Commit 7307de80510a70e5e5aa98de1e80ccbb7d90a3a8
1 parent
607d44aa3f
Exists in
master
and in
20 other branches
ocfs2: shared writeable mmap
Implement cluster consistent shared writeable mappings using the ->page_mkwrite() callback. Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Showing 4 changed files with 200 additions and 39 deletions Inline Diff
fs/ocfs2/aops.c
1 | /* -*- mode: c; c-basic-offset: 8; -*- | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
3 | * | 3 | * |
4 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | 4 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
5 | * | 5 | * |
6 | * This program is free software; you can redistribute it and/or | 6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public | 7 | * modify it under the terms of the GNU General Public |
8 | * License as published by the Free Software Foundation; either | 8 | * License as published by the Free Software Foundation; either |
9 | * version 2 of the License, or (at your option) any later version. | 9 | * version 2 of the License, or (at your option) any later version. |
10 | * | 10 | * |
11 | * This program is distributed in the hope that it will be useful, | 11 | * This program is distributed in the hope that it will be useful, |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | * General Public License for more details. | 14 | * General Public License for more details. |
15 | * | 15 | * |
16 | * You should have received a copy of the GNU General Public | 16 | * You should have received a copy of the GNU General Public |
17 | * License along with this program; if not, write to the | 17 | * License along with this program; if not, write to the |
18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | 18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
19 | * Boston, MA 021110-1307, USA. | 19 | * Boston, MA 021110-1307, USA. |
20 | */ | 20 | */ |
21 | 21 | ||
22 | #include <linux/fs.h> | 22 | #include <linux/fs.h> |
23 | #include <linux/slab.h> | 23 | #include <linux/slab.h> |
24 | #include <linux/highmem.h> | 24 | #include <linux/highmem.h> |
25 | #include <linux/pagemap.h> | 25 | #include <linux/pagemap.h> |
26 | #include <asm/byteorder.h> | 26 | #include <asm/byteorder.h> |
27 | #include <linux/swap.h> | 27 | #include <linux/swap.h> |
28 | #include <linux/pipe_fs_i.h> | 28 | #include <linux/pipe_fs_i.h> |
29 | 29 | ||
30 | #define MLOG_MASK_PREFIX ML_FILE_IO | 30 | #define MLOG_MASK_PREFIX ML_FILE_IO |
31 | #include <cluster/masklog.h> | 31 | #include <cluster/masklog.h> |
32 | 32 | ||
33 | #include "ocfs2.h" | 33 | #include "ocfs2.h" |
34 | 34 | ||
35 | #include "alloc.h" | 35 | #include "alloc.h" |
36 | #include "aops.h" | 36 | #include "aops.h" |
37 | #include "dlmglue.h" | 37 | #include "dlmglue.h" |
38 | #include "extent_map.h" | 38 | #include "extent_map.h" |
39 | #include "file.h" | 39 | #include "file.h" |
40 | #include "inode.h" | 40 | #include "inode.h" |
41 | #include "journal.h" | 41 | #include "journal.h" |
42 | #include "suballoc.h" | 42 | #include "suballoc.h" |
43 | #include "super.h" | 43 | #include "super.h" |
44 | #include "symlink.h" | 44 | #include "symlink.h" |
45 | 45 | ||
46 | #include "buffer_head_io.h" | 46 | #include "buffer_head_io.h" |
47 | 47 | ||
48 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, | 48 | static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock, |
49 | struct buffer_head *bh_result, int create) | 49 | struct buffer_head *bh_result, int create) |
50 | { | 50 | { |
51 | int err = -EIO; | 51 | int err = -EIO; |
52 | int status; | 52 | int status; |
53 | struct ocfs2_dinode *fe = NULL; | 53 | struct ocfs2_dinode *fe = NULL; |
54 | struct buffer_head *bh = NULL; | 54 | struct buffer_head *bh = NULL; |
55 | struct buffer_head *buffer_cache_bh = NULL; | 55 | struct buffer_head *buffer_cache_bh = NULL; |
56 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 56 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
57 | void *kaddr; | 57 | void *kaddr; |
58 | 58 | ||
59 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, | 59 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, |
60 | (unsigned long long)iblock, bh_result, create); | 60 | (unsigned long long)iblock, bh_result, create); |
61 | 61 | ||
62 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); | 62 | BUG_ON(ocfs2_inode_is_fast_symlink(inode)); |
63 | 63 | ||
64 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { | 64 | if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) { |
65 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", | 65 | mlog(ML_ERROR, "block offset > PATH_MAX: %llu", |
66 | (unsigned long long)iblock); | 66 | (unsigned long long)iblock); |
67 | goto bail; | 67 | goto bail; |
68 | } | 68 | } |
69 | 69 | ||
70 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), | 70 | status = ocfs2_read_block(OCFS2_SB(inode->i_sb), |
71 | OCFS2_I(inode)->ip_blkno, | 71 | OCFS2_I(inode)->ip_blkno, |
72 | &bh, OCFS2_BH_CACHED, inode); | 72 | &bh, OCFS2_BH_CACHED, inode); |
73 | if (status < 0) { | 73 | if (status < 0) { |
74 | mlog_errno(status); | 74 | mlog_errno(status); |
75 | goto bail; | 75 | goto bail; |
76 | } | 76 | } |
77 | fe = (struct ocfs2_dinode *) bh->b_data; | 77 | fe = (struct ocfs2_dinode *) bh->b_data; |
78 | 78 | ||
79 | if (!OCFS2_IS_VALID_DINODE(fe)) { | 79 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
80 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", | 80 | mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n", |
81 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, | 81 | (unsigned long long)le64_to_cpu(fe->i_blkno), 7, |
82 | fe->i_signature); | 82 | fe->i_signature); |
83 | goto bail; | 83 | goto bail; |
84 | } | 84 | } |
85 | 85 | ||
86 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, | 86 | if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb, |
87 | le32_to_cpu(fe->i_clusters))) { | 87 | le32_to_cpu(fe->i_clusters))) { |
88 | mlog(ML_ERROR, "block offset is outside the allocated size: " | 88 | mlog(ML_ERROR, "block offset is outside the allocated size: " |
89 | "%llu\n", (unsigned long long)iblock); | 89 | "%llu\n", (unsigned long long)iblock); |
90 | goto bail; | 90 | goto bail; |
91 | } | 91 | } |
92 | 92 | ||
93 | /* We don't use the page cache to create symlink data, so if | 93 | /* We don't use the page cache to create symlink data, so if |
94 | * need be, copy it over from the buffer cache. */ | 94 | * need be, copy it over from the buffer cache. */ |
95 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { | 95 | if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) { |
96 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + | 96 | u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + |
97 | iblock; | 97 | iblock; |
98 | buffer_cache_bh = sb_getblk(osb->sb, blkno); | 98 | buffer_cache_bh = sb_getblk(osb->sb, blkno); |
99 | if (!buffer_cache_bh) { | 99 | if (!buffer_cache_bh) { |
100 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); | 100 | mlog(ML_ERROR, "couldn't getblock for symlink!\n"); |
101 | goto bail; | 101 | goto bail; |
102 | } | 102 | } |
103 | 103 | ||
104 | /* we haven't locked out transactions, so a commit | 104 | /* we haven't locked out transactions, so a commit |
105 | * could've happened. Since we've got a reference on | 105 | * could've happened. Since we've got a reference on |
106 | * the bh, even if it commits while we're doing the | 106 | * the bh, even if it commits while we're doing the |
107 | * copy, the data is still good. */ | 107 | * copy, the data is still good. */ |
108 | if (buffer_jbd(buffer_cache_bh) | 108 | if (buffer_jbd(buffer_cache_bh) |
109 | && ocfs2_inode_is_new(inode)) { | 109 | && ocfs2_inode_is_new(inode)) { |
110 | kaddr = kmap_atomic(bh_result->b_page, KM_USER0); | 110 | kaddr = kmap_atomic(bh_result->b_page, KM_USER0); |
111 | if (!kaddr) { | 111 | if (!kaddr) { |
112 | mlog(ML_ERROR, "couldn't kmap!\n"); | 112 | mlog(ML_ERROR, "couldn't kmap!\n"); |
113 | goto bail; | 113 | goto bail; |
114 | } | 114 | } |
115 | memcpy(kaddr + (bh_result->b_size * iblock), | 115 | memcpy(kaddr + (bh_result->b_size * iblock), |
116 | buffer_cache_bh->b_data, | 116 | buffer_cache_bh->b_data, |
117 | bh_result->b_size); | 117 | bh_result->b_size); |
118 | kunmap_atomic(kaddr, KM_USER0); | 118 | kunmap_atomic(kaddr, KM_USER0); |
119 | set_buffer_uptodate(bh_result); | 119 | set_buffer_uptodate(bh_result); |
120 | } | 120 | } |
121 | brelse(buffer_cache_bh); | 121 | brelse(buffer_cache_bh); |
122 | } | 122 | } |
123 | 123 | ||
124 | map_bh(bh_result, inode->i_sb, | 124 | map_bh(bh_result, inode->i_sb, |
125 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); | 125 | le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock); |
126 | 126 | ||
127 | err = 0; | 127 | err = 0; |
128 | 128 | ||
129 | bail: | 129 | bail: |
130 | if (bh) | 130 | if (bh) |
131 | brelse(bh); | 131 | brelse(bh); |
132 | 132 | ||
133 | mlog_exit(err); | 133 | mlog_exit(err); |
134 | return err; | 134 | return err; |
135 | } | 135 | } |
136 | 136 | ||
137 | static int ocfs2_get_block(struct inode *inode, sector_t iblock, | 137 | static int ocfs2_get_block(struct inode *inode, sector_t iblock, |
138 | struct buffer_head *bh_result, int create) | 138 | struct buffer_head *bh_result, int create) |
139 | { | 139 | { |
140 | int err = 0; | 140 | int err = 0; |
141 | unsigned int ext_flags; | 141 | unsigned int ext_flags; |
142 | u64 p_blkno, past_eof; | 142 | u64 p_blkno, past_eof; |
143 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 143 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
144 | 144 | ||
