB_PAGES_PER_WR_BIO];
94 #else 94 #else
95 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO]; 95 struct scrub_page *pagev[SCRUB_PAGES_PER_RD_BIO];
96 #endif 96 #endif
97 int page_count; 97 int page_count;
98 int next_free; 98 int next_free;
99 struct btrfs_work work; 99 struct btrfs_work work;
100 }; 100 };
101 101
102 struct scrub_block { 102 struct scrub_block {
103 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK]; 103 struct scrub_page *pagev[SCRUB_MAX_PAGES_PER_BLOCK];
104 int page_count; 104 int page_count;
105 atomic_t outstanding_pages; 105 atomic_t outstanding_pages;
106 atomic_t ref_count; /* free mem on transition to zero */ 106 atomic_t ref_count; /* free mem on transition to zero */
107 struct scrub_ctx *sctx; 107 struct scrub_ctx *sctx;
108 struct { 108 struct {
109 unsigned int header_error:1; 109 unsigned int header_error:1;
110 unsigned int checksum_error:1; 110 unsigned int checksum_error:1;
111 unsigned int no_io_error_seen:1; 111 unsigned int no_io_error_seen:1;
112 unsigned int generation_error:1; /* also sets header_error */ 112 unsigned int generation_error:1; /* also sets header_error */
113 }; 113 };
114 }; 114 };
115 115
116 struct scrub_wr_ctx { 116 struct scrub_wr_ctx {
117 struct scrub_bio *wr_curr_bio; 117 struct scrub_bio *wr_curr_bio;
118 struct btrfs_device *tgtdev; 118 struct btrfs_device *tgtdev;
119 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */ 119 int pages_per_wr_bio; /* <= SCRUB_PAGES_PER_WR_BIO */
120 atomic_t flush_all_writes; 120 atomic_t flush_all_writes;
121 struct mutex wr_lock; 121 struct mutex wr_lock;
122 }; 122 };
123 123
124 struct scrub_ctx { 124 struct scrub_ctx {
125 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX]; 125 struct scrub_bio *bios[SCRUB_BIOS_PER_SCTX];
126 struct btrfs_root *dev_root; 126 struct btrfs_root *dev_root;
127 int first_free; 127 int first_free;
128 int curr; 128 int curr;
129 atomic_t bios_in_flight; 129 atomic_t bios_in_flight;
130 atomic_t workers_pending; 130 atomic_t workers_pending;
131 spinlock_t list_lock; 131 spinlock_t list_lock;
132 wait_queue_head_t list_wait; 132 wait_queue_head_t list_wait;
133 u16 csum_size; 133 u16 csum_size;
134 struct list_head csum_list; 134 struct list_head csum_list;
135 atomic_t cancel_req; 135 atomic_t cancel_req;
136 int readonly; 136 int readonly;
137 int pages_per_rd_bio; 137 int pages_per_rd_bio;
138 u32 sectorsize; 138 u32 sectorsize;
139 u32 nodesize; 139 u32 nodesize;
140 u32 leafsize; 140 u32 leafsize;
141 141
142 int is_dev_replace; 142 int is_dev_replace;
143 struct scrub_wr_ctx wr_ctx; 143 struct scrub_wr_ctx wr_ctx;
144 144
145 /* 145 /*
146 * statistics 146 * statistics
147 */ 147 */
148 struct btrfs_scrub_progress stat; 148 struct btrfs_scrub_progress stat;
149 spinlock_t stat_lock; 149 spinlock_t stat_lock;
150 }; 150 };
151 151
152 struct scrub_fixup_nodatasum { 152 struct scrub_fixup_nodatasum {
153 struct scrub_ctx *sctx; 153 struct scrub_ctx *sctx;
154 struct btrfs_device *dev; 154 struct btrfs_device *dev;
155 u64 logical; 155 u64 logical;
156 struct btrfs_root *root; 156 struct btrfs_root *root;
157 struct btrfs_work work; 157 struct btrfs_work work;
158 int mirror_num; 158 int mirror_num;
159 }; 159 };
160 160
161 struct scrub_nocow_inode { 161 struct scrub_nocow_inode {
162 u64 inum; 162 u64 inum;
163 u64 offset; 163 u64 offset;
164 u64 root; 164 u64 root;
165 struct list_head list; 165 struct list_head list;
166 }; 166 };
167 167
168 struct scrub_copy_nocow_ctx { 168 struct scrub_copy_nocow_ctx {
169 struct scrub_ctx *sctx; 169 struct scrub_ctx *sctx;
170 u64 logical; 170 u64 logical;
171 u64 len; 171 u64 len;
172 int mirror_num; 172 int mirror_num;
173 u64 physical_for_dev_replace; 173 u64 physical_for_dev_replace;
174 struct list_head inodes; 174 struct list_head inodes;
175 struct btrfs_work work; 175 struct btrfs_work work;
176 }; 176 };
177 177
178 struct scrub_warning { 178 struct scrub_warning {
179 struct btrfs_path *path; 179 struct btrfs_path *path;
180 u64 extent_item_size; 180 u64 extent_item_size;
181 char *scratch_buf; 181 char *scratch_buf;
182 char *msg_buf; 182 char *msg_buf;
183 const char *errstr; 183 const char *errstr;
184 sector_t sector; 184 sector_t sector;
185 u64 logical; 185 u64 logical;
186 struct btrfs_device *dev; 186 struct btrfs_device *dev;
187 int msg_bufsize; 187 int msg_bufsize;
188 int scratch_bufsize; 188 int scratch_bufsize;
189 }; 189 };
190 190
191 191
192 static void scrub_pending_bio_inc(struct scrub_ctx *sctx); 192 static void scrub_pending_bio_inc(struct scrub_ctx *sctx);
193 static void scrub_pending_bio_dec(struct scrub_ctx *sctx); 193 static void scrub_pending_bio_dec(struct scrub_ctx *sctx);
194 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx); 194 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx);
195 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx); 195 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx);
196 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check); 196 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check);
197 static int scrub_setup_recheck_block(struct scrub_ctx *sctx, 197 static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
198 struct btrfs_fs_info *fs_info, 198 struct btrfs_fs_info *fs_info,
199 struct scrub_block *original_sblock, 199 struct scrub_block *original_sblock,
200 u64 length, u64 logical, 200 u64 length, u64 logical,
201 struct scrub_block *sblocks_for_recheck); 201 struct scrub_block *sblocks_for_recheck);
202 static void scrub_recheck_block(struct btrfs_fs_info *fs_info, 202 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
203 struct scrub_block *sblock, int is_metadata, 203 struct scrub_block *sblock, int is_metadata,
204 int have_csum, u8 *csum, u64 generation, 204 int have_csum, u8 *csum, u64 generation,
205 u16 csum_size); 205 u16 csum_size);
206 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, 206 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
207 struct scrub_block *sblock, 207 struct scrub_block *sblock,
208 int is_metadata, int have_csum, 208 int is_metadata, int have_csum,
209 const u8 *csum, u64 generation, 209 const u8 *csum, u64 generation,
210 u16 csum_size); 210 u16 csum_size);
211 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, 211 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
212 struct scrub_block *sblock_good, 212 struct scrub_block *sblock_good,
213 int force_write); 213 int force_write);
214 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, 214 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
215 struct scrub_block *sblock_good, 215 struct scrub_block *sblock_good,
216 int page_num, int force_write); 216 int page_num, int force_write);
217 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock); 217 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock);
218 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, 218 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
219 int page_num); 219 int page_num);
220 static int scrub_checksum_data(struct scrub_block *sblock); 220 static int scrub_checksum_data(struct scrub_block *sblock);
221 static int scrub_checksum_tree_block(struct scrub_block *sblock); 221 static int scrub_checksum_tree_block(struct scrub_block *sblock);
222 static int scrub_checksum_super(struct scrub_block *sblock); 222 static int scrub_checksum_super(struct scrub_block *sblock);
223 static void scrub_block_get(struct scrub_block *sblock); 223 static void scrub_block_get(struct scrub_block *sblock);
224 static void scrub_block_put(struct scrub_block *sblock); 224 static void scrub_block_put(struct scrub_block *sblock);
225 static void scrub_page_get(struct scrub_page *spage); 225 static void scrub_page_get(struct scrub_page *spage);
226 static void scrub_page_put(struct scrub_page *spage); 226 static void scrub_page_put(struct scrub_page *spage);
227 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, 227 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
228 struct scrub_page *spage); 228 struct scrub_page *spage);
229 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, 229 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
230 u64 physical, struct btrfs_device *dev, u64 flags, 230 u64 physical, struct btrfs_device *dev, u64 flags,
231 u64 gen, int mirror_num, u8 *csum, int force, 231 u64 gen, int mirror_num, u8 *csum, int force,
232 u64 physical_for_dev_replace); 232 u64 physical_for_dev_replace);
233 static void scrub_bio_end_io(struct bio *bio, int err); 233 static void scrub_bio_end_io(struct bio *bio, int err);
234 static void scrub_bio_end_io_worker(struct btrfs_work *work); 234 static void scrub_bio_end_io_worker(struct btrfs_work *work);
235 static void scrub_block_complete(struct scrub_block *sblock); 235 static void scrub_block_complete(struct scrub_block *sblock);
236 static void scrub_remap_extent(struct btrfs_fs_info *fs_info, 236 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
237 u64 extent_logical, u64 extent_len, 237 u64 extent_logical, u64 extent_len,
238 u64 *extent_physical, 238 u64 *extent_physical,
239 struct btrfs_device **extent_dev, 239 struct btrfs_device **extent_dev,
240 int *extent_mirror_num); 240 int *extent_mirror_num);
241 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, 241 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
242 struct scrub_wr_ctx *wr_ctx, 242 struct scrub_wr_ctx *wr_ctx,
243 struct btrfs_fs_info *fs_info, 243 struct btrfs_fs_info *fs_info,
244 struct btrfs_device *dev, 244 struct btrfs_device *dev,
245 int is_dev_replace); 245 int is_dev_replace);
246 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx); 246 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx);
247 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, 247 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
248 struct scrub_page *spage); 248 struct scrub_page *spage);
249 static void scrub_wr_submit(struct scrub_ctx *sctx); 249 static void scrub_wr_submit(struct scrub_ctx *sctx);
250 static void scrub_wr_bio_end_io(struct bio *bio, int err); 250 static void scrub_wr_bio_end_io(struct bio *bio, int err);
251 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work); 251 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work);
252 static int write_page_nocow(struct scrub_ctx *sctx, 252 static int write_page_nocow(struct scrub_ctx *sctx,
253 u64 physical_for_dev_replace, struct page *page); 253 u64 physical_for_dev_replace, struct page *page);
254 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, 254 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
255 struct scrub_copy_nocow_ctx *ctx); 255 struct scrub_copy_nocow_ctx *ctx);
256 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, 256 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
257 int mirror_num, u64 physical_for_dev_replace); 257 int mirror_num, u64 physical_for_dev_replace);
258 static void copy_nocow_pages_worker(struct btrfs_work *work); 258 static void copy_nocow_pages_worker(struct btrfs_work *work);
259 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); 259 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
260 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info); 260 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info);
261 261
262 262
263 static void scrub_pending_bio_inc(struct scrub_ctx *sctx) 263 static void scrub_pending_bio_inc(struct scrub_ctx *sctx)
264 { 264 {
265 atomic_inc(&sctx->bios_in_flight); 265 atomic_inc(&sctx->bios_in_flight);
266 } 266 }
267 267
268 static void scrub_pending_bio_dec(struct scrub_ctx *sctx) 268 static void scrub_pending_bio_dec(struct scrub_ctx *sctx)
269 { 269 {
270 atomic_dec(&sctx->bios_in_flight); 270 atomic_dec(&sctx->bios_in_flight);
271 wake_up(&sctx->list_wait); 271 wake_up(&sctx->list_wait);
272 } 272 }
273 273
274 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) 274 static void __scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
275 { 275 {
276 while (atomic_read(&fs_info->scrub_pause_req)) { 276 while (atomic_read(&fs_info->scrub_pause_req)) {
277 mutex_unlock(&fs_info->scrub_lock); 277 mutex_unlock(&fs_info->scrub_lock);
278 wait_event(fs_info->scrub_pause_wait, 278 wait_event(fs_info->scrub_pause_wait,
279 atomic_read(&fs_info->scrub_pause_req) == 0); 279 atomic_read(&fs_info->scrub_pause_req) == 0);
280 mutex_lock(&fs_info->scrub_lock); 280 mutex_lock(&fs_info->scrub_lock);
281 } 281 }
282 } 282 }
283 283
284 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info) 284 static void scrub_blocked_if_needed(struct btrfs_fs_info *fs_info)
285 { 285 {
286 atomic_inc(&fs_info->scrubs_paused); 286 atomic_inc(&fs_info->scrubs_paused);
287 wake_up(&fs_info->scrub_pause_wait); 287 wake_up(&fs_info->scrub_pause_wait);
288 288
289 mutex_lock(&fs_info->scrub_lock); 289 mutex_lock(&fs_info->scrub_lock);
290 __scrub_blocked_if_needed(fs_info); 290 __scrub_blocked_if_needed(fs_info);
291 atomic_dec(&fs_info->scrubs_paused); 291 atomic_dec(&fs_info->scrubs_paused);
292 mutex_unlock(&fs_info->scrub_lock); 292 mutex_unlock(&fs_info->scrub_lock);
293 293
294 wake_up(&fs_info->scrub_pause_wait); 294 wake_up(&fs_info->scrub_pause_wait);
295 } 295 }
296 296
297 /* 297 /*
298 * used for workers that require transaction commits (i.e., for the 298 * used for workers that require transaction commits (i.e., for the
299 * NOCOW case) 299 * NOCOW case)
300 */ 300 */
301 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx) 301 static void scrub_pending_trans_workers_inc(struct scrub_ctx *sctx)
302 { 302 {
303 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; 303 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
304 304
305 /* 305 /*
306 * increment scrubs_running to prevent cancel requests from 306 * increment scrubs_running to prevent cancel requests from
307 * completing as long as a worker is running. we must also 307 * completing as long as a worker is running. we must also
308 * increment scrubs_paused to prevent deadlocking on pause 308 * increment scrubs_paused to prevent deadlocking on pause
309 * requests used for transactions commits (as the worker uses a 309 * requests used for transactions commits (as the worker uses a
310 * transaction context). it is safe to regard the worker 310 * transaction context). it is safe to regard the worker
311 * as paused for all matters practical. effectively, we only 311 * as paused for all matters practical. effectively, we only
312 * avoid cancellation requests from completing. 312 * avoid cancellation requests from completing.
313 */ 313 */
314 mutex_lock(&fs_info->scrub_lock); 314 mutex_lock(&fs_info->scrub_lock);
315 atomic_inc(&fs_info->scrubs_running); 315 atomic_inc(&fs_info->scrubs_running);
316 atomic_inc(&fs_info->scrubs_paused); 316 atomic_inc(&fs_info->scrubs_paused);
317 mutex_unlock(&fs_info->scrub_lock); 317 mutex_unlock(&fs_info->scrub_lock);
318 318
319 /* 319 /*
320 * check if @scrubs_running=@scrubs_paused condition 320 * check if @scrubs_running=@scrubs_paused condition
321 * inside wait_event() is not an atomic operation. 321 * inside wait_event() is not an atomic operation.
322 * which means we may inc/dec @scrub_running/paused 322 * which means we may inc/dec @scrub_running/paused
323 * at any time. Let's wake up @scrub_pause_wait as 323 * at any time. Let's wake up @scrub_pause_wait as
324 * much as we can to let commit transaction blocked less. 324 * much as we can to let commit transaction blocked less.
325 */ 325 */
326 wake_up(&fs_info->scrub_pause_wait); 326 wake_up(&fs_info->scrub_pause_wait);
327 327
328 atomic_inc(&sctx->workers_pending); 328 atomic_inc(&sctx->workers_pending);
329 } 329 }
330 330
331 /* used for workers that require transaction commits */ 331 /* used for workers that require transaction commits */
332 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx) 332 static void scrub_pending_trans_workers_dec(struct scrub_ctx *sctx)
333 { 333 {
334 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; 334 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
335 335
336 /* 336 /*
337 * see scrub_pending_trans_workers_inc() why we're pretending 337 * see scrub_pending_trans_workers_inc() why we're pretending
338 * to be paused in the scrub counters 338 * to be paused in the scrub counters
339 */ 339 */
340 mutex_lock(&fs_info->scrub_lock); 340 mutex_lock(&fs_info->scrub_lock);
341 atomic_dec(&fs_info->scrubs_running); 341 atomic_dec(&fs_info->scrubs_running);
342 atomic_dec(&fs_info->scrubs_paused); 342 atomic_dec(&fs_info->scrubs_paused);
343 mutex_unlock(&fs_info->scrub_lock); 343 mutex_unlock(&fs_info->scrub_lock);
344 atomic_dec(&sctx->workers_pending); 344 atomic_dec(&sctx->workers_pending);
345 wake_up(&fs_info->scrub_pause_wait); 345 wake_up(&fs_info->scrub_pause_wait);
346 wake_up(&sctx->list_wait); 346 wake_up(&sctx->list_wait);
347 } 347 }
348 348
349 static void scrub_free_csums(struct scrub_ctx *sctx) 349 static void scrub_free_csums(struct scrub_ctx *sctx)
350 { 350 {
351 while (!list_empty(&sctx->csum_list)) { 351 while (!list_empty(&sctx->csum_list)) {
352 struct btrfs_ordered_sum *sum; 352 struct btrfs_ordered_sum *sum;
353 sum = list_first_entry(&sctx->csum_list, 353 sum = list_first_entry(&sctx->csum_list,
354 struct btrfs_ordered_sum, list); 354 struct btrfs_ordered_sum, list);
355 list_del(&sum->list); 355 list_del(&sum->list);
356 kfree(sum); 356 kfree(sum);
357 } 357 }
358 } 358 }
359 359
360 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx) 360 static noinline_for_stack void scrub_free_ctx(struct scrub_ctx *sctx)
361 { 361 {
362 int i; 362 int i;
363 363
364 if (!sctx) 364 if (!sctx)
365 return; 365 return;
366 366
367 scrub_free_wr_ctx(&sctx->wr_ctx); 367 scrub_free_wr_ctx(&sctx->wr_ctx);
368 368
369 /* this can happen when scrub is cancelled */ 369 /* this can happen when scrub is cancelled */
370 if (sctx->curr != -1) { 370 if (sctx->curr != -1) {
371 struct scrub_bio *sbio = sctx->bios[sctx->curr]; 371 struct scrub_bio *sbio = sctx->bios[sctx->curr];
372 372
373 for (i = 0; i < sbio->page_count; i++) { 373 for (i = 0; i < sbio->page_count; i++) {
374 WARN_ON(!sbio->pagev[i]->page); 374 WARN_ON(!sbio->pagev[i]->page);
375 scrub_block_put(sbio->pagev[i]->sblock); 375 scrub_block_put(sbio->pagev[i]->sblock);
376 } 376 }
377 bio_put(sbio->bio); 377 bio_put(sbio->bio);
378 } 378 }
379 379
380 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { 380 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
381 struct scrub_bio *sbio = sctx->bios[i]; 381 struct scrub_bio *sbio = sctx->bios[i];
382 382
383 if (!sbio) 383 if (!sbio)
384 break; 384 break;
385 kfree(sbio); 385 kfree(sbio);
386 } 386 }
387 387
388 scrub_free_csums(sctx); 388 scrub_free_csums(sctx);
389 kfree(sctx); 389 kfree(sctx);
390 } 390 }
391 391
392 static noinline_for_stack 392 static noinline_for_stack
393 struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace) 393 struct scrub_ctx *scrub_setup_ctx(struct btrfs_device *dev, int is_dev_replace)
394 { 394 {
395 struct scrub_ctx *sctx; 395 struct scrub_ctx *sctx;
396 int i; 396 int i;
397 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info; 397 struct btrfs_fs_info *fs_info = dev->dev_root->fs_info;
398 int pages_per_rd_bio; 398 int pages_per_rd_bio;
399 int ret; 399 int ret;
400 400
401 /* 401 /*
402 * the setting of pages_per_rd_bio is correct for scrub but might 402 * the setting of pages_per_rd_bio is correct for scrub but might
403 * be wrong for the dev_replace code where we might read from 403 * be wrong for the dev_replace code where we might read from
404 * different devices in the initial huge bios. However, that 404 * different devices in the initial huge bios. However, that
405 * code is able to correctly handle the case when adding a page 405 * code is able to correctly handle the case when adding a page
406 * to a bio fails. 406 * to a bio fails.
407 */ 407 */
408 if (dev->bdev) 408 if (dev->bdev)
409 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO, 409 pages_per_rd_bio = min_t(int, SCRUB_PAGES_PER_RD_BIO,
410 bio_get_nr_vecs(dev->bdev)); 410 bio_get_nr_vecs(dev->bdev));
411 else 411 else
412 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO; 412 pages_per_rd_bio = SCRUB_PAGES_PER_RD_BIO;
413 sctx = kzalloc(sizeof(*sctx), GFP_NOFS); 413 sctx = kzalloc(sizeof(*sctx), GFP_NOFS);
414 if (!sctx) 414 if (!sctx)
415 goto nomem; 415 goto nomem;
416 sctx->is_dev_replace = is_dev_replace; 416 sctx->is_dev_replace = is_dev_replace;
417 sctx->pages_per_rd_bio = pages_per_rd_bio; 417 sctx->pages_per_rd_bio = pages_per_rd_bio;
418 sctx->curr = -1; 418 sctx->curr = -1;
419 sctx->dev_root = dev->dev_root; 419 sctx->dev_root = dev->dev_root;
420 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) { 420 for (i = 0; i < SCRUB_BIOS_PER_SCTX; ++i) {
421 struct scrub_bio *sbio; 421 struct scrub_bio *sbio;
422 422
423 sbio = kzalloc(sizeof(*sbio), GFP_NOFS); 423 sbio = kzalloc(sizeof(*sbio), GFP_NOFS);
424 if (!sbio) 424 if (!sbio)
425 goto nomem; 425 goto nomem;
426 sctx->bios[i] = sbio; 426 sctx->bios[i] = sbio;
427 427
428 sbio->index = i; 428 sbio->index = i;
429 sbio->sctx = sctx; 429 sbio->sctx = sctx;
430 sbio->page_count = 0; 430 sbio->page_count = 0;
431 btrfs_init_work(&sbio->work, scrub_bio_end_io_worker, 431 btrfs_init_work(&sbio->work, scrub_bio_end_io_worker,
432 NULL, NULL); 432 NULL, NULL);
433 433
434 if (i != SCRUB_BIOS_PER_SCTX - 1) 434 if (i != SCRUB_BIOS_PER_SCTX - 1)
435 sctx->bios[i]->next_free = i + 1; 435 sctx->bios[i]->next_free = i + 1;
436 else 436 else
437 sctx->bios[i]->next_free = -1; 437 sctx->bios[i]->next_free = -1;
438 } 438 }
439 sctx->first_free = 0; 439 sctx->first_free = 0;
440 sctx->nodesize = dev->dev_root->nodesize; 440 sctx->nodesize = dev->dev_root->nodesize;
441 sctx->leafsize = dev->dev_root->leafsize; 441 sctx->leafsize = dev->dev_root->leafsize;
442 sctx->sectorsize = dev->dev_root->sectorsize; 442 sctx->sectorsize = dev->dev_root->sectorsize;
443 atomic_set(&sctx->bios_in_flight, 0); 443 atomic_set(&sctx->bios_in_flight, 0);
444 atomic_set(&sctx->workers_pending, 0); 444 atomic_set(&sctx->workers_pending, 0);
445 atomic_set(&sctx->cancel_req, 0); 445 atomic_set(&sctx->cancel_req, 0);
446 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy); 446 sctx->csum_size = btrfs_super_csum_size(fs_info->super_copy);
447 INIT_LIST_HEAD(&sctx->csum_list); 447 INIT_LIST_HEAD(&sctx->csum_list);
448 448
449 spin_lock_init(&sctx->list_lock); 449 spin_lock_init(&sctx->list_lock);
450 spin_lock_init(&sctx->stat_lock); 450 spin_lock_init(&sctx->stat_lock);
451 init_waitqueue_head(&sctx->list_wait); 451 init_waitqueue_head(&sctx->list_wait);
452 452
453 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info, 453 ret = scrub_setup_wr_ctx(sctx, &sctx->wr_ctx, fs_info,
454 fs_info->dev_replace.tgtdev, is_dev_replace); 454 fs_info->dev_replace.tgtdev, is_dev_replace);
455 if (ret) { 455 if (ret) {
456 scrub_free_ctx(sctx); 456 scrub_free_ctx(sctx);
457 return ERR_PTR(ret); 457 return ERR_PTR(ret);
458 } 458 }
459 return sctx; 459 return sctx;
460 460
461 nomem: 461 nomem:
462 scrub_free_ctx(sctx); 462 scrub_free_ctx(sctx);
463 return ERR_PTR(-ENOMEM); 463 return ERR_PTR(-ENOMEM);
464 } 464 }
465 465
466 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root, 466 static int scrub_print_warning_inode(u64 inum, u64 offset, u64 root,
467 void *warn_ctx) 467 void *warn_ctx)
468 { 468 {
469 u64 isize; 469 u64 isize;
470 u32 nlink; 470 u32 nlink;
471 int ret; 471 int ret;
472 int i; 472 int i;
473 struct extent_buffer *eb; 473 struct extent_buffer *eb;
474 struct btrfs_inode_item *inode_item; 474 struct btrfs_inode_item *inode_item;
475 struct scrub_warning *swarn = warn_ctx; 475 struct scrub_warning *swarn = warn_ctx;
476 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info; 476 struct btrfs_fs_info *fs_info = swarn->dev->dev_root->fs_info;
477 struct inode_fs_paths *ipath = NULL; 477 struct inode_fs_paths *ipath = NULL;
478 struct btrfs_root *local_root; 478 struct btrfs_root *local_root;
479 struct btrfs_key root_key; 479 struct btrfs_key root_key;
480 480
481 root_key.objectid = root; 481 root_key.objectid = root;
482 root_key.type = BTRFS_ROOT_ITEM_KEY; 482 root_key.type = BTRFS_ROOT_ITEM_KEY;
483 root_key.offset = (u64)-1; 483 root_key.offset = (u64)-1;
484 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key); 484 local_root = btrfs_read_fs_root_no_name(fs_info, &root_key);
485 if (IS_ERR(local_root)) { 485 if (IS_ERR(local_root)) {
486 ret = PTR_ERR(local_root); 486 ret = PTR_ERR(local_root);
487 goto err; 487 goto err;
488 } 488 }
489 489
490 ret = inode_item_info(inum, 0, local_root, swarn->path); 490 ret = inode_item_info(inum, 0, local_root, swarn->path);
491 if (ret) { 491 if (ret) {
492 btrfs_release_path(swarn->path); 492 btrfs_release_path(swarn->path);
493 goto err; 493 goto err;
494 } 494 }
495 495
496 eb = swarn->path->nodes[0]; 496 eb = swarn->path->nodes[0];
497 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0], 497 inode_item = btrfs_item_ptr(eb, swarn->path->slots[0],
498 struct btrfs_inode_item); 498 struct btrfs_inode_item);
499 isize = btrfs_inode_size(eb, inode_item); 499 isize = btrfs_inode_size(eb, inode_item);
500 nlink = btrfs_inode_nlink(eb, inode_item); 500 nlink = btrfs_inode_nlink(eb, inode_item);
501 btrfs_release_path(swarn->path); 501 btrfs_release_path(swarn->path);
502 502
503 ipath = init_ipath(4096, local_root, swarn->path); 503 ipath = init_ipath(4096, local_root, swarn->path);
504 if (IS_ERR(ipath)) { 504 if (IS_ERR(ipath)) {
505 ret = PTR_ERR(ipath); 505 ret = PTR_ERR(ipath);
506 ipath = NULL; 506 ipath = NULL;
507 goto err; 507 goto err;
508 } 508 }
509 ret = paths_from_inode(inum, ipath); 509 ret = paths_from_inode(inum, ipath);
510 510
511 if (ret < 0) 511 if (ret < 0)
512 goto err; 512 goto err;
513 513
514 /* 514 /*
515 * we deliberately ignore the bit ipath might have been too small to 515 * we deliberately ignore the bit ipath might have been too small to
516 * hold all of the paths here 516 * hold all of the paths here
517 */ 517 */
518 for (i = 0; i < ipath->fspath->elem_cnt; ++i) 518 for (i = 0; i < ipath->fspath->elem_cnt; ++i)
519 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev " 519 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
520 "%s, sector %llu, root %llu, inode %llu, offset %llu, " 520 "%s, sector %llu, root %llu, inode %llu, offset %llu, "
521 "length %llu, links %u (path: %s)\n", swarn->errstr, 521 "length %llu, links %u (path: %s)\n", swarn->errstr,
522 swarn->logical, rcu_str_deref(swarn->dev->name), 522 swarn->logical, rcu_str_deref(swarn->dev->name),
523 (unsigned long long)swarn->sector, root, inum, offset, 523 (unsigned long long)swarn->sector, root, inum, offset,
524 min(isize - offset, (u64)PAGE_SIZE), nlink, 524 min(isize - offset, (u64)PAGE_SIZE), nlink,
525 (char *)(unsigned long)ipath->fspath->val[i]); 525 (char *)(unsigned long)ipath->fspath->val[i]);
526 526
527 free_ipath(ipath); 527 free_ipath(ipath);
528 return 0; 528 return 0;
529 529
530 err: 530 err:
531 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev " 531 printk_in_rcu(KERN_WARNING "BTRFS: %s at logical %llu on dev "
532 "%s, sector %llu, root %llu, inode %llu, offset %llu: path " 532 "%s, sector %llu, root %llu, inode %llu, offset %llu: path "
533 "resolving failed with ret=%d\n", swarn->errstr, 533 "resolving failed with ret=%d\n", swarn->errstr,
534 swarn->logical, rcu_str_deref(swarn->dev->name), 534 swarn->logical, rcu_str_deref(swarn->dev->name),
535 (unsigned long long)swarn->sector, root, inum, offset, ret); 535 (unsigned long long)swarn->sector, root, inum, offset, ret);
536 536
537 free_ipath(ipath); 537 free_ipath(ipath);
538 return 0; 538 return 0;
539 } 539 }
540 540
541 static void scrub_print_warning(const char *errstr, struct scrub_block *sblock) 541 static void scrub_print_warning(const char *errstr, struct scrub_block *sblock)
542 { 542 {
543 struct btrfs_device *dev; 543 struct btrfs_device *dev;
544 struct btrfs_fs_info *fs_info; 544 struct btrfs_fs_info *fs_info;
545 struct btrfs_path *path; 545 struct btrfs_path *path;
546 struct btrfs_key found_key; 546 struct btrfs_key found_key;
547 struct extent_buffer *eb; 547 struct extent_buffer *eb;
548 struct btrfs_extent_item *ei; 548 struct btrfs_extent_item *ei;
549 struct scrub_warning swarn; 549 struct scrub_warning swarn;