145 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, | 145 | mlog_entry("(0x%p, %llu, 0x%p, %d)\n", inode, |
146 | (unsigned long long)iblock, bh_result, create); | 146 | (unsigned long long)iblock, bh_result, create); |
147 | 147 | ||
148 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) | 148 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) |
149 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", | 149 | mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n", |
150 | inode, inode->i_ino); | 150 | inode, inode->i_ino); |
151 | 151 | ||
152 | if (S_ISLNK(inode->i_mode)) { | 152 | if (S_ISLNK(inode->i_mode)) { |
153 | /* this always does I/O for some reason. */ | 153 | /* this always does I/O for some reason. */ |
154 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); | 154 | err = ocfs2_symlink_get_block(inode, iblock, bh_result, create); |
155 | goto bail; | 155 | goto bail; |
156 | } | 156 | } |
157 | 157 | ||
158 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, NULL, | 158 | err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, NULL, |
159 | &ext_flags); | 159 | &ext_flags); |
160 | if (err) { | 160 | if (err) { |
161 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " | 161 | mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, " |
162 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, | 162 | "%llu, NULL)\n", err, inode, (unsigned long long)iblock, |
163 | (unsigned long long)p_blkno); | 163 | (unsigned long long)p_blkno); |
164 | goto bail; | 164 | goto bail; |
165 | } | 165 | } |
166 | 166 | ||
167 | /* | 167 | /* |
168 | * ocfs2 never allocates in this function - the only time we | 168 | * ocfs2 never allocates in this function - the only time we |
169 | * need to use BH_New is when we're extending i_size on a file | 169 | * need to use BH_New is when we're extending i_size on a file |
170 | * system which doesn't support holes, in which case BH_New | 170 | * system which doesn't support holes, in which case BH_New |
171 | * allows block_prepare_write() to zero. | 171 | * allows block_prepare_write() to zero. |
172 | */ | 172 | */ |
173 | mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb), | 173 | mlog_bug_on_msg(create && p_blkno == 0 && ocfs2_sparse_alloc(osb), |
174 | "ino %lu, iblock %llu\n", inode->i_ino, | 174 | "ino %lu, iblock %llu\n", inode->i_ino, |
175 | (unsigned long long)iblock); | 175 | (unsigned long long)iblock); |
176 | 176 | ||
177 | /* Treat the unwritten extent as a hole for zeroing purposes. */ | 177 | /* Treat the unwritten extent as a hole for zeroing purposes. */ |
178 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | 178 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
179 | map_bh(bh_result, inode->i_sb, p_blkno); | 179 | map_bh(bh_result, inode->i_sb, p_blkno); |
180 | 180 | ||
181 | if (!ocfs2_sparse_alloc(osb)) { | 181 | if (!ocfs2_sparse_alloc(osb)) { |
182 | if (p_blkno == 0) { | 182 | if (p_blkno == 0) { |
183 | err = -EIO; | 183 | err = -EIO; |
184 | mlog(ML_ERROR, | 184 | mlog(ML_ERROR, |
185 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", | 185 | "iblock = %llu p_blkno = %llu blkno=(%llu)\n", |
186 | (unsigned long long)iblock, | 186 | (unsigned long long)iblock, |
187 | (unsigned long long)p_blkno, | 187 | (unsigned long long)p_blkno, |
188 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | 188 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
189 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); | 189 | mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters); |
190 | dump_stack(); | 190 | dump_stack(); |
191 | } | 191 | } |
192 | 192 | ||
193 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); | 193 | past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
194 | mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino, | 194 | mlog(0, "Inode %lu, past_eof = %llu\n", inode->i_ino, |
195 | (unsigned long long)past_eof); | 195 | (unsigned long long)past_eof); |
196 | 196 | ||
197 | if (create && (iblock >= past_eof)) | 197 | if (create && (iblock >= past_eof)) |
198 | set_buffer_new(bh_result); | 198 | set_buffer_new(bh_result); |
199 | } | 199 | } |
200 | 200 | ||
201 | bail: | 201 | bail: |
202 | if (err < 0) | 202 | if (err < 0) |
203 | err = -EIO; | 203 | err = -EIO; |
204 | 204 | ||
205 | mlog_exit(err); | 205 | mlog_exit(err); |
206 | return err; | 206 | return err; |
207 | } | 207 | } |
208 | 208 | ||
209 | static int ocfs2_readpage(struct file *file, struct page *page) | 209 | static int ocfs2_readpage(struct file *file, struct page *page) |
210 | { | 210 | { |
211 | struct inode *inode = page->mapping->host; | 211 | struct inode *inode = page->mapping->host; |
212 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; | 212 | loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT; |
213 | int ret, unlock = 1; | 213 | int ret, unlock = 1; |
214 | 214 | ||
215 | mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0)); | 215 | mlog_entry("(0x%p, %lu)\n", file, (page ? page->index : 0)); |
216 | 216 | ||
217 | ret = ocfs2_meta_lock_with_page(inode, NULL, 0, page); | 217 | ret = ocfs2_meta_lock_with_page(inode, NULL, 0, page); |
218 | if (ret != 0) { | 218 | if (ret != 0) { |
219 | if (ret == AOP_TRUNCATED_PAGE) | 219 | if (ret == AOP_TRUNCATED_PAGE) |
220 | unlock = 0; | 220 | unlock = 0; |
221 | mlog_errno(ret); | 221 | mlog_errno(ret); |
222 | goto out; | 222 | goto out; |
223 | } | 223 | } |
224 | 224 | ||
225 | if (down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem) == 0) { | 225 | if (down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem) == 0) { |
226 | ret = AOP_TRUNCATED_PAGE; | 226 | ret = AOP_TRUNCATED_PAGE; |
227 | goto out_meta_unlock; | 227 | goto out_meta_unlock; |
228 | } | 228 | } |
229 | 229 | ||
230 | /* | 230 | /* |
231 | * i_size might have just been updated as we grabed the meta lock. We | 231 | * i_size might have just been updated as we grabed the meta lock. We |
232 | * might now be discovering a truncate that hit on another node. | 232 | * might now be discovering a truncate that hit on another node. |
233 | * block_read_full_page->get_block freaks out if it is asked to read | 233 | * block_read_full_page->get_block freaks out if it is asked to read |
234 | * beyond the end of a file, so we check here. Callers | 234 | * beyond the end of a file, so we check here. Callers |
235 | * (generic_file_read, fault->nopage) are clever enough to check i_size | 235 | * (generic_file_read, fault->nopage) are clever enough to check i_size |
236 | * and notice that the page they just read isn't needed. | 236 | * and notice that the page they just read isn't needed. |
237 | * | 237 | * |
238 | * XXX sys_readahead() seems to get that wrong? | 238 | * XXX sys_readahead() seems to get that wrong? |
239 | */ | 239 | */ |
240 | if (start >= i_size_read(inode)) { | 240 | if (start >= i_size_read(inode)) { |
241 | zero_user_page(page, 0, PAGE_SIZE, KM_USER0); | 241 | zero_user_page(page, 0, PAGE_SIZE, KM_USER0); |
242 | SetPageUptodate(page); | 242 | SetPageUptodate(page); |
243 | ret = 0; | 243 | ret = 0; |
244 | goto out_alloc; | 244 | goto out_alloc; |
245 | } | 245 | } |
246 | 246 | ||
247 | ret = ocfs2_data_lock_with_page(inode, 0, page); | 247 | ret = ocfs2_data_lock_with_page(inode, 0, page); |
248 | if (ret != 0) { | 248 | if (ret != 0) { |
249 | if (ret == AOP_TRUNCATED_PAGE) | 249 | if (ret == AOP_TRUNCATED_PAGE) |
250 | unlock = 0; | 250 | unlock = 0; |
251 | mlog_errno(ret); | 251 | mlog_errno(ret); |
252 | goto out_alloc; | 252 | goto out_alloc; |
253 | } | 253 | } |
254 | 254 | ||
255 | ret = block_read_full_page(page, ocfs2_get_block); | 255 | ret = block_read_full_page(page, ocfs2_get_block); |
256 | unlock = 0; | 256 | unlock = 0; |
257 | 257 | ||
258 | ocfs2_data_unlock(inode, 0); | 258 | ocfs2_data_unlock(inode, 0); |
259 | out_alloc: | 259 | out_alloc: |
260 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | 260 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
261 | out_meta_unlock: | 261 | out_meta_unlock: |
262 | ocfs2_meta_unlock(inode, 0); | 262 | ocfs2_meta_unlock(inode, 0); |
263 | out: | 263 | out: |
264 | if (unlock) | 264 | if (unlock) |
265 | unlock_page(page); | 265 | unlock_page(page); |
266 | mlog_exit(ret); | 266 | mlog_exit(ret); |
267 | return ret; | 267 | return ret; |
268 | } | 268 | } |
269 | 269 | ||
270 | /* Note: Because we don't support holes, our allocation has | 270 | /* Note: Because we don't support holes, our allocation has |
271 | * already happened (allocation writes zeros to the file data) | 271 | * already happened (allocation writes zeros to the file data) |
272 | * so we don't have to worry about ordered writes in | 272 | * so we don't have to worry about ordered writes in |
273 | * ocfs2_writepage. | 273 | * ocfs2_writepage. |
274 | * | 274 | * |
275 | * ->writepage is called during the process of invalidating the page cache | 275 | * ->writepage is called during the process of invalidating the page cache |
276 | * during blocked lock processing. It can't block on any cluster locks | 276 | * during blocked lock processing. It can't block on any cluster locks |
277 | * to during block mapping. It's relying on the fact that the block | 277 | * to during block mapping. It's relying on the fact that the block |
278 | * mapping can't have disappeared under the dirty pages that it is | 278 | * mapping can't have disappeared under the dirty pages that it is |
279 | * being asked to write back. | 279 | * being asked to write back. |
280 | */ | 280 | */ |
281 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) | 281 | static int ocfs2_writepage(struct page *page, struct writeback_control *wbc) |
282 | { | 282 | { |
283 | int ret; | 283 | int ret; |
284 | 284 | ||
285 | mlog_entry("(0x%p)\n", page); | 285 | mlog_entry("(0x%p)\n", page); |
286 | 286 | ||
287 | ret = block_write_full_page(page, ocfs2_get_block, wbc); | 287 | ret = block_write_full_page(page, ocfs2_get_block, wbc); |
288 | 288 | ||
289 | mlog_exit(ret); | 289 | mlog_exit(ret); |
290 | 290 | ||
291 | return ret; | 291 | return ret; |
292 | } | 292 | } |
293 | 293 | ||
294 | /* | 294 | /* |
295 | * This is called from ocfs2_write_zero_page() which has handled it's | 295 | * This is called from ocfs2_write_zero_page() which has handled it's |
296 | * own cluster locking and has ensured allocation exists for those | 296 | * own cluster locking and has ensured allocation exists for those |
297 | * blocks to be written. | 297 | * blocks to be written. |
298 | */ | 298 | */ |
299 | int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page, | 299 | int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page, |
300 | unsigned from, unsigned to) | 300 | unsigned from, unsigned to) |
301 | { | 301 | { |
302 | int ret; | 302 | int ret; |
303 | 303 | ||
304 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | 304 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
305 | 305 | ||
306 | ret = block_prepare_write(page, from, to, ocfs2_get_block); | 306 | ret = block_prepare_write(page, from, to, ocfs2_get_block); |
307 | 307 | ||
308 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | 308 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
309 | 309 | ||
310 | return ret; | 310 | return ret; |
311 | } | 311 | } |
312 | 312 | ||
313 | /* Taken from ext3. We don't necessarily need the full blown | 313 | /* Taken from ext3. We don't necessarily need the full blown |
314 | * functionality yet, but IMHO it's better to cut and paste the whole | 314 | * functionality yet, but IMHO it's better to cut and paste the whole |
315 | * thing so we can avoid introducing our own bugs (and easily pick up | 315 | * thing so we can avoid introducing our own bugs (and easily pick up |
316 | * their fixes when they happen) --Mark */ | 316 | * their fixes when they happen) --Mark */ |
317 | int walk_page_buffers( handle_t *handle, | 317 | int walk_page_buffers( handle_t *handle, |
318 | struct buffer_head *head, | 318 | struct buffer_head *head, |
319 | unsigned from, | 319 | unsigned from, |
320 | unsigned to, | 320 | unsigned to, |
321 | int *partial, | 321 | int *partial, |
322 | int (*fn)( handle_t *handle, | 322 | int (*fn)( handle_t *handle, |
323 | struct buffer_head *bh)) | 323 | struct buffer_head *bh)) |
324 | { | 324 | { |
325 | struct buffer_head *bh; | 325 | struct buffer_head *bh; |
326 | unsigned block_start, block_end; | 326 | unsigned block_start, block_end; |
327 | unsigned blocksize = head->b_size; | 327 | unsigned blocksize = head->b_size; |
328 | int err, ret = 0; | 328 | int err, ret = 0; |
329 | struct buffer_head *next; | 329 | struct buffer_head *next; |
330 | 330 | ||
331 | for ( bh = head, block_start = 0; | 331 | for ( bh = head, block_start = 0; |
332 | ret == 0 && (bh != head || !block_start); | 332 | ret == 0 && (bh != head || !block_start); |
333 | block_start = block_end, bh = next) | 333 | block_start = block_end, bh = next) |
334 | { | 334 | { |
335 | next = bh->b_this_page; | 335 | next = bh->b_this_page; |
336 | block_end = block_start + blocksize; | 336 | block_end = block_start + blocksize; |
337 | if (block_end <= from || block_start >= to) { | 337 | if (block_end <= from || block_start >= to) { |
338 | if (partial && !buffer_uptodate(bh)) | 338 | if (partial && !buffer_uptodate(bh)) |
339 | *partial = 1; | 339 | *partial = 1; |
340 | continue; | 340 | continue; |
341 | } | 341 | } |
342 | err = (*fn)(handle, bh); | 342 | err = (*fn)(handle, bh); |
343 | if (!ret) | 343 | if (!ret) |
344 | ret = err; | 344 | ret = err; |
345 | } | 345 | } |
346 | return ret; | 346 | return ret; |
347 | } | 347 | } |
348 | 348 | ||
349 | handle_t *ocfs2_start_walk_page_trans(struct inode *inode, | 349 | handle_t *ocfs2_start_walk_page_trans(struct inode *inode, |
350 | struct page *page, | 350 | struct page *page, |
351 | unsigned from, | 351 | unsigned from, |
352 | unsigned to) | 352 | unsigned to) |
353 | { | 353 | { |
354 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 354 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
355 | handle_t *handle = NULL; | 355 | handle_t *handle = NULL; |
356 | int ret = 0; | 356 | int ret = 0; |
357 | 357 | ||
358 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | 358 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
359 | if (!handle) { | 359 | if (!handle) { |
360 | ret = -ENOMEM; | 360 | ret = -ENOMEM; |
361 | mlog_errno(ret); | 361 | mlog_errno(ret); |
362 | goto out; | 362 | goto out; |
363 | } | 363 | } |
364 | 364 | ||
365 | if (ocfs2_should_order_data(inode)) { | 365 | if (ocfs2_should_order_data(inode)) { |
366 | ret = walk_page_buffers(handle, | 366 | ret = walk_page_buffers(handle, |
367 | page_buffers(page), | 367 | page_buffers(page), |
368 | from, to, NULL, | 368 | from, to, NULL, |
369 | ocfs2_journal_dirty_data); | 369 | ocfs2_journal_dirty_data); |
370 | if (ret < 0) | 370 | if (ret < 0) |
371 | mlog_errno(ret); | 371 | mlog_errno(ret); |
372 | } | 372 | } |
373 | out: | 373 | out: |
374 | if (ret) { | 374 | if (ret) { |
375 | if (handle) | 375 | if (handle) |
376 | ocfs2_commit_trans(osb, handle); | 376 | ocfs2_commit_trans(osb, handle); |
377 | handle = ERR_PTR(ret); | 377 | handle = ERR_PTR(ret); |
378 | } | 378 | } |
379 | return handle; | 379 | return handle; |
380 | } | 380 | } |
381 | 381 | ||
382 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) | 382 | static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block) |
383 | { | 383 | { |
384 | sector_t status; | 384 | sector_t status; |
385 | u64 p_blkno = 0; | 385 | u64 p_blkno = 0; |
386 | int err = 0; | 386 | int err = 0; |
387 | struct inode *inode = mapping->host; | 387 | struct inode *inode = mapping->host; |
388 | 388 | ||
389 | mlog_entry("(block = %llu)\n", (unsigned long long)block); | 389 | mlog_entry("(block = %llu)\n", (unsigned long long)block); |
390 | 390 | ||
391 | /* We don't need to lock journal system files, since they aren't | 391 | /* We don't need to lock journal system files, since they aren't |
392 | * accessed concurrently from multiple nodes. | 392 | * accessed concurrently from multiple nodes. |
393 | */ | 393 | */ |
394 | if (!INODE_JOURNAL(inode)) { | 394 | if (!INODE_JOURNAL(inode)) { |
395 | err = ocfs2_meta_lock(inode, NULL, 0); | 395 | err = ocfs2_meta_lock(inode, NULL, 0); |
396 | if (err) { | 396 | if (err) { |
397 | if (err != -ENOENT) | 397 | if (err != -ENOENT) |
398 | mlog_errno(err); | 398 | mlog_errno(err); |
399 | goto bail; | 399 | goto bail; |
400 | } | 400 | } |
401 | down_read(&OCFS2_I(inode)->ip_alloc_sem); | 401 | down_read(&OCFS2_I(inode)->ip_alloc_sem); |
402 | } | 402 | } |
403 | 403 | ||
404 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, NULL); | 404 | err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL, NULL); |
405 | 405 | ||
406 | if (!INODE_JOURNAL(inode)) { | 406 | if (!INODE_JOURNAL(inode)) { |
407 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | 407 | up_read(&OCFS2_I(inode)->ip_alloc_sem); |
408 | ocfs2_meta_unlock(inode, 0); | 408 | ocfs2_meta_unlock(inode, 0); |
409 | } | 409 | } |
410 | 410 | ||
411 | if (err) { | 411 | if (err) { |
412 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", | 412 | mlog(ML_ERROR, "get_blocks() failed, block = %llu\n", |
413 | (unsigned long long)block); | 413 | (unsigned long long)block); |
414 | mlog_errno(err); | 414 | mlog_errno(err); |
415 | goto bail; | 415 | goto bail; |
416 | } | 416 | } |
417 | 417 | ||
418 | 418 | ||
419 | bail: | 419 | bail: |
420 | status = err ? 0 : p_blkno; | 420 | status = err ? 0 : p_blkno; |
421 | 421 | ||
422 | mlog_exit((int)status); | 422 | mlog_exit((int)status); |
423 | 423 | ||
424 | return status; | 424 | return status; |
425 | } | 425 | } |
426 | 426 | ||
427 | /* | 427 | /* |
428 | * TODO: Make this into a generic get_blocks function. | 428 | * TODO: Make this into a generic get_blocks function. |
429 | * | 429 | * |
430 | * From do_direct_io in direct-io.c: | 430 | * From do_direct_io in direct-io.c: |
431 | * "So what we do is to permit the ->get_blocks function to populate | 431 | * "So what we do is to permit the ->get_blocks function to populate |
432 | * bh.b_size with the size of IO which is permitted at this offset and | 432 | * bh.b_size with the size of IO which is permitted at this offset and |
433 | * this i_blkbits." | 433 | * this i_blkbits." |
434 | * | 434 | * |
435 | * This function is called directly from get_more_blocks in direct-io.c. | 435 | * This function is called directly from get_more_blocks in direct-io.c. |
436 | * | 436 | * |
437 | * called like this: dio->get_blocks(dio->inode, fs_startblk, | 437 | * called like this: dio->get_blocks(dio->inode, fs_startblk, |
438 | * fs_count, map_bh, dio->rw == WRITE); | 438 | * fs_count, map_bh, dio->rw == WRITE); |
439 | */ | 439 | */ |
440 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, | 440 | static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock, |
441 | struct buffer_head *bh_result, int create) | 441 | struct buffer_head *bh_result, int create) |
442 | { | 442 | { |
443 | int ret; | 443 | int ret; |
444 | u64 p_blkno, inode_blocks, contig_blocks; | 444 | u64 p_blkno, inode_blocks, contig_blocks; |
445 | unsigned int ext_flags; | 445 | unsigned int ext_flags; |