550 unsigned long ptr = 0; 550 unsigned long ptr = 0;
551 u64 extent_item_pos; 551 u64 extent_item_pos;
552 u64 flags = 0; 552 u64 flags = 0;
553 u64 ref_root; 553 u64 ref_root;
554 u32 item_size; 554 u32 item_size;
555 u8 ref_level; 555 u8 ref_level;
556 const int bufsize = 4096; 556 const int bufsize = 4096;
557 int ret; 557 int ret;
558 558
559 WARN_ON(sblock->page_count < 1); 559 WARN_ON(sblock->page_count < 1);
560 dev = sblock->pagev[0]->dev; 560 dev = sblock->pagev[0]->dev;
561 fs_info = sblock->sctx->dev_root->fs_info; 561 fs_info = sblock->sctx->dev_root->fs_info;
562 562
563 path = btrfs_alloc_path(); 563 path = btrfs_alloc_path();
564 564
565 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS); 565 swarn.scratch_buf = kmalloc(bufsize, GFP_NOFS);
566 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS); 566 swarn.msg_buf = kmalloc(bufsize, GFP_NOFS);
567 swarn.sector = (sblock->pagev[0]->physical) >> 9; 567 swarn.sector = (sblock->pagev[0]->physical) >> 9;
568 swarn.logical = sblock->pagev[0]->logical; 568 swarn.logical = sblock->pagev[0]->logical;
569 swarn.errstr = errstr; 569 swarn.errstr = errstr;
570 swarn.dev = NULL; 570 swarn.dev = NULL;
571 swarn.msg_bufsize = bufsize; 571 swarn.msg_bufsize = bufsize;
572 swarn.scratch_bufsize = bufsize; 572 swarn.scratch_bufsize = bufsize;
573 573
574 if (!path || !swarn.scratch_buf || !swarn.msg_buf) 574 if (!path || !swarn.scratch_buf || !swarn.msg_buf)
575 goto out; 575 goto out;
576 576
577 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key, 577 ret = extent_from_logical(fs_info, swarn.logical, path, &found_key,
578 &flags); 578 &flags);
579 if (ret < 0) 579 if (ret < 0)
580 goto out; 580 goto out;
581 581
582 extent_item_pos = swarn.logical - found_key.objectid; 582 extent_item_pos = swarn.logical - found_key.objectid;
583 swarn.extent_item_size = found_key.offset; 583 swarn.extent_item_size = found_key.offset;
584 584
585 eb = path->nodes[0]; 585 eb = path->nodes[0];
586 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); 586 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
587 item_size = btrfs_item_size_nr(eb, path->slots[0]); 587 item_size = btrfs_item_size_nr(eb, path->slots[0]);
588 588
589 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 589 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
590 do { 590 do {
591 ret = tree_backref_for_extent(&ptr, eb, ei, item_size, 591 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
592 &ref_root, &ref_level); 592 item_size, &ref_root,
593 &ref_level);
593 printk_in_rcu(KERN_WARNING 594 printk_in_rcu(KERN_WARNING
594 "BTRFS: %s at logical %llu on dev %s, " 595 "BTRFS: %s at logical %llu on dev %s, "
595 "sector %llu: metadata %s (level %d) in tree " 596 "sector %llu: metadata %s (level %d) in tree "
596 "%llu\n", errstr, swarn.logical, 597 "%llu\n", errstr, swarn.logical,
597 rcu_str_deref(dev->name), 598 rcu_str_deref(dev->name),
598 (unsigned long long)swarn.sector, 599 (unsigned long long)swarn.sector,
599 ref_level ? "node" : "leaf", 600 ref_level ? "node" : "leaf",
600 ret < 0 ? -1 : ref_level, 601 ret < 0 ? -1 : ref_level,
601 ret < 0 ? -1 : ref_root); 602 ret < 0 ? -1 : ref_root);
602 } while (ret != 1); 603 } while (ret != 1);
603 btrfs_release_path(path); 604 btrfs_release_path(path);
604 } else { 605 } else {
605 btrfs_release_path(path); 606 btrfs_release_path(path);
606 swarn.path = path; 607 swarn.path = path;
607 swarn.dev = dev; 608 swarn.dev = dev;
608 iterate_extent_inodes(fs_info, found_key.objectid, 609 iterate_extent_inodes(fs_info, found_key.objectid,
609 extent_item_pos, 1, 610 extent_item_pos, 1,
610 scrub_print_warning_inode, &swarn); 611 scrub_print_warning_inode, &swarn);
611 } 612 }
612 613
613 out: 614 out:
614 btrfs_free_path(path); 615 btrfs_free_path(path);
615 kfree(swarn.scratch_buf); 616 kfree(swarn.scratch_buf);
616 kfree(swarn.msg_buf); 617 kfree(swarn.msg_buf);
617 } 618 }
618 619
619 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx) 620 static int scrub_fixup_readpage(u64 inum, u64 offset, u64 root, void *fixup_ctx)
620 { 621 {
621 struct page *page = NULL; 622 struct page *page = NULL;
622 unsigned long index; 623 unsigned long index;
623 struct scrub_fixup_nodatasum *fixup = fixup_ctx; 624 struct scrub_fixup_nodatasum *fixup = fixup_ctx;
624 int ret; 625 int ret;
625 int corrected = 0; 626 int corrected = 0;
626 struct btrfs_key key; 627 struct btrfs_key key;
627 struct inode *inode = NULL; 628 struct inode *inode = NULL;
628 struct btrfs_fs_info *fs_info; 629 struct btrfs_fs_info *fs_info;
629 u64 end = offset + PAGE_SIZE - 1; 630 u64 end = offset + PAGE_SIZE - 1;
630 struct btrfs_root *local_root; 631 struct btrfs_root *local_root;
631 int srcu_index; 632 int srcu_index;
632 633
633 key.objectid = root; 634 key.objectid = root;
634 key.type = BTRFS_ROOT_ITEM_KEY; 635 key.type = BTRFS_ROOT_ITEM_KEY;
635 key.offset = (u64)-1; 636 key.offset = (u64)-1;
636 637
637 fs_info = fixup->root->fs_info; 638 fs_info = fixup->root->fs_info;
638 srcu_index = srcu_read_lock(&fs_info->subvol_srcu); 639 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
639 640
640 local_root = btrfs_read_fs_root_no_name(fs_info, &key); 641 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
641 if (IS_ERR(local_root)) { 642 if (IS_ERR(local_root)) {
642 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); 643 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
643 return PTR_ERR(local_root); 644 return PTR_ERR(local_root);
644 } 645 }
645 646
646 key.type = BTRFS_INODE_ITEM_KEY; 647 key.type = BTRFS_INODE_ITEM_KEY;
647 key.objectid = inum; 648 key.objectid = inum;
648 key.offset = 0; 649 key.offset = 0;
649 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); 650 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
650 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); 651 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
651 if (IS_ERR(inode)) 652 if (IS_ERR(inode))
652 return PTR_ERR(inode); 653 return PTR_ERR(inode);
653 654
654 index = offset >> PAGE_CACHE_SHIFT; 655 index = offset >> PAGE_CACHE_SHIFT;
655 656
656 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); 657 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
657 if (!page) { 658 if (!page) {
658 ret = -ENOMEM; 659 ret = -ENOMEM;
659 goto out; 660 goto out;
660 } 661 }
661 662
662 if (PageUptodate(page)) { 663 if (PageUptodate(page)) {
663 if (PageDirty(page)) { 664 if (PageDirty(page)) {
664 /* 665 /*
665 * we need to write the data to the defect sector. the 666 * we need to write the data to the defect sector. the
666 * data that was in that sector is not in memory, 667 * data that was in that sector is not in memory,
667 * because the page was modified. we must not write the 668 * because the page was modified. we must not write the
668 * modified page to that sector. 669 * modified page to that sector.
669 * 670 *
670 * TODO: what could be done here: wait for the delalloc 671 * TODO: what could be done here: wait for the delalloc
671 * runner to write out that page (might involve 672 * runner to write out that page (might involve
672 * COW) and see whether the sector is still 673 * COW) and see whether the sector is still
673 * referenced afterwards. 674 * referenced afterwards.
674 * 675 *
675 * For the meantime, we'll treat this error 676 * For the meantime, we'll treat this error
676 * incorrectable, although there is a chance that a 677 * incorrectable, although there is a chance that a
677 * later scrub will find the bad sector again and that 678 * later scrub will find the bad sector again and that
678 * there's no dirty page in memory, then. 679 * there's no dirty page in memory, then.
679 */ 680 */
680 ret = -EIO; 681 ret = -EIO;
681 goto out; 682 goto out;
682 } 683 }
683 fs_info = BTRFS_I(inode)->root->fs_info; 684 fs_info = BTRFS_I(inode)->root->fs_info;
684 ret = repair_io_failure(fs_info, offset, PAGE_SIZE, 685 ret = repair_io_failure(fs_info, offset, PAGE_SIZE,
685 fixup->logical, page, 686 fixup->logical, page,
686 fixup->mirror_num); 687 fixup->mirror_num);
687 unlock_page(page); 688 unlock_page(page);
688 corrected = !ret; 689 corrected = !ret;
689 } else { 690 } else {
690 /* 691 /*
691 * we need to get good data first. the general readpage path 692 * we need to get good data first. the general readpage path
692 * will call repair_io_failure for us, we just have to make 693 * will call repair_io_failure for us, we just have to make
693 * sure we read the bad mirror. 694 * sure we read the bad mirror.
694 */ 695 */
695 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, 696 ret = set_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
696 EXTENT_DAMAGED, GFP_NOFS); 697 EXTENT_DAMAGED, GFP_NOFS);
697 if (ret) { 698 if (ret) {
698 /* set_extent_bits should give proper error */ 699 /* set_extent_bits should give proper error */
699 WARN_ON(ret > 0); 700 WARN_ON(ret > 0);
700 if (ret > 0) 701 if (ret > 0)
701 ret = -EFAULT; 702 ret = -EFAULT;
702 goto out; 703 goto out;
703 } 704 }
704 705
705 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page, 706 ret = extent_read_full_page(&BTRFS_I(inode)->io_tree, page,
706 btrfs_get_extent, 707 btrfs_get_extent,
707 fixup->mirror_num); 708 fixup->mirror_num);
708 wait_on_page_locked(page); 709 wait_on_page_locked(page);
709 710
710 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset, 711 corrected = !test_range_bit(&BTRFS_I(inode)->io_tree, offset,
711 end, EXTENT_DAMAGED, 0, NULL); 712 end, EXTENT_DAMAGED, 0, NULL);
712 if (!corrected) 713 if (!corrected)
713 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end, 714 clear_extent_bits(&BTRFS_I(inode)->io_tree, offset, end,
714 EXTENT_DAMAGED, GFP_NOFS); 715 EXTENT_DAMAGED, GFP_NOFS);
715 } 716 }
716 717
717 out: 718 out:
718 if (page) 719 if (page)
719 put_page(page); 720 put_page(page);
720 if (inode) 721 if (inode)
721 iput(inode); 722 iput(inode);
722 723
723 if (ret < 0) 724 if (ret < 0)
724 return ret; 725 return ret;
725 726
726 if (ret == 0 && corrected) { 727 if (ret == 0 && corrected) {
727 /* 728 /*
728 * we only need to call readpage for one of the inodes belonging 729 * we only need to call readpage for one of the inodes belonging
729 * to this extent. so make iterate_extent_inodes stop 730 * to this extent. so make iterate_extent_inodes stop
730 */ 731 */
731 return 1; 732 return 1;
732 } 733 }
733 734
734 return -EIO; 735 return -EIO;
735 } 736 }
736 737
737 static void scrub_fixup_nodatasum(struct btrfs_work *work) 738 static void scrub_fixup_nodatasum(struct btrfs_work *work)
738 { 739 {
739 int ret; 740 int ret;
740 struct scrub_fixup_nodatasum *fixup; 741 struct scrub_fixup_nodatasum *fixup;
741 struct scrub_ctx *sctx; 742 struct scrub_ctx *sctx;
742 struct btrfs_trans_handle *trans = NULL; 743 struct btrfs_trans_handle *trans = NULL;
743 struct btrfs_path *path; 744 struct btrfs_path *path;
744 int uncorrectable = 0; 745 int uncorrectable = 0;
745 746
746 fixup = container_of(work, struct scrub_fixup_nodatasum, work); 747 fixup = container_of(work, struct scrub_fixup_nodatasum, work);
747 sctx = fixup->sctx; 748 sctx = fixup->sctx;
748 749
749 path = btrfs_alloc_path(); 750 path = btrfs_alloc_path();
750 if (!path) { 751 if (!path) {
751 spin_lock(&sctx->stat_lock); 752 spin_lock(&sctx->stat_lock);
752 ++sctx->stat.malloc_errors; 753 ++sctx->stat.malloc_errors;
753 spin_unlock(&sctx->stat_lock); 754 spin_unlock(&sctx->stat_lock);
754 uncorrectable = 1; 755 uncorrectable = 1;
755 goto out; 756 goto out;
756 } 757 }
757 758
758 trans = btrfs_join_transaction(fixup->root); 759 trans = btrfs_join_transaction(fixup->root);
759 if (IS_ERR(trans)) { 760 if (IS_ERR(trans)) {
760 uncorrectable = 1; 761 uncorrectable = 1;
761 goto out; 762 goto out;
762 } 763 }
763 764
764 /* 765 /*
765 * the idea is to trigger a regular read through the standard path. we 766 * the idea is to trigger a regular read through the standard path. we
766 * read a page from the (failed) logical address by specifying the 767 * read a page from the (failed) logical address by specifying the
767 * corresponding copynum of the failed sector. thus, that readpage is 768 * corresponding copynum of the failed sector. thus, that readpage is
768 * expected to fail. 769 * expected to fail.
769 * that is the point where on-the-fly error correction will kick in 770 * that is the point where on-the-fly error correction will kick in
770 * (once it's finished) and rewrite the failed sector if a good copy 771 * (once it's finished) and rewrite the failed sector if a good copy
771 * can be found. 772 * can be found.
772 */ 773 */
773 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info, 774 ret = iterate_inodes_from_logical(fixup->logical, fixup->root->fs_info,
774 path, scrub_fixup_readpage, 775 path, scrub_fixup_readpage,
775 fixup); 776 fixup);
776 if (ret < 0) { 777 if (ret < 0) {
777 uncorrectable = 1; 778 uncorrectable = 1;
778 goto out; 779 goto out;
779 } 780 }
780 WARN_ON(ret != 1); 781 WARN_ON(ret != 1);
781 782
782 spin_lock(&sctx->stat_lock); 783 spin_lock(&sctx->stat_lock);
783 ++sctx->stat.corrected_errors; 784 ++sctx->stat.corrected_errors;
784 spin_unlock(&sctx->stat_lock); 785 spin_unlock(&sctx->stat_lock);
785 786
786 out: 787 out:
787 if (trans && !IS_ERR(trans)) 788 if (trans && !IS_ERR(trans))
788 btrfs_end_transaction(trans, fixup->root); 789 btrfs_end_transaction(trans, fixup->root);
789 if (uncorrectable) { 790 if (uncorrectable) {
790 spin_lock(&sctx->stat_lock); 791 spin_lock(&sctx->stat_lock);
791 ++sctx->stat.uncorrectable_errors; 792 ++sctx->stat.uncorrectable_errors;
792 spin_unlock(&sctx->stat_lock); 793 spin_unlock(&sctx->stat_lock);
793 btrfs_dev_replace_stats_inc( 794 btrfs_dev_replace_stats_inc(
794 &sctx->dev_root->fs_info->dev_replace. 795 &sctx->dev_root->fs_info->dev_replace.
795 num_uncorrectable_read_errors); 796 num_uncorrectable_read_errors);
796 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " 797 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
797 "unable to fixup (nodatasum) error at logical %llu on dev %s\n", 798 "unable to fixup (nodatasum) error at logical %llu on dev %s\n",
798 fixup->logical, rcu_str_deref(fixup->dev->name)); 799 fixup->logical, rcu_str_deref(fixup->dev->name));
799 } 800 }
800 801
801 btrfs_free_path(path); 802 btrfs_free_path(path);
802 kfree(fixup); 803 kfree(fixup);
803 804
804 scrub_pending_trans_workers_dec(sctx); 805 scrub_pending_trans_workers_dec(sctx);
805 } 806 }
806 807
807 /* 808 /*
808 * scrub_handle_errored_block gets called when either verification of the 809 * scrub_handle_errored_block gets called when either verification of the
809 * pages failed or the bio failed to read, e.g. with EIO. In the latter 810 * pages failed or the bio failed to read, e.g. with EIO. In the latter
810 * case, this function handles all pages in the bio, even though only one 811 * case, this function handles all pages in the bio, even though only one
811 * may be bad. 812 * may be bad.
812 * The goal of this function is to repair the errored block by using the 813 * The goal of this function is to repair the errored block by using the
813 * contents of one of the mirrors. 814 * contents of one of the mirrors.
814 */ 815 */
815 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check) 816 static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
816 { 817 {
817 struct scrub_ctx *sctx = sblock_to_check->sctx; 818 struct scrub_ctx *sctx = sblock_to_check->sctx;
818 struct btrfs_device *dev; 819 struct btrfs_device *dev;
819 struct btrfs_fs_info *fs_info; 820 struct btrfs_fs_info *fs_info;
820 u64 length; 821 u64 length;
821 u64 logical; 822 u64 logical;
822 u64 generation; 823 u64 generation;
823 unsigned int failed_mirror_index; 824 unsigned int failed_mirror_index;
824 unsigned int is_metadata; 825 unsigned int is_metadata;
825 unsigned int have_csum; 826 unsigned int have_csum;
826 u8 *csum; 827 u8 *csum;
827 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */ 828 struct scrub_block *sblocks_for_recheck; /* holds one for each mirror */
828 struct scrub_block *sblock_bad; 829 struct scrub_block *sblock_bad;
829 int ret; 830 int ret;
830 int mirror_index; 831 int mirror_index;
831 int page_num; 832 int page_num;
832 int success; 833 int success;
833 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL, 834 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
834 DEFAULT_RATELIMIT_BURST); 835 DEFAULT_RATELIMIT_BURST);
835 836
836 BUG_ON(sblock_to_check->page_count < 1); 837 BUG_ON(sblock_to_check->page_count < 1);
837 fs_info = sctx->dev_root->fs_info; 838 fs_info = sctx->dev_root->fs_info;
838 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) { 839 if (sblock_to_check->pagev[0]->flags & BTRFS_EXTENT_FLAG_SUPER) {
839 /* 840 /*
840 * if we find an error in a super block, we just report it. 841 * if we find an error in a super block, we just report it.
841 * They will get written with the next transaction commit 842 * They will get written with the next transaction commit
842 * anyway 843 * anyway
843 */ 844 */
844 spin_lock(&sctx->stat_lock); 845 spin_lock(&sctx->stat_lock);
845 ++sctx->stat.super_errors; 846 ++sctx->stat.super_errors;
846 spin_unlock(&sctx->stat_lock); 847 spin_unlock(&sctx->stat_lock);
847 return 0; 848 return 0;
848 } 849 }
849 length = sblock_to_check->page_count * PAGE_SIZE; 850 length = sblock_to_check->page_count * PAGE_SIZE;
850 logical = sblock_to_check->pagev[0]->logical; 851 logical = sblock_to_check->pagev[0]->logical;
851 generation = sblock_to_check->pagev[0]->generation; 852 generation = sblock_to_check->pagev[0]->generation;
852 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1); 853 BUG_ON(sblock_to_check->pagev[0]->mirror_num < 1);
853 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1; 854 failed_mirror_index = sblock_to_check->pagev[0]->mirror_num - 1;
854 is_metadata = !(sblock_to_check->pagev[0]->flags & 855 is_metadata = !(sblock_to_check->pagev[0]->flags &
855 BTRFS_EXTENT_FLAG_DATA); 856 BTRFS_EXTENT_FLAG_DATA);
856 have_csum = sblock_to_check->pagev[0]->have_csum; 857 have_csum = sblock_to_check->pagev[0]->have_csum;
857 csum = sblock_to_check->pagev[0]->csum; 858 csum = sblock_to_check->pagev[0]->csum;
858 dev = sblock_to_check->pagev[0]->dev; 859 dev = sblock_to_check->pagev[0]->dev;
859 860
860 if (sctx->is_dev_replace && !is_metadata && !have_csum) { 861 if (sctx->is_dev_replace && !is_metadata && !have_csum) {
861 sblocks_for_recheck = NULL; 862 sblocks_for_recheck = NULL;
862 goto nodatasum_case; 863 goto nodatasum_case;
863 } 864 }
864 865
865 /* 866 /*
866 * read all mirrors one after the other. This includes to 867 * read all mirrors one after the other. This includes to
867 * re-read the extent or metadata block that failed (that was 868 * re-read the extent or metadata block that failed (that was
868 * the cause that this fixup code is called) another time, 869 * the cause that this fixup code is called) another time,
869 * page by page this time in order to know which pages 870 * page by page this time in order to know which pages
870 * caused I/O errors and which ones are good (for all mirrors). 871 * caused I/O errors and which ones are good (for all mirrors).
871 * It is the goal to handle the situation when more than one 872 * It is the goal to handle the situation when more than one
872 * mirror contains I/O errors, but the errors do not 873 * mirror contains I/O errors, but the errors do not
873 * overlap, i.e. the data can be repaired by selecting the 874 * overlap, i.e. the data can be repaired by selecting the
874 * pages from those mirrors without I/O error on the 875 * pages from those mirrors without I/O error on the
875 * particular pages. One example (with blocks >= 2 * PAGE_SIZE) 876 * particular pages. One example (with blocks >= 2 * PAGE_SIZE)
876 * would be that mirror #1 has an I/O error on the first page, 877 * would be that mirror #1 has an I/O error on the first page,
877 * the second page is good, and mirror #2 has an I/O error on 878 * the second page is good, and mirror #2 has an I/O error on
878 * the second page, but the first page is good. 879 * the second page, but the first page is good.
879 * Then the first page of the first mirror can be repaired by 880 * Then the first page of the first mirror can be repaired by
880 * taking the first page of the second mirror, and the 881 * taking the first page of the second mirror, and the
881 * second page of the second mirror can be repaired by 882 * second page of the second mirror can be repaired by
882 * copying the contents of the 2nd page of the 1st mirror. 883 * copying the contents of the 2nd page of the 1st mirror.
883 * One more note: if the pages of one mirror contain I/O 884 * One more note: if the pages of one mirror contain I/O
884 * errors, the checksum cannot be verified. In order to get 885 * errors, the checksum cannot be verified. In order to get
885 * the best data for repairing, the first attempt is to find 886 * the best data for repairing, the first attempt is to find
886 * a mirror without I/O errors and with a validated checksum. 887 * a mirror without I/O errors and with a validated checksum.
887 * Only if this is not possible, the pages are picked from 888 * Only if this is not possible, the pages are picked from
888 * mirrors with I/O errors without considering the checksum. 889 * mirrors with I/O errors without considering the checksum.
889 * If the latter is the case, at the end, the checksum of the 890 * If the latter is the case, at the end, the checksum of the
890 * repaired area is verified in order to correctly maintain 891 * repaired area is verified in order to correctly maintain
891 * the statistics. 892 * the statistics.
892 */ 893 */
893 894
894 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS * 895 sblocks_for_recheck = kzalloc(BTRFS_MAX_MIRRORS *
895 sizeof(*sblocks_for_recheck), 896 sizeof(*sblocks_for_recheck),
896 GFP_NOFS); 897 GFP_NOFS);
897 if (!sblocks_for_recheck) { 898 if (!sblocks_for_recheck) {
898 spin_lock(&sctx->stat_lock); 899 spin_lock(&sctx->stat_lock);
899 sctx->stat.malloc_errors++; 900 sctx->stat.malloc_errors++;
900 sctx->stat.read_errors++; 901 sctx->stat.read_errors++;
901 sctx->stat.uncorrectable_errors++; 902 sctx->stat.uncorrectable_errors++;
902 spin_unlock(&sctx->stat_lock); 903 spin_unlock(&sctx->stat_lock);
903 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); 904 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
904 goto out; 905 goto out;
905 } 906 }
906 907
907 /* setup the context, map the logical blocks and alloc the pages */ 908 /* setup the context, map the logical blocks and alloc the pages */
908 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length, 909 ret = scrub_setup_recheck_block(sctx, fs_info, sblock_to_check, length,
909 logical, sblocks_for_recheck); 910 logical, sblocks_for_recheck);
910 if (ret) { 911 if (ret) {
911 spin_lock(&sctx->stat_lock); 912 spin_lock(&sctx->stat_lock);
912 sctx->stat.read_errors++; 913 sctx->stat.read_errors++;
913 sctx->stat.uncorrectable_errors++; 914 sctx->stat.uncorrectable_errors++;
914 spin_unlock(&sctx->stat_lock); 915 spin_unlock(&sctx->stat_lock);
915 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); 916 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
916 goto out; 917 goto out;
917 } 918 }
918 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS); 919 BUG_ON(failed_mirror_index >= BTRFS_MAX_MIRRORS);
919 sblock_bad = sblocks_for_recheck + failed_mirror_index; 920 sblock_bad = sblocks_for_recheck + failed_mirror_index;
920 921
921 /* build and submit the bios for the failed mirror, check checksums */ 922 /* build and submit the bios for the failed mirror, check checksums */
922 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum, 923 scrub_recheck_block(fs_info, sblock_bad, is_metadata, have_csum,
923 csum, generation, sctx->csum_size); 924 csum, generation, sctx->csum_size);
924 925
925 if (!sblock_bad->header_error && !sblock_bad->checksum_error && 926 if (!sblock_bad->header_error && !sblock_bad->checksum_error &&
926 sblock_bad->no_io_error_seen) { 927 sblock_bad->no_io_error_seen) {
927 /* 928 /*
928 * the error disappeared after reading page by page, or 929 * the error disappeared after reading page by page, or
929 * the area was part of a huge bio and other parts of the 930 * the area was part of a huge bio and other parts of the
930 * bio caused I/O errors, or the block layer merged several 931 * bio caused I/O errors, or the block layer merged several
931 * read requests into one and the error is caused by a 932 * read requests into one and the error is caused by a
932 * different bio (usually one of the two latter cases is 933 * different bio (usually one of the two latter cases is
933 * the cause) 934 * the cause)
934 */ 935 */
935 spin_lock(&sctx->stat_lock); 936 spin_lock(&sctx->stat_lock);
936 sctx->stat.unverified_errors++; 937 sctx->stat.unverified_errors++;
937 spin_unlock(&sctx->stat_lock); 938 spin_unlock(&sctx->stat_lock);
938 939
939 if (sctx->is_dev_replace) 940 if (sctx->is_dev_replace)
940 scrub_write_block_to_dev_replace(sblock_bad); 941 scrub_write_block_to_dev_replace(sblock_bad);
941 goto out; 942 goto out;
942 } 943 }
943 944
944 if (!sblock_bad->no_io_error_seen) { 945 if (!sblock_bad->no_io_error_seen) {
945 spin_lock(&sctx->stat_lock); 946 spin_lock(&sctx->stat_lock);
946 sctx->stat.read_errors++; 947 sctx->stat.read_errors++;
947 spin_unlock(&sctx->stat_lock); 948 spin_unlock(&sctx->stat_lock);
948 if (__ratelimit(&_rs)) 949 if (__ratelimit(&_rs))
949 scrub_print_warning("i/o error", sblock_to_check); 950 scrub_print_warning("i/o error", sblock_to_check);
950 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS); 951 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_READ_ERRS);
951 } else if (sblock_bad->checksum_error) { 952 } else if (sblock_bad->checksum_error) {
952 spin_lock(&sctx->stat_lock); 953 spin_lock(&sctx->stat_lock);
953 sctx->stat.csum_errors++; 954 sctx->stat.csum_errors++;
954 spin_unlock(&sctx->stat_lock); 955 spin_unlock(&sctx->stat_lock);
955 if (__ratelimit(&_rs)) 956 if (__ratelimit(&_rs))
956 scrub_print_warning("checksum error", sblock_to_check); 957 scrub_print_warning("checksum error", sblock_to_check);
957 btrfs_dev_stat_inc_and_print(dev, 958 btrfs_dev_stat_inc_and_print(dev,
958 BTRFS_DEV_STAT_CORRUPTION_ERRS); 959 BTRFS_DEV_STAT_CORRUPTION_ERRS);
959 } else if (sblock_bad->header_error) { 960 } else if (sblock_bad->header_error) {
960 spin_lock(&sctx->stat_lock); 961 spin_lock(&sctx->stat_lock);
961 sctx->stat.verify_errors++; 962 sctx->stat.verify_errors++;
962 spin_unlock(&sctx->stat_lock); 963 spin_unlock(&sctx->stat_lock);
963 if (__ratelimit(&_rs)) 964 if (__ratelimit(&_rs))
964 scrub_print_warning("checksum/header error", 965 scrub_print_warning("checksum/header error",
965 sblock_to_check); 966 sblock_to_check);
966 if (sblock_bad->generation_error) 967 if (sblock_bad->generation_error)
967 btrfs_dev_stat_inc_and_print(dev, 968 btrfs_dev_stat_inc_and_print(dev,
968 BTRFS_DEV_STAT_GENERATION_ERRS); 969 BTRFS_DEV_STAT_GENERATION_ERRS);
969 else 970 else
970 btrfs_dev_stat_inc_and_print(dev, 971 btrfs_dev_stat_inc_and_print(dev,
971 BTRFS_DEV_STAT_CORRUPTION_ERRS); 972 BTRFS_DEV_STAT_CORRUPTION_ERRS);
972 } 973 }
973 974
974 if (sctx->readonly) { 975 if (sctx->readonly) {
975 ASSERT(!sctx->is_dev_replace); 976 ASSERT(!sctx->is_dev_replace);
976 goto out; 977 goto out;
977 } 978 }
978 979
979 if (!is_metadata && !have_csum) { 980 if (!is_metadata && !have_csum) {
980 struct scrub_fixup_nodatasum *fixup_nodatasum; 981 struct scrub_fixup_nodatasum *fixup_nodatasum;
981 982
982 nodatasum_case: 983 nodatasum_case:
983 WARN_ON(sctx->is_dev_replace); 984 WARN_ON(sctx->is_dev_replace);
984 985
985 /* 986 /*
986 * !is_metadata and !have_csum, this means that the data 987 * !is_metadata and !have_csum, this means that the data
987 * might not be COW'ed, that it might be modified 988 * might not be COW'ed, that it might be modified
988 * concurrently. The general strategy to work on the 989 * concurrently. The general strategy to work on the
989 * commit root does not help in the case when COW is not 990 * commit root does not help in the case when COW is not
990 * used. 991 * used.