446 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; | 446 | unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits; |
447 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; | 447 | unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits; |
448 | 448 | ||
449 | /* This function won't even be called if the request isn't all | 449 | /* This function won't even be called if the request isn't all |
450 | * nicely aligned and of the right size, so there's no need | 450 | * nicely aligned and of the right size, so there's no need |
451 | * for us to check any of that. */ | 451 | * for us to check any of that. */ |
452 | 452 | ||
453 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); | 453 | inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode)); |
454 | 454 | ||
455 | /* | 455 | /* |
456 | * Any write past EOF is not allowed because we'd be extending. | 456 | * Any write past EOF is not allowed because we'd be extending. |
457 | */ | 457 | */ |
458 | if (create && (iblock + max_blocks) > inode_blocks) { | 458 | if (create && (iblock + max_blocks) > inode_blocks) { |
459 | ret = -EIO; | 459 | ret = -EIO; |
460 | goto bail; | 460 | goto bail; |
461 | } | 461 | } |
462 | 462 | ||
463 | /* This figures out the size of the next contiguous block, and | 463 | /* This figures out the size of the next contiguous block, and |
464 | * our logical offset */ | 464 | * our logical offset */ |
465 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, | 465 | ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, |
466 | &contig_blocks, &ext_flags); | 466 | &contig_blocks, &ext_flags); |
467 | if (ret) { | 467 | if (ret) { |
468 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", | 468 | mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n", |
469 | (unsigned long long)iblock); | 469 | (unsigned long long)iblock); |
470 | ret = -EIO; | 470 | ret = -EIO; |
471 | goto bail; | 471 | goto bail; |
472 | } | 472 | } |
473 | 473 | ||
474 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) { | 474 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)) && !p_blkno) { |
475 | ocfs2_error(inode->i_sb, | 475 | ocfs2_error(inode->i_sb, |
476 | "Inode %llu has a hole at block %llu\n", | 476 | "Inode %llu has a hole at block %llu\n", |
477 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | 477 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
478 | (unsigned long long)iblock); | 478 | (unsigned long long)iblock); |
479 | ret = -EROFS; | 479 | ret = -EROFS; |
480 | goto bail; | 480 | goto bail; |
481 | } | 481 | } |
482 | 482 | ||
483 | /* | 483 | /* |
484 | * get_more_blocks() expects us to describe a hole by clearing | 484 | * get_more_blocks() expects us to describe a hole by clearing |
485 | * the mapped bit on bh_result(). | 485 | * the mapped bit on bh_result(). |
486 | * | 486 | * |
487 | * Consider an unwritten extent as a hole. | 487 | * Consider an unwritten extent as a hole. |
488 | */ | 488 | */ |
489 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) | 489 | if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN)) |
490 | map_bh(bh_result, inode->i_sb, p_blkno); | 490 | map_bh(bh_result, inode->i_sb, p_blkno); |
491 | else { | 491 | else { |
492 | /* | 492 | /* |
493 | * ocfs2_prepare_inode_for_write() should have caught | 493 | * ocfs2_prepare_inode_for_write() should have caught |
494 | * the case where we'd be filling a hole and triggered | 494 | * the case where we'd be filling a hole and triggered |
495 | * a buffered write instead. | 495 | * a buffered write instead. |
496 | */ | 496 | */ |
497 | if (create) { | 497 | if (create) { |
498 | ret = -EIO; | 498 | ret = -EIO; |
499 | mlog_errno(ret); | 499 | mlog_errno(ret); |
500 | goto bail; | 500 | goto bail; |
501 | } | 501 | } |
502 | 502 | ||
503 | clear_buffer_mapped(bh_result); | 503 | clear_buffer_mapped(bh_result); |
504 | } | 504 | } |
505 | 505 | ||
506 | /* make sure we don't map more than max_blocks blocks here as | 506 | /* make sure we don't map more than max_blocks blocks here as |
507 | that's all the kernel will handle at this point. */ | 507 | that's all the kernel will handle at this point. */ |
508 | if (max_blocks < contig_blocks) | 508 | if (max_blocks < contig_blocks) |
509 | contig_blocks = max_blocks; | 509 | contig_blocks = max_blocks; |
510 | bh_result->b_size = contig_blocks << blocksize_bits; | 510 | bh_result->b_size = contig_blocks << blocksize_bits; |
511 | bail: | 511 | bail: |
512 | return ret; | 512 | return ret; |
513 | } | 513 | } |
514 | 514 | ||
515 | /* | 515 | /* |
516 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're | 516 | * ocfs2_dio_end_io is called by the dio core when a dio is finished. We're |
517 | * particularly interested in the aio/dio case. Like the core uses | 517 | * particularly interested in the aio/dio case. Like the core uses |
518 | * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from | 518 | * i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from |
519 | * truncation on another. | 519 | * truncation on another. |
520 | */ | 520 | */ |
521 | static void ocfs2_dio_end_io(struct kiocb *iocb, | 521 | static void ocfs2_dio_end_io(struct kiocb *iocb, |
522 | loff_t offset, | 522 | loff_t offset, |
523 | ssize_t bytes, | 523 | ssize_t bytes, |
524 | void *private) | 524 | void *private) |
525 | { | 525 | { |
526 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; | 526 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
527 | int level; | 527 | int level; |
528 | 528 | ||
529 | /* this io's submitter should not have unlocked this before we could */ | 529 | /* this io's submitter should not have unlocked this before we could */ |
530 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); | 530 | BUG_ON(!ocfs2_iocb_is_rw_locked(iocb)); |
531 | 531 | ||
532 | ocfs2_iocb_clear_rw_locked(iocb); | 532 | ocfs2_iocb_clear_rw_locked(iocb); |
533 | 533 | ||
534 | level = ocfs2_iocb_rw_locked_level(iocb); | 534 | level = ocfs2_iocb_rw_locked_level(iocb); |
535 | if (!level) | 535 | if (!level) |
536 | up_read(&inode->i_alloc_sem); | 536 | up_read(&inode->i_alloc_sem); |
537 | ocfs2_rw_unlock(inode, level); | 537 | ocfs2_rw_unlock(inode, level); |
538 | } | 538 | } |
539 | 539 | ||
540 | /* | 540 | /* |
541 | * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen | 541 | * ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen |
542 | * from ext3. PageChecked() bits have been removed as OCFS2 does not | 542 | * from ext3. PageChecked() bits have been removed as OCFS2 does not |
543 | * do journalled data. | 543 | * do journalled data. |
544 | */ | 544 | */ |
545 | static void ocfs2_invalidatepage(struct page *page, unsigned long offset) | 545 | static void ocfs2_invalidatepage(struct page *page, unsigned long offset) |
546 | { | 546 | { |
547 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | 547 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; |
548 | 548 | ||
549 | journal_invalidatepage(journal, page, offset); | 549 | journal_invalidatepage(journal, page, offset); |
550 | } | 550 | } |
551 | 551 | ||
552 | static int ocfs2_releasepage(struct page *page, gfp_t wait) | 552 | static int ocfs2_releasepage(struct page *page, gfp_t wait) |
553 | { | 553 | { |
554 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; | 554 | journal_t *journal = OCFS2_SB(page->mapping->host->i_sb)->journal->j_journal; |
555 | 555 | ||
556 | if (!page_has_buffers(page)) | 556 | if (!page_has_buffers(page)) |
557 | return 0; | 557 | return 0; |
558 | return journal_try_to_free_buffers(journal, page, wait); | 558 | return journal_try_to_free_buffers(journal, page, wait); |
559 | } | 559 | } |
560 | 560 | ||
561 | static ssize_t ocfs2_direct_IO(int rw, | 561 | static ssize_t ocfs2_direct_IO(int rw, |
562 | struct kiocb *iocb, | 562 | struct kiocb *iocb, |
563 | const struct iovec *iov, | 563 | const struct iovec *iov, |
564 | loff_t offset, | 564 | loff_t offset, |
565 | unsigned long nr_segs) | 565 | unsigned long nr_segs) |
566 | { | 566 | { |
567 | struct file *file = iocb->ki_filp; | 567 | struct file *file = iocb->ki_filp; |
568 | struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host; | 568 | struct inode *inode = file->f_path.dentry->d_inode->i_mapping->host; |
569 | int ret; | 569 | int ret; |
570 | 570 | ||
571 | mlog_entry_void(); | 571 | mlog_entry_void(); |
572 | 572 | ||
573 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { | 573 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { |
574 | /* | 574 | /* |
575 | * We get PR data locks even for O_DIRECT. This | 575 | * We get PR data locks even for O_DIRECT. This |
576 | * allows concurrent O_DIRECT I/O but doesn't let | 576 | * allows concurrent O_DIRECT I/O but doesn't let |
577 | * O_DIRECT with extending and buffered zeroing writes | 577 | * O_DIRECT with extending and buffered zeroing writes |
578 | * race. If they did race then the buffered zeroing | 578 | * race. If they did race then the buffered zeroing |
579 | * could be written back after the O_DIRECT I/O. It's | 579 | * could be written back after the O_DIRECT I/O. It's |
580 | * one thing to tell people not to mix buffered and | 580 | * one thing to tell people not to mix buffered and |
581 | * O_DIRECT writes, but expecting them to understand | 581 | * O_DIRECT writes, but expecting them to understand |
582 | * that file extension is also an implicit buffered | 582 | * that file extension is also an implicit buffered |
583 | * write is too much. By getting the PR we force | 583 | * write is too much. By getting the PR we force |
584 | * writeback of the buffered zeroing before | 584 | * writeback of the buffered zeroing before |
585 | * proceeding. | 585 | * proceeding. |
586 | */ | 586 | */ |
587 | ret = ocfs2_data_lock(inode, 0); | 587 | ret = ocfs2_data_lock(inode, 0); |
588 | if (ret < 0) { | 588 | if (ret < 0) { |
589 | mlog_errno(ret); | 589 | mlog_errno(ret); |
590 | goto out; | 590 | goto out; |
591 | } | 591 | } |
592 | ocfs2_data_unlock(inode, 0); | 592 | ocfs2_data_unlock(inode, 0); |
593 | } | 593 | } |
594 | 594 | ||
595 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, | 595 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
596 | inode->i_sb->s_bdev, iov, offset, | 596 | inode->i_sb->s_bdev, iov, offset, |
597 | nr_segs, | 597 | nr_segs, |
598 | ocfs2_direct_IO_get_blocks, | 598 | ocfs2_direct_IO_get_blocks, |
599 | ocfs2_dio_end_io); | 599 | ocfs2_dio_end_io); |
600 | out: | 600 | out: |
601 | mlog_exit(ret); | 601 | mlog_exit(ret); |
602 | return ret; | 602 | return ret; |
603 | } | 603 | } |
604 | 604 | ||
605 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, | 605 | static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb, |
606 | u32 cpos, | 606 | u32 cpos, |
607 | unsigned int *start, | 607 | unsigned int *start, |
608 | unsigned int *end) | 608 | unsigned int *end) |
609 | { | 609 | { |
610 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; | 610 | unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE; |
611 | 611 | ||
612 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { | 612 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) { |
613 | unsigned int cpp; | 613 | unsigned int cpp; |
614 | 614 | ||
615 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); | 615 | cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits); |
616 | 616 | ||
617 | cluster_start = cpos % cpp; | 617 | cluster_start = cpos % cpp; |
618 | cluster_start = cluster_start << osb->s_clustersize_bits; | 618 | cluster_start = cluster_start << osb->s_clustersize_bits; |
619 | 619 | ||
620 | cluster_end = cluster_start + osb->s_clustersize; | 620 | cluster_end = cluster_start + osb->s_clustersize; |
621 | } | 621 | } |
622 | 622 | ||
623 | BUG_ON(cluster_start > PAGE_SIZE); | 623 | BUG_ON(cluster_start > PAGE_SIZE); |
624 | BUG_ON(cluster_end > PAGE_SIZE); | 624 | BUG_ON(cluster_end > PAGE_SIZE); |
625 | 625 | ||
626 | if (start) | 626 | if (start) |
627 | *start = cluster_start; | 627 | *start = cluster_start; |
628 | if (end) | 628 | if (end) |
629 | *end = cluster_end; | 629 | *end = cluster_end; |
630 | } | 630 | } |
631 | 631 | ||
632 | /* | 632 | /* |
633 | * 'from' and 'to' are the region in the page to avoid zeroing. | 633 | * 'from' and 'to' are the region in the page to avoid zeroing. |
634 | * | 634 | * |
635 | * If pagesize > clustersize, this function will avoid zeroing outside | 635 | * If pagesize > clustersize, this function will avoid zeroing outside |
636 | * of the cluster boundary. | 636 | * of the cluster boundary. |
637 | * | 637 | * |
638 | * from == to == 0 is code for "zero the entire cluster region" | 638 | * from == to == 0 is code for "zero the entire cluster region" |
639 | */ | 639 | */ |
640 | static void ocfs2_clear_page_regions(struct page *page, | 640 | static void ocfs2_clear_page_regions(struct page *page, |
641 | struct ocfs2_super *osb, u32 cpos, | 641 | struct ocfs2_super *osb, u32 cpos, |
642 | unsigned from, unsigned to) | 642 | unsigned from, unsigned to) |
643 | { | 643 | { |
644 | void *kaddr; | 644 | void *kaddr; |
645 | unsigned int cluster_start, cluster_end; | 645 | unsigned int cluster_start, cluster_end; |
646 | 646 | ||
647 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); | 647 | ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end); |
648 | 648 | ||
649 | kaddr = kmap_atomic(page, KM_USER0); | 649 | kaddr = kmap_atomic(page, KM_USER0); |
650 | 650 | ||
651 | if (from || to) { | 651 | if (from || to) { |
652 | if (from > cluster_start) | 652 | if (from > cluster_start) |
653 | memset(kaddr + cluster_start, 0, from - cluster_start); | 653 | memset(kaddr + cluster_start, 0, from - cluster_start); |
654 | if (to < cluster_end) | 654 | if (to < cluster_end) |
655 | memset(kaddr + to, 0, cluster_end - to); | 655 | memset(kaddr + to, 0, cluster_end - to); |
656 | } else { | 656 | } else { |
657 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); | 657 | memset(kaddr + cluster_start, 0, cluster_end - cluster_start); |
658 | } | 658 | } |
659 | 659 | ||
660 | kunmap_atomic(kaddr, KM_USER0); | 660 | kunmap_atomic(kaddr, KM_USER0); |
661 | } | 661 | } |
662 | 662 | ||
663 | /* | 663 | /* |
664 | * Some of this taken from block_prepare_write(). We already have our | 664 | * Some of this taken from block_prepare_write(). We already have our |
665 | * mapping by now though, and the entire write will be allocating or | 665 | * mapping by now though, and the entire write will be allocating or |
666 | * it won't, so not much need to use BH_New. | 666 | * it won't, so not much need to use BH_New. |
667 | * | 667 | * |
668 | * This will also skip zeroing, which is handled externally. | 668 | * This will also skip zeroing, which is handled externally. |
669 | */ | 669 | */ |
670 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, | 670 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
671 | struct inode *inode, unsigned int from, | 671 | struct inode *inode, unsigned int from, |
672 | unsigned int to, int new) | 672 | unsigned int to, int new) |
673 | { | 673 | { |
674 | int ret = 0; | 674 | int ret = 0; |
675 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; | 675 | struct buffer_head *head, *bh, *wait[2], **wait_bh = wait; |
676 | unsigned int block_end, block_start; | 676 | unsigned int block_end, block_start; |
677 | unsigned int bsize = 1 << inode->i_blkbits; | 677 | unsigned int bsize = 1 << inode->i_blkbits; |
678 | 678 | ||
679 | if (!page_has_buffers(page)) | 679 | if (!page_has_buffers(page)) |
680 | create_empty_buffers(page, bsize, 0); | 680 | create_empty_buffers(page, bsize, 0); |
681 | 681 | ||
682 | head = page_buffers(page); | 682 | head = page_buffers(page); |
683 | for (bh = head, block_start = 0; bh != head || !block_start; | 683 | for (bh = head, block_start = 0; bh != head || !block_start; |
684 | bh = bh->b_this_page, block_start += bsize) { | 684 | bh = bh->b_this_page, block_start += bsize) { |
685 | block_end = block_start + bsize; | 685 | block_end = block_start + bsize; |
686 | 686 | ||
687 | clear_buffer_new(bh); | 687 | clear_buffer_new(bh); |
688 | 688 | ||
689 | /* | 689 | /* |
690 | * Ignore blocks outside of our i/o range - | 690 | * Ignore blocks outside of our i/o range - |
691 | * they may belong to unallocated clusters. | 691 | * they may belong to unallocated clusters. |
692 | */ | 692 | */ |
693 | if (block_start >= to || block_end <= from) { | 693 | if (block_start >= to || block_end <= from) { |
694 | if (PageUptodate(page)) | 694 | if (PageUptodate(page)) |
695 | set_buffer_uptodate(bh); | 695 | set_buffer_uptodate(bh); |
696 | continue; | 696 | continue; |
697 | } | 697 | } |
698 | 698 | ||
699 | /* | 699 | /* |
700 | * For an allocating write with cluster size >= page | 700 | * For an allocating write with cluster size >= page |
701 | * size, we always write the entire page. | 701 | * size, we always write the entire page. |
702 | */ | 702 | */ |
703 | if (new) | 703 | if (new) |
704 | set_buffer_new(bh); | 704 | set_buffer_new(bh); |
705 | 705 | ||
706 | if (!buffer_mapped(bh)) { | 706 | if (!buffer_mapped(bh)) { |
707 | map_bh(bh, inode->i_sb, *p_blkno); | 707 | map_bh(bh, inode->i_sb, *p_blkno); |
708 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | 708 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); |
709 | } | 709 | } |
710 | 710 | ||
711 | if (PageUptodate(page)) { | 711 | if (PageUptodate(page)) { |
712 | if (!buffer_uptodate(bh)) | 712 | if (!buffer_uptodate(bh)) |
713 | set_buffer_uptodate(bh); | 713 | set_buffer_uptodate(bh); |
714 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && | 714 | } else if (!buffer_uptodate(bh) && !buffer_delay(bh) && |
715 | (block_start < from || block_end > to)) { | 715 | (block_start < from || block_end > to)) { |
716 | ll_rw_block(READ, 1, &bh); | 716 | ll_rw_block(READ, 1, &bh); |
717 | *wait_bh++=bh; | 717 | *wait_bh++=bh; |
718 | } | 718 | } |
719 | 719 | ||
720 | *p_blkno = *p_blkno + 1; | 720 | *p_blkno = *p_blkno + 1; |
721 | } | 721 | } |
722 | 722 | ||
723 | /* | 723 | /* |
724 | * If we issued read requests - let them complete. | 724 | * If we issued read requests - let them complete. |
725 | */ | 725 | */ |
726 | while(wait_bh > wait) { | 726 | while(wait_bh > wait) { |
727 | wait_on_buffer(*--wait_bh); | 727 | wait_on_buffer(*--wait_bh); |
728 | if (!buffer_uptodate(*wait_bh)) | 728 | if (!buffer_uptodate(*wait_bh)) |
729 | ret = -EIO; | 729 | ret = -EIO; |
730 | } | 730 | } |
731 | 731 | ||
732 | if (ret == 0 || !new) | 732 | if (ret == 0 || !new) |
733 | return ret; | 733 | return ret; |
734 | 734 | ||
735 | /* | 735 | /* |
736 | * If we get -EIO above, zero out any newly allocated blocks | 736 | * If we get -EIO above, zero out any newly allocated blocks |
737 | * to avoid exposing stale data. | 737 | * to avoid exposing stale data. |
738 | */ | 738 | */ |
739 | bh = head; | 739 | bh = head; |
740 | block_start = 0; | 740 | block_start = 0; |
741 | do { | 741 | do { |
742 | void *kaddr; | 742 | void *kaddr; |
743 | 743 | ||
744 | block_end = block_start + bsize; | 744 | block_end = block_start + bsize; |
745 | if (block_end <= from) | 745 | if (block_end <= from) |
746 | goto next_bh; | 746 | goto next_bh; |