991 */ 992 */
992 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS); 993 fixup_nodatasum = kzalloc(sizeof(*fixup_nodatasum), GFP_NOFS);
993 if (!fixup_nodatasum) 994 if (!fixup_nodatasum)
994 goto did_not_correct_error; 995 goto did_not_correct_error;
995 fixup_nodatasum->sctx = sctx; 996 fixup_nodatasum->sctx = sctx;
996 fixup_nodatasum->dev = dev; 997 fixup_nodatasum->dev = dev;
997 fixup_nodatasum->logical = logical; 998 fixup_nodatasum->logical = logical;
998 fixup_nodatasum->root = fs_info->extent_root; 999 fixup_nodatasum->root = fs_info->extent_root;
999 fixup_nodatasum->mirror_num = failed_mirror_index + 1; 1000 fixup_nodatasum->mirror_num = failed_mirror_index + 1;
1000 scrub_pending_trans_workers_inc(sctx); 1001 scrub_pending_trans_workers_inc(sctx);
1001 btrfs_init_work(&fixup_nodatasum->work, scrub_fixup_nodatasum, 1002 btrfs_init_work(&fixup_nodatasum->work, scrub_fixup_nodatasum,
1002 NULL, NULL); 1003 NULL, NULL);
1003 btrfs_queue_work(fs_info->scrub_workers, 1004 btrfs_queue_work(fs_info->scrub_workers,
1004 &fixup_nodatasum->work); 1005 &fixup_nodatasum->work);
1005 goto out; 1006 goto out;
1006 } 1007 }
1007 1008
1008 /* 1009 /*
1009 * now build and submit the bios for the other mirrors, check 1010 * now build and submit the bios for the other mirrors, check
1010 * checksums. 1011 * checksums.
1011 * First try to pick the mirror which is completely without I/O 1012 * First try to pick the mirror which is completely without I/O
1012 * errors and also does not have a checksum error. 1013 * errors and also does not have a checksum error.
1013 * If one is found, and if a checksum is present, the full block 1014 * If one is found, and if a checksum is present, the full block
1014 * that is known to contain an error is rewritten. Afterwards 1015 * that is known to contain an error is rewritten. Afterwards
1015 * the block is known to be corrected. 1016 * the block is known to be corrected.
1016 * If a mirror is found which is completely correct, and no 1017 * If a mirror is found which is completely correct, and no
1017 * checksum is present, only those pages are rewritten that had 1018 * checksum is present, only those pages are rewritten that had
1018 * an I/O error in the block to be repaired, since it cannot be 1019 * an I/O error in the block to be repaired, since it cannot be
1019 * determined, which copy of the other pages is better (and it 1020 * determined, which copy of the other pages is better (and it
1020 * could happen otherwise that a correct page would be 1021 * could happen otherwise that a correct page would be
1021 * overwritten by a bad one). 1022 * overwritten by a bad one).
1022 */ 1023 */
1023 for (mirror_index = 0; 1024 for (mirror_index = 0;
1024 mirror_index < BTRFS_MAX_MIRRORS && 1025 mirror_index < BTRFS_MAX_MIRRORS &&
1025 sblocks_for_recheck[mirror_index].page_count > 0; 1026 sblocks_for_recheck[mirror_index].page_count > 0;
1026 mirror_index++) { 1027 mirror_index++) {
1027 struct scrub_block *sblock_other; 1028 struct scrub_block *sblock_other;
1028 1029
1029 if (mirror_index == failed_mirror_index) 1030 if (mirror_index == failed_mirror_index)
1030 continue; 1031 continue;
1031 sblock_other = sblocks_for_recheck + mirror_index; 1032 sblock_other = sblocks_for_recheck + mirror_index;
1032 1033
1033 /* build and submit the bios, check checksums */ 1034 /* build and submit the bios, check checksums */
1034 scrub_recheck_block(fs_info, sblock_other, is_metadata, 1035 scrub_recheck_block(fs_info, sblock_other, is_metadata,
1035 have_csum, csum, generation, 1036 have_csum, csum, generation,
1036 sctx->csum_size); 1037 sctx->csum_size);
1037 1038
1038 if (!sblock_other->header_error && 1039 if (!sblock_other->header_error &&
1039 !sblock_other->checksum_error && 1040 !sblock_other->checksum_error &&
1040 sblock_other->no_io_error_seen) { 1041 sblock_other->no_io_error_seen) {
1041 if (sctx->is_dev_replace) { 1042 if (sctx->is_dev_replace) {
1042 scrub_write_block_to_dev_replace(sblock_other); 1043 scrub_write_block_to_dev_replace(sblock_other);
1043 } else { 1044 } else {
1044 int force_write = is_metadata || have_csum; 1045 int force_write = is_metadata || have_csum;
1045 1046
1046 ret = scrub_repair_block_from_good_copy( 1047 ret = scrub_repair_block_from_good_copy(
1047 sblock_bad, sblock_other, 1048 sblock_bad, sblock_other,
1048 force_write); 1049 force_write);
1049 } 1050 }
1050 if (0 == ret) 1051 if (0 == ret)
1051 goto corrected_error; 1052 goto corrected_error;
1052 } 1053 }
1053 } 1054 }
1054 1055
1055 /* 1056 /*
1056 * for dev_replace, pick good pages and write to the target device. 1057 * for dev_replace, pick good pages and write to the target device.
1057 */ 1058 */
1058 if (sctx->is_dev_replace) { 1059 if (sctx->is_dev_replace) {
1059 success = 1; 1060 success = 1;
1060 for (page_num = 0; page_num < sblock_bad->page_count; 1061 for (page_num = 0; page_num < sblock_bad->page_count;
1061 page_num++) { 1062 page_num++) {
1062 int sub_success; 1063 int sub_success;
1063 1064
1064 sub_success = 0; 1065 sub_success = 0;
1065 for (mirror_index = 0; 1066 for (mirror_index = 0;
1066 mirror_index < BTRFS_MAX_MIRRORS && 1067 mirror_index < BTRFS_MAX_MIRRORS &&
1067 sblocks_for_recheck[mirror_index].page_count > 0; 1068 sblocks_for_recheck[mirror_index].page_count > 0;
1068 mirror_index++) { 1069 mirror_index++) {
1069 struct scrub_block *sblock_other = 1070 struct scrub_block *sblock_other =
1070 sblocks_for_recheck + mirror_index; 1071 sblocks_for_recheck + mirror_index;
1071 struct scrub_page *page_other = 1072 struct scrub_page *page_other =
1072 sblock_other->pagev[page_num]; 1073 sblock_other->pagev[page_num];
1073 1074
1074 if (!page_other->io_error) { 1075 if (!page_other->io_error) {
1075 ret = scrub_write_page_to_dev_replace( 1076 ret = scrub_write_page_to_dev_replace(
1076 sblock_other, page_num); 1077 sblock_other, page_num);
1077 if (ret == 0) { 1078 if (ret == 0) {
1078 /* succeeded for this page */ 1079 /* succeeded for this page */
1079 sub_success = 1; 1080 sub_success = 1;
1080 break; 1081 break;
1081 } else { 1082 } else {
1082 btrfs_dev_replace_stats_inc( 1083 btrfs_dev_replace_stats_inc(
1083 &sctx->dev_root-> 1084 &sctx->dev_root->
1084 fs_info->dev_replace. 1085 fs_info->dev_replace.
1085 num_write_errors); 1086 num_write_errors);
1086 } 1087 }
1087 } 1088 }
1088 } 1089 }
1089 1090
1090 if (!sub_success) { 1091 if (!sub_success) {
1091 /* 1092 /*
1092 * did not find a mirror to fetch the page 1093 * did not find a mirror to fetch the page
1093 * from. scrub_write_page_to_dev_replace() 1094 * from. scrub_write_page_to_dev_replace()
1094 * handles this case (page->io_error), by 1095 * handles this case (page->io_error), by
1095 * filling the block with zeros before 1096 * filling the block with zeros before
1096 * submitting the write request 1097 * submitting the write request
1097 */ 1098 */
1098 success = 0; 1099 success = 0;
1099 ret = scrub_write_page_to_dev_replace( 1100 ret = scrub_write_page_to_dev_replace(
1100 sblock_bad, page_num); 1101 sblock_bad, page_num);
1101 if (ret) 1102 if (ret)
1102 btrfs_dev_replace_stats_inc( 1103 btrfs_dev_replace_stats_inc(
1103 &sctx->dev_root->fs_info-> 1104 &sctx->dev_root->fs_info->
1104 dev_replace.num_write_errors); 1105 dev_replace.num_write_errors);
1105 } 1106 }
1106 } 1107 }
1107 1108
1108 goto out; 1109 goto out;
1109 } 1110 }
1110 1111
1111 /* 1112 /*
1112 * for regular scrub, repair those pages that are errored. 1113 * for regular scrub, repair those pages that are errored.
1113 * In case of I/O errors in the area that is supposed to be 1114 * In case of I/O errors in the area that is supposed to be
1114 * repaired, continue by picking good copies of those pages. 1115 * repaired, continue by picking good copies of those pages.
1115 * Select the good pages from mirrors to rewrite bad pages from 1116 * Select the good pages from mirrors to rewrite bad pages from
1116 * the area to fix. Afterwards verify the checksum of the block 1117 * the area to fix. Afterwards verify the checksum of the block
1117 * that is supposed to be repaired. This verification step is 1118 * that is supposed to be repaired. This verification step is
1118 * only done for the purpose of statistic counting and for the 1119 * only done for the purpose of statistic counting and for the
1119 * final scrub report, whether errors remain. 1120 * final scrub report, whether errors remain.
1120 * A perfect algorithm could make use of the checksum and try 1121 * A perfect algorithm could make use of the checksum and try
1121 * all possible combinations of pages from the different mirrors 1122 * all possible combinations of pages from the different mirrors
1122 * until the checksum verification succeeds. For example, when 1123 * until the checksum verification succeeds. For example, when
1123 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page 1124 * the 2nd page of mirror #1 faces I/O errors, and the 2nd page
1124 * of mirror #2 is readable but the final checksum test fails, 1125 * of mirror #2 is readable but the final checksum test fails,
1125 * then the 2nd page of mirror #3 could be tried, whether now 1126 * then the 2nd page of mirror #3 could be tried, whether now
1126 * the final checksum succeedes. But this would be a rare 1127 * the final checksum succeedes. But this would be a rare
1127 * exception and is therefore not implemented. At least it is 1128 * exception and is therefore not implemented. At least it is
1128 * avoided that the good copy is overwritten. 1129 * avoided that the good copy is overwritten.
1129 * A more useful improvement would be to pick the sectors 1130 * A more useful improvement would be to pick the sectors
1130 * without I/O error based on sector sizes (512 bytes on legacy 1131 * without I/O error based on sector sizes (512 bytes on legacy
1131 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one 1132 * disks) instead of on PAGE_SIZE. Then maybe 512 byte of one
1132 * mirror could be repaired by taking 512 byte of a different 1133 * mirror could be repaired by taking 512 byte of a different
1133 * mirror, even if other 512 byte sectors in the same PAGE_SIZE 1134 * mirror, even if other 512 byte sectors in the same PAGE_SIZE
1134 * area are unreadable. 1135 * area are unreadable.
1135 */ 1136 */
1136 1137
1137 /* can only fix I/O errors from here on */ 1138 /* can only fix I/O errors from here on */
1138 if (sblock_bad->no_io_error_seen) 1139 if (sblock_bad->no_io_error_seen)
1139 goto did_not_correct_error; 1140 goto did_not_correct_error;
1140 1141
1141 success = 1; 1142 success = 1;
1142 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { 1143 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1143 struct scrub_page *page_bad = sblock_bad->pagev[page_num]; 1144 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1144 1145
1145 if (!page_bad->io_error) 1146 if (!page_bad->io_error)
1146 continue; 1147 continue;
1147 1148
1148 for (mirror_index = 0; 1149 for (mirror_index = 0;
1149 mirror_index < BTRFS_MAX_MIRRORS && 1150 mirror_index < BTRFS_MAX_MIRRORS &&
1150 sblocks_for_recheck[mirror_index].page_count > 0; 1151 sblocks_for_recheck[mirror_index].page_count > 0;
1151 mirror_index++) { 1152 mirror_index++) {
1152 struct scrub_block *sblock_other = sblocks_for_recheck + 1153 struct scrub_block *sblock_other = sblocks_for_recheck +
1153 mirror_index; 1154 mirror_index;
1154 struct scrub_page *page_other = sblock_other->pagev[ 1155 struct scrub_page *page_other = sblock_other->pagev[
1155 page_num]; 1156 page_num];
1156 1157
1157 if (!page_other->io_error) { 1158 if (!page_other->io_error) {
1158 ret = scrub_repair_page_from_good_copy( 1159 ret = scrub_repair_page_from_good_copy(
1159 sblock_bad, sblock_other, page_num, 0); 1160 sblock_bad, sblock_other, page_num, 0);
1160 if (0 == ret) { 1161 if (0 == ret) {
1161 page_bad->io_error = 0; 1162 page_bad->io_error = 0;
1162 break; /* succeeded for this page */ 1163 break; /* succeeded for this page */
1163 } 1164 }
1164 } 1165 }
1165 } 1166 }
1166 1167
1167 if (page_bad->io_error) { 1168 if (page_bad->io_error) {
1168 /* did not find a mirror to copy the page from */ 1169 /* did not find a mirror to copy the page from */
1169 success = 0; 1170 success = 0;
1170 } 1171 }
1171 } 1172 }
1172 1173
1173 if (success) { 1174 if (success) {
1174 if (is_metadata || have_csum) { 1175 if (is_metadata || have_csum) {
1175 /* 1176 /*
1176 * need to verify the checksum now that all 1177 * need to verify the checksum now that all
1177 * sectors on disk are repaired (the write 1178 * sectors on disk are repaired (the write
1178 * request for data to be repaired is on its way). 1179 * request for data to be repaired is on its way).
1179 * Just be lazy and use scrub_recheck_block() 1180 * Just be lazy and use scrub_recheck_block()
1180 * which re-reads the data before the checksum 1181 * which re-reads the data before the checksum
1181 * is verified, but most likely the data comes out 1182 * is verified, but most likely the data comes out
1182 * of the page cache. 1183 * of the page cache.
1183 */ 1184 */
1184 scrub_recheck_block(fs_info, sblock_bad, 1185 scrub_recheck_block(fs_info, sblock_bad,
1185 is_metadata, have_csum, csum, 1186 is_metadata, have_csum, csum,
1186 generation, sctx->csum_size); 1187 generation, sctx->csum_size);
1187 if (!sblock_bad->header_error && 1188 if (!sblock_bad->header_error &&
1188 !sblock_bad->checksum_error && 1189 !sblock_bad->checksum_error &&
1189 sblock_bad->no_io_error_seen) 1190 sblock_bad->no_io_error_seen)
1190 goto corrected_error; 1191 goto corrected_error;
1191 else 1192 else
1192 goto did_not_correct_error; 1193 goto did_not_correct_error;
1193 } else { 1194 } else {
1194 corrected_error: 1195 corrected_error:
1195 spin_lock(&sctx->stat_lock); 1196 spin_lock(&sctx->stat_lock);
1196 sctx->stat.corrected_errors++; 1197 sctx->stat.corrected_errors++;
1197 spin_unlock(&sctx->stat_lock); 1198 spin_unlock(&sctx->stat_lock);
1198 printk_ratelimited_in_rcu(KERN_ERR 1199 printk_ratelimited_in_rcu(KERN_ERR
1199 "BTRFS: fixed up error at logical %llu on dev %s\n", 1200 "BTRFS: fixed up error at logical %llu on dev %s\n",
1200 logical, rcu_str_deref(dev->name)); 1201 logical, rcu_str_deref(dev->name));
1201 } 1202 }
1202 } else { 1203 } else {
1203 did_not_correct_error: 1204 did_not_correct_error:
1204 spin_lock(&sctx->stat_lock); 1205 spin_lock(&sctx->stat_lock);
1205 sctx->stat.uncorrectable_errors++; 1206 sctx->stat.uncorrectable_errors++;
1206 spin_unlock(&sctx->stat_lock); 1207 spin_unlock(&sctx->stat_lock);
1207 printk_ratelimited_in_rcu(KERN_ERR 1208 printk_ratelimited_in_rcu(KERN_ERR
1208 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n", 1209 "BTRFS: unable to fixup (regular) error at logical %llu on dev %s\n",
1209 logical, rcu_str_deref(dev->name)); 1210 logical, rcu_str_deref(dev->name));
1210 } 1211 }
1211 1212
1212 out: 1213 out:
1213 if (sblocks_for_recheck) { 1214 if (sblocks_for_recheck) {
1214 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS; 1215 for (mirror_index = 0; mirror_index < BTRFS_MAX_MIRRORS;
1215 mirror_index++) { 1216 mirror_index++) {
1216 struct scrub_block *sblock = sblocks_for_recheck + 1217 struct scrub_block *sblock = sblocks_for_recheck +
1217 mirror_index; 1218 mirror_index;
1218 int page_index; 1219 int page_index;
1219 1220
1220 for (page_index = 0; page_index < sblock->page_count; 1221 for (page_index = 0; page_index < sblock->page_count;
1221 page_index++) { 1222 page_index++) {
1222 sblock->pagev[page_index]->sblock = NULL; 1223 sblock->pagev[page_index]->sblock = NULL;
1223 scrub_page_put(sblock->pagev[page_index]); 1224 scrub_page_put(sblock->pagev[page_index]);
1224 } 1225 }
1225 } 1226 }
1226 kfree(sblocks_for_recheck); 1227 kfree(sblocks_for_recheck);
1227 } 1228 }
1228 1229
1229 return 0; 1230 return 0;
1230 } 1231 }
1231 1232
1232 static int scrub_setup_recheck_block(struct scrub_ctx *sctx, 1233 static int scrub_setup_recheck_block(struct scrub_ctx *sctx,
1233 struct btrfs_fs_info *fs_info, 1234 struct btrfs_fs_info *fs_info,
1234 struct scrub_block *original_sblock, 1235 struct scrub_block *original_sblock,
1235 u64 length, u64 logical, 1236 u64 length, u64 logical,
1236 struct scrub_block *sblocks_for_recheck) 1237 struct scrub_block *sblocks_for_recheck)
1237 { 1238 {
1238 int page_index; 1239 int page_index;
1239 int mirror_index; 1240 int mirror_index;
1240 int ret; 1241 int ret;
1241 1242
1242 /* 1243 /*
1243 * note: the two members ref_count and outstanding_pages 1244 * note: the two members ref_count and outstanding_pages
1244 * are not used (and not set) in the blocks that are used for 1245 * are not used (and not set) in the blocks that are used for
1245 * the recheck procedure 1246 * the recheck procedure
1246 */ 1247 */
1247 1248
1248 page_index = 0; 1249 page_index = 0;
1249 while (length > 0) { 1250 while (length > 0) {
1250 u64 sublen = min_t(u64, length, PAGE_SIZE); 1251 u64 sublen = min_t(u64, length, PAGE_SIZE);
1251 u64 mapped_length = sublen; 1252 u64 mapped_length = sublen;
1252 struct btrfs_bio *bbio = NULL; 1253 struct btrfs_bio *bbio = NULL;
1253 1254
1254 /* 1255 /*
1255 * with a length of PAGE_SIZE, each returned stripe 1256 * with a length of PAGE_SIZE, each returned stripe
1256 * represents one mirror 1257 * represents one mirror
1257 */ 1258 */
1258 ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical, 1259 ret = btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, logical,
1259 &mapped_length, &bbio, 0); 1260 &mapped_length, &bbio, 0);
1260 if (ret || !bbio || mapped_length < sublen) { 1261 if (ret || !bbio || mapped_length < sublen) {
1261 kfree(bbio); 1262 kfree(bbio);
1262 return -EIO; 1263 return -EIO;
1263 } 1264 }
1264 1265
1265 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO); 1266 BUG_ON(page_index >= SCRUB_PAGES_PER_RD_BIO);
1266 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes; 1267 for (mirror_index = 0; mirror_index < (int)bbio->num_stripes;
1267 mirror_index++) { 1268 mirror_index++) {
1268 struct scrub_block *sblock; 1269 struct scrub_block *sblock;
1269 struct scrub_page *page; 1270 struct scrub_page *page;
1270 1271
1271 if (mirror_index >= BTRFS_MAX_MIRRORS) 1272 if (mirror_index >= BTRFS_MAX_MIRRORS)
1272 continue; 1273 continue;
1273 1274
1274 sblock = sblocks_for_recheck + mirror_index; 1275 sblock = sblocks_for_recheck + mirror_index;
1275 sblock->sctx = sctx; 1276 sblock->sctx = sctx;
1276 page = kzalloc(sizeof(*page), GFP_NOFS); 1277 page = kzalloc(sizeof(*page), GFP_NOFS);
1277 if (!page) { 1278 if (!page) {
1278 leave_nomem: 1279 leave_nomem:
1279 spin_lock(&sctx->stat_lock); 1280 spin_lock(&sctx->stat_lock);
1280 sctx->stat.malloc_errors++; 1281 sctx->stat.malloc_errors++;
1281 spin_unlock(&sctx->stat_lock); 1282 spin_unlock(&sctx->stat_lock);
1282 kfree(bbio); 1283 kfree(bbio);
1283 return -ENOMEM; 1284 return -ENOMEM;
1284 } 1285 }
1285 scrub_page_get(page); 1286 scrub_page_get(page);
1286 sblock->pagev[page_index] = page; 1287 sblock->pagev[page_index] = page;
1287 page->logical = logical; 1288 page->logical = logical;
1288 page->physical = bbio->stripes[mirror_index].physical; 1289 page->physical = bbio->stripes[mirror_index].physical;
1289 BUG_ON(page_index >= original_sblock->page_count); 1290 BUG_ON(page_index >= original_sblock->page_count);
1290 page->physical_for_dev_replace = 1291 page->physical_for_dev_replace =
1291 original_sblock->pagev[page_index]-> 1292 original_sblock->pagev[page_index]->
1292 physical_for_dev_replace; 1293 physical_for_dev_replace;
1293 /* for missing devices, dev->bdev is NULL */ 1294 /* for missing devices, dev->bdev is NULL */
1294 page->dev = bbio->stripes[mirror_index].dev; 1295 page->dev = bbio->stripes[mirror_index].dev;
1295 page->mirror_num = mirror_index + 1; 1296 page->mirror_num = mirror_index + 1;
1296 sblock->page_count++; 1297 sblock->page_count++;
1297 page->page = alloc_page(GFP_NOFS); 1298 page->page = alloc_page(GFP_NOFS);
1298 if (!page->page) 1299 if (!page->page)
1299 goto leave_nomem; 1300 goto leave_nomem;
1300 } 1301 }
1301 kfree(bbio); 1302 kfree(bbio);
1302 length -= sublen; 1303 length -= sublen;
1303 logical += sublen; 1304 logical += sublen;
1304 page_index++; 1305 page_index++;
1305 } 1306 }
1306 1307
1307 return 0; 1308 return 0;
1308 } 1309 }
1309 1310
1310 /* 1311 /*
1311 * this function will check the on disk data for checksum errors, header 1312 * this function will check the on disk data for checksum errors, header
1312 * errors and read I/O errors. If any I/O errors happen, the exact pages 1313 * errors and read I/O errors. If any I/O errors happen, the exact pages
1313 * which are errored are marked as being bad. The goal is to enable scrub 1314 * which are errored are marked as being bad. The goal is to enable scrub
1314 * to take those pages that are not errored from all the mirrors so that 1315 * to take those pages that are not errored from all the mirrors so that
1315 * the pages that are errored in the just handled mirror can be repaired. 1316 * the pages that are errored in the just handled mirror can be repaired.