747 | if (block_start >= to) | 747 | if (block_start >= to) |
748 | break; | 748 | break; |
749 | 749 | ||
750 | kaddr = kmap_atomic(page, KM_USER0); | 750 | kaddr = kmap_atomic(page, KM_USER0); |
751 | memset(kaddr+block_start, 0, bh->b_size); | 751 | memset(kaddr+block_start, 0, bh->b_size); |
752 | flush_dcache_page(page); | 752 | flush_dcache_page(page); |
753 | kunmap_atomic(kaddr, KM_USER0); | 753 | kunmap_atomic(kaddr, KM_USER0); |
754 | set_buffer_uptodate(bh); | 754 | set_buffer_uptodate(bh); |
755 | mark_buffer_dirty(bh); | 755 | mark_buffer_dirty(bh); |
756 | 756 | ||
757 | next_bh: | 757 | next_bh: |
758 | block_start = block_end; | 758 | block_start = block_end; |
759 | bh = bh->b_this_page; | 759 | bh = bh->b_this_page; |
760 | } while (bh != head); | 760 | } while (bh != head); |
761 | 761 | ||
762 | return ret; | 762 | return ret; |
763 | } | 763 | } |
764 | 764 | ||
765 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) | 765 | #if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE) |
766 | #define OCFS2_MAX_CTXT_PAGES 1 | 766 | #define OCFS2_MAX_CTXT_PAGES 1 |
767 | #else | 767 | #else |
768 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) | 768 | #define OCFS2_MAX_CTXT_PAGES (OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE) |
769 | #endif | 769 | #endif |
770 | 770 | ||
771 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) | 771 | #define OCFS2_MAX_CLUSTERS_PER_PAGE (PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE) |
772 | 772 | ||
773 | /* | 773 | /* |
774 | * Describe the state of a single cluster to be written to. | 774 | * Describe the state of a single cluster to be written to. |
775 | */ | 775 | */ |
776 | struct ocfs2_write_cluster_desc { | 776 | struct ocfs2_write_cluster_desc { |
777 | u32 c_cpos; | 777 | u32 c_cpos; |
778 | u32 c_phys; | 778 | u32 c_phys; |
779 | /* | 779 | /* |
780 | * Give this a unique field because c_phys eventually gets | 780 | * Give this a unique field because c_phys eventually gets |
781 | * filled. | 781 | * filled. |
782 | */ | 782 | */ |
783 | unsigned c_new; | 783 | unsigned c_new; |
784 | }; | 784 | }; |
785 | 785 | ||
786 | struct ocfs2_write_ctxt { | 786 | struct ocfs2_write_ctxt { |
787 | /* Logical cluster position / len of write */ | 787 | /* Logical cluster position / len of write */ |
788 | u32 w_cpos; | 788 | u32 w_cpos; |
789 | u32 w_clen; | 789 | u32 w_clen; |
790 | 790 | ||
791 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; | 791 | struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE]; |
792 | 792 | ||
793 | /* | 793 | /* |
794 | * This is true if page_size > cluster_size. | 794 | * This is true if page_size > cluster_size. |
795 | * | 795 | * |
796 | * It triggers a set of special cases during write which might | 796 | * It triggers a set of special cases during write which might |
797 | * have to deal with allocating writes to partial pages. | 797 | * have to deal with allocating writes to partial pages. |
798 | */ | 798 | */ |
799 | unsigned int w_large_pages; | 799 | unsigned int w_large_pages; |
800 | 800 | ||
801 | /* | 801 | /* |
802 | * Pages involved in this write. | 802 | * Pages involved in this write. |
803 | * | 803 | * |
804 | * w_target_page is the page being written to by the user. | 804 | * w_target_page is the page being written to by the user. |
805 | * | 805 | * |
806 | * w_pages is an array of pages which always contains | 806 | * w_pages is an array of pages which always contains |
807 | * w_target_page, and in the case of an allocating write with | 807 | * w_target_page, and in the case of an allocating write with |
808 | * page_size < cluster size, it will contain zero'd and mapped | 808 | * page_size < cluster size, it will contain zero'd and mapped |
809 | * pages adjacent to w_target_page which need to be written | 809 | * pages adjacent to w_target_page which need to be written |
810 | * out in so that future reads from that region will get | 810 | * out in so that future reads from that region will get |
811 | * zero's. | 811 | * zero's. |
812 | */ | 812 | */ |
813 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; | 813 | struct page *w_pages[OCFS2_MAX_CTXT_PAGES]; |
814 | unsigned int w_num_pages; | 814 | unsigned int w_num_pages; |
815 | struct page *w_target_page; | 815 | struct page *w_target_page; |
816 | 816 | ||
817 | /* | 817 | /* |
818 | * ocfs2_write_end() uses this to know what the real range to | 818 | * ocfs2_write_end() uses this to know what the real range to |
819 | * write in the target should be. | 819 | * write in the target should be. |
820 | */ | 820 | */ |
821 | unsigned int w_target_from; | 821 | unsigned int w_target_from; |
822 | unsigned int w_target_to; | 822 | unsigned int w_target_to; |
823 | 823 | ||
824 | /* | 824 | /* |
825 | * We could use journal_current_handle() but this is cleaner, | 825 | * We could use journal_current_handle() but this is cleaner, |
826 | * IMHO -Mark | 826 | * IMHO -Mark |
827 | */ | 827 | */ |
828 | handle_t *w_handle; | 828 | handle_t *w_handle; |
829 | 829 | ||
830 | struct buffer_head *w_di_bh; | 830 | struct buffer_head *w_di_bh; |
831 | }; | 831 | }; |
832 | 832 | ||
833 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) | 833 | static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc) |
834 | { | 834 | { |
835 | int i; | 835 | int i; |
836 | 836 | ||
837 | for(i = 0; i < wc->w_num_pages; i++) { | 837 | for(i = 0; i < wc->w_num_pages; i++) { |
838 | if (wc->w_pages[i] == NULL) | 838 | if (wc->w_pages[i] == NULL) |
839 | continue; | 839 | continue; |
840 | 840 | ||
841 | unlock_page(wc->w_pages[i]); | 841 | unlock_page(wc->w_pages[i]); |
842 | mark_page_accessed(wc->w_pages[i]); | 842 | mark_page_accessed(wc->w_pages[i]); |
843 | page_cache_release(wc->w_pages[i]); | 843 | page_cache_release(wc->w_pages[i]); |
844 | } | 844 | } |
845 | 845 | ||
846 | brelse(wc->w_di_bh); | 846 | brelse(wc->w_di_bh); |
847 | kfree(wc); | 847 | kfree(wc); |
848 | } | 848 | } |
849 | 849 | ||
850 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, | 850 | static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp, |
851 | struct ocfs2_super *osb, loff_t pos, | 851 | struct ocfs2_super *osb, loff_t pos, |
852 | unsigned len, struct buffer_head *di_bh) | 852 | unsigned len, struct buffer_head *di_bh) |
853 | { | 853 | { |
854 | struct ocfs2_write_ctxt *wc; | 854 | struct ocfs2_write_ctxt *wc; |
855 | 855 | ||
856 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); | 856 | wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS); |
857 | if (!wc) | 857 | if (!wc) |
858 | return -ENOMEM; | 858 | return -ENOMEM; |
859 | 859 | ||
860 | wc->w_cpos = pos >> osb->s_clustersize_bits; | 860 | wc->w_cpos = pos >> osb->s_clustersize_bits; |
861 | wc->w_clen = ocfs2_clusters_for_bytes(osb->sb, len); | 861 | wc->w_clen = ocfs2_clusters_for_bytes(osb->sb, len); |
862 | get_bh(di_bh); | 862 | get_bh(di_bh); |
863 | wc->w_di_bh = di_bh; | 863 | wc->w_di_bh = di_bh; |
864 | 864 | ||
865 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) | 865 | if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) |
866 | wc->w_large_pages = 1; | 866 | wc->w_large_pages = 1; |
867 | else | 867 | else |
868 | wc->w_large_pages = 0; | 868 | wc->w_large_pages = 0; |
869 | 869 | ||
870 | *wcp = wc; | 870 | *wcp = wc; |
871 | 871 | ||
872 | return 0; | 872 | return 0; |
873 | } | 873 | } |
874 | 874 | ||
875 | /* | 875 | /* |
876 | * If a page has any new buffers, zero them out here, and mark them uptodate | 876 | * If a page has any new buffers, zero them out here, and mark them uptodate |
877 | * and dirty so they'll be written out (in order to prevent uninitialised | 877 | * and dirty so they'll be written out (in order to prevent uninitialised |
878 | * block data from leaking). And clear the new bit. | 878 | * block data from leaking). And clear the new bit. |
879 | */ | 879 | */ |
880 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) | 880 | static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to) |
881 | { | 881 | { |
882 | unsigned int block_start, block_end; | 882 | unsigned int block_start, block_end; |
883 | struct buffer_head *head, *bh; | 883 | struct buffer_head *head, *bh; |
884 | 884 | ||
885 | BUG_ON(!PageLocked(page)); | 885 | BUG_ON(!PageLocked(page)); |
886 | if (!page_has_buffers(page)) | 886 | if (!page_has_buffers(page)) |
887 | return; | 887 | return; |
888 | 888 | ||
889 | bh = head = page_buffers(page); | 889 | bh = head = page_buffers(page); |
890 | block_start = 0; | 890 | block_start = 0; |
891 | do { | 891 | do { |
892 | block_end = block_start + bh->b_size; | 892 | block_end = block_start + bh->b_size; |
893 | 893 | ||
894 | if (buffer_new(bh)) { | 894 | if (buffer_new(bh)) { |
895 | if (block_end > from && block_start < to) { | 895 | if (block_end > from && block_start < to) { |
896 | if (!PageUptodate(page)) { | 896 | if (!PageUptodate(page)) { |
897 | unsigned start, end; | 897 | unsigned start, end; |
898 | void *kaddr; | 898 | void *kaddr; |
899 | 899 | ||
900 | start = max(from, block_start); | 900 | start = max(from, block_start); |
901 | end = min(to, block_end); | 901 | end = min(to, block_end); |
902 | 902 | ||
903 | kaddr = kmap_atomic(page, KM_USER0); | 903 | kaddr = kmap_atomic(page, KM_USER0); |
904 | memset(kaddr+start, 0, end - start); | 904 | memset(kaddr+start, 0, end - start); |
905 | flush_dcache_page(page); | 905 | flush_dcache_page(page); |
906 | kunmap_atomic(kaddr, KM_USER0); | 906 | kunmap_atomic(kaddr, KM_USER0); |
907 | set_buffer_uptodate(bh); | 907 | set_buffer_uptodate(bh); |
908 | } | 908 | } |
909 | 909 | ||
910 | clear_buffer_new(bh); | 910 | clear_buffer_new(bh); |