1316 */ 1317 */
1317 static void scrub_recheck_block(struct btrfs_fs_info *fs_info, 1318 static void scrub_recheck_block(struct btrfs_fs_info *fs_info,
1318 struct scrub_block *sblock, int is_metadata, 1319 struct scrub_block *sblock, int is_metadata,
1319 int have_csum, u8 *csum, u64 generation, 1320 int have_csum, u8 *csum, u64 generation,
1320 u16 csum_size) 1321 u16 csum_size)
1321 { 1322 {
1322 int page_num; 1323 int page_num;
1323 1324
1324 sblock->no_io_error_seen = 1; 1325 sblock->no_io_error_seen = 1;
1325 sblock->header_error = 0; 1326 sblock->header_error = 0;
1326 sblock->checksum_error = 0; 1327 sblock->checksum_error = 0;
1327 1328
1328 for (page_num = 0; page_num < sblock->page_count; page_num++) { 1329 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1329 struct bio *bio; 1330 struct bio *bio;
1330 struct scrub_page *page = sblock->pagev[page_num]; 1331 struct scrub_page *page = sblock->pagev[page_num];
1331 1332
1332 if (page->dev->bdev == NULL) { 1333 if (page->dev->bdev == NULL) {
1333 page->io_error = 1; 1334 page->io_error = 1;
1334 sblock->no_io_error_seen = 0; 1335 sblock->no_io_error_seen = 0;
1335 continue; 1336 continue;
1336 } 1337 }
1337 1338
1338 WARN_ON(!page->page); 1339 WARN_ON(!page->page);
1339 bio = btrfs_io_bio_alloc(GFP_NOFS, 1); 1340 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1340 if (!bio) { 1341 if (!bio) {
1341 page->io_error = 1; 1342 page->io_error = 1;
1342 sblock->no_io_error_seen = 0; 1343 sblock->no_io_error_seen = 0;
1343 continue; 1344 continue;
1344 } 1345 }
1345 bio->bi_bdev = page->dev->bdev; 1346 bio->bi_bdev = page->dev->bdev;
1346 bio->bi_iter.bi_sector = page->physical >> 9; 1347 bio->bi_iter.bi_sector = page->physical >> 9;
1347 1348
1348 bio_add_page(bio, page->page, PAGE_SIZE, 0); 1349 bio_add_page(bio, page->page, PAGE_SIZE, 0);
1349 if (btrfsic_submit_bio_wait(READ, bio)) 1350 if (btrfsic_submit_bio_wait(READ, bio))
1350 sblock->no_io_error_seen = 0; 1351 sblock->no_io_error_seen = 0;
1351 1352
1352 bio_put(bio); 1353 bio_put(bio);
1353 } 1354 }
1354 1355
1355 if (sblock->no_io_error_seen) 1356 if (sblock->no_io_error_seen)
1356 scrub_recheck_block_checksum(fs_info, sblock, is_metadata, 1357 scrub_recheck_block_checksum(fs_info, sblock, is_metadata,
1357 have_csum, csum, generation, 1358 have_csum, csum, generation,
1358 csum_size); 1359 csum_size);
1359 1360
1360 return; 1361 return;
1361 } 1362 }
1362 1363
1363 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info, 1364 static void scrub_recheck_block_checksum(struct btrfs_fs_info *fs_info,
1364 struct scrub_block *sblock, 1365 struct scrub_block *sblock,
1365 int is_metadata, int have_csum, 1366 int is_metadata, int have_csum,
1366 const u8 *csum, u64 generation, 1367 const u8 *csum, u64 generation,
1367 u16 csum_size) 1368 u16 csum_size)
1368 { 1369 {
1369 int page_num; 1370 int page_num;
1370 u8 calculated_csum[BTRFS_CSUM_SIZE]; 1371 u8 calculated_csum[BTRFS_CSUM_SIZE];
1371 u32 crc = ~(u32)0; 1372 u32 crc = ~(u32)0;
1372 void *mapped_buffer; 1373 void *mapped_buffer;
1373 1374
1374 WARN_ON(!sblock->pagev[0]->page); 1375 WARN_ON(!sblock->pagev[0]->page);
1375 if (is_metadata) { 1376 if (is_metadata) {
1376 struct btrfs_header *h; 1377 struct btrfs_header *h;
1377 1378
1378 mapped_buffer = kmap_atomic(sblock->pagev[0]->page); 1379 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1379 h = (struct btrfs_header *)mapped_buffer; 1380 h = (struct btrfs_header *)mapped_buffer;
1380 1381
1381 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) || 1382 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h) ||
1382 memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) || 1383 memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE) ||
1383 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, 1384 memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1384 BTRFS_UUID_SIZE)) { 1385 BTRFS_UUID_SIZE)) {
1385 sblock->header_error = 1; 1386 sblock->header_error = 1;
1386 } else if (generation != btrfs_stack_header_generation(h)) { 1387 } else if (generation != btrfs_stack_header_generation(h)) {
1387 sblock->header_error = 1; 1388 sblock->header_error = 1;
1388 sblock->generation_error = 1; 1389 sblock->generation_error = 1;
1389 } 1390 }
1390 csum = h->csum; 1391 csum = h->csum;
1391 } else { 1392 } else {
1392 if (!have_csum) 1393 if (!have_csum)
1393 return; 1394 return;
1394 1395
1395 mapped_buffer = kmap_atomic(sblock->pagev[0]->page); 1396 mapped_buffer = kmap_atomic(sblock->pagev[0]->page);
1396 } 1397 }
1397 1398
1398 for (page_num = 0;;) { 1399 for (page_num = 0;;) {
1399 if (page_num == 0 && is_metadata) 1400 if (page_num == 0 && is_metadata)
1400 crc = btrfs_csum_data( 1401 crc = btrfs_csum_data(
1401 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE, 1402 ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE,
1402 crc, PAGE_SIZE - BTRFS_CSUM_SIZE); 1403 crc, PAGE_SIZE - BTRFS_CSUM_SIZE);
1403 else 1404 else
1404 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE); 1405 crc = btrfs_csum_data(mapped_buffer, crc, PAGE_SIZE);
1405 1406
1406 kunmap_atomic(mapped_buffer); 1407 kunmap_atomic(mapped_buffer);
1407 page_num++; 1408 page_num++;
1408 if (page_num >= sblock->page_count) 1409 if (page_num >= sblock->page_count)
1409 break; 1410 break;
1410 WARN_ON(!sblock->pagev[page_num]->page); 1411 WARN_ON(!sblock->pagev[page_num]->page);
1411 1412
1412 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page); 1413 mapped_buffer = kmap_atomic(sblock->pagev[page_num]->page);
1413 } 1414 }
1414 1415
1415 btrfs_csum_final(crc, calculated_csum); 1416 btrfs_csum_final(crc, calculated_csum);
1416 if (memcmp(calculated_csum, csum, csum_size)) 1417 if (memcmp(calculated_csum, csum, csum_size))
1417 sblock->checksum_error = 1; 1418 sblock->checksum_error = 1;
1418 } 1419 }
1419 1420
1420 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad, 1421 static int scrub_repair_block_from_good_copy(struct scrub_block *sblock_bad,
1421 struct scrub_block *sblock_good, 1422 struct scrub_block *sblock_good,
1422 int force_write) 1423 int force_write)
1423 { 1424 {
1424 int page_num; 1425 int page_num;
1425 int ret = 0; 1426 int ret = 0;
1426 1427
1427 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) { 1428 for (page_num = 0; page_num < sblock_bad->page_count; page_num++) {
1428 int ret_sub; 1429 int ret_sub;
1429 1430
1430 ret_sub = scrub_repair_page_from_good_copy(sblock_bad, 1431 ret_sub = scrub_repair_page_from_good_copy(sblock_bad,
1431 sblock_good, 1432 sblock_good,
1432 page_num, 1433 page_num,
1433 force_write); 1434 force_write);
1434 if (ret_sub) 1435 if (ret_sub)
1435 ret = ret_sub; 1436 ret = ret_sub;
1436 } 1437 }
1437 1438
1438 return ret; 1439 return ret;
1439 } 1440 }
1440 1441
1441 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad, 1442 static int scrub_repair_page_from_good_copy(struct scrub_block *sblock_bad,
1442 struct scrub_block *sblock_good, 1443 struct scrub_block *sblock_good,
1443 int page_num, int force_write) 1444 int page_num, int force_write)
1444 { 1445 {
1445 struct scrub_page *page_bad = sblock_bad->pagev[page_num]; 1446 struct scrub_page *page_bad = sblock_bad->pagev[page_num];
1446 struct scrub_page *page_good = sblock_good->pagev[page_num]; 1447 struct scrub_page *page_good = sblock_good->pagev[page_num];
1447 1448
1448 BUG_ON(page_bad->page == NULL); 1449 BUG_ON(page_bad->page == NULL);
1449 BUG_ON(page_good->page == NULL); 1450 BUG_ON(page_good->page == NULL);
1450 if (force_write || sblock_bad->header_error || 1451 if (force_write || sblock_bad->header_error ||
1451 sblock_bad->checksum_error || page_bad->io_error) { 1452 sblock_bad->checksum_error || page_bad->io_error) {
1452 struct bio *bio; 1453 struct bio *bio;
1453 int ret; 1454 int ret;
1454 1455
1455 if (!page_bad->dev->bdev) { 1456 if (!page_bad->dev->bdev) {
1456 printk_ratelimited(KERN_WARNING "BTRFS: " 1457 printk_ratelimited(KERN_WARNING "BTRFS: "
1457 "scrub_repair_page_from_good_copy(bdev == NULL) " 1458 "scrub_repair_page_from_good_copy(bdev == NULL) "
1458 "is unexpected!\n"); 1459 "is unexpected!\n");
1459 return -EIO; 1460 return -EIO;
1460 } 1461 }
1461 1462
1462 bio = btrfs_io_bio_alloc(GFP_NOFS, 1); 1463 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
1463 if (!bio) 1464 if (!bio)
1464 return -EIO; 1465 return -EIO;
1465 bio->bi_bdev = page_bad->dev->bdev; 1466 bio->bi_bdev = page_bad->dev->bdev;
1466 bio->bi_iter.bi_sector = page_bad->physical >> 9; 1467 bio->bi_iter.bi_sector = page_bad->physical >> 9;
1467 1468
1468 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0); 1469 ret = bio_add_page(bio, page_good->page, PAGE_SIZE, 0);
1469 if (PAGE_SIZE != ret) { 1470 if (PAGE_SIZE != ret) {
1470 bio_put(bio); 1471 bio_put(bio);
1471 return -EIO; 1472 return -EIO;
1472 } 1473 }
1473 1474
1474 if (btrfsic_submit_bio_wait(WRITE, bio)) { 1475 if (btrfsic_submit_bio_wait(WRITE, bio)) {
1475 btrfs_dev_stat_inc_and_print(page_bad->dev, 1476 btrfs_dev_stat_inc_and_print(page_bad->dev,
1476 BTRFS_DEV_STAT_WRITE_ERRS); 1477 BTRFS_DEV_STAT_WRITE_ERRS);
1477 btrfs_dev_replace_stats_inc( 1478 btrfs_dev_replace_stats_inc(
1478 &sblock_bad->sctx->dev_root->fs_info-> 1479 &sblock_bad->sctx->dev_root->fs_info->
1479 dev_replace.num_write_errors); 1480 dev_replace.num_write_errors);
1480 bio_put(bio); 1481 bio_put(bio);
1481 return -EIO; 1482 return -EIO;
1482 } 1483 }
1483 bio_put(bio); 1484 bio_put(bio);
1484 } 1485 }
1485 1486
1486 return 0; 1487 return 0;
1487 } 1488 }
1488 1489
1489 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock) 1490 static void scrub_write_block_to_dev_replace(struct scrub_block *sblock)
1490 { 1491 {
1491 int page_num; 1492 int page_num;
1492 1493
1493 for (page_num = 0; page_num < sblock->page_count; page_num++) { 1494 for (page_num = 0; page_num < sblock->page_count; page_num++) {
1494 int ret; 1495 int ret;
1495 1496
1496 ret = scrub_write_page_to_dev_replace(sblock, page_num); 1497 ret = scrub_write_page_to_dev_replace(sblock, page_num);
1497 if (ret) 1498 if (ret)
1498 btrfs_dev_replace_stats_inc( 1499 btrfs_dev_replace_stats_inc(
1499 &sblock->sctx->dev_root->fs_info->dev_replace. 1500 &sblock->sctx->dev_root->fs_info->dev_replace.
1500 num_write_errors); 1501 num_write_errors);
1501 } 1502 }
1502 } 1503 }
1503 1504
1504 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock, 1505 static int scrub_write_page_to_dev_replace(struct scrub_block *sblock,
1505 int page_num) 1506 int page_num)
1506 { 1507 {
1507 struct scrub_page *spage = sblock->pagev[page_num]; 1508 struct scrub_page *spage = sblock->pagev[page_num];
1508 1509
1509 BUG_ON(spage->page == NULL); 1510 BUG_ON(spage->page == NULL);
1510 if (spage->io_error) { 1511 if (spage->io_error) {
1511 void *mapped_buffer = kmap_atomic(spage->page); 1512 void *mapped_buffer = kmap_atomic(spage->page);
1512 1513
1513 memset(mapped_buffer, 0, PAGE_CACHE_SIZE); 1514 memset(mapped_buffer, 0, PAGE_CACHE_SIZE);
1514 flush_dcache_page(spage->page); 1515 flush_dcache_page(spage->page);
1515 kunmap_atomic(mapped_buffer); 1516 kunmap_atomic(mapped_buffer);
1516 } 1517 }
1517 return scrub_add_page_to_wr_bio(sblock->sctx, spage); 1518 return scrub_add_page_to_wr_bio(sblock->sctx, spage);
1518 } 1519 }
1519 1520
1520 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx, 1521 static int scrub_add_page_to_wr_bio(struct scrub_ctx *sctx,
1521 struct scrub_page *spage) 1522 struct scrub_page *spage)
1522 { 1523 {
1523 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; 1524 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1524 struct scrub_bio *sbio; 1525 struct scrub_bio *sbio;
1525 int ret; 1526 int ret;
1526 1527
1527 mutex_lock(&wr_ctx->wr_lock); 1528 mutex_lock(&wr_ctx->wr_lock);
1528 again: 1529 again:
1529 if (!wr_ctx->wr_curr_bio) { 1530 if (!wr_ctx->wr_curr_bio) {
1530 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio), 1531 wr_ctx->wr_curr_bio = kzalloc(sizeof(*wr_ctx->wr_curr_bio),
1531 GFP_NOFS); 1532 GFP_NOFS);
1532 if (!wr_ctx->wr_curr_bio) { 1533 if (!wr_ctx->wr_curr_bio) {
1533 mutex_unlock(&wr_ctx->wr_lock); 1534 mutex_unlock(&wr_ctx->wr_lock);
1534 return -ENOMEM; 1535 return -ENOMEM;
1535 } 1536 }
1536 wr_ctx->wr_curr_bio->sctx = sctx; 1537 wr_ctx->wr_curr_bio->sctx = sctx;
1537 wr_ctx->wr_curr_bio->page_count = 0; 1538 wr_ctx->wr_curr_bio->page_count = 0;
1538 } 1539 }
1539 sbio = wr_ctx->wr_curr_bio; 1540 sbio = wr_ctx->wr_curr_bio;
1540 if (sbio->page_count == 0) { 1541 if (sbio->page_count == 0) {
1541 struct bio *bio; 1542 struct bio *bio;
1542 1543
1543 sbio->physical = spage->physical_for_dev_replace; 1544 sbio->physical = spage->physical_for_dev_replace;
1544 sbio->logical = spage->logical; 1545 sbio->logical = spage->logical;
1545 sbio->dev = wr_ctx->tgtdev; 1546 sbio->dev = wr_ctx->tgtdev;
1546 bio = sbio->bio; 1547 bio = sbio->bio;
1547 if (!bio) { 1548 if (!bio) {
1548 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio); 1549 bio = btrfs_io_bio_alloc(GFP_NOFS, wr_ctx->pages_per_wr_bio);
1549 if (!bio) { 1550 if (!bio) {
1550 mutex_unlock(&wr_ctx->wr_lock); 1551 mutex_unlock(&wr_ctx->wr_lock);
1551 return -ENOMEM; 1552 return -ENOMEM;
1552 } 1553 }
1553 sbio->bio = bio; 1554 sbio->bio = bio;
1554 } 1555 }
1555 1556
1556 bio->bi_private = sbio; 1557 bio->bi_private = sbio;
1557 bio->bi_end_io = scrub_wr_bio_end_io; 1558 bio->bi_end_io = scrub_wr_bio_end_io;
1558 bio->bi_bdev = sbio->dev->bdev; 1559 bio->bi_bdev = sbio->dev->bdev;
1559 bio->bi_iter.bi_sector = sbio->physical >> 9; 1560 bio->bi_iter.bi_sector = sbio->physical >> 9;
1560 sbio->err = 0; 1561 sbio->err = 0;
1561 } else if (sbio->physical + sbio->page_count * PAGE_SIZE != 1562 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1562 spage->physical_for_dev_replace || 1563 spage->physical_for_dev_replace ||
1563 sbio->logical + sbio->page_count * PAGE_SIZE != 1564 sbio->logical + sbio->page_count * PAGE_SIZE !=
1564 spage->logical) { 1565 spage->logical) {
1565 scrub_wr_submit(sctx); 1566 scrub_wr_submit(sctx);
1566 goto again; 1567 goto again;
1567 } 1568 }
1568 1569
1569 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); 1570 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1570 if (ret != PAGE_SIZE) { 1571 if (ret != PAGE_SIZE) {
1571 if (sbio->page_count < 1) { 1572 if (sbio->page_count < 1) {
1572 bio_put(sbio->bio); 1573 bio_put(sbio->bio);
1573 sbio->bio = NULL; 1574 sbio->bio = NULL;
1574 mutex_unlock(&wr_ctx->wr_lock); 1575 mutex_unlock(&wr_ctx->wr_lock);
1575 return -EIO; 1576 return -EIO;
1576 } 1577 }
1577 scrub_wr_submit(sctx); 1578 scrub_wr_submit(sctx);
1578 goto again; 1579 goto again;
1579 } 1580 }
1580 1581
1581 sbio->pagev[sbio->page_count] = spage; 1582 sbio->pagev[sbio->page_count] = spage;
1582 scrub_page_get(spage); 1583 scrub_page_get(spage);
1583 sbio->page_count++; 1584 sbio->page_count++;
1584 if (sbio->page_count == wr_ctx->pages_per_wr_bio) 1585 if (sbio->page_count == wr_ctx->pages_per_wr_bio)
1585 scrub_wr_submit(sctx); 1586 scrub_wr_submit(sctx);
1586 mutex_unlock(&wr_ctx->wr_lock); 1587 mutex_unlock(&wr_ctx->wr_lock);
1587 1588
1588 return 0; 1589 return 0;
1589 } 1590 }
1590 1591
1591 static void scrub_wr_submit(struct scrub_ctx *sctx) 1592 static void scrub_wr_submit(struct scrub_ctx *sctx)
1592 { 1593 {
1593 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx; 1594 struct scrub_wr_ctx *wr_ctx = &sctx->wr_ctx;
1594 struct scrub_bio *sbio; 1595 struct scrub_bio *sbio;
1595 1596
1596 if (!wr_ctx->wr_curr_bio) 1597 if (!wr_ctx->wr_curr_bio)
1597 return; 1598 return;
1598 1599
1599 sbio = wr_ctx->wr_curr_bio; 1600 sbio = wr_ctx->wr_curr_bio;
1600 wr_ctx->wr_curr_bio = NULL; 1601 wr_ctx->wr_curr_bio = NULL;
1601 WARN_ON(!sbio->bio->bi_bdev); 1602 WARN_ON(!sbio->bio->bi_bdev);
1602 scrub_pending_bio_inc(sctx); 1603 scrub_pending_bio_inc(sctx);
1603 /* process all writes in a single worker thread. Then the block layer 1604 /* process all writes in a single worker thread. Then the block layer
1604 * orders the requests before sending them to the driver which 1605 * orders the requests before sending them to the driver which
1605 * doubled the write performance on spinning disks when measured 1606 * doubled the write performance on spinning disks when measured
1606 * with Linux 3.5 */ 1607 * with Linux 3.5 */
1607 btrfsic_submit_bio(WRITE, sbio->bio); 1608 btrfsic_submit_bio(WRITE, sbio->bio);
1608 } 1609 }
1609 1610
1610 static void scrub_wr_bio_end_io(struct bio *bio, int err) 1611 static void scrub_wr_bio_end_io(struct bio *bio, int err)
1611 { 1612 {
1612 struct scrub_bio *sbio = bio->bi_private; 1613 struct scrub_bio *sbio = bio->bi_private;
1613 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; 1614 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
1614 1615
1615 sbio->err = err; 1616 sbio->err = err;
1616 sbio->bio = bio; 1617 sbio->bio = bio;
1617 1618
1618 btrfs_init_work(&sbio->work, scrub_wr_bio_end_io_worker, NULL, NULL); 1619 btrfs_init_work(&sbio->work, scrub_wr_bio_end_io_worker, NULL, NULL);
1619 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work); 1620 btrfs_queue_work(fs_info->scrub_wr_completion_workers, &sbio->work);
1620 } 1621 }
1621 1622
1622 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work) 1623 static void scrub_wr_bio_end_io_worker(struct btrfs_work *work)
1623 { 1624 {
1624 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); 1625 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
1625 struct scrub_ctx *sctx = sbio->sctx; 1626 struct scrub_ctx *sctx = sbio->sctx;
1626 int i; 1627 int i;
1627 1628
1628 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO); 1629 WARN_ON(sbio->page_count > SCRUB_PAGES_PER_WR_BIO);
1629 if (sbio->err) { 1630 if (sbio->err) {
1630 struct btrfs_dev_replace *dev_replace = 1631 struct btrfs_dev_replace *dev_replace =
1631 &sbio->sctx->dev_root->fs_info->dev_replace; 1632 &sbio->sctx->dev_root->fs_info->dev_replace;
1632 1633
1633 for (i = 0; i < sbio->page_count; i++) { 1634 for (i = 0; i < sbio->page_count; i++) {
1634 struct scrub_page *spage = sbio->pagev[i]; 1635 struct scrub_page *spage = sbio->pagev[i];
1635 1636
1636 spage->io_error = 1; 1637 spage->io_error = 1;
1637 btrfs_dev_replace_stats_inc(&dev_replace-> 1638 btrfs_dev_replace_stats_inc(&dev_replace->
1638 num_write_errors); 1639 num_write_errors);
1639 } 1640 }
1640 } 1641 }
1641 1642
1642 for (i = 0; i < sbio->page_count; i++) 1643 for (i = 0; i < sbio->page_count; i++)
1643 scrub_page_put(sbio->pagev[i]); 1644 scrub_page_put(sbio->pagev[i]);
1644 1645
1645 bio_put(sbio->bio); 1646 bio_put(sbio->bio);
1646 kfree(sbio); 1647 kfree(sbio);
1647 scrub_pending_bio_dec(sctx); 1648 scrub_pending_bio_dec(sctx);
1648 } 1649 }
1649 1650
1650 static int scrub_checksum(struct scrub_block *sblock) 1651 static int scrub_checksum(struct scrub_block *sblock)
1651 { 1652 {
1652 u64 flags; 1653 u64 flags;
1653 int ret; 1654 int ret;
1654 1655
1655 WARN_ON(sblock->page_count < 1); 1656 WARN_ON(sblock->page_count < 1);
1656 flags = sblock->pagev[0]->flags; 1657 flags = sblock->pagev[0]->flags;
1657 ret = 0; 1658 ret = 0;
1658 if (flags & BTRFS_EXTENT_FLAG_DATA) 1659 if (flags & BTRFS_EXTENT_FLAG_DATA)
1659 ret = scrub_checksum_data(sblock); 1660 ret = scrub_checksum_data(sblock);
1660 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) 1661 else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1661 ret = scrub_checksum_tree_block(sblock); 1662 ret = scrub_checksum_tree_block(sblock);
1662 else if (flags & BTRFS_EXTENT_FLAG_SUPER) 1663 else if (flags & BTRFS_EXTENT_FLAG_SUPER)
1663 (void)scrub_checksum_super(sblock); 1664 (void)scrub_checksum_super(sblock);
1664 else 1665 else
1665 WARN_ON(1); 1666 WARN_ON(1);
1666 if (ret) 1667 if (ret)
1667 scrub_handle_errored_block(sblock); 1668 scrub_handle_errored_block(sblock);
1668 1669
1669 return ret; 1670 return ret;
1670 } 1671 }
1671 1672
1672 static int scrub_checksum_data(struct scrub_block *sblock) 1673 static int scrub_checksum_data(struct scrub_block *sblock)
1673 { 1674 {
1674 struct scrub_ctx *sctx = sblock->sctx; 1675 struct scrub_ctx *sctx = sblock->sctx;
1675 u8 csum[BTRFS_CSUM_SIZE]; 1676 u8 csum[BTRFS_CSUM_SIZE];
1676 u8 *on_disk_csum; 1677 u8 *on_disk_csum;
1677 struct page *page; 1678 struct page *page;
1678 void *buffer; 1679 void *buffer;
1679 u32 crc = ~(u32)0; 1680 u32 crc = ~(u32)0;
1680 int fail = 0; 1681 int fail = 0;
1681 u64 len; 1682 u64 len;
1682 int index; 1683 int index;
1683 1684
1684 BUG_ON(sblock->page_count < 1); 1685 BUG_ON(sblock->page_count < 1);
1685 if (!sblock->pagev[0]->have_csum) 1686 if (!sblock->pagev[0]->have_csum)
1686 return 0; 1687 return 0;
1687 1688
1688 on_disk_csum = sblock->pagev[0]->csum; 1689 on_disk_csum = sblock->pagev[0]->csum;
1689 page = sblock->pagev[0]->page; 1690 page = sblock->pagev[0]->page;
1690 buffer = kmap_atomic(page); 1691 buffer = kmap_atomic(page);
1691 1692
1692 len = sctx->sectorsize; 1693 len = sctx->sectorsize;
1693 index = 0; 1694 index = 0;
1694 for (;;) { 1695 for (;;) {
1695 u64 l = min_t(u64, len, PAGE_SIZE); 1696 u64 l = min_t(u64, len, PAGE_SIZE);
1696 1697
1697 crc = btrfs_csum_data(buffer, crc, l); 1698 crc = btrfs_csum_data(buffer, crc, l);
1698 kunmap_atomic(buffer); 1699 kunmap_atomic(buffer);
1699 len -= l; 1700 len -= l;
1700 if (len == 0) 1701 if (len == 0)
1701 break; 1702 break;
1702 index++; 1703 index++;
1703 BUG_ON(index >= sblock->page_count); 1704 BUG_ON(index >= sblock->page_count);
1704 BUG_ON(!sblock->pagev[index]->page); 1705 BUG_ON(!sblock->pagev[index]->page);
1705 page = sblock->pagev[index]->page; 1706 page = sblock->pagev[index]->page;
1706 buffer = kmap_atomic(page); 1707 buffer = kmap_atomic(page);
1707 } 1708 }
1708 1709
1709 btrfs_csum_final(crc, csum); 1710 btrfs_csum_final(crc, csum);
1710 if (memcmp(csum, on_disk_csum, sctx->csum_size)) 1711 if (memcmp(csum, on_disk_csum, sctx->csum_size))
1711 fail = 1; 1712 fail = 1;
1712 1713
1713 return fail; 1714 return fail;
1714 } 1715 }
1715 1716
1716 static int scrub_checksum_tree_block(struct scrub_block *sblock) 1717 static int scrub_checksum_tree_block(struct scrub_block *sblock)
1717 { 1718 {
1718 struct scrub_ctx *sctx = sblock->sctx; 1719 struct scrub_ctx *sctx = sblock->sctx;
1719 struct btrfs_header *h; 1720 struct btrfs_header *h;
1720 struct btrfs_root *root = sctx->dev_root; 1721 struct btrfs_root *root = sctx->dev_root;
1721 struct btrfs_fs_info *fs_info = root->fs_info; 1722 struct btrfs_fs_info *fs_info = root->fs_info;
1722 u8 calculated_csum[BTRFS_CSUM_SIZE]; 1723 u8 calculated_csum[BTRFS_CSUM_SIZE];
1723 u8 on_disk_csum[BTRFS_CSUM_SIZE]; 1724 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1724 struct page *page; 1725 struct page *page;
1725 void *mapped_buffer; 1726 void *mapped_buffer;
1726 u64 mapped_size; 1727 u64 mapped_size;
1727 void *p; 1728 void *p;
1728 u32 crc = ~(u32)0; 1729 u32 crc = ~(u32)0;
1729 int fail = 0; 1730 int fail = 0;
1730 int crc_fail = 0; 1731 int crc_fail = 0;
1731 u64 len; 1732 u64 len;
1732 int index; 1733 int index;
1733 1734
1734 BUG_ON(sblock->page_count < 1); 1735 BUG_ON(sblock->page_count < 1);
1735 page = sblock->pagev[0]->page; 1736 page = sblock->pagev[0]->page;
1736 mapped_buffer = kmap_atomic(page); 1737 mapped_buffer = kmap_atomic(page);
1737 h = (struct btrfs_header *)mapped_buffer; 1738 h = (struct btrfs_header *)mapped_buffer;
1738 memcpy(on_disk_csum, h->csum, sctx->csum_size); 1739 memcpy(on_disk_csum, h->csum, sctx->csum_size);
1739 1740
1740 /* 1741 /*
1741 * we don't use the getter functions here, as we 1742 * we don't use the getter functions here, as we
1742 * a) don't have an extent buffer and 1743 * a) don't have an extent buffer and
1743 * b) the page is already kmapped 1744 * b) the page is already kmapped
1744 */ 1745 */
1745 1746
1746 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h)) 1747 if (sblock->pagev[0]->logical != btrfs_stack_header_bytenr(h))
1747 ++fail; 1748 ++fail;
1748 1749
1749 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h)) 1750 if (sblock->pagev[0]->generation != btrfs_stack_header_generation(h))
1750 ++fail; 1751 ++fail;
1751 1752
1752 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) 1753 if (memcmp(h->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1753 ++fail; 1754 ++fail;
1754 1755
1755 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid, 1756 if (memcmp(h->chunk_tree_uuid, fs_info->chunk_tree_uuid,
1756 BTRFS_UUID_SIZE)) 1757 BTRFS_UUID_SIZE))
1757 ++fail; 1758 ++fail;
1758 1759
1759 WARN_ON(sctx->nodesize != sctx->leafsize); 1760 WARN_ON(sctx->nodesize != sctx->leafsize);
1760 len = sctx->nodesize - BTRFS_CSUM_SIZE; 1761 len = sctx->nodesize - BTRFS_CSUM_SIZE;
1761 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; 1762 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1762 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; 1763 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1763 index = 0; 1764 index = 0;
1764 for (;;) { 1765 for (;;) {
1765 u64 l = min_t(u64, len, mapped_size); 1766 u64 l = min_t(u64, len, mapped_size);
1766 1767
1767 crc = btrfs_csum_data(p, crc, l); 1768 crc = btrfs_csum_data(p, crc, l);
1768 kunmap_atomic(mapped_buffer); 1769 kunmap_atomic(mapped_buffer);
1769 len -= l; 1770 len -= l;
1770 if (len == 0) 1771 if (len == 0)
1771 break; 1772 break;
1772 index++; 1773 index++;
1773 BUG_ON(index >= sblock->page_count); 1774 BUG_ON(index >= sblock->page_count);
1774 BUG_ON(!sblock->pagev[index]->page); 1775 BUG_ON(!sblock->pagev[index]->page);
1775 page = sblock->pagev[index]->page; 1776 page = sblock->pagev[index]->page;
1776 mapped_buffer = kmap_atomic(page); 1777 mapped_buffer = kmap_atomic(page);
1777 mapped_size = PAGE_SIZE; 1778 mapped_size = PAGE_SIZE;
1778 p = mapped_buffer; 1779 p = mapped_buffer;
1779 } 1780 }
1780 1781
1781 btrfs_csum_final(crc, calculated_csum); 1782 btrfs_csum_final(crc, calculated_csum);
1782 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) 1783 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1783 ++crc_fail; 1784 ++crc_fail;
1784 1785
1785 return fail || crc_fail; 1786 return fail || crc_fail;
1786 } 1787 }
1787 1788
1788 static int scrub_checksum_super(struct scrub_block *sblock) 1789 static int scrub_checksum_super(struct scrub_block *sblock)
1789 { 1790 {
1790 struct btrfs_super_block *s; 1791 struct btrfs_super_block *s;
1791 struct scrub_ctx *sctx = sblock->sctx; 1792 struct scrub_ctx *sctx = sblock->sctx;
1792 struct btrfs_root *root = sctx->dev_root; 1793 struct btrfs_root *root = sctx->dev_root;
1793 struct btrfs_fs_info *fs_info = root->fs_info; 1794 struct btrfs_fs_info *fs_info = root->fs_info;
1794 u8 calculated_csum[BTRFS_CSUM_SIZE]; 1795 u8 calculated_csum[BTRFS_CSUM_SIZE];
1795 u8 on_disk_csum[BTRFS_CSUM_SIZE]; 1796 u8 on_disk_csum[BTRFS_CSUM_SIZE];
1796 struct page *page; 1797 struct page *page;
1797 void *mapped_buffer; 1798 void *mapped_buffer;
1798 u64 mapped_size; 1799 u64 mapped_size;
1799 void *p; 1800 void *p;
1800 u32 crc = ~(u32)0; 1801 u32 crc = ~(u32)0;
1801 int fail_gen = 0; 1802 int fail_gen = 0;
1802 int fail_cor = 0; 1803 int fail_cor = 0;
1803 u64 len; 1804 u64 len;
1804 int index; 1805 int index;
1805 1806
1806 BUG_ON(sblock->page_count < 1); 1807 BUG_ON(sblock->page_count < 1);
1807 page = sblock->pagev[0]->page; 1808 page = sblock->pagev[0]->page;
1808 mapped_buffer = kmap_atomic(page); 1809 mapped_buffer = kmap_atomic(page);
1809 s = (struct btrfs_super_block *)mapped_buffer; 1810 s = (struct btrfs_super_block *)mapped_buffer;
1810 memcpy(on_disk_csum, s->csum, sctx->csum_size); 1811 memcpy(on_disk_csum, s->csum, sctx->csum_size);
1811 1812
1812 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s)) 1813 if (sblock->pagev[0]->logical != btrfs_super_bytenr(s))
1813 ++fail_cor; 1814 ++fail_cor;
1814 1815
1815 if (sblock->pagev[0]->generation != btrfs_super_generation(s)) 1816 if (sblock->pagev[0]->generation != btrfs_super_generation(s))
1816 ++fail_gen; 1817 ++fail_gen;
1817 1818
1818 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE)) 1819 if (memcmp(s->fsid, fs_info->fsid, BTRFS_UUID_SIZE))
1819 ++fail_cor; 1820 ++fail_cor;
1820 1821
1821 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE; 1822 len = BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE;
1822 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE; 1823 mapped_size = PAGE_SIZE - BTRFS_CSUM_SIZE;
1823 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE; 1824 p = ((u8 *)mapped_buffer) + BTRFS_CSUM_SIZE;
1824 index = 0; 1825 index = 0;
1825 for (;;) { 1826 for (;;) {
1826 u64 l = min_t(u64, len, mapped_size); 1827 u64 l = min_t(u64, len, mapped_size);
1827 1828
1828 crc = btrfs_csum_data(p, crc, l); 1829 crc = btrfs_csum_data(p, crc, l);
1829 kunmap_atomic(mapped_buffer); 1830 kunmap_atomic(mapped_buffer);
1830 len -= l; 1831 len -= l;
1831 if (len == 0) 1832 if (len == 0)
1832 break; 1833 break;
1833 index++; 1834 index++;
1834 BUG_ON(index >= sblock->page_count); 1835 BUG_ON(index >= sblock->page_count);
1835 BUG_ON(!sblock->pagev[index]->page); 1836 BUG_ON(!sblock->pagev[index]->page);
1836 page = sblock->pagev[index]->page; 1837 page = sblock->pagev[index]->page;
1837 mapped_buffer = kmap_atomic(page); 1838 mapped_buffer = kmap_atomic(page);
1838 mapped_size = PAGE_SIZE; 1839 mapped_size = PAGE_SIZE;
1839 p = mapped_buffer; 1840 p = mapped_buffer;
1840 } 1841 }
1841 1842
1842 btrfs_csum_final(crc, calculated_csum); 1843 btrfs_csum_final(crc, calculated_csum);
1843 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size)) 1844 if (memcmp(calculated_csum, on_disk_csum, sctx->csum_size))
1844 ++fail_cor; 1845 ++fail_cor;
1845 1846
1846 if (fail_cor + fail_gen) { 1847 if (fail_cor + fail_gen) {
1847 /* 1848 /*
1848 * if we find an error in a super block, we just report it. 1849 * if we find an error in a super block, we just report it.
1849 * They will get written with the next transaction commit 1850 * They will get written with the next transaction commit
1850 * anyway 1851 * anyway
1851 */ 1852 */
1852 spin_lock(&sctx->stat_lock); 1853 spin_lock(&sctx->stat_lock);
1853 ++sctx->stat.super_errors; 1854 ++sctx->stat.super_errors;
1854 spin_unlock(&sctx->stat_lock); 1855 spin_unlock(&sctx->stat_lock);
1855 if (fail_cor) 1856 if (fail_cor)
1856 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, 1857 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1857 BTRFS_DEV_STAT_CORRUPTION_ERRS); 1858 BTRFS_DEV_STAT_CORRUPTION_ERRS);
1858 else 1859 else
1859 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev, 1860 btrfs_dev_stat_inc_and_print(sblock->pagev[0]->dev,
1860 BTRFS_DEV_STAT_GENERATION_ERRS); 1861 BTRFS_DEV_STAT_GENERATION_ERRS);
1861 } 1862 }
1862 1863
1863 return fail_cor + fail_gen; 1864 return fail_cor + fail_gen;
1864 } 1865 }
1865 1866
1866 static void scrub_block_get(struct scrub_block *sblock) 1867 static void scrub_block_get(struct scrub_block *sblock)
1867 { 1868 {
1868 atomic_inc(&sblock->ref_count); 1869 atomic_inc(&sblock->ref_count);
1869 } 1870 }
1870 1871
1871 static void scrub_block_put(struct scrub_block *sblock) 1872 static void scrub_block_put(struct scrub_block *sblock)
1872 { 1873 {
1873 if (atomic_dec_and_test(&sblock->ref_count)) { 1874 if (atomic_dec_and_test(&sblock->ref_count)) {
1874 int i; 1875 int i;
1875 1876
1876 for (i = 0; i < sblock->page_count; i++) 1877 for (i = 0; i < sblock->page_count; i++)
1877 scrub_page_put(sblock->pagev[i]); 1878 scrub_page_put(sblock->pagev[i]);
1878 kfree(sblock); 1879 kfree(sblock);
1879 } 1880 }
1880 } 1881 }
1881 1882
1882 static void scrub_page_get(struct scrub_page *spage) 1883 static void scrub_page_get(struct scrub_page *spage)
1883 { 1884 {
1884 atomic_inc(&spage->ref_count); 1885 atomic_inc(&spage->ref_count);
1885 } 1886 }
1886 1887
1887 static void scrub_page_put(struct scrub_page *spage) 1888 static void scrub_page_put(struct scrub_page *spage)
1888 { 1889 {
1889 if (atomic_dec_and_test(&spage->ref_count)) { 1890 if (atomic_dec_and_test(&spage->ref_count)) {
1890 if (spage->page) 1891 if (spage->page)
1891 __free_page(spage->page); 1892 __free_page(spage->page);
1892 kfree(spage); 1893 kfree(spage);
1893 } 1894 }
1894 } 1895 }
1895 1896
1896 static void scrub_submit(struct scrub_ctx *sctx) 1897 static void scrub_submit(struct scrub_ctx *sctx)
1897 { 1898 {
1898 struct scrub_bio *sbio; 1899 struct scrub_bio *sbio;
1899 1900
1900 if (sctx->curr == -1) 1901 if (sctx->curr == -1)
1901 return; 1902 return;
1902 1903
1903 sbio = sctx->bios[sctx->curr]; 1904 sbio = sctx->bios[sctx->curr];
1904 sctx->curr = -1; 1905 sctx->curr = -1;
1905 scrub_pending_bio_inc(sctx); 1906 scrub_pending_bio_inc(sctx);
1906 1907
1907 if (!sbio->bio->bi_bdev) { 1908 if (!sbio->bio->bi_bdev) {
1908 /* 1909 /*
1909 * this case should not happen. If btrfs_map_block() is 1910 * this case should not happen. If btrfs_map_block() is
1910 * wrong, it could happen for dev-replace operations on 1911 * wrong, it could happen for dev-replace operations on
1911 * missing devices when no mirrors are available, but in 1912 * missing devices when no mirrors are available, but in
1912 * this case it should already fail the mount. 1913 * this case it should already fail the mount.
1913 * This case is handled correctly (but _very_ slowly). 1914 * This case is handled correctly (but _very_ slowly).
1914 */ 1915 */
1915 printk_ratelimited(KERN_WARNING 1916 printk_ratelimited(KERN_WARNING
1916 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n"); 1917 "BTRFS: scrub_submit(bio bdev == NULL) is unexpected!\n");
1917 bio_endio(sbio->bio, -EIO); 1918 bio_endio(sbio->bio, -EIO);
1918 } else { 1919 } else {
1919 btrfsic_submit_bio(READ, sbio->bio); 1920 btrfsic_submit_bio(READ, sbio->bio);
1920 } 1921 }
1921 } 1922 }
1922 1923
1923 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx, 1924 static int scrub_add_page_to_rd_bio(struct scrub_ctx *sctx,
1924 struct scrub_page *spage) 1925 struct scrub_page *spage)
1925 { 1926 {
1926 struct scrub_block *sblock = spage->sblock; 1927 struct scrub_block *sblock = spage->sblock;
1927 struct scrub_bio *sbio; 1928 struct scrub_bio *sbio;
1928 int ret; 1929 int ret;
1929 1930
1930 again: 1931 again:
1931 /* 1932 /*
1932 * grab a fresh bio or wait for one to become available 1933 * grab a fresh bio or wait for one to become available
1933 */ 1934 */
1934 while (sctx->curr == -1) { 1935 while (sctx->curr == -1) {
1935 spin_lock(&sctx->list_lock); 1936 spin_lock(&sctx->list_lock);
1936 sctx->curr = sctx->first_free; 1937 sctx->curr = sctx->first_free;
1937 if (sctx->curr != -1) { 1938 if (sctx->curr != -1) {
1938 sctx->first_free = sctx->bios[sctx->curr]->next_free; 1939 sctx->first_free = sctx->bios[sctx->curr]->next_free;
1939 sctx->bios[sctx->curr]->next_free = -1; 1940 sctx->bios[sctx->curr]->next_free = -1;
1940 sctx->bios[sctx->curr]->page_count = 0; 1941 sctx->bios[sctx->curr]->page_count = 0;
1941 spin_unlock(&sctx->list_lock); 1942 spin_unlock(&sctx->list_lock);
1942 } else { 1943 } else {
1943 spin_unlock(&sctx->list_lock); 1944 spin_unlock(&sctx->list_lock);
1944 wait_event(sctx->list_wait, sctx->first_free != -1); 1945 wait_event(sctx->list_wait, sctx->first_free != -1);
1945 } 1946 }
1946 } 1947 }
1947 sbio = sctx->bios[sctx->curr]; 1948 sbio = sctx->bios[sctx->curr];
1948 if (sbio->page_count == 0) { 1949 if (sbio->page_count == 0) {
1949 struct bio *bio; 1950 struct bio *bio;
1950 1951
1951 sbio->physical = spage->physical; 1952 sbio->physical = spage->physical;
1952 sbio->logical = spage->logical; 1953 sbio->logical = spage->logical;
1953 sbio->dev = spage->dev; 1954 sbio->dev = spage->dev;
1954 bio = sbio->bio; 1955 bio = sbio->bio;
1955 if (!bio) { 1956 if (!bio) {
1956 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio); 1957 bio = btrfs_io_bio_alloc(GFP_NOFS, sctx->pages_per_rd_bio);
1957 if (!bio) 1958 if (!bio)
1958 return -ENOMEM; 1959 return -ENOMEM;
1959 sbio->bio = bio; 1960 sbio->bio = bio;
1960 } 1961 }
1961 1962
1962 bio->bi_private = sbio; 1963 bio->bi_private = sbio;
1963 bio->bi_end_io = scrub_bio_end_io; 1964 bio->bi_end_io = scrub_bio_end_io;
1964 bio->bi_bdev = sbio->dev->bdev; 1965 bio->bi_bdev = sbio->dev->bdev;
1965 bio->bi_iter.bi_sector = sbio->physical >> 9; 1966 bio->bi_iter.bi_sector = sbio->physical >> 9;
1966 sbio->err = 0; 1967 sbio->err = 0;
1967 } else if (sbio->physical + sbio->page_count * PAGE_SIZE != 1968 } else if (sbio->physical + sbio->page_count * PAGE_SIZE !=
1968 spage->physical || 1969 spage->physical ||
1969 sbio->logical + sbio->page_count * PAGE_SIZE != 1970 sbio->logical + sbio->page_count * PAGE_SIZE !=
1970 spage->logical || 1971 spage->logical ||
1971 sbio->dev != spage->dev) { 1972 sbio->dev != spage->dev) {
1972 scrub_submit(sctx); 1973 scrub_submit(sctx);
1973 goto again; 1974 goto again;
1974 } 1975 }
1975 1976
1976 sbio->pagev[sbio->page_count] = spage; 1977 sbio->pagev[sbio->page_count] = spage;
1977 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0); 1978 ret = bio_add_page(sbio->bio, spage->page, PAGE_SIZE, 0);
1978 if (ret != PAGE_SIZE) { 1979 if (ret != PAGE_SIZE) {
1979 if (sbio->page_count < 1) { 1980 if (sbio->page_count < 1) {
1980 bio_put(sbio->bio); 1981 bio_put(sbio->bio);
1981 sbio->bio = NULL; 1982 sbio->bio = NULL;
1982 return -EIO; 1983 return -EIO;
1983 } 1984 }
1984 scrub_submit(sctx); 1985 scrub_submit(sctx);
1985 goto again; 1986 goto again;
1986 } 1987 }
1987 1988
1988 scrub_block_get(sblock); /* one for the page added to the bio */ 1989 scrub_block_get(sblock); /* one for the page added to the bio */
1989 atomic_inc(&sblock->outstanding_pages); 1990 atomic_inc(&sblock->outstanding_pages);
1990 sbio->page_count++; 1991 sbio->page_count++;
1991 if (sbio->page_count == sctx->pages_per_rd_bio) 1992 if (sbio->page_count == sctx->pages_per_rd_bio)
1992 scrub_submit(sctx); 1993 scrub_submit(sctx);
1993 1994
1994 return 0; 1995 return 0;
1995 } 1996 }
1996 1997
1997 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len, 1998 static int scrub_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
1998 u64 physical, struct btrfs_device *dev, u64 flags, 1999 u64 physical, struct btrfs_device *dev, u64 flags,
1999 u64 gen, int mirror_num, u8 *csum, int force, 2000 u64 gen, int mirror_num, u8 *csum, int force,
2000 u64 physical_for_dev_replace) 2001 u64 physical_for_dev_replace)
2001 { 2002 {
2002 struct scrub_block *sblock; 2003 struct scrub_block *sblock;
2003 int index; 2004 int index;
2004 2005
2005 sblock = kzalloc(sizeof(*sblock), GFP_NOFS); 2006 sblock = kzalloc(sizeof(*sblock), GFP_NOFS);
2006 if (!sblock) { 2007 if (!sblock) {
2007 spin_lock(&sctx->stat_lock); 2008 spin_lock(&sctx->stat_lock);
2008 sctx->stat.malloc_errors++; 2009 sctx->stat.malloc_errors++;
2009 spin_unlock(&sctx->stat_lock); 2010 spin_unlock(&sctx->stat_lock);
2010 return -ENOMEM; 2011 return -ENOMEM;
2011 } 2012 }
2012 2013
2013 /* one ref inside this function, plus one for each page added to 2014 /* one ref inside this function, plus one for each page added to
2014 * a bio later on */ 2015 * a bio later on */
2015 atomic_set(&sblock->ref_count, 1); 2016 atomic_set(&sblock->ref_count, 1);
2016 sblock->sctx = sctx; 2017 sblock->sctx = sctx;
2017 sblock->no_io_error_seen = 1; 2018 sblock->no_io_error_seen = 1;
2018 2019
2019 for (index = 0; len > 0; index++) { 2020 for (index = 0; len > 0; index++) {
2020 struct scrub_page *spage; 2021 struct scrub_page *spage;
2021 u64 l = min_t(u64, len, PAGE_SIZE); 2022 u64 l = min_t(u64, len, PAGE_SIZE);
2022 2023
2023 spage = kzalloc(sizeof(*spage), GFP_NOFS); 2024 spage = kzalloc(sizeof(*spage), GFP_NOFS);
2024 if (!spage) { 2025 if (!spage) {
2025 leave_nomem: 2026 leave_nomem:
2026 spin_lock(&sctx->stat_lock); 2027 spin_lock(&sctx->stat_lock);
2027 sctx->stat.malloc_errors++; 2028 sctx->stat.malloc_errors++;
2028 spin_unlock(&sctx->stat_lock); 2029 spin_unlock(&sctx->stat_lock);
2029 scrub_block_put(sblock); 2030 scrub_block_put(sblock);
2030 return -ENOMEM; 2031 return -ENOMEM;
2031 } 2032 }
2032 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK); 2033 BUG_ON(index >= SCRUB_MAX_PAGES_PER_BLOCK);
2033 scrub_page_get(spage); 2034 scrub_page_get(spage);
2034 sblock->pagev[index] = spage; 2035 sblock->pagev[index] = spage;
2035 spage->sblock = sblock; 2036 spage->sblock = sblock;
2036 spage->dev = dev; 2037 spage->dev = dev;
2037 spage->flags = flags; 2038 spage->flags = flags;
2038 spage->generation = gen; 2039 spage->generation = gen;
2039 spage->logical = logical; 2040 spage->logical = logical;
2040 spage->physical = physical; 2041 spage->physical = physical;
2041 spage->physical_for_dev_replace = physical_for_dev_replace; 2042 spage->physical_for_dev_replace = physical_for_dev_replace;
2042 spage->mirror_num = mirror_num; 2043 spage->mirror_num = mirror_num;
2043 if (csum) { 2044 if (csum) {
2044 spage->have_csum = 1; 2045 spage->have_csum = 1;
2045 memcpy(spage->csum, csum, sctx->csum_size); 2046 memcpy(spage->csum, csum, sctx->csum_size);
2046 } else { 2047 } else {
2047 spage->have_csum = 0; 2048 spage->have_csum = 0;
2048 } 2049 }
2049 sblock->page_count++; 2050 sblock->page_count++;
2050 spage->page = alloc_page(GFP_NOFS); 2051 spage->page = alloc_page(GFP_NOFS);
2051 if (!spage->page) 2052 if (!spage->page)
2052 goto leave_nomem; 2053 goto leave_nomem;
2053 len -= l; 2054 len -= l;
2054 logical += l; 2055 logical += l;
2055 physical += l; 2056 physical += l;
2056 physical_for_dev_replace += l; 2057 physical_for_dev_replace += l;
2057 } 2058 }
2058 2059
2059 WARN_ON(sblock->page_count == 0); 2060 WARN_ON(sblock->page_count == 0);
2060 for (index = 0; index < sblock->page_count; index++) { 2061 for (index = 0; index < sblock->page_count; index++) {
2061 struct scrub_page *spage = sblock->pagev[index]; 2062 struct scrub_page *spage = sblock->pagev[index];
2062 int ret; 2063 int ret;
2063 2064
2064 ret = scrub_add_page_to_rd_bio(sctx, spage); 2065 ret = scrub_add_page_to_rd_bio(sctx, spage);
2065 if (ret) { 2066 if (ret) {
2066 scrub_block_put(sblock); 2067 scrub_block_put(sblock);
2067 return ret; 2068 return ret;
2068 } 2069 }
2069 } 2070 }
2070 2071
2071 if (force) 2072 if (force)
2072 scrub_submit(sctx); 2073 scrub_submit(sctx);
2073 2074
2074 /* last one frees, either here or in bio completion for last page */ 2075 /* last one frees, either here or in bio completion for last page */
2075 scrub_block_put(sblock); 2076 scrub_block_put(sblock);
2076 return 0; 2077 return 0;
2077 } 2078 }
2078 2079
2079 static void scrub_bio_end_io(struct bio *bio, int err) 2080 static void scrub_bio_end_io(struct bio *bio, int err)
2080 { 2081 {
2081 struct scrub_bio *sbio = bio->bi_private; 2082 struct scrub_bio *sbio = bio->bi_private;
2082 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info; 2083 struct btrfs_fs_info *fs_info = sbio->dev->dev_root->fs_info;
2083 2084
2084 sbio->err = err; 2085 sbio->err = err;
2085 sbio->bio = bio; 2086 sbio->bio = bio;
2086 2087
2087 btrfs_queue_work(fs_info->scrub_workers, &sbio->work); 2088 btrfs_queue_work(fs_info->scrub_workers, &sbio->work);
2088 } 2089 }
2089 2090
2090 static void scrub_bio_end_io_worker(struct btrfs_work *work) 2091 static void scrub_bio_end_io_worker(struct btrfs_work *work)
2091 { 2092 {
2092 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work); 2093 struct scrub_bio *sbio = container_of(work, struct scrub_bio, work);
2093 struct scrub_ctx *sctx = sbio->sctx; 2094 struct scrub_ctx *sctx = sbio->sctx;
2094 int i; 2095 int i;
2095 2096
2096 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO); 2097 BUG_ON(sbio->page_count > SCRUB_PAGES_PER_RD_BIO);
2097 if (sbio->err) { 2098 if (sbio->err) {
2098 for (i = 0; i < sbio->page_count; i++) { 2099 for (i = 0; i < sbio->page_count; i++) {
2099 struct scrub_page *spage = sbio->pagev[i]; 2100 struct scrub_page *spage = sbio->pagev[i];
2100 2101
2101 spage->io_error = 1; 2102 spage->io_error = 1;
2102 spage->sblock->no_io_error_seen = 0; 2103 spage->sblock->no_io_error_seen = 0;
2103 } 2104 }
2104 } 2105 }
2105 2106
2106 /* now complete the scrub_block items that have all pages completed */ 2107 /* now complete the scrub_block items that have all pages completed */
2107 for (i = 0; i < sbio->page_count; i++) { 2108 for (i = 0; i < sbio->page_count; i++) {
2108 struct scrub_page *spage = sbio->pagev[i]; 2109 struct scrub_page *spage = sbio->pagev[i];
2109 struct scrub_block *sblock = spage->sblock; 2110 struct scrub_block *sblock = spage->sblock;
2110 2111
2111 if (atomic_dec_and_test(&sblock->outstanding_pages)) 2112 if (atomic_dec_and_test(&sblock->outstanding_pages))
2112 scrub_block_complete(sblock); 2113 scrub_block_complete(sblock);
2113 scrub_block_put(sblock); 2114 scrub_block_put(sblock);
2114 } 2115 }
2115 2116
2116 bio_put(sbio->bio); 2117 bio_put(sbio->bio);
2117 sbio->bio = NULL; 2118 sbio->bio = NULL;
2118 spin_lock(&sctx->list_lock); 2119 spin_lock(&sctx->list_lock);
2119 sbio->next_free = sctx->first_free; 2120 sbio->next_free = sctx->first_free;
2120 sctx->first_free = sbio->index; 2121 sctx->first_free = sbio->index;
2121 spin_unlock(&sctx->list_lock); 2122 spin_unlock(&sctx->list_lock);
2122 2123
2123 if (sctx->is_dev_replace && 2124 if (sctx->is_dev_replace &&
2124 atomic_read(&sctx->wr_ctx.flush_all_writes)) { 2125 atomic_read(&sctx->wr_ctx.flush_all_writes)) {
2125 mutex_lock(&sctx->wr_ctx.wr_lock); 2126 mutex_lock(&sctx->wr_ctx.wr_lock);
2126 scrub_wr_submit(sctx); 2127 scrub_wr_submit(sctx);
2127 mutex_unlock(&sctx->wr_ctx.wr_lock); 2128 mutex_unlock(&sctx->wr_ctx.wr_lock);
2128 } 2129 }
2129 2130
2130 scrub_pending_bio_dec(sctx); 2131 scrub_pending_bio_dec(sctx);
2131 } 2132 }
2132 2133
2133 static void scrub_block_complete(struct scrub_block *sblock) 2134 static void scrub_block_complete(struct scrub_block *sblock)
2134 { 2135 {
2135 if (!sblock->no_io_error_seen) { 2136 if (!sblock->no_io_error_seen) {
2136 scrub_handle_errored_block(sblock); 2137 scrub_handle_errored_block(sblock);
2137 } else { 2138 } else {
2138 /* 2139 /*
2139 * if has checksum error, write via repair mechanism in 2140 * if has checksum error, write via repair mechanism in
2140 * dev replace case, otherwise write here in dev replace 2141 * dev replace case, otherwise write here in dev replace
2141 * case. 2142 * case.
2142 */ 2143 */
2143 if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace) 2144 if (!scrub_checksum(sblock) && sblock->sctx->is_dev_replace)
2144 scrub_write_block_to_dev_replace(sblock); 2145 scrub_write_block_to_dev_replace(sblock);
2145 } 2146 }
2146 } 2147 }
2147 2148
2148 static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len, 2149 static int scrub_find_csum(struct scrub_ctx *sctx, u64 logical, u64 len,
2149 u8 *csum) 2150 u8 *csum)
2150 { 2151 {
2151 struct btrfs_ordered_sum *sum = NULL; 2152 struct btrfs_ordered_sum *sum = NULL;
2152 unsigned long index; 2153 unsigned long index;
2153 unsigned long num_sectors; 2154 unsigned long num_sectors;
2154 2155
2155 while (!list_empty(&sctx->csum_list)) { 2156 while (!list_empty(&sctx->csum_list)) {
2156 sum = list_first_entry(&sctx->csum_list, 2157 sum = list_first_entry(&sctx->csum_list,
2157 struct btrfs_ordered_sum, list); 2158 struct btrfs_ordered_sum, list);
2158 if (sum->bytenr > logical) 2159 if (sum->bytenr > logical)
2159 return 0; 2160 return 0;
2160 if (sum->bytenr + sum->len > logical) 2161 if (sum->bytenr + sum->len > logical)
2161 break; 2162 break;
2162 2163
2163 ++sctx->stat.csum_discards; 2164 ++sctx->stat.csum_discards;
2164 list_del(&sum->list); 2165 list_del(&sum->list);
2165 kfree(sum); 2166 kfree(sum);
2166 sum = NULL; 2167 sum = NULL;
2167 } 2168 }
2168 if (!sum) 2169 if (!sum)
2169 return 0; 2170 return 0;
2170 2171
2171 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize; 2172 index = ((u32)(logical - sum->bytenr)) / sctx->sectorsize;
2172 num_sectors = sum->len / sctx->sectorsize; 2173 num_sectors = sum->len / sctx->sectorsize;
2173 memcpy(csum, sum->sums + index, sctx->csum_size); 2174 memcpy(csum, sum->sums + index, sctx->csum_size);
2174 if (index == num_sectors - 1) { 2175 if (index == num_sectors - 1) {
2175 list_del(&sum->list); 2176 list_del(&sum->list);
2176 kfree(sum); 2177 kfree(sum);
2177 } 2178 }
2178 return 1; 2179 return 1;
2179 } 2180 }
2180 2181
2181 /* scrub extent tries to collect up to 64 kB for each bio */ 2182 /* scrub extent tries to collect up to 64 kB for each bio */
2182 static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len, 2183 static int scrub_extent(struct scrub_ctx *sctx, u64 logical, u64 len,
2183 u64 physical, struct btrfs_device *dev, u64 flags, 2184 u64 physical, struct btrfs_device *dev, u64 flags,
2184 u64 gen, int mirror_num, u64 physical_for_dev_replace) 2185 u64 gen, int mirror_num, u64 physical_for_dev_replace)
2185 { 2186 {
2186 int ret; 2187 int ret;
2187 u8 csum[BTRFS_CSUM_SIZE]; 2188 u8 csum[BTRFS_CSUM_SIZE];
2188 u32 blocksize; 2189 u32 blocksize;
2189 2190
2190 if (flags & BTRFS_EXTENT_FLAG_DATA) { 2191 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2191 blocksize = sctx->sectorsize; 2192 blocksize = sctx->sectorsize;
2192 spin_lock(&sctx->stat_lock); 2193 spin_lock(&sctx->stat_lock);
2193 sctx->stat.data_extents_scrubbed++; 2194 sctx->stat.data_extents_scrubbed++;
2194 sctx->stat.data_bytes_scrubbed += len; 2195 sctx->stat.data_bytes_scrubbed += len;
2195 spin_unlock(&sctx->stat_lock); 2196 spin_unlock(&sctx->stat_lock);
2196 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 2197 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2197 WARN_ON(sctx->nodesize != sctx->leafsize); 2198 WARN_ON(sctx->nodesize != sctx->leafsize);
2198 blocksize = sctx->nodesize; 2199 blocksize = sctx->nodesize;
2199 spin_lock(&sctx->stat_lock); 2200 spin_lock(&sctx->stat_lock);
2200 sctx->stat.tree_extents_scrubbed++; 2201 sctx->stat.tree_extents_scrubbed++;
2201 sctx->stat.tree_bytes_scrubbed += len; 2202 sctx->stat.tree_bytes_scrubbed += len;
2202 spin_unlock(&sctx->stat_lock); 2203 spin_unlock(&sctx->stat_lock);
2203 } else { 2204 } else {
2204 blocksize = sctx->sectorsize; 2205 blocksize = sctx->sectorsize;
2205 WARN_ON(1); 2206 WARN_ON(1);
2206 } 2207 }
2207 2208
2208 while (len) { 2209 while (len) {
2209 u64 l = min_t(u64, len, blocksize); 2210 u64 l = min_t(u64, len, blocksize);
2210 int have_csum = 0; 2211 int have_csum = 0;
2211 2212
2212 if (flags & BTRFS_EXTENT_FLAG_DATA) { 2213 if (flags & BTRFS_EXTENT_FLAG_DATA) {
2213 /* push csums to sbio */ 2214 /* push csums to sbio */
2214 have_csum = scrub_find_csum(sctx, logical, l, csum); 2215 have_csum = scrub_find_csum(sctx, logical, l, csum);
2215 if (have_csum == 0) 2216 if (have_csum == 0)
2216 ++sctx->stat.no_csum; 2217 ++sctx->stat.no_csum;
2217 if (sctx->is_dev_replace && !have_csum) { 2218 if (sctx->is_dev_replace && !have_csum) {
2218 ret = copy_nocow_pages(sctx, logical, l, 2219 ret = copy_nocow_pages(sctx, logical, l,
2219 mirror_num, 2220 mirror_num,
2220 physical_for_dev_replace); 2221 physical_for_dev_replace);
2221 goto behind_scrub_pages; 2222 goto behind_scrub_pages;
2222 } 2223 }
2223 } 2224 }
2224 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen, 2225 ret = scrub_pages(sctx, logical, l, physical, dev, flags, gen,
2225 mirror_num, have_csum ? csum : NULL, 0, 2226 mirror_num, have_csum ? csum : NULL, 0,
2226 physical_for_dev_replace); 2227 physical_for_dev_replace);
2227 behind_scrub_pages: 2228 behind_scrub_pages:
2228 if (ret) 2229 if (ret)
2229 return ret; 2230 return ret;
2230 len -= l; 2231 len -= l;
2231 logical += l; 2232 logical += l;
2232 physical += l; 2233 physical += l;
2233 physical_for_dev_replace += l; 2234 physical_for_dev_replace += l;
2234 } 2235 }
2235 return 0; 2236 return 0;
2236 } 2237 }
2237 2238
2238 /* 2239 /*
2239 * Given a physical address, this will calculate it's 2240 * Given a physical address, this will calculate it's
2240 * logical offset. if this is a parity stripe, it will return 2241 * logical offset. if this is a parity stripe, it will return
2241 * the most left data stripe's logical offset. 2242 * the most left data stripe's logical offset.