911 | mark_buffer_dirty(bh); | 911 | mark_buffer_dirty(bh); |
912 | } | 912 | } |
913 | } | 913 | } |
914 | 914 | ||
915 | block_start = block_end; | 915 | block_start = block_end; |
916 | bh = bh->b_this_page; | 916 | bh = bh->b_this_page; |
917 | } while (bh != head); | 917 | } while (bh != head); |
918 | } | 918 | } |
919 | 919 | ||
920 | /* | 920 | /* |
921 | * Only called when we have a failure during allocating write to write | 921 | * Only called when we have a failure during allocating write to write |
922 | * zero's to the newly allocated region. | 922 | * zero's to the newly allocated region. |
923 | */ | 923 | */ |
924 | static void ocfs2_write_failure(struct inode *inode, | 924 | static void ocfs2_write_failure(struct inode *inode, |
925 | struct ocfs2_write_ctxt *wc, | 925 | struct ocfs2_write_ctxt *wc, |
926 | loff_t user_pos, unsigned user_len) | 926 | loff_t user_pos, unsigned user_len) |
927 | { | 927 | { |
928 | int i; | 928 | int i; |
929 | unsigned from, to; | 929 | unsigned from, to; |
930 | struct page *tmppage; | 930 | struct page *tmppage; |
931 | 931 | ||
932 | ocfs2_zero_new_buffers(wc->w_target_page, user_pos, user_len); | 932 | ocfs2_zero_new_buffers(wc->w_target_page, user_pos, user_len); |
933 | 933 | ||
934 | if (wc->w_large_pages) { | 934 | if (wc->w_large_pages) { |
935 | from = wc->w_target_from; | 935 | from = wc->w_target_from; |
936 | to = wc->w_target_to; | 936 | to = wc->w_target_to; |
937 | } else { | 937 | } else { |
938 | from = 0; | 938 | from = 0; |
939 | to = PAGE_CACHE_SIZE; | 939 | to = PAGE_CACHE_SIZE; |
940 | } | 940 | } |
941 | 941 | ||
942 | for(i = 0; i < wc->w_num_pages; i++) { | 942 | for(i = 0; i < wc->w_num_pages; i++) { |
943 | tmppage = wc->w_pages[i]; | 943 | tmppage = wc->w_pages[i]; |
944 | 944 | ||
945 | if (ocfs2_should_order_data(inode)) | 945 | if (ocfs2_should_order_data(inode)) |
946 | walk_page_buffers(wc->w_handle, page_buffers(tmppage), | 946 | walk_page_buffers(wc->w_handle, page_buffers(tmppage), |
947 | from, to, NULL, | 947 | from, to, NULL, |
948 | ocfs2_journal_dirty_data); | 948 | ocfs2_journal_dirty_data); |
949 | 949 | ||
950 | block_commit_write(tmppage, from, to); | 950 | block_commit_write(tmppage, from, to); |
951 | } | 951 | } |
952 | } | 952 | } |
953 | 953 | ||
954 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, | 954 | static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno, |
955 | struct ocfs2_write_ctxt *wc, | 955 | struct ocfs2_write_ctxt *wc, |
956 | struct page *page, u32 cpos, | 956 | struct page *page, u32 cpos, |
957 | loff_t user_pos, unsigned user_len, | 957 | loff_t user_pos, unsigned user_len, |
958 | int new) | 958 | int new) |
959 | { | 959 | { |
960 | int ret; | 960 | int ret; |
961 | unsigned int map_from = 0, map_to = 0; | 961 | unsigned int map_from = 0, map_to = 0; |
962 | unsigned int cluster_start, cluster_end; | 962 | unsigned int cluster_start, cluster_end; |
963 | unsigned int user_data_from = 0, user_data_to = 0; | 963 | unsigned int user_data_from = 0, user_data_to = 0; |
964 | 964 | ||
965 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, | 965 | ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos, |
966 | &cluster_start, &cluster_end); | 966 | &cluster_start, &cluster_end); |
967 | 967 | ||
968 | if (page == wc->w_target_page) { | 968 | if (page == wc->w_target_page) { |
969 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); | 969 | map_from = user_pos & (PAGE_CACHE_SIZE - 1); |
970 | map_to = map_from + user_len; | 970 | map_to = map_from + user_len; |
971 | 971 | ||
972 | if (new) | 972 | if (new) |
973 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | 973 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, |
974 | cluster_start, cluster_end, | 974 | cluster_start, cluster_end, |
975 | new); | 975 | new); |
976 | else | 976 | else |
977 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | 977 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, |
978 | map_from, map_to, new); | 978 | map_from, map_to, new); |
979 | if (ret) { | 979 | if (ret) { |
980 | mlog_errno(ret); | 980 | mlog_errno(ret); |
981 | goto out; | 981 | goto out; |
982 | } | 982 | } |
983 | 983 | ||
984 | user_data_from = map_from; | 984 | user_data_from = map_from; |
985 | user_data_to = map_to; | 985 | user_data_to = map_to; |
986 | if (new) { | 986 | if (new) { |
987 | map_from = cluster_start; | 987 | map_from = cluster_start; |
988 | map_to = cluster_end; | 988 | map_to = cluster_end; |
989 | } | 989 | } |
990 | 990 | ||
991 | wc->w_target_from = map_from; | 991 | wc->w_target_from = map_from; |
992 | wc->w_target_to = map_to; | 992 | wc->w_target_to = map_to; |
993 | } else { | 993 | } else { |
994 | /* | 994 | /* |
995 | * If we haven't allocated the new page yet, we | 995 | * If we haven't allocated the new page yet, we |
996 | * shouldn't be writing it out without copying user | 996 | * shouldn't be writing it out without copying user |
997 | * data. This is likely a math error from the caller. | 997 | * data. This is likely a math error from the caller. |
998 | */ | 998 | */ |
999 | BUG_ON(!new); | 999 | BUG_ON(!new); |
1000 | 1000 | ||
1001 | map_from = cluster_start; | 1001 | map_from = cluster_start; |
1002 | map_to = cluster_end; | 1002 | map_to = cluster_end; |
1003 | 1003 | ||
1004 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, | 1004 | ret = ocfs2_map_page_blocks(page, p_blkno, inode, |
1005 | cluster_start, cluster_end, new); | 1005 | cluster_start, cluster_end, new); |
1006 | if (ret) { | 1006 | if (ret) { |
1007 | mlog_errno(ret); | 1007 | mlog_errno(ret); |
1008 | goto out; | 1008 | goto out; |
1009 | } | 1009 | } |
1010 | } | 1010 | } |
1011 | 1011 | ||
1012 | /* | 1012 | /* |
1013 | * Parts of newly allocated pages need to be zero'd. | 1013 | * Parts of newly allocated pages need to be zero'd. |
1014 | * | 1014 | * |
1015 | * Above, we have also rewritten 'to' and 'from' - as far as | 1015 | * Above, we have also rewritten 'to' and 'from' - as far as |
1016 | * the rest of the function is concerned, the entire cluster | 1016 | * the rest of the function is concerned, the entire cluster |
1017 | * range inside of a page needs to be written. | 1017 | * range inside of a page needs to be written. |
1018 | * | 1018 | * |
1019 | * We can skip this if the page is up to date - it's already | 1019 | * We can skip this if the page is up to date - it's already |
1020 | * been zero'd from being read in as a hole. | 1020 | * been zero'd from being read in as a hole. |
1021 | */ | 1021 | */ |
1022 | if (new && !PageUptodate(page)) | 1022 | if (new && !PageUptodate(page)) |
1023 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), | 1023 | ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb), |
1024 | cpos, user_data_from, user_data_to); | 1024 | cpos, user_data_from, user_data_to); |
1025 | 1025 | ||
1026 | flush_dcache_page(page); | 1026 | flush_dcache_page(page); |
1027 | 1027 | ||
1028 | out: | 1028 | out: |
1029 | return ret; | 1029 | return ret; |
1030 | } | 1030 | } |
1031 | 1031 | ||
1032 | /* | 1032 | /* |
1033 | * This function will only grab one clusters worth of pages. | 1033 | * This function will only grab one clusters worth of pages. |
1034 | */ | 1034 | */ |
1035 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, | 1035 | static int ocfs2_grab_pages_for_write(struct address_space *mapping, |
1036 | struct ocfs2_write_ctxt *wc, | 1036 | struct ocfs2_write_ctxt *wc, |
1037 | u32 cpos, loff_t user_pos, int new) | 1037 | u32 cpos, loff_t user_pos, int new, |
1038 | struct page *mmap_page) | ||
1038 | { | 1039 | { |
1039 | int ret = 0, i; | 1040 | int ret = 0, i; |
1040 | unsigned long start, target_index, index; | 1041 | unsigned long start, target_index, index; |
1041 | struct inode *inode = mapping->host; | 1042 | struct inode *inode = mapping->host; |
1042 | 1043 | ||
1043 | target_index = user_pos >> PAGE_CACHE_SHIFT; | 1044 | target_index = user_pos >> PAGE_CACHE_SHIFT; |
1044 | 1045 | ||
1045 | /* | 1046 | /* |
1046 | * Figure out how many pages we'll be manipulating here. For | 1047 | * Figure out how many pages we'll be manipulating here. For |
1047 | * non allocating write, we just change the one | 1048 | * non allocating write, we just change the one |
1048 | * page. Otherwise, we'll need a whole clusters worth. | 1049 | * page. Otherwise, we'll need a whole clusters worth. |
1049 | */ | 1050 | */ |
1050 | if (new) { | 1051 | if (new) { |
1051 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); | 1052 | wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb); |
1052 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); | 1053 | start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos); |
1053 | } else { | 1054 | } else { |
1054 | wc->w_num_pages = 1; | 1055 | wc->w_num_pages = 1; |
1055 | start = target_index; | 1056 | start = target_index; |
1056 | } | 1057 | } |
1057 | 1058 | ||
1058 | for(i = 0; i < wc->w_num_pages; i++) { | 1059 | for(i = 0; i < wc->w_num_pages; i++) { |
1059 | index = start + i; | 1060 | index = start + i; |
1060 | 1061 | ||
1061 | wc->w_pages[i] = find_or_create_page(mapping, index, GFP_NOFS); | 1062 | if (index == target_index && mmap_page) { |
1062 | if (!wc->w_pages[i]) { | 1063 | /* |
1063 | ret = -ENOMEM; | 1064 | * ocfs2_pagemkwrite() is a little different |
1064 | mlog_errno(ret); | 1065 | * and wants us to directly use the page |
1065 | goto out; | 1066 | * passed in. |
1067 | */ | ||
1068 | lock_page(mmap_page); | ||
1069 | |||
1070 | if (mmap_page->mapping != mapping) { | ||
1071 | unlock_page(mmap_page); | ||