2242 * 2243 *
2243 * return 0 if it is a data stripe, 1 means parity stripe. 2244 * return 0 if it is a data stripe, 1 means parity stripe.
2244 */ 2245 */
2245 static int get_raid56_logic_offset(u64 physical, int num, 2246 static int get_raid56_logic_offset(u64 physical, int num,
2246 struct map_lookup *map, u64 *offset) 2247 struct map_lookup *map, u64 *offset)
2247 { 2248 {
2248 int i; 2249 int i;
2249 int j = 0; 2250 int j = 0;
2250 u64 stripe_nr; 2251 u64 stripe_nr;
2251 u64 last_offset; 2252 u64 last_offset;
2252 int stripe_index; 2253 int stripe_index;
2253 int rot; 2254 int rot;
2254 2255
2255 last_offset = (physical - map->stripes[num].physical) * 2256 last_offset = (physical - map->stripes[num].physical) *
2256 nr_data_stripes(map); 2257 nr_data_stripes(map);
2257 *offset = last_offset; 2258 *offset = last_offset;
2258 for (i = 0; i < nr_data_stripes(map); i++) { 2259 for (i = 0; i < nr_data_stripes(map); i++) {
2259 *offset = last_offset + i * map->stripe_len; 2260 *offset = last_offset + i * map->stripe_len;
2260 2261
2261 stripe_nr = *offset; 2262 stripe_nr = *offset;
2262 do_div(stripe_nr, map->stripe_len); 2263 do_div(stripe_nr, map->stripe_len);
2263 do_div(stripe_nr, nr_data_stripes(map)); 2264 do_div(stripe_nr, nr_data_stripes(map));
2264 2265
2265 /* Work out the disk rotation on this stripe-set */ 2266 /* Work out the disk rotation on this stripe-set */
2266 rot = do_div(stripe_nr, map->num_stripes); 2267 rot = do_div(stripe_nr, map->num_stripes);
2267 /* calculate which stripe this data locates */ 2268 /* calculate which stripe this data locates */
2268 rot += i; 2269 rot += i;
2269 stripe_index = rot % map->num_stripes; 2270 stripe_index = rot % map->num_stripes;
2270 if (stripe_index == num) 2271 if (stripe_index == num)
2271 return 0; 2272 return 0;
2272 if (stripe_index < num) 2273 if (stripe_index < num)
2273 j++; 2274 j++;
2274 } 2275 }
2275 *offset = last_offset + j * map->stripe_len; 2276 *offset = last_offset + j * map->stripe_len;
2276 return 1; 2277 return 1;
2277 } 2278 }
2278 2279
2279 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx, 2280 static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
2280 struct map_lookup *map, 2281 struct map_lookup *map,
2281 struct btrfs_device *scrub_dev, 2282 struct btrfs_device *scrub_dev,
2282 int num, u64 base, u64 length, 2283 int num, u64 base, u64 length,
2283 int is_dev_replace) 2284 int is_dev_replace)
2284 { 2285 {
2285 struct btrfs_path *path; 2286 struct btrfs_path *path;
2286 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; 2287 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
2287 struct btrfs_root *root = fs_info->extent_root; 2288 struct btrfs_root *root = fs_info->extent_root;
2288 struct btrfs_root *csum_root = fs_info->csum_root; 2289 struct btrfs_root *csum_root = fs_info->csum_root;
2289 struct btrfs_extent_item *extent; 2290 struct btrfs_extent_item *extent;
2290 struct blk_plug plug; 2291 struct blk_plug plug;
2291 u64 flags; 2292 u64 flags;
2292 int ret; 2293 int ret;
2293 int slot; 2294 int slot;
2294 u64 nstripes; 2295 u64 nstripes;
2295 struct extent_buffer *l; 2296 struct extent_buffer *l;
2296 struct btrfs_key key; 2297 struct btrfs_key key;
2297 u64 physical; 2298 u64 physical;
2298 u64 logical; 2299 u64 logical;
2299 u64 logic_end; 2300 u64 logic_end;
2300 u64 physical_end; 2301 u64 physical_end;
2301 u64 generation; 2302 u64 generation;
2302 int mirror_num; 2303 int mirror_num;
2303 struct reada_control *reada1; 2304 struct reada_control *reada1;
2304 struct reada_control *reada2; 2305 struct reada_control *reada2;
2305 struct btrfs_key key_start; 2306 struct btrfs_key key_start;
2306 struct btrfs_key key_end; 2307 struct btrfs_key key_end;
2307 u64 increment = map->stripe_len; 2308 u64 increment = map->stripe_len;
2308 u64 offset; 2309 u64 offset;
2309 u64 extent_logical; 2310 u64 extent_logical;
2310 u64 extent_physical; 2311 u64 extent_physical;
2311 u64 extent_len; 2312 u64 extent_len;
2312 struct btrfs_device *extent_dev; 2313 struct btrfs_device *extent_dev;
2313 int extent_mirror_num; 2314 int extent_mirror_num;
2314 int stop_loop = 0; 2315 int stop_loop = 0;
2315 2316
2316 nstripes = length; 2317 nstripes = length;
2317 physical = map->stripes[num].physical; 2318 physical = map->stripes[num].physical;
2318 offset = 0; 2319 offset = 0;
2319 do_div(nstripes, map->stripe_len); 2320 do_div(nstripes, map->stripe_len);
2320 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 2321 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2321 offset = map->stripe_len * num; 2322 offset = map->stripe_len * num;
2322 increment = map->stripe_len * map->num_stripes; 2323 increment = map->stripe_len * map->num_stripes;
2323 mirror_num = 1; 2324 mirror_num = 1;
2324 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 2325 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2325 int factor = map->num_stripes / map->sub_stripes; 2326 int factor = map->num_stripes / map->sub_stripes;
2326 offset = map->stripe_len * (num / map->sub_stripes); 2327 offset = map->stripe_len * (num / map->sub_stripes);
2327 increment = map->stripe_len * factor; 2328 increment = map->stripe_len * factor;
2328 mirror_num = num % map->sub_stripes + 1; 2329 mirror_num = num % map->sub_stripes + 1;
2329 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { 2330 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2330 increment = map->stripe_len; 2331 increment = map->stripe_len;
2331 mirror_num = num % map->num_stripes + 1; 2332 mirror_num = num % map->num_stripes + 1;
2332 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 2333 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2333 increment = map->stripe_len; 2334 increment = map->stripe_len;
2334 mirror_num = num % map->num_stripes + 1; 2335 mirror_num = num % map->num_stripes + 1;
2335 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | 2336 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2336 BTRFS_BLOCK_GROUP_RAID6)) { 2337 BTRFS_BLOCK_GROUP_RAID6)) {
2337 get_raid56_logic_offset(physical, num, map, &offset); 2338 get_raid56_logic_offset(physical, num, map, &offset);
2338 increment = map->stripe_len * nr_data_stripes(map); 2339 increment = map->stripe_len * nr_data_stripes(map);
2339 mirror_num = 1; 2340 mirror_num = 1;
2340 } else { 2341 } else {
2341 increment = map->stripe_len; 2342 increment = map->stripe_len;
2342 mirror_num = 1; 2343 mirror_num = 1;
2343 } 2344 }
2344 2345
2345 path = btrfs_alloc_path(); 2346 path = btrfs_alloc_path();
2346 if (!path) 2347 if (!path)
2347 return -ENOMEM; 2348 return -ENOMEM;
2348 2349
2349 /* 2350 /*
2350 * work on commit root. The related disk blocks are static as 2351 * work on commit root. The related disk blocks are static as
2351 * long as COW is applied. This means, it is save to rewrite 2352 * long as COW is applied. This means, it is save to rewrite
2352 * them to repair disk errors without any race conditions 2353 * them to repair disk errors without any race conditions
2353 */ 2354 */
2354 path->search_commit_root = 1; 2355 path->search_commit_root = 1;
2355 path->skip_locking = 1; 2356 path->skip_locking = 1;
2356 2357
2357 /* 2358 /*
2358 * trigger the readahead for extent tree csum tree and wait for 2359 * trigger the readahead for extent tree csum tree and wait for
2359 * completion. During readahead, the scrub is officially paused 2360 * completion. During readahead, the scrub is officially paused
2360 * to not hold off transaction commits 2361 * to not hold off transaction commits
2361 */ 2362 */
2362 logical = base + offset; 2363 logical = base + offset;
2363 physical_end = physical + nstripes * map->stripe_len; 2364 physical_end = physical + nstripes * map->stripe_len;
2364 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | 2365 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2365 BTRFS_BLOCK_GROUP_RAID6)) { 2366 BTRFS_BLOCK_GROUP_RAID6)) {
2366 get_raid56_logic_offset(physical_end, num, 2367 get_raid56_logic_offset(physical_end, num,
2367 map, &logic_end); 2368 map, &logic_end);
2368 logic_end += base; 2369 logic_end += base;
2369 } else { 2370 } else {
2370 logic_end = logical + increment * nstripes; 2371 logic_end = logical + increment * nstripes;
2371 } 2372 }
2372 wait_event(sctx->list_wait, 2373 wait_event(sctx->list_wait,
2373 atomic_read(&sctx->bios_in_flight) == 0); 2374 atomic_read(&sctx->bios_in_flight) == 0);
2374 scrub_blocked_if_needed(fs_info); 2375 scrub_blocked_if_needed(fs_info);
2375 2376
2376 /* FIXME it might be better to start readahead at commit root */ 2377 /* FIXME it might be better to start readahead at commit root */
2377 key_start.objectid = logical; 2378 key_start.objectid = logical;
2378 key_start.type = BTRFS_EXTENT_ITEM_KEY; 2379 key_start.type = BTRFS_EXTENT_ITEM_KEY;
2379 key_start.offset = (u64)0; 2380 key_start.offset = (u64)0;
2380 key_end.objectid = logic_end; 2381 key_end.objectid = logic_end;
2381 key_end.type = BTRFS_METADATA_ITEM_KEY; 2382 key_end.type = BTRFS_METADATA_ITEM_KEY;
2382 key_end.offset = (u64)-1; 2383 key_end.offset = (u64)-1;
2383 reada1 = btrfs_reada_add(root, &key_start, &key_end); 2384 reada1 = btrfs_reada_add(root, &key_start, &key_end);
2384 2385
2385 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 2386 key_start.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2386 key_start.type = BTRFS_EXTENT_CSUM_KEY; 2387 key_start.type = BTRFS_EXTENT_CSUM_KEY;
2387 key_start.offset = logical; 2388 key_start.offset = logical;
2388 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID; 2389 key_end.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
2389 key_end.type = BTRFS_EXTENT_CSUM_KEY; 2390 key_end.type = BTRFS_EXTENT_CSUM_KEY;
2390 key_end.offset = logic_end; 2391 key_end.offset = logic_end;
2391 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end); 2392 reada2 = btrfs_reada_add(csum_root, &key_start, &key_end);
2392 2393
2393 if (!IS_ERR(reada1)) 2394 if (!IS_ERR(reada1))
2394 btrfs_reada_wait(reada1); 2395 btrfs_reada_wait(reada1);
2395 if (!IS_ERR(reada2)) 2396 if (!IS_ERR(reada2))
2396 btrfs_reada_wait(reada2); 2397 btrfs_reada_wait(reada2);
2397 2398
2398 2399
2399 /* 2400 /*
2400 * collect all data csums for the stripe to avoid seeking during 2401 * collect all data csums for the stripe to avoid seeking during
2401 * the scrub. This might currently (crc32) end up to be about 1MB 2402 * the scrub. This might currently (crc32) end up to be about 1MB
2402 */ 2403 */
2403 blk_start_plug(&plug); 2404 blk_start_plug(&plug);
2404 2405
2405 /* 2406 /*
2406 * now find all extents for each stripe and scrub them 2407 * now find all extents for each stripe and scrub them
2407 */ 2408 */
2408 ret = 0; 2409 ret = 0;
2409 while (physical < physical_end) { 2410 while (physical < physical_end) {
2410 /* for raid56, we skip parity stripe */ 2411 /* for raid56, we skip parity stripe */
2411 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | 2412 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2412 BTRFS_BLOCK_GROUP_RAID6)) { 2413 BTRFS_BLOCK_GROUP_RAID6)) {
2413 ret = get_raid56_logic_offset(physical, num, 2414 ret = get_raid56_logic_offset(physical, num,
2414 map, &logical); 2415 map, &logical);
2415 logical += base; 2416 logical += base;
2416 if (ret) 2417 if (ret)
2417 goto skip; 2418 goto skip;
2418 } 2419 }
2419 /* 2420 /*
2420 * canceled? 2421 * canceled?
2421 */ 2422 */
2422 if (atomic_read(&fs_info->scrub_cancel_req) || 2423 if (atomic_read(&fs_info->scrub_cancel_req) ||
2423 atomic_read(&sctx->cancel_req)) { 2424 atomic_read(&sctx->cancel_req)) {
2424 ret = -ECANCELED; 2425 ret = -ECANCELED;
2425 goto out; 2426 goto out;
2426 } 2427 }
2427 /* 2428 /*
2428 * check to see if we have to pause 2429 * check to see if we have to pause
2429 */ 2430 */
2430 if (atomic_read(&fs_info->scrub_pause_req)) { 2431 if (atomic_read(&fs_info->scrub_pause_req)) {
2431 /* push queued extents */ 2432 /* push queued extents */
2432 atomic_set(&sctx->wr_ctx.flush_all_writes, 1); 2433 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2433 scrub_submit(sctx); 2434 scrub_submit(sctx);
2434 mutex_lock(&sctx->wr_ctx.wr_lock); 2435 mutex_lock(&sctx->wr_ctx.wr_lock);
2435 scrub_wr_submit(sctx); 2436 scrub_wr_submit(sctx);
2436 mutex_unlock(&sctx->wr_ctx.wr_lock); 2437 mutex_unlock(&sctx->wr_ctx.wr_lock);
2437 wait_event(sctx->list_wait, 2438 wait_event(sctx->list_wait,
2438 atomic_read(&sctx->bios_in_flight) == 0); 2439 atomic_read(&sctx->bios_in_flight) == 0);
2439 atomic_set(&sctx->wr_ctx.flush_all_writes, 0); 2440 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2440 scrub_blocked_if_needed(fs_info); 2441 scrub_blocked_if_needed(fs_info);
2441 } 2442 }
2442 2443
2443 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) 2444 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2444 key.type = BTRFS_METADATA_ITEM_KEY; 2445 key.type = BTRFS_METADATA_ITEM_KEY;
2445 else 2446 else
2446 key.type = BTRFS_EXTENT_ITEM_KEY; 2447 key.type = BTRFS_EXTENT_ITEM_KEY;
2447 key.objectid = logical; 2448 key.objectid = logical;
2448 key.offset = (u64)-1; 2449 key.offset = (u64)-1;
2449 2450
2450 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2451 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2451 if (ret < 0) 2452 if (ret < 0)
2452 goto out; 2453 goto out;
2453 2454
2454 if (ret > 0) { 2455 if (ret > 0) {
2455 ret = btrfs_previous_extent_item(root, path, 0); 2456 ret = btrfs_previous_extent_item(root, path, 0);
2456 if (ret < 0) 2457 if (ret < 0)
2457 goto out; 2458 goto out;
2458 if (ret > 0) { 2459 if (ret > 0) {
2459 /* there's no smaller item, so stick with the 2460 /* there's no smaller item, so stick with the
2460 * larger one */ 2461 * larger one */
2461 btrfs_release_path(path); 2462 btrfs_release_path(path);
2462 ret = btrfs_search_slot(NULL, root, &key, 2463 ret = btrfs_search_slot(NULL, root, &key,
2463 path, 0, 0); 2464 path, 0, 0);
2464 if (ret < 0) 2465 if (ret < 0)
2465 goto out; 2466 goto out;
2466 } 2467 }
2467 } 2468 }
2468 2469
2469 stop_loop = 0; 2470 stop_loop = 0;
2470 while (1) { 2471 while (1) {
2471 u64 bytes; 2472 u64 bytes;
2472 2473
2473 l = path->nodes[0]; 2474 l = path->nodes[0];
2474 slot = path->slots[0]; 2475 slot = path->slots[0];
2475 if (slot >= btrfs_header_nritems(l)) { 2476 if (slot >= btrfs_header_nritems(l)) {
2476 ret = btrfs_next_leaf(root, path); 2477 ret = btrfs_next_leaf(root, path);
2477 if (ret == 0) 2478 if (ret == 0)
2478 continue; 2479 continue;
2479 if (ret < 0) 2480 if (ret < 0)
2480 goto out; 2481 goto out;
2481 2482
2482 stop_loop = 1; 2483 stop_loop = 1;
2483 break; 2484 break;
2484 } 2485 }
2485 btrfs_item_key_to_cpu(l, &key, slot); 2486 btrfs_item_key_to_cpu(l, &key, slot);
2486 2487
2487 if (key.type == BTRFS_METADATA_ITEM_KEY) 2488 if (key.type == BTRFS_METADATA_ITEM_KEY)
2488 bytes = root->leafsize; 2489 bytes = root->leafsize;
2489 else 2490 else
2490 bytes = key.offset; 2491 bytes = key.offset;
2491 2492
2492 if (key.objectid + bytes <= logical) 2493 if (key.objectid + bytes <= logical)
2493 goto next; 2494 goto next;
2494 2495
2495 if (key.type != BTRFS_EXTENT_ITEM_KEY && 2496 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
2496 key.type != BTRFS_METADATA_ITEM_KEY) 2497 key.type != BTRFS_METADATA_ITEM_KEY)
2497 goto next; 2498 goto next;
2498 2499
2499 if (key.objectid >= logical + map->stripe_len) { 2500 if (key.objectid >= logical + map->stripe_len) {
2500 /* out of this device extent */ 2501 /* out of this device extent */
2501 if (key.objectid >= logic_end) 2502 if (key.objectid >= logic_end)
2502 stop_loop = 1; 2503 stop_loop = 1;
2503 break; 2504 break;
2504 } 2505 }
2505 2506
2506 extent = btrfs_item_ptr(l, slot, 2507 extent = btrfs_item_ptr(l, slot,
2507 struct btrfs_extent_item); 2508 struct btrfs_extent_item);
2508 flags = btrfs_extent_flags(l, extent); 2509 flags = btrfs_extent_flags(l, extent);
2509 generation = btrfs_extent_generation(l, extent); 2510 generation = btrfs_extent_generation(l, extent);
2510 2511
2511 if (key.objectid < logical && 2512 if (key.objectid < logical &&
2512 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) { 2513 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
2513 btrfs_err(fs_info, 2514 btrfs_err(fs_info,
2514 "scrub: tree block %llu spanning " 2515 "scrub: tree block %llu spanning "
2515 "stripes, ignored. logical=%llu", 2516 "stripes, ignored. logical=%llu",
2516 key.objectid, logical); 2517 key.objectid, logical);
2517 goto next; 2518 goto next;
2518 } 2519 }
2519 2520
2520 again: 2521 again:
2521 extent_logical = key.objectid; 2522 extent_logical = key.objectid;
2522 extent_len = bytes; 2523 extent_len = bytes;
2523 2524
2524 /* 2525 /*
2525 * trim extent to this stripe 2526 * trim extent to this stripe
2526 */ 2527 */
2527 if (extent_logical < logical) { 2528 if (extent_logical < logical) {
2528 extent_len -= logical - extent_logical; 2529 extent_len -= logical - extent_logical;
2529 extent_logical = logical; 2530 extent_logical = logical;
2530 } 2531 }
2531 if (extent_logical + extent_len > 2532 if (extent_logical + extent_len >
2532 logical + map->stripe_len) { 2533 logical + map->stripe_len) {
2533 extent_len = logical + map->stripe_len - 2534 extent_len = logical + map->stripe_len -
2534 extent_logical; 2535 extent_logical;
2535 } 2536 }
2536 2537
2537 extent_physical = extent_logical - logical + physical; 2538 extent_physical = extent_logical - logical + physical;
2538 extent_dev = scrub_dev; 2539 extent_dev = scrub_dev;
2539 extent_mirror_num = mirror_num; 2540 extent_mirror_num = mirror_num;
2540 if (is_dev_replace) 2541 if (is_dev_replace)
2541 scrub_remap_extent(fs_info, extent_logical, 2542 scrub_remap_extent(fs_info, extent_logical,
2542 extent_len, &extent_physical, 2543 extent_len, &extent_physical,
2543 &extent_dev, 2544 &extent_dev,
2544 &extent_mirror_num); 2545 &extent_mirror_num);
2545 2546
2546 ret = btrfs_lookup_csums_range(csum_root, logical, 2547 ret = btrfs_lookup_csums_range(csum_root, logical,
2547 logical + map->stripe_len - 1, 2548 logical + map->stripe_len - 1,
2548 &sctx->csum_list, 1); 2549 &sctx->csum_list, 1);
2549 if (ret) 2550 if (ret)
2550 goto out; 2551 goto out;
2551 2552
2552 ret = scrub_extent(sctx, extent_logical, extent_len, 2553 ret = scrub_extent(sctx, extent_logical, extent_len,
2553 extent_physical, extent_dev, flags, 2554 extent_physical, extent_dev, flags,
2554 generation, extent_mirror_num, 2555 generation, extent_mirror_num,
2555 extent_logical - logical + physical); 2556 extent_logical - logical + physical);
2556 if (ret) 2557 if (ret)
2557 goto out; 2558 goto out;
2558 2559
2559 scrub_free_csums(sctx); 2560 scrub_free_csums(sctx);
2560 if (extent_logical + extent_len < 2561 if (extent_logical + extent_len <
2561 key.objectid + bytes) { 2562 key.objectid + bytes) {
2562 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | 2563 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
2563 BTRFS_BLOCK_GROUP_RAID6)) { 2564 BTRFS_BLOCK_GROUP_RAID6)) {
2564 /* 2565 /*
2565 * loop until we find next data stripe 2566 * loop until we find next data stripe
2566 * or we have finished all stripes. 2567 * or we have finished all stripes.
2567 */ 2568 */
2568 do { 2569 do {
2569 physical += map->stripe_len; 2570 physical += map->stripe_len;
2570 ret = get_raid56_logic_offset( 2571 ret = get_raid56_logic_offset(
2571 physical, num, 2572 physical, num,
2572 map, &logical); 2573 map, &logical);
2573 logical += base; 2574 logical += base;
2574 } while (physical < physical_end && ret); 2575 } while (physical < physical_end && ret);
2575 } else { 2576 } else {
2576 physical += map->stripe_len; 2577 physical += map->stripe_len;
2577 logical += increment; 2578 logical += increment;
2578 } 2579 }
2579 if (logical < key.objectid + bytes) { 2580 if (logical < key.objectid + bytes) {
2580 cond_resched(); 2581 cond_resched();
2581 goto again; 2582 goto again;
2582 } 2583 }
2583 2584
2584 if (physical >= physical_end) { 2585 if (physical >= physical_end) {
2585 stop_loop = 1; 2586 stop_loop = 1;
2586 break; 2587 break;
2587 } 2588 }
2588 } 2589 }
2589 next: 2590 next:
2590 path->slots[0]++; 2591 path->slots[0]++;
2591 } 2592 }
2592 btrfs_release_path(path); 2593 btrfs_release_path(path);
2593 skip: 2594 skip:
2594 logical += increment; 2595 logical += increment;
2595 physical += map->stripe_len; 2596 physical += map->stripe_len;
2596 spin_lock(&sctx->stat_lock); 2597 spin_lock(&sctx->stat_lock);
2597 if (stop_loop) 2598 if (stop_loop)
2598 sctx->stat.last_physical = map->stripes[num].physical + 2599 sctx->stat.last_physical = map->stripes[num].physical +
2599 length; 2600 length;
2600 else 2601 else
2601 sctx->stat.last_physical = physical; 2602 sctx->stat.last_physical = physical;
2602 spin_unlock(&sctx->stat_lock); 2603 spin_unlock(&sctx->stat_lock);
2603 if (stop_loop) 2604 if (stop_loop)
2604 break; 2605 break;
2605 } 2606 }
2606 out: 2607 out:
2607 /* push queued extents */ 2608 /* push queued extents */
2608 scrub_submit(sctx); 2609 scrub_submit(sctx);
2609 mutex_lock(&sctx->wr_ctx.wr_lock); 2610 mutex_lock(&sctx->wr_ctx.wr_lock);
2610 scrub_wr_submit(sctx); 2611 scrub_wr_submit(sctx);
2611 mutex_unlock(&sctx->wr_ctx.wr_lock); 2612 mutex_unlock(&sctx->wr_ctx.wr_lock);
2612 2613
2613 blk_finish_plug(&plug); 2614 blk_finish_plug(&plug);
2614 btrfs_free_path(path); 2615 btrfs_free_path(path);
2615 return ret < 0 ? ret : 0; 2616 return ret < 0 ? ret : 0;
2616 } 2617 }
2617 2618
2618 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, 2619 static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx,
2619 struct btrfs_device *scrub_dev, 2620 struct btrfs_device *scrub_dev,
2620 u64 chunk_tree, u64 chunk_objectid, 2621 u64 chunk_tree, u64 chunk_objectid,
2621 u64 chunk_offset, u64 length, 2622 u64 chunk_offset, u64 length,
2622 u64 dev_offset, int is_dev_replace) 2623 u64 dev_offset, int is_dev_replace)
2623 { 2624 {
2624 struct btrfs_mapping_tree *map_tree = 2625 struct btrfs_mapping_tree *map_tree =
2625 &sctx->dev_root->fs_info->mapping_tree; 2626 &sctx->dev_root->fs_info->mapping_tree;
2626 struct map_lookup *map; 2627 struct map_lookup *map;
2627 struct extent_map *em; 2628 struct extent_map *em;
2628 int i; 2629 int i;
2629 int ret = 0; 2630 int ret = 0;
2630 2631
2631 read_lock(&map_tree->map_tree.lock); 2632 read_lock(&map_tree->map_tree.lock);
2632 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); 2633 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2633 read_unlock(&map_tree->map_tree.lock); 2634 read_unlock(&map_tree->map_tree.lock);
2634 2635
2635 if (!em) 2636 if (!em)
2636 return -EINVAL; 2637 return -EINVAL;
2637 2638
2638 map = (struct map_lookup *)em->bdev; 2639 map = (struct map_lookup *)em->bdev;
2639 if (em->start != chunk_offset) 2640 if (em->start != chunk_offset)
2640 goto out; 2641 goto out;
2641 2642
2642 if (em->len < length) 2643 if (em->len < length)
2643 goto out; 2644 goto out;
2644 2645
2645 for (i = 0; i < map->num_stripes; ++i) { 2646 for (i = 0; i < map->num_stripes; ++i) {
2646 if (map->stripes[i].dev->bdev == scrub_dev->bdev && 2647 if (map->stripes[i].dev->bdev == scrub_dev->bdev &&
2647 map->stripes[i].physical == dev_offset) { 2648 map->stripes[i].physical == dev_offset) {
2648 ret = scrub_stripe(sctx, map, scrub_dev, i, 2649 ret = scrub_stripe(sctx, map, scrub_dev, i,
2649 chunk_offset, length, 2650 chunk_offset, length,
2650 is_dev_replace); 2651 is_dev_replace);
2651 if (ret) 2652 if (ret)
2652 goto out; 2653 goto out;
2653 } 2654 }
2654 } 2655 }
2655 out: 2656 out:
2656 free_extent_map(em); 2657 free_extent_map(em);
2657 2658
2658 return ret; 2659 return ret;
2659 } 2660 }
2660 2661
2661 static noinline_for_stack 2662 static noinline_for_stack
2662 int scrub_enumerate_chunks(struct scrub_ctx *sctx, 2663 int scrub_enumerate_chunks(struct scrub_ctx *sctx,
2663 struct btrfs_device *scrub_dev, u64 start, u64 end, 2664 struct btrfs_device *scrub_dev, u64 start, u64 end,
2664 int is_dev_replace) 2665 int is_dev_replace)
2665 { 2666 {
2666 struct btrfs_dev_extent *dev_extent = NULL; 2667 struct btrfs_dev_extent *dev_extent = NULL;
2667 struct btrfs_path *path; 2668 struct btrfs_path *path;
2668 struct btrfs_root *root = sctx->dev_root; 2669 struct btrfs_root *root = sctx->dev_root;
2669 struct btrfs_fs_info *fs_info = root->fs_info; 2670 struct btrfs_fs_info *fs_info = root->fs_info;
2670 u64 length; 2671 u64 length;
2671 u64 chunk_tree; 2672 u64 chunk_tree;
2672 u64 chunk_objectid; 2673 u64 chunk_objectid;
2673 u64 chunk_offset; 2674 u64 chunk_offset;
2674 int ret; 2675 int ret;
2675 int slot; 2676 int slot;
2676 struct extent_buffer *l; 2677 struct extent_buffer *l;
2677 struct btrfs_key key; 2678 struct btrfs_key key;
2678 struct btrfs_key found_key; 2679 struct btrfs_key found_key;
2679 struct btrfs_block_group_cache *cache; 2680 struct btrfs_block_group_cache *cache;
2680 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 2681 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
2681 2682
2682 path = btrfs_alloc_path(); 2683 path = btrfs_alloc_path();
2683 if (!path) 2684 if (!path)
2684 return -ENOMEM; 2685 return -ENOMEM;
2685 2686
2686 path->reada = 2; 2687 path->reada = 2;
2687 path->search_commit_root = 1; 2688 path->search_commit_root = 1;
2688 path->skip_locking = 1; 2689 path->skip_locking = 1;
2689 2690
2690 key.objectid = scrub_dev->devid; 2691 key.objectid = scrub_dev->devid;
2691 key.offset = 0ull; 2692 key.offset = 0ull;
2692 key.type = BTRFS_DEV_EXTENT_KEY; 2693 key.type = BTRFS_DEV_EXTENT_KEY;
2693 2694
2694 while (1) { 2695 while (1) {
2695 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2696 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2696 if (ret < 0) 2697 if (ret < 0)
2697 break; 2698 break;
2698 if (ret > 0) { 2699 if (ret > 0) {
2699 if (path->slots[0] >= 2700 if (path->slots[0] >=
2700 btrfs_header_nritems(path->nodes[0])) { 2701 btrfs_header_nritems(path->nodes[0])) {
2701 ret = btrfs_next_leaf(root, path); 2702 ret = btrfs_next_leaf(root, path);
2702 if (ret) 2703 if (ret)
2703 break; 2704 break;
2704 } 2705 }
2705 } 2706 }
2706 2707
2707 l = path->nodes[0]; 2708 l = path->nodes[0];
2708 slot = path->slots[0]; 2709 slot = path->slots[0];
2709 2710
2710 btrfs_item_key_to_cpu(l, &found_key, slot); 2711 btrfs_item_key_to_cpu(l, &found_key, slot);
2711 2712
2712 if (found_key.objectid != scrub_dev->devid) 2713 if (found_key.objectid != scrub_dev->devid)
2713 break; 2714 break;
2714 2715
2715 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY) 2716 if (btrfs_key_type(&found_key) != BTRFS_DEV_EXTENT_KEY)
2716 break; 2717 break;
2717 2718
2718 if (found_key.offset >= end) 2719 if (found_key.offset >= end)
2719 break; 2720 break;
2720 2721
2721 if (found_key.offset < key.offset) 2722 if (found_key.offset < key.offset)
2722 break; 2723 break;
2723 2724
2724 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 2725 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2725 length = btrfs_dev_extent_length(l, dev_extent); 2726 length = btrfs_dev_extent_length(l, dev_extent);
2726 2727
2727 if (found_key.offset + length <= start) { 2728 if (found_key.offset + length <= start) {
2728 key.offset = found_key.offset + length; 2729 key.offset = found_key.offset + length;
2729 btrfs_release_path(path); 2730 btrfs_release_path(path);
2730 continue; 2731 continue;
2731 } 2732 }
2732 2733
2733 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); 2734 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2734 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); 2735 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2735 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 2736 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2736 2737
2737 /* 2738 /*
2738 * get a reference on the corresponding block group to prevent 2739 * get a reference on the corresponding block group to prevent
2739 * the chunk from going away while we scrub it 2740 * the chunk from going away while we scrub it
2740 */ 2741 */
2741 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 2742 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2742 if (!cache) { 2743 if (!cache) {
2743 ret = -ENOENT; 2744 ret = -ENOENT;
2744 break; 2745 break;
2745 } 2746 }
2746 dev_replace->cursor_right = found_key.offset + length; 2747 dev_replace->cursor_right = found_key.offset + length;
2747 dev_replace->cursor_left = found_key.offset; 2748 dev_replace->cursor_left = found_key.offset;
2748 dev_replace->item_needs_writeback = 1; 2749 dev_replace->item_needs_writeback = 1;
2749 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid, 2750 ret = scrub_chunk(sctx, scrub_dev, chunk_tree, chunk_objectid,
2750 chunk_offset, length, found_key.offset, 2751 chunk_offset, length, found_key.offset,
2751 is_dev_replace); 2752 is_dev_replace);
2752 2753
2753 /* 2754 /*
2754 * flush, submit all pending read and write bios, afterwards 2755 * flush, submit all pending read and write bios, afterwards
2755 * wait for them. 2756 * wait for them.
2756 * Note that in the dev replace case, a read request causes 2757 * Note that in the dev replace case, a read request causes
2757 * write requests that are submitted in the read completion 2758 * write requests that are submitted in the read completion
2758 * worker. Therefore in the current situation, it is required 2759 * worker. Therefore in the current situation, it is required
2759 * that all write requests are flushed, so that all read and 2760 * that all write requests are flushed, so that all read and
2760 * write requests are really completed when bios_in_flight 2761 * write requests are really completed when bios_in_flight
2761 * changes to 0. 2762 * changes to 0.
2762 */ 2763 */
2763 atomic_set(&sctx->wr_ctx.flush_all_writes, 1); 2764 atomic_set(&sctx->wr_ctx.flush_all_writes, 1);
2764 scrub_submit(sctx); 2765 scrub_submit(sctx);
2765 mutex_lock(&sctx->wr_ctx.wr_lock); 2766 mutex_lock(&sctx->wr_ctx.wr_lock);
2766 scrub_wr_submit(sctx); 2767 scrub_wr_submit(sctx);
2767 mutex_unlock(&sctx->wr_ctx.wr_lock); 2768 mutex_unlock(&sctx->wr_ctx.wr_lock);
2768 2769
2769 wait_event(sctx->list_wait, 2770 wait_event(sctx->list_wait,
2770 atomic_read(&sctx->bios_in_flight) == 0); 2771 atomic_read(&sctx->bios_in_flight) == 0);
2771 atomic_inc(&fs_info->scrubs_paused); 2772 atomic_inc(&fs_info->scrubs_paused);
2772 wake_up(&fs_info->scrub_pause_wait); 2773 wake_up(&fs_info->scrub_pause_wait);
2773 2774
2774 /* 2775 /*
2775 * must be called before we decrease @scrub_paused. 2776 * must be called before we decrease @scrub_paused.
2776 * make sure we don't block transaction commit while 2777 * make sure we don't block transaction commit while
2777 * we are waiting pending workers finished. 2778 * we are waiting pending workers finished.
2778 */ 2779 */
2779 wait_event(sctx->list_wait, 2780 wait_event(sctx->list_wait,
2780 atomic_read(&sctx->workers_pending) == 0); 2781 atomic_read(&sctx->workers_pending) == 0);
2781 atomic_set(&sctx->wr_ctx.flush_all_writes, 0); 2782 atomic_set(&sctx->wr_ctx.flush_all_writes, 0);
2782 2783
2783 mutex_lock(&fs_info->scrub_lock); 2784 mutex_lock(&fs_info->scrub_lock);
2784 __scrub_blocked_if_needed(fs_info); 2785 __scrub_blocked_if_needed(fs_info);
2785 atomic_dec(&fs_info->scrubs_paused); 2786 atomic_dec(&fs_info->scrubs_paused);
2786 mutex_unlock(&fs_info->scrub_lock); 2787 mutex_unlock(&fs_info->scrub_lock);
2787 wake_up(&fs_info->scrub_pause_wait); 2788 wake_up(&fs_info->scrub_pause_wait);
2788 2789
2789 btrfs_put_block_group(cache); 2790 btrfs_put_block_group(cache);
2790 if (ret) 2791 if (ret)
2791 break; 2792 break;
2792 if (is_dev_replace && 2793 if (is_dev_replace &&
2793 atomic64_read(&dev_replace->num_write_errors) > 0) { 2794 atomic64_read(&dev_replace->num_write_errors) > 0) {
2794 ret = -EIO; 2795 ret = -EIO;
2795 break; 2796 break;
2796 } 2797 }
2797 if (sctx->stat.malloc_errors > 0) { 2798 if (sctx->stat.malloc_errors > 0) {
2798 ret = -ENOMEM; 2799 ret = -ENOMEM;
2799 break; 2800 break;
2800 } 2801 }
2801 2802
2802 dev_replace->cursor_left = dev_replace->cursor_right; 2803 dev_replace->cursor_left = dev_replace->cursor_right;
2803 dev_replace->item_needs_writeback = 1; 2804 dev_replace->item_needs_writeback = 1;
2804 2805
2805 key.offset = found_key.offset + length; 2806 key.offset = found_key.offset + length;
2806 btrfs_release_path(path); 2807 btrfs_release_path(path);
2807 } 2808 }
2808 2809
2809 btrfs_free_path(path); 2810 btrfs_free_path(path);
2810 2811
2811 /* 2812 /*
2812 * ret can still be 1 from search_slot or next_leaf, 2813 * ret can still be 1 from search_slot or next_leaf,
2813 * that's not an error 2814 * that's not an error
2814 */ 2815 */
2815 return ret < 0 ? ret : 0; 2816 return ret < 0 ? ret : 0;
2816 } 2817 }
2817 2818
2818 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx, 2819 static noinline_for_stack int scrub_supers(struct scrub_ctx *sctx,
2819 struct btrfs_device *scrub_dev) 2820 struct btrfs_device *scrub_dev)
2820 { 2821 {
2821 int i; 2822 int i;
2822 u64 bytenr; 2823 u64 bytenr;
2823 u64 gen; 2824 u64 gen;
2824 int ret; 2825 int ret;
2825 struct btrfs_root *root = sctx->dev_root; 2826 struct btrfs_root *root = sctx->dev_root;
2826 2827
2827 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) 2828 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
2828 return -EIO; 2829 return -EIO;
2829 2830
2830 gen = root->fs_info->last_trans_committed; 2831 gen = root->fs_info->last_trans_committed;
2831 2832
2832 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 2833 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2833 bytenr = btrfs_sb_offset(i); 2834 bytenr = btrfs_sb_offset(i);
2834 if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes) 2835 if (bytenr + BTRFS_SUPER_INFO_SIZE > scrub_dev->total_bytes)
2835 break; 2836 break;
2836 2837
2837 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr, 2838 ret = scrub_pages(sctx, bytenr, BTRFS_SUPER_INFO_SIZE, bytenr,
2838 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i, 2839 scrub_dev, BTRFS_EXTENT_FLAG_SUPER, gen, i,
2839 NULL, 1, bytenr); 2840 NULL, 1, bytenr);
2840 if (ret) 2841 if (ret)
2841 return ret; 2842 return ret;
2842 } 2843 }
2843 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); 2844 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
2844 2845
2845 return 0; 2846 return 0;
2846 } 2847 }
2847 2848
2848 /* 2849 /*
2849 * get a reference count on fs_info->scrub_workers. start worker if necessary 2850 * get a reference count on fs_info->scrub_workers. start worker if necessary
2850 */ 2851 */
2851 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info, 2852 static noinline_for_stack int scrub_workers_get(struct btrfs_fs_info *fs_info,
2852 int is_dev_replace) 2853 int is_dev_replace)
2853 { 2854 {
2854 int ret = 0; 2855 int ret = 0;
2855 int flags = WQ_FREEZABLE | WQ_UNBOUND; 2856 int flags = WQ_FREEZABLE | WQ_UNBOUND;
2856 int max_active = fs_info->thread_pool_size; 2857 int max_active = fs_info->thread_pool_size;
2857 2858
2858 if (fs_info->scrub_workers_refcnt == 0) { 2859 if (fs_info->scrub_workers_refcnt == 0) {
2859 if (is_dev_replace) 2860 if (is_dev_replace)
2860 fs_info->scrub_workers = 2861 fs_info->scrub_workers =
2861 btrfs_alloc_workqueue("btrfs-scrub", flags, 2862 btrfs_alloc_workqueue("btrfs-scrub", flags,
2862 1, 4); 2863 1, 4);
2863 else 2864 else
2864 fs_info->scrub_workers = 2865 fs_info->scrub_workers =
2865 btrfs_alloc_workqueue("btrfs-scrub", flags, 2866 btrfs_alloc_workqueue("btrfs-scrub", flags,
2866 max_active, 4); 2867 max_active, 4);
2867 if (!fs_info->scrub_workers) { 2868 if (!fs_info->scrub_workers) {
2868 ret = -ENOMEM; 2869 ret = -ENOMEM;
2869 goto out; 2870 goto out;
2870 } 2871 }
2871 fs_info->scrub_wr_completion_workers = 2872 fs_info->scrub_wr_completion_workers =
2872 btrfs_alloc_workqueue("btrfs-scrubwrc", flags, 2873 btrfs_alloc_workqueue("btrfs-scrubwrc", flags,
2873 max_active, 2); 2874 max_active, 2);
2874 if (!fs_info->scrub_wr_completion_workers) { 2875 if (!fs_info->scrub_wr_completion_workers) {
2875 ret = -ENOMEM; 2876 ret = -ENOMEM;
2876 goto out; 2877 goto out;
2877 } 2878 }
2878 fs_info->scrub_nocow_workers = 2879 fs_info->scrub_nocow_workers =
2879 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0); 2880 btrfs_alloc_workqueue("btrfs-scrubnc", flags, 1, 0);
2880 if (!fs_info->scrub_nocow_workers) { 2881 if (!fs_info->scrub_nocow_workers) {
2881 ret = -ENOMEM; 2882 ret = -ENOMEM;
2882 goto out; 2883 goto out;
2883 } 2884 }
2884 } 2885 }
2885 ++fs_info->scrub_workers_refcnt; 2886 ++fs_info->scrub_workers_refcnt;
2886 out: 2887 out:
2887 return ret; 2888 return ret;
2888 } 2889 }
2889 2890
2890 static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info) 2891 static noinline_for_stack void scrub_workers_put(struct btrfs_fs_info *fs_info)
2891 { 2892 {
2892 if (--fs_info->scrub_workers_refcnt == 0) { 2893 if (--fs_info->scrub_workers_refcnt == 0) {
2893 btrfs_destroy_workqueue(fs_info->scrub_workers); 2894 btrfs_destroy_workqueue(fs_info->scrub_workers);
2894 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers); 2895 btrfs_destroy_workqueue(fs_info->scrub_wr_completion_workers);
2895 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers); 2896 btrfs_destroy_workqueue(fs_info->scrub_nocow_workers);
2896 } 2897 }
2897 WARN_ON(fs_info->scrub_workers_refcnt < 0); 2898 WARN_ON(fs_info->scrub_workers_refcnt < 0);
2898 } 2899 }
2899 2900
2900 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start, 2901 int btrfs_scrub_dev(struct btrfs_fs_info *fs_info, u64 devid, u64 start,
2901 u64 end, struct btrfs_scrub_progress *progress, 2902 u64 end, struct btrfs_scrub_progress *progress,
2902 int readonly, int is_dev_replace) 2903 int readonly, int is_dev_replace)
2903 { 2904 {
2904 struct scrub_ctx *sctx; 2905 struct scrub_ctx *sctx;
2905 int ret; 2906 int ret;
2906 struct btrfs_device *dev; 2907 struct btrfs_device *dev;
2907 2908
2908 if (btrfs_fs_closing(fs_info)) 2909 if (btrfs_fs_closing(fs_info))
2909 return -EINVAL; 2910 return -EINVAL;
2910 2911
2911 /* 2912 /*
2912 * check some assumptions 2913 * check some assumptions
2913 */ 2914 */
2914 if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) { 2915 if (fs_info->chunk_root->nodesize != fs_info->chunk_root->leafsize) {
2915 btrfs_err(fs_info, 2916 btrfs_err(fs_info,
2916 "scrub: size assumption nodesize == leafsize (%d == %d) fails", 2917 "scrub: size assumption nodesize == leafsize (%d == %d) fails",
2917 fs_info->chunk_root->nodesize, 2918 fs_info->chunk_root->nodesize,
2918 fs_info->chunk_root->leafsize); 2919 fs_info->chunk_root->leafsize);
2919 return -EINVAL; 2920 return -EINVAL;
2920 } 2921 }
2921 2922
2922 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) { 2923 if (fs_info->chunk_root->nodesize > BTRFS_STRIPE_LEN) {
2923 /* 2924 /*
2924 * in this case scrub is unable to calculate the checksum 2925 * in this case scrub is unable to calculate the checksum
2925 * the way scrub is implemented. Do not handle this 2926 * the way scrub is implemented. Do not handle this
2926 * situation at all because it won't ever happen. 2927 * situation at all because it won't ever happen.
2927 */ 2928 */
2928 btrfs_err(fs_info, 2929 btrfs_err(fs_info,
2929 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails", 2930 "scrub: size assumption nodesize <= BTRFS_STRIPE_LEN (%d <= %d) fails",
2930 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN); 2931 fs_info->chunk_root->nodesize, BTRFS_STRIPE_LEN);
2931 return -EINVAL; 2932 return -EINVAL;
2932 } 2933 }
2933 2934
2934 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) { 2935 if (fs_info->chunk_root->sectorsize != PAGE_SIZE) {
2935 /* not supported for data w/o checksums */ 2936 /* not supported for data w/o checksums */
2936 btrfs_err(fs_info, 2937 btrfs_err(fs_info,
2937 "scrub: size assumption sectorsize != PAGE_SIZE " 2938 "scrub: size assumption sectorsize != PAGE_SIZE "
2938 "(%d != %lu) fails", 2939 "(%d != %lu) fails",
2939 fs_info->chunk_root->sectorsize, PAGE_SIZE); 2940 fs_info->chunk_root->sectorsize, PAGE_SIZE);
2940 return -EINVAL; 2941 return -EINVAL;
2941 } 2942 }
2942 2943
2943 if (fs_info->chunk_root->nodesize > 2944 if (fs_info->chunk_root->nodesize >
2944 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK || 2945 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK ||
2945 fs_info->chunk_root->sectorsize > 2946 fs_info->chunk_root->sectorsize >
2946 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) { 2947 PAGE_SIZE * SCRUB_MAX_PAGES_PER_BLOCK) {
2947 /* 2948 /*
2948 * would exhaust the array bounds of pagev member in 2949 * would exhaust the array bounds of pagev member in
2949 * struct scrub_block 2950 * struct scrub_block
2950 */ 2951 */
2951 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize " 2952 btrfs_err(fs_info, "scrub: size assumption nodesize and sectorsize "
2952 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails", 2953 "<= SCRUB_MAX_PAGES_PER_BLOCK (%d <= %d && %d <= %d) fails",
2953 fs_info->chunk_root->nodesize, 2954 fs_info->chunk_root->nodesize,
2954 SCRUB_MAX_PAGES_PER_BLOCK, 2955 SCRUB_MAX_PAGES_PER_BLOCK,
2955 fs_info->chunk_root->sectorsize, 2956 fs_info->chunk_root->sectorsize,
2956 SCRUB_MAX_PAGES_PER_BLOCK); 2957 SCRUB_MAX_PAGES_PER_BLOCK);
2957 return -EINVAL; 2958 return -EINVAL;
2958 } 2959 }
2959 2960
2960 2961
2961 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2962 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2962 dev = btrfs_find_device(fs_info, devid, NULL, NULL); 2963 dev = btrfs_find_device(fs_info, devid, NULL, NULL);
2963 if (!dev || (dev->missing && !is_dev_replace)) { 2964 if (!dev || (dev->missing && !is_dev_replace)) {
2964 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2965 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2965 return -ENODEV; 2966 return -ENODEV;
2966 } 2967 }
2967 2968
2968 mutex_lock(&fs_info->scrub_lock); 2969 mutex_lock(&fs_info->scrub_lock);
2969 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) { 2970 if (!dev->in_fs_metadata || dev->is_tgtdev_for_dev_replace) {
2970 mutex_unlock(&fs_info->scrub_lock); 2971 mutex_unlock(&fs_info->scrub_lock);
2971 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2972 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2972 return -EIO; 2973 return -EIO;
2973 } 2974 }
2974 2975
2975 btrfs_dev_replace_lock(&fs_info->dev_replace); 2976 btrfs_dev_replace_lock(&fs_info->dev_replace);
2976 if (dev->scrub_device || 2977 if (dev->scrub_device ||
2977 (!is_dev_replace && 2978 (!is_dev_replace &&
2978 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) { 2979 btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))) {
2979 btrfs_dev_replace_unlock(&fs_info->dev_replace); 2980 btrfs_dev_replace_unlock(&fs_info->dev_replace);
2980 mutex_unlock(&fs_info->scrub_lock); 2981 mutex_unlock(&fs_info->scrub_lock);
2981 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2982 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2982 return -EINPROGRESS; 2983 return -EINPROGRESS;
2983 } 2984 }
2984 btrfs_dev_replace_unlock(&fs_info->dev_replace); 2985 btrfs_dev_replace_unlock(&fs_info->dev_replace);
2985 2986
2986 ret = scrub_workers_get(fs_info, is_dev_replace); 2987 ret = scrub_workers_get(fs_info, is_dev_replace);
2987 if (ret) { 2988 if (ret) {
2988 mutex_unlock(&fs_info->scrub_lock); 2989 mutex_unlock(&fs_info->scrub_lock);
2989 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2990 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2990 return ret; 2991 return ret;
2991 } 2992 }
2992 2993
2993 sctx = scrub_setup_ctx(dev, is_dev_replace); 2994 sctx = scrub_setup_ctx(dev, is_dev_replace);
2994 if (IS_ERR(sctx)) { 2995 if (IS_ERR(sctx)) {
2995 mutex_unlock(&fs_info->scrub_lock); 2996 mutex_unlock(&fs_info->scrub_lock);
2996 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2997 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2997 scrub_workers_put(fs_info); 2998 scrub_workers_put(fs_info);
2998 return PTR_ERR(sctx); 2999 return PTR_ERR(sctx);
2999 } 3000 }
3000 sctx->readonly = readonly; 3001 sctx->readonly = readonly;
3001 dev->scrub_device = sctx; 3002 dev->scrub_device = sctx;
3002 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3003 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3003 3004
3004 /* 3005 /*
3005 * checking @scrub_pause_req here, we can avoid 3006 * checking @scrub_pause_req here, we can avoid
3006 * race between committing transaction and scrubbing. 3007 * race between committing transaction and scrubbing.
3007 */ 3008 */
3008 __scrub_blocked_if_needed(fs_info); 3009 __scrub_blocked_if_needed(fs_info);
3009 atomic_inc(&fs_info->scrubs_running); 3010 atomic_inc(&fs_info->scrubs_running);
3010 mutex_unlock(&fs_info->scrub_lock); 3011 mutex_unlock(&fs_info->scrub_lock);
3011 3012
3012 if (!is_dev_replace) { 3013 if (!is_dev_replace) {
3013 /* 3014 /*
3014 * by holding device list mutex, we can 3015 * by holding device list mutex, we can
3015 * kick off writing super in log tree sync. 3016 * kick off writing super in log tree sync.
3016 */ 3017 */
3017 mutex_lock(&fs_info->fs_devices->device_list_mutex); 3018 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3018 ret = scrub_supers(sctx, dev); 3019 ret = scrub_supers(sctx, dev);
3019 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 3020 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3020 } 3021 }
3021 3022
3022 if (!ret) 3023 if (!ret)
3023 ret = scrub_enumerate_chunks(sctx, dev, start, end, 3024 ret = scrub_enumerate_chunks(sctx, dev, start, end,
3024 is_dev_replace); 3025 is_dev_replace);
3025 3026
3026 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0); 3027 wait_event(sctx->list_wait, atomic_read(&sctx->bios_in_flight) == 0);
3027 atomic_dec(&fs_info->scrubs_running); 3028 atomic_dec(&fs_info->scrubs_running);
3028 wake_up(&fs_info->scrub_pause_wait); 3029 wake_up(&fs_info->scrub_pause_wait);
3029 3030
3030 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0); 3031 wait_event(sctx->list_wait, atomic_read(&sctx->workers_pending) == 0);
3031 3032
3032 if (progress) 3033 if (progress)
3033 memcpy(progress, &sctx->stat, sizeof(*progress)); 3034 memcpy(progress, &sctx->stat, sizeof(*progress));
3034 3035
3035 mutex_lock(&fs_info->scrub_lock); 3036 mutex_lock(&fs_info->scrub_lock);
3036 dev->scrub_device = NULL; 3037 dev->scrub_device = NULL;
3037 scrub_workers_put(fs_info); 3038 scrub_workers_put(fs_info);
3038 mutex_unlock(&fs_info->scrub_lock); 3039 mutex_unlock(&fs_info->scrub_lock);
3039 3040
3040 scrub_free_ctx(sctx); 3041 scrub_free_ctx(sctx);
3041 3042
3042 return ret; 3043 return ret;
3043 } 3044 }
3044 3045
3045 void btrfs_scrub_pause(struct btrfs_root *root) 3046 void btrfs_scrub_pause(struct btrfs_root *root)
3046 { 3047 {
3047 struct btrfs_fs_info *fs_info = root->fs_info; 3048 struct btrfs_fs_info *fs_info = root->fs_info;
3048 3049
3049 mutex_lock(&fs_info->scrub_lock); 3050 mutex_lock(&fs_info->scrub_lock);
3050 atomic_inc(&fs_info->scrub_pause_req); 3051 atomic_inc(&fs_info->scrub_pause_req);
3051 while (atomic_read(&fs_info->scrubs_paused) != 3052 while (atomic_read(&fs_info->scrubs_paused) !=
3052 atomic_read(&fs_info->scrubs_running)) { 3053 atomic_read(&fs_info->scrubs_running)) {
3053 mutex_unlock(&fs_info->scrub_lock); 3054 mutex_unlock(&fs_info->scrub_lock);
3054 wait_event(fs_info->scrub_pause_wait, 3055 wait_event(fs_info->scrub_pause_wait,
3055 atomic_read(&fs_info->scrubs_paused) == 3056 atomic_read(&fs_info->scrubs_paused) ==
3056 atomic_read(&fs_info->scrubs_running)); 3057 atomic_read(&fs_info->scrubs_running));
3057 mutex_lock(&fs_info->scrub_lock); 3058 mutex_lock(&fs_info->scrub_lock);
3058 } 3059 }
3059 mutex_unlock(&fs_info->scrub_lock); 3060 mutex_unlock(&fs_info->scrub_lock);
3060 } 3061 }
3061 3062
3062 void btrfs_scrub_continue(struct btrfs_root *root) 3063 void btrfs_scrub_continue(struct btrfs_root *root)
3063 { 3064 {
3064 struct btrfs_fs_info *fs_info = root->fs_info; 3065 struct btrfs_fs_info *fs_info = root->fs_info;
3065 3066
3066 atomic_dec(&fs_info->scrub_pause_req); 3067 atomic_dec(&fs_info->scrub_pause_req);
3067 wake_up(&fs_info->scrub_pause_wait); 3068 wake_up(&fs_info->scrub_pause_wait);
3068 } 3069 }
3069 3070
3070 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info) 3071 int btrfs_scrub_cancel(struct btrfs_fs_info *fs_info)
3071 { 3072 {
3072 mutex_lock(&fs_info->scrub_lock); 3073 mutex_lock(&fs_info->scrub_lock);
3073 if (!atomic_read(&fs_info->scrubs_running)) { 3074 if (!atomic_read(&fs_info->scrubs_running)) {
3074 mutex_unlock(&fs_info->scrub_lock); 3075 mutex_unlock(&fs_info->scrub_lock);
3075 return -ENOTCONN; 3076 return -ENOTCONN;
3076 } 3077 }
3077 3078
3078 atomic_inc(&fs_info->scrub_cancel_req); 3079 atomic_inc(&fs_info->scrub_cancel_req);
3079 while (atomic_read(&fs_info->scrubs_running)) { 3080 while (atomic_read(&fs_info->scrubs_running)) {
3080 mutex_unlock(&fs_info->scrub_lock); 3081 mutex_unlock(&fs_info->scrub_lock);
3081 wait_event(fs_info->scrub_pause_wait, 3082 wait_event(fs_info->scrub_pause_wait,
3082 atomic_read(&fs_info->scrubs_running) == 0); 3083 atomic_read(&fs_info->scrubs_running) == 0);
3083 mutex_lock(&fs_info->scrub_lock); 3084 mutex_lock(&fs_info->scrub_lock);
3084 } 3085 }
3085 atomic_dec(&fs_info->scrub_cancel_req); 3086 atomic_dec(&fs_info->scrub_cancel_req);
3086 mutex_unlock(&fs_info->scrub_lock); 3087 mutex_unlock(&fs_info->scrub_lock);
3087 3088
3088 return 0; 3089 return 0;
3089 } 3090 }
3090 3091
3091 int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info, 3092 int btrfs_scrub_cancel_dev(struct btrfs_fs_info *fs_info,
3092 struct btrfs_device *dev) 3093 struct btrfs_device *dev)
3093 { 3094 {
3094 struct scrub_ctx *sctx; 3095 struct scrub_ctx *sctx;
3095 3096
3096 mutex_lock(&fs_info->scrub_lock); 3097 mutex_lock(&fs_info->scrub_lock);
3097 sctx = dev->scrub_device; 3098 sctx = dev->scrub_device;
3098 if (!sctx) { 3099 if (!sctx) {
3099 mutex_unlock(&fs_info->scrub_lock); 3100 mutex_unlock(&fs_info->scrub_lock);
3100 return -ENOTCONN; 3101 return -ENOTCONN;
3101 } 3102 }
3102 atomic_inc(&sctx->cancel_req); 3103 atomic_inc(&sctx->cancel_req);
3103 while (dev->scrub_device) { 3104 while (dev->scrub_device) {
3104 mutex_unlock(&fs_info->scrub_lock); 3105 mutex_unlock(&fs_info->scrub_lock);
3105 wait_event(fs_info->scrub_pause_wait, 3106 wait_event(fs_info->scrub_pause_wait,
3106 dev->scrub_device == NULL); 3107 dev->scrub_device == NULL);
3107 mutex_lock(&fs_info->scrub_lock); 3108 mutex_lock(&fs_info->scrub_lock);
3108 } 3109 }
3109 mutex_unlock(&fs_info->scrub_lock); 3110 mutex_unlock(&fs_info->scrub_lock);
3110 3111
3111 return 0; 3112 return 0;
3112 } 3113 }
3113 3114
3114 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid, 3115 int btrfs_scrub_progress(struct btrfs_root *root, u64 devid,
3115 struct btrfs_scrub_progress *progress) 3116 struct btrfs_scrub_progress *progress)
3116 { 3117 {
3117 struct btrfs_device *dev; 3118 struct btrfs_device *dev;
3118 struct scrub_ctx *sctx = NULL; 3119 struct scrub_ctx *sctx = NULL;
3119 3120
3120 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3121 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3121 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL); 3122 dev = btrfs_find_device(root->fs_info, devid, NULL, NULL);
3122 if (dev) 3123 if (dev)
3123 sctx = dev->scrub_device; 3124 sctx = dev->scrub_device;
3124 if (sctx) 3125 if (sctx)
3125 memcpy(progress, &sctx->stat, sizeof(*progress)); 3126 memcpy(progress, &sctx->stat, sizeof(*progress));
3126 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3127 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3127 3128
3128 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV; 3129 return dev ? (sctx ? 0 : -ENOTCONN) : -ENODEV;
3129 } 3130 }
3130 3131
3131 static void scrub_remap_extent(struct btrfs_fs_info *fs_info, 3132 static void scrub_remap_extent(struct btrfs_fs_info *fs_info,
3132 u64 extent_logical, u64 extent_len, 3133 u64 extent_logical, u64 extent_len,
3133 u64 *extent_physical, 3134 u64 *extent_physical,
3134 struct btrfs_device **extent_dev, 3135 struct btrfs_device **extent_dev,
3135 int *extent_mirror_num) 3136 int *extent_mirror_num)
3136 { 3137 {
3137 u64 mapped_length; 3138 u64 mapped_length;
3138 struct btrfs_bio *bbio = NULL; 3139 struct btrfs_bio *bbio = NULL;
3139 int ret; 3140 int ret;
3140 3141
3141 mapped_length = extent_len; 3142 mapped_length = extent_len;
3142 ret = btrfs_map_block(fs_info, READ, extent_logical, 3143 ret = btrfs_map_block(fs_info, READ, extent_logical,
3143 &mapped_length, &bbio, 0); 3144 &mapped_length, &bbio, 0);
3144 if (ret || !bbio || mapped_length < extent_len || 3145 if (ret || !bbio || mapped_length < extent_len ||
3145 !bbio->stripes[0].dev->bdev) { 3146 !bbio->stripes[0].dev->bdev) {
3146 kfree(bbio); 3147 kfree(bbio);
3147 return; 3148 return;
3148 } 3149 }
3149 3150
3150 *extent_physical = bbio->stripes[0].physical; 3151 *extent_physical = bbio->stripes[0].physical;
3151 *extent_mirror_num = bbio->mirror_num; 3152 *extent_mirror_num = bbio->mirror_num;
3152 *extent_dev = bbio->stripes[0].dev; 3153 *extent_dev = bbio->stripes[0].dev;
3153 kfree(bbio); 3154 kfree(bbio);
3154 } 3155 }
3155 3156
3156 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx, 3157 static int scrub_setup_wr_ctx(struct scrub_ctx *sctx,
3157 struct scrub_wr_ctx *wr_ctx, 3158 struct scrub_wr_ctx *wr_ctx,
3158 struct btrfs_fs_info *fs_info, 3159 struct btrfs_fs_info *fs_info,
3159 struct btrfs_device *dev, 3160 struct btrfs_device *dev,
3160 int is_dev_replace) 3161 int is_dev_replace)
3161 { 3162 {
3162 WARN_ON(wr_ctx->wr_curr_bio != NULL); 3163 WARN_ON(wr_ctx->wr_curr_bio != NULL);
3163 3164
3164 mutex_init(&wr_ctx->wr_lock); 3165 mutex_init(&wr_ctx->wr_lock);
3165 wr_ctx->wr_curr_bio = NULL; 3166 wr_ctx->wr_curr_bio = NULL;
3166 if (!is_dev_replace) 3167 if (!is_dev_replace)
3167 return 0; 3168 return 0;
3168 3169
3169 WARN_ON(!dev->bdev); 3170 WARN_ON(!dev->bdev);
3170 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO, 3171 wr_ctx->pages_per_wr_bio = min_t(int, SCRUB_PAGES_PER_WR_BIO,
3171 bio_get_nr_vecs(dev->bdev)); 3172 bio_get_nr_vecs(dev->bdev));
3172 wr_ctx->tgtdev = dev; 3173 wr_ctx->tgtdev = dev;
3173 atomic_set(&wr_ctx->flush_all_writes, 0); 3174 atomic_set(&wr_ctx->flush_all_writes, 0);
3174 return 0; 3175 return 0;
3175 } 3176 }
3176 3177
3177 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx) 3178 static void scrub_free_wr_ctx(struct scrub_wr_ctx *wr_ctx)
3178 { 3179 {
3179 mutex_lock(&wr_ctx->wr_lock); 3180 mutex_lock(&wr_ctx->wr_lock);
3180 kfree(wr_ctx->wr_curr_bio); 3181 kfree(wr_ctx->wr_curr_bio);
3181 wr_ctx->wr_curr_bio = NULL; 3182 wr_ctx->wr_curr_bio = NULL;
3182 mutex_unlock(&wr_ctx->wr_lock); 3183 mutex_unlock(&wr_ctx->wr_lock);
3183 } 3184 }
3184 3185
3185 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len, 3186 static int copy_nocow_pages(struct scrub_ctx *sctx, u64 logical, u64 len,
3186 int mirror_num, u64 physical_for_dev_replace) 3187 int mirror_num, u64 physical_for_dev_replace)
3187 { 3188 {
3188 struct scrub_copy_nocow_ctx *nocow_ctx; 3189 struct scrub_copy_nocow_ctx *nocow_ctx;
3189 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info; 3190 struct btrfs_fs_info *fs_info = sctx->dev_root->fs_info;
3190 3191
3191 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS); 3192 nocow_ctx = kzalloc(sizeof(*nocow_ctx), GFP_NOFS);
3192 if (!nocow_ctx) { 3193 if (!nocow_ctx) {
3193 spin_lock(&sctx->stat_lock); 3194 spin_lock(&sctx->stat_lock);
3194 sctx->stat.malloc_errors++; 3195 sctx->stat.malloc_errors++;
3195 spin_unlock(&sctx->stat_lock); 3196 spin_unlock(&sctx->stat_lock);
3196 return -ENOMEM; 3197 return -ENOMEM;
3197 } 3198 }
3198 3199
3199 scrub_pending_trans_workers_inc(sctx); 3200 scrub_pending_trans_workers_inc(sctx);
3200 3201
3201 nocow_ctx->sctx = sctx; 3202 nocow_ctx->sctx = sctx;
3202 nocow_ctx->logical = logical; 3203 nocow_ctx->logical = logical;
3203 nocow_ctx->len = len; 3204 nocow_ctx->len = len;
3204 nocow_ctx->mirror_num = mirror_num; 3205 nocow_ctx->mirror_num = mirror_num;
3205 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace; 3206 nocow_ctx->physical_for_dev_replace = physical_for_dev_replace;
3206 btrfs_init_work(&nocow_ctx->work, copy_nocow_pages_worker, NULL, NULL); 3207 btrfs_init_work(&nocow_ctx->work, copy_nocow_pages_worker, NULL, NULL);
3207 INIT_LIST_HEAD(&nocow_ctx->inodes); 3208 INIT_LIST_HEAD(&nocow_ctx->inodes);
3208 btrfs_queue_work(fs_info->scrub_nocow_workers, 3209 btrfs_queue_work(fs_info->scrub_nocow_workers,
3209 &nocow_ctx->work); 3210 &nocow_ctx->work);
3210 3211
3211 return 0; 3212 return 0;
3212 } 3213 }
3213 3214
3214 static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx) 3215 static int record_inode_for_nocow(u64 inum, u64 offset, u64 root, void *ctx)
3215 { 3216 {
3216 struct scrub_copy_nocow_ctx *nocow_ctx = ctx; 3217 struct scrub_copy_nocow_ctx *nocow_ctx = ctx;
3217 struct scrub_nocow_inode *nocow_inode; 3218 struct scrub_nocow_inode *nocow_inode;
3218 3219
3219 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS); 3220 nocow_inode = kzalloc(sizeof(*nocow_inode), GFP_NOFS);
3220 if (!nocow_inode) 3221 if (!nocow_inode)
3221 return -ENOMEM; 3222 return -ENOMEM;
3222 nocow_inode->inum = inum; 3223 nocow_inode->inum = inum;
3223 nocow_inode->offset = offset; 3224 nocow_inode->offset = offset;
3224 nocow_inode->root = root; 3225 nocow_inode->root = root;
3225 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes); 3226 list_add_tail(&nocow_inode->list, &nocow_ctx->inodes);
3226 return 0; 3227 return 0;
3227 } 3228 }
3228 3229
3229 #define COPY_COMPLETE 1 3230 #define COPY_COMPLETE 1
3230 3231
3231 static void copy_nocow_pages_worker(struct btrfs_work *work) 3232 static void copy_nocow_pages_worker(struct btrfs_work *work)
3232 { 3233 {
3233 struct scrub_copy_nocow_ctx *nocow_ctx = 3234 struct scrub_copy_nocow_ctx *nocow_ctx =
3234 container_of(work, struct scrub_copy_nocow_ctx, work); 3235 container_of(work, struct scrub_copy_nocow_ctx, work);
3235 struct scrub_ctx *sctx = nocow_ctx->sctx; 3236 struct scrub_ctx *sctx = nocow_ctx->sctx;
3236 u64 logical = nocow_ctx->logical; 3237 u64 logical = nocow_ctx->logical;
3237 u64 len = nocow_ctx->len; 3238 u64 len = nocow_ctx->len;
3238 int mirror_num = nocow_ctx->mirror_num; 3239 int mirror_num = nocow_ctx->mirror_num;
3239 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; 3240 u64 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3240 int ret; 3241 int ret;
3241 struct btrfs_trans_handle *trans = NULL; 3242 struct btrfs_trans_handle *trans = NULL;
3242 struct btrfs_fs_info *fs_info; 3243 struct btrfs_fs_info *fs_info;
3243 struct btrfs_path *path; 3244 struct btrfs_path *path;
3244 struct btrfs_root *root; 3245 struct btrfs_root *root;
3245 int not_written = 0; 3246 int not_written = 0;
3246 3247
3247 fs_info = sctx->dev_root->fs_info; 3248 fs_info = sctx->dev_root->fs_info;
3248 root = fs_info->extent_root; 3249 root = fs_info->extent_root;
3249 3250
3250 path = btrfs_alloc_path(); 3251 path = btrfs_alloc_path();
3251 if (!path) { 3252 if (!path) {
3252 spin_lock(&sctx->stat_lock); 3253 spin_lock(&sctx->stat_lock);
3253 sctx->stat.malloc_errors++; 3254 sctx->stat.malloc_errors++;
3254 spin_unlock(&sctx->stat_lock); 3255 spin_unlock(&sctx->stat_lock);
3255 not_written = 1; 3256 not_written = 1;
3256 goto out; 3257 goto out;
3257 } 3258 }
3258 3259
3259 trans = btrfs_join_transaction(root); 3260 trans = btrfs_join_transaction(root);
3260 if (IS_ERR(trans)) { 3261 if (IS_ERR(trans)) {
3261 not_written = 1; 3262 not_written = 1;
3262 goto out; 3263 goto out;
3263 } 3264 }
3264 3265
3265 ret = iterate_inodes_from_logical(logical, fs_info, path, 3266 ret = iterate_inodes_from_logical(logical, fs_info, path,
3266 record_inode_for_nocow, nocow_ctx); 3267 record_inode_for_nocow, nocow_ctx);
3267 if (ret != 0 && ret != -ENOENT) { 3268 if (ret != 0 && ret != -ENOENT) {
3268 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, " 3269 btrfs_warn(fs_info, "iterate_inodes_from_logical() failed: log %llu, "
3269 "phys %llu, len %llu, mir %u, ret %d", 3270 "phys %llu, len %llu, mir %u, ret %d",
3270 logical, physical_for_dev_replace, len, mirror_num, 3271 logical, physical_for_dev_replace, len, mirror_num,
3271 ret); 3272 ret);
3272 not_written = 1; 3273 not_written = 1;
3273 goto out; 3274 goto out;
3274 } 3275 }
3275 3276
3276 btrfs_end_transaction(trans, root); 3277 btrfs_end_transaction(trans, root);
3277 trans = NULL; 3278 trans = NULL;
3278 while (!list_empty(&nocow_ctx->inodes)) { 3279 while (!list_empty(&nocow_ctx->inodes)) {
3279 struct scrub_nocow_inode *entry; 3280 struct scrub_nocow_inode *entry;
3280 entry = list_first_entry(&nocow_ctx->inodes, 3281 entry = list_first_entry(&nocow_ctx->inodes,
3281 struct scrub_nocow_inode, 3282 struct scrub_nocow_inode,
3282 list); 3283 list);
3283 list_del_init(&entry->list); 3284 list_del_init(&entry->list);
3284 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset, 3285 ret = copy_nocow_pages_for_inode(entry->inum, entry->offset,
3285 entry->root, nocow_ctx); 3286 entry->root, nocow_ctx);
3286 kfree(entry); 3287 kfree(entry);
3287 if (ret == COPY_COMPLETE) { 3288 if (ret == COPY_COMPLETE) {
3288 ret = 0; 3289 ret = 0;
3289 break; 3290 break;
3290 } else if (ret) { 3291 } else if (ret) {
3291 break; 3292 break;
3292 } 3293 }
3293 } 3294 }
3294 out: 3295 out:
3295 while (!list_empty(&nocow_ctx->inodes)) { 3296 while (!list_empty(&nocow_ctx->inodes)) {
3296 struct scrub_nocow_inode *entry; 3297 struct scrub_nocow_inode *entry;
3297 entry = list_first_entry(&nocow_ctx->inodes, 3298 entry = list_first_entry(&nocow_ctx->inodes,
3298 struct scrub_nocow_inode, 3299 struct scrub_nocow_inode,
3299 list); 3300 list);
3300 list_del_init(&entry->list); 3301 list_del_init(&entry->list);
3301 kfree(entry); 3302 kfree(entry);
3302 } 3303 }
3303 if (trans && !IS_ERR(trans)) 3304 if (trans && !IS_ERR(trans))
3304 btrfs_end_transaction(trans, root); 3305 btrfs_end_transaction(trans, root);
3305 if (not_written) 3306 if (not_written)
3306 btrfs_dev_replace_stats_inc(&fs_info->dev_replace. 3307 btrfs_dev_replace_stats_inc(&fs_info->dev_replace.
3307 num_uncorrectable_read_errors); 3308 num_uncorrectable_read_errors);
3308 3309
3309 btrfs_free_path(path); 3310 btrfs_free_path(path);
3310 kfree(nocow_ctx); 3311 kfree(nocow_ctx);
3311 3312
3312 scrub_pending_trans_workers_dec(sctx); 3313 scrub_pending_trans_workers_dec(sctx);
3313 } 3314 }
3314 3315
3315 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root, 3316 static int copy_nocow_pages_for_inode(u64 inum, u64 offset, u64 root,
3316 struct scrub_copy_nocow_ctx *nocow_ctx) 3317 struct scrub_copy_nocow_ctx *nocow_ctx)
3317 { 3318 {
3318 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info; 3319 struct btrfs_fs_info *fs_info = nocow_ctx->sctx->dev_root->fs_info;
3319 struct btrfs_key key; 3320 struct btrfs_key key;
3320 struct inode *inode; 3321 struct inode *inode;
3321 struct page *page; 3322 struct page *page;
3322 struct btrfs_root *local_root; 3323 struct btrfs_root *local_root;
3323 struct btrfs_ordered_extent *ordered; 3324 struct btrfs_ordered_extent *ordered;
3324 struct extent_map *em; 3325 struct extent_map *em;
3325 struct extent_state *cached_state = NULL; 3326 struct extent_state *cached_state = NULL;
3326 struct extent_io_tree *io_tree; 3327 struct extent_io_tree *io_tree;
3327 u64 physical_for_dev_replace; 3328 u64 physical_for_dev_replace;
3328 u64 len = nocow_ctx->len; 3329 u64 len = nocow_ctx->len;
3329 u64 lockstart = offset, lockend = offset + len - 1; 3330 u64 lockstart = offset, lockend = offset + len - 1;
3330 unsigned long index; 3331 unsigned long index;
3331 int srcu_index; 3332 int srcu_index;
3332 int ret = 0; 3333 int ret = 0;
3333 int err = 0; 3334 int err = 0;
3334 3335
3335 key.objectid = root; 3336 key.objectid = root;
3336 key.type = BTRFS_ROOT_ITEM_KEY; 3337 key.type = BTRFS_ROOT_ITEM_KEY;
3337 key.offset = (u64)-1; 3338 key.offset = (u64)-1;
3338 3339
3339 srcu_index = srcu_read_lock(&fs_info->subvol_srcu); 3340 srcu_index = srcu_read_lock(&fs_info->subvol_srcu);
3340 3341
3341 local_root = btrfs_read_fs_root_no_name(fs_info, &key); 3342 local_root = btrfs_read_fs_root_no_name(fs_info, &key);
3342 if (IS_ERR(local_root)) { 3343 if (IS_ERR(local_root)) {
3343 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); 3344 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
3344 return PTR_ERR(local_root); 3345 return PTR_ERR(local_root);
3345 } 3346 }
3346 3347
3347 key.type = BTRFS_INODE_ITEM_KEY; 3348 key.type = BTRFS_INODE_ITEM_KEY;
3348 key.objectid = inum; 3349 key.objectid = inum;
3349 key.offset = 0; 3350 key.offset = 0;
3350 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL); 3351 inode = btrfs_iget(fs_info->sb, &key, local_root, NULL);
3351 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index); 3352 srcu_read_unlock(&fs_info->subvol_srcu, srcu_index);
3352 if (IS_ERR(inode)) 3353 if (IS_ERR(inode))
3353 return PTR_ERR(inode); 3354 return PTR_ERR(inode);
3354 3355
3355 /* Avoid truncate/dio/punch hole.. */ 3356 /* Avoid truncate/dio/punch hole.. */
3356 mutex_lock(&inode->i_mutex); 3357 mutex_lock(&inode->i_mutex);
3357 inode_dio_wait(inode); 3358 inode_dio_wait(inode);
3358 3359
3359 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace; 3360 physical_for_dev_replace = nocow_ctx->physical_for_dev_replace;
3360 io_tree = &BTRFS_I(inode)->io_tree; 3361 io_tree = &BTRFS_I(inode)->io_tree;
3361 3362
3362 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state); 3363 lock_extent_bits(io_tree, lockstart, lockend, 0, &cached_state);
3363 ordered = btrfs_lookup_ordered_range(inode, lockstart, len); 3364 ordered = btrfs_lookup_ordered_range(inode, lockstart, len);
3364 if (ordered) { 3365 if (ordered) {
3365 btrfs_put_ordered_extent(ordered); 3366 btrfs_put_ordered_extent(ordered);
3366 goto out_unlock; 3367 goto out_unlock;
3367 } 3368 }
3368 3369
3369 em = btrfs_get_extent(inode, NULL, 0, lockstart, len, 0); 3370 em = btrfs_get_extent(inode, NULL, 0, lockstart, len, 0);
3370 if (IS_ERR(em)) { 3371 if (IS_ERR(em)) {
3371 ret = PTR_ERR(em); 3372 ret = PTR_ERR(em);
3372 goto out_unlock; 3373 goto out_unlock;
3373 } 3374 }
3374 3375
3375 /* 3376 /*
3376 * This extent does not actually cover the logical extent anymore, 3377 * This extent does not actually cover the logical extent anymore,
3377 * move on to the next inode. 3378 * move on to the next inode.
3378 */ 3379 */
3379 if (em->block_start > nocow_ctx->logical || 3380 if (em->block_start > nocow_ctx->logical ||
3380 em->block_start + em->block_len < nocow_ctx->logical + len) { 3381 em->block_start + em->block_len < nocow_ctx->logical + len) {
3381 free_extent_map(em); 3382 free_extent_map(em);
3382 goto out_unlock; 3383 goto out_unlock;
3383 } 3384 }
3384 free_extent_map(em); 3385 free_extent_map(em);
3385 3386
3386 while (len >= PAGE_CACHE_SIZE) { 3387 while (len >= PAGE_CACHE_SIZE) {
3387 index = offset >> PAGE_CACHE_SHIFT; 3388 index = offset >> PAGE_CACHE_SHIFT;
3388 again: 3389 again:
3389 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); 3390 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
3390 if (!page) { 3391 if (!page) {
3391 btrfs_err(fs_info, "find_or_create_page() failed"); 3392 btrfs_err(fs_info, "find_or_create_page() failed");
3392 ret = -ENOMEM; 3393 ret = -ENOMEM;
3393 goto out; 3394 goto out;
3394 } 3395 }
3395 3396
3396 if (PageUptodate(page)) { 3397 if (PageUptodate(page)) {
3397 if (PageDirty(page)) 3398 if (PageDirty(page))
3398 goto next_page; 3399 goto next_page;
3399 } else { 3400 } else {
3400 ClearPageError(page); 3401 ClearPageError(page);
3401 err = extent_read_full_page_nolock(io_tree, page, 3402 err = extent_read_full_page_nolock(io_tree, page,
3402 btrfs_get_extent, 3403 btrfs_get_extent,
3403 nocow_ctx->mirror_num); 3404 nocow_ctx->mirror_num);
3404 if (err) { 3405 if (err) {
3405 ret = err; 3406 ret = err;
3406 goto next_page; 3407 goto next_page;
3407 } 3408 }
3408 3409
3409 lock_page(page); 3410 lock_page(page);
3410 /* 3411 /*
3411 * If the page has been remove from the page cache, 3412 * If the page has been remove from the page cache,
3412 * the data on it is meaningless, because it may be 3413 * the data on it is meaningless, because it may be
3413 * old one, the new data may be written into the new 3414 * old one, the new data may be written into the new
3414 * page in the page cache. 3415 * page in the page cache.
3415 */ 3416 */
3416 if (page->mapping != inode->i_mapping) { 3417 if (page->mapping != inode->i_mapping) {
3417 unlock_page(page); 3418 unlock_page(page);
3418 page_cache_release(page); 3419 page_cache_release(page);
3419 goto again; 3420 goto again;
3420 } 3421 }
3421 if (!PageUptodate(page)) { 3422 if (!PageUptodate(page)) {
3422 ret = -EIO; 3423 ret = -EIO;
3423 goto next_page; 3424 goto next_page;
3424 } 3425 }
3425 } 3426 }
3426 err = write_page_nocow(nocow_ctx->sctx, 3427 err = write_page_nocow(nocow_ctx->sctx,
3427 physical_for_dev_replace, page); 3428 physical_for_dev_replace, page);
3428 if (err) 3429 if (err)
3429 ret = err; 3430 ret = err;
3430 next_page: 3431 next_page:
3431 unlock_page(page); 3432 unlock_page(page);
3432 page_cache_release(page); 3433 page_cache_release(page);
3433 3434
3434 if (ret) 3435 if (ret)
3435 break; 3436 break;
3436 3437
3437 offset += PAGE_CACHE_SIZE; 3438 offset += PAGE_CACHE_SIZE;
3438 physical_for_dev_replace += PAGE_CACHE_SIZE; 3439 physical_for_dev_replace += PAGE_CACHE_SIZE;
3439 len -= PAGE_CACHE_SIZE; 3440 len -= PAGE_CACHE_SIZE;
3440 } 3441 }
3441 ret = COPY_COMPLETE; 3442 ret = COPY_COMPLETE;
3442 out_unlock: 3443 out_unlock:
3443 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state, 3444 unlock_extent_cached(io_tree, lockstart, lockend, &cached_state,
3444 GFP_NOFS); 3445 GFP_NOFS);
3445 out: 3446 out:
3446 mutex_unlock(&inode->i_mutex); 3447 mutex_unlock(&inode->i_mutex);
3447 iput(inode); 3448 iput(inode);
3448 return ret; 3449 return ret;
3449 } 3450 }
3450 3451
3451 static int write_page_nocow(struct scrub_ctx *sctx, 3452 static int write_page_nocow(struct scrub_ctx *sctx,
3452 u64 physical_for_dev_replace, struct page *page) 3453 u64 physical_for_dev_replace, struct page *page)
3453 { 3454 {
3454 struct bio *bio; 3455 struct bio *bio;
3455 struct btrfs_device *dev; 3456 struct btrfs_device *dev;
3456 int ret; 3457 int ret;
3457 3458
3458 dev = sctx->wr_ctx.tgtdev; 3459 dev = sctx->wr_ctx.tgtdev;
3459 if (!dev) 3460 if (!dev)
3460 return -EIO; 3461 return -EIO;
3461 if (!dev->bdev) { 3462 if (!dev->bdev) {
3462 printk_ratelimited(KERN_WARNING 3463 printk_ratelimited(KERN_WARNING
3463 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n"); 3464 "BTRFS: scrub write_page_nocow(bdev == NULL) is unexpected!\n");
3464 return -EIO; 3465 return -EIO;
3465 } 3466 }
3466 bio = btrfs_io_bio_alloc(GFP_NOFS, 1); 3467 bio = btrfs_io_bio_alloc(GFP_NOFS, 1);
3467 if (!bio) { 3468 if (!bio) {
3468 spin_lock(&sctx->stat_lock); 3469 spin_lock(&sctx->stat_lock);
3469 sctx->stat.malloc_errors++; 3470 sctx->stat.malloc_errors++;
3470 spin_unlock(&sctx->stat_lock); 3471 spin_unlock(&sctx->stat_lock);
3471 return -ENOMEM; 3472 return -ENOMEM;
3472 } 3473 }
3473 bio->bi_iter.bi_size = 0; 3474 bio->bi_iter.bi_size = 0;
3474 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9; 3475 bio->bi_iter.bi_sector = physical_for_dev_replace >> 9;
3475 bio->bi_bdev = dev->bdev; 3476 bio->bi_bdev = dev->bdev;
3476 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); 3477 ret = bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
3477 if (ret != PAGE_CACHE_SIZE) { 3478 if (ret != PAGE_CACHE_SIZE) {
3478 leave_with_eio: 3479 leave_with_eio:
3479 bio_put(bio); 3480 bio_put(bio);
3480 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); 3481 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS);
3481 return -EIO; 3482 return -EIO;
3482 } 3483 }
3483 3484
3484 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio)) 3485 if (btrfsic_submit_bio_wait(WRITE_SYNC, bio))
3485 goto leave_with_eio; 3486 goto leave_with_eio;
3486 3487
3487 bio_put(bio); 3488 bio_put(bio);
3488 return 0; 3489 return 0;
3489 } 3490 }
3490 3491