1072 | /* | ||
1073 | * Sanity check - the locking in | ||
1074 | * ocfs2_pagemkwrite() should ensure | ||
1075 | * that this code doesn't trigger. | ||
1076 | */ | ||
1077 | ret = -EINVAL; | ||
1078 | mlog_errno(ret); | ||
1079 | goto out; | ||
1080 | } | ||
1081 | |||
1082 | page_cache_get(mmap_page); | ||
1083 | wc->w_pages[i] = mmap_page; | ||
1084 | } else { | ||
1085 | wc->w_pages[i] = find_or_create_page(mapping, index, | ||
1086 | GFP_NOFS); | ||
1087 | if (!wc->w_pages[i]) { | ||
1088 | ret = -ENOMEM; | ||
1089 | mlog_errno(ret); | ||
1090 | goto out; | ||
1091 | } | ||
1066 | } | 1092 | } |
1067 | 1093 | ||
1068 | if (index == target_index) | 1094 | if (index == target_index) |
1069 | wc->w_target_page = wc->w_pages[i]; | 1095 | wc->w_target_page = wc->w_pages[i]; |
1070 | } | 1096 | } |
1071 | out: | 1097 | out: |
1072 | return ret; | 1098 | return ret; |
1073 | } | 1099 | } |
1074 | 1100 | ||
1075 | /* | 1101 | /* |
1076 | * Prepare a single cluster for write one cluster into the file. | 1102 | * Prepare a single cluster for write one cluster into the file. |
1077 | */ | 1103 | */ |
1078 | static int ocfs2_write_cluster(struct address_space *mapping, | 1104 | static int ocfs2_write_cluster(struct address_space *mapping, |
1079 | u32 phys, struct ocfs2_alloc_context *data_ac, | 1105 | u32 phys, struct ocfs2_alloc_context *data_ac, |
1080 | struct ocfs2_alloc_context *meta_ac, | 1106 | struct ocfs2_alloc_context *meta_ac, |
1081 | struct ocfs2_write_ctxt *wc, u32 cpos, | 1107 | struct ocfs2_write_ctxt *wc, u32 cpos, |
1082 | loff_t user_pos, unsigned user_len) | 1108 | loff_t user_pos, unsigned user_len) |
1083 | { | 1109 | { |
1084 | int ret, i, new; | 1110 | int ret, i, new; |
1085 | u64 v_blkno, p_blkno; | 1111 | u64 v_blkno, p_blkno; |
1086 | struct inode *inode = mapping->host; | 1112 | struct inode *inode = mapping->host; |
1087 | 1113 | ||
1088 | new = phys == 0 ? 1 : 0; | 1114 | new = phys == 0 ? 1 : 0; |
1089 | 1115 | ||
1090 | if (new) { | 1116 | if (new) { |
1091 | u32 tmp_pos; | 1117 | u32 tmp_pos; |
1092 | 1118 | ||
1093 | /* | 1119 | /* |
1094 | * This is safe to call with the page locks - it won't take | 1120 | * This is safe to call with the page locks - it won't take |
1095 | * any additional semaphores or cluster locks. | 1121 | * any additional semaphores or cluster locks. |
1096 | */ | 1122 | */ |
1097 | tmp_pos = cpos; | 1123 | tmp_pos = cpos; |
1098 | ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode, | 1124 | ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode, |
1099 | &tmp_pos, 1, wc->w_di_bh, | 1125 | &tmp_pos, 1, wc->w_di_bh, |
1100 | wc->w_handle, data_ac, | 1126 | wc->w_handle, data_ac, |
1101 | meta_ac, NULL); | 1127 | meta_ac, NULL); |
1102 | /* | 1128 | /* |
1103 | * This shouldn't happen because we must have already | 1129 | * This shouldn't happen because we must have already |
1104 | * calculated the correct meta data allocation required. The | 1130 | * calculated the correct meta data allocation required. The |
1105 | * internal tree allocation code should know how to increase | 1131 | * internal tree allocation code should know how to increase |
1106 | * transaction credits itself. | 1132 | * transaction credits itself. |
1107 | * | 1133 | * |
1108 | * If need be, we could handle -EAGAIN for a | 1134 | * If need be, we could handle -EAGAIN for a |
1109 | * RESTART_TRANS here. | 1135 | * RESTART_TRANS here. |
1110 | */ | 1136 | */ |
1111 | mlog_bug_on_msg(ret == -EAGAIN, | 1137 | mlog_bug_on_msg(ret == -EAGAIN, |
1112 | "Inode %llu: EAGAIN return during allocation.\n", | 1138 | "Inode %llu: EAGAIN return during allocation.\n", |
1113 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | 1139 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
1114 | if (ret < 0) { | 1140 | if (ret < 0) { |
1115 | mlog_errno(ret); | 1141 | mlog_errno(ret); |
1116 | goto out; | 1142 | goto out; |
1117 | } | 1143 | } |
1118 | 1144 | ||
1119 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); | 1145 | v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos); |
1120 | } else { | 1146 | } else { |
1121 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; | 1147 | v_blkno = user_pos >> inode->i_sb->s_blocksize_bits; |
1122 | } | 1148 | } |
1123 | 1149 | ||
1124 | /* | 1150 | /* |
1125 | * The only reason this should fail is due to an inability to | 1151 | * The only reason this should fail is due to an inability to |
1126 | * find the extent added. | 1152 | * find the extent added. |
1127 | */ | 1153 | */ |
1128 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, | 1154 | ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL, |
1129 | NULL); | 1155 | NULL); |
1130 | if (ret < 0) { | 1156 | if (ret < 0) { |
1131 | ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, " | 1157 | ocfs2_error(inode->i_sb, "Corrupting extend for inode %llu, " |
1132 | "at logical block %llu", | 1158 | "at logical block %llu", |
1133 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | 1159 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
1134 | (unsigned long long)v_blkno); | 1160 | (unsigned long long)v_blkno); |
1135 | goto out; | 1161 | goto out; |
1136 | } | 1162 | } |
1137 | 1163 | ||
1138 | BUG_ON(p_blkno == 0); | 1164 | BUG_ON(p_blkno == 0); |
1139 | 1165 | ||
1140 | for(i = 0; i < wc->w_num_pages; i++) { | 1166 | for(i = 0; i < wc->w_num_pages; i++) { |
1141 | int tmpret; | 1167 | int tmpret; |
1142 | 1168 | ||
1143 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, | 1169 | tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc, |
1144 | wc->w_pages[i], cpos, | 1170 | wc->w_pages[i], cpos, |
1145 | user_pos, user_len, new); | 1171 | user_pos, user_len, new); |
1146 | if (tmpret) { | 1172 | if (tmpret) { |
1147 | mlog_errno(tmpret); | 1173 | mlog_errno(tmpret); |
1148 | if (ret == 0) | 1174 | if (ret == 0) |
1149 | tmpret = ret; | 1175 | tmpret = ret; |
1150 | } | 1176 | } |
1151 | } | 1177 | } |
1152 | 1178 | ||
1153 | /* | 1179 | /* |
1154 | * We only have cleanup to do in case of allocating write. | 1180 | * We only have cleanup to do in case of allocating write. |
1155 | */ | 1181 | */ |
1156 | if (ret && new) | 1182 | if (ret && new) |
1157 | ocfs2_write_failure(inode, wc, user_pos, user_len); | 1183 | ocfs2_write_failure(inode, wc, user_pos, user_len); |
1158 | 1184 | ||
1159 | out: | 1185 | out: |
1160 | 1186 | ||
1161 | return ret; | 1187 | return ret; |
1162 | } | 1188 | } |
1163 | 1189 | ||
1164 | /* | 1190 | /* |
1165 | * ocfs2_write_end() wants to know which parts of the target page it | 1191 | * ocfs2_write_end() wants to know which parts of the target page it |
1166 | * should complete the write on. It's easiest to compute them ahead of | 1192 | * should complete the write on. It's easiest to compute them ahead of |
1167 | * time when a more complete view of the write is available. | 1193 | * time when a more complete view of the write is available. |
1168 | */ | 1194 | */ |
1169 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, | 1195 | static void ocfs2_set_target_boundaries(struct ocfs2_super *osb, |
1170 | struct ocfs2_write_ctxt *wc, | 1196 | struct ocfs2_write_ctxt *wc, |
1171 | loff_t pos, unsigned len, int alloc) | 1197 | loff_t pos, unsigned len, int alloc) |
1172 | { | 1198 | { |
1173 | struct ocfs2_write_cluster_desc *desc; | 1199 | struct ocfs2_write_cluster_desc *desc; |
1174 | 1200 | ||
1175 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); | 1201 | wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1); |
1176 | wc->w_target_to = wc->w_target_from + len; | 1202 | wc->w_target_to = wc->w_target_from + len; |
1177 | 1203 | ||
1178 | if (alloc == 0) | 1204 | if (alloc == 0) |
1179 | return; | 1205 | return; |
1180 | 1206 | ||
1181 | /* | 1207 | /* |
1182 | * Allocating write - we may have different boundaries based | 1208 | * Allocating write - we may have different boundaries based |
1183 | * on page size and cluster size. | 1209 | * on page size and cluster size. |
1184 | * | 1210 | * |
1185 | * NOTE: We can no longer compute one value from the other as | 1211 | * NOTE: We can no longer compute one value from the other as |
1186 | * the actual write length and user provided length may be | 1212 | * the actual write length and user provided length may be |
1187 | * different. | 1213 | * different. |
1188 | */ | 1214 | */ |
1189 | 1215 | ||
1190 | if (wc->w_large_pages) { | 1216 | if (wc->w_large_pages) { |
1191 | /* | 1217 | /* |
1192 | * We only care about the 1st and last cluster within | 1218 | * We only care about the 1st and last cluster within |
1193 | * our range and whether they are holes or not. Either | 1219 | * our range and whether they are holes or not. Either |
1194 | * value may be extended out to the start/end of a | 1220 | * value may be extended out to the start/end of a |
1195 | * newly allocated cluster. | 1221 | * newly allocated cluster. |
1196 | */ | 1222 | */ |
1197 | desc = &wc->w_desc[0]; | 1223 | desc = &wc->w_desc[0]; |
1198 | if (desc->c_new) | 1224 | if (desc->c_new) |
1199 | ocfs2_figure_cluster_boundaries(osb, | 1225 | ocfs2_figure_cluster_boundaries(osb, |
1200 | desc->c_cpos, | 1226 | desc->c_cpos, |
1201 | &wc->w_target_from, | 1227 | &wc->w_target_from, |
1202 | NULL); | 1228 | NULL); |
1203 | 1229 | ||
1204 | desc = &wc->w_desc[wc->w_clen - 1]; | 1230 | desc = &wc->w_desc[wc->w_clen - 1]; |
1205 | if (desc->c_new) | 1231 | if (desc->c_new) |
1206 | ocfs2_figure_cluster_boundaries(osb, | 1232 | ocfs2_figure_cluster_boundaries(osb, |
1207 | desc->c_cpos, | 1233 | desc->c_cpos, |
1208 | NULL, | 1234 | NULL, |
1209 | &wc->w_target_to); | 1235 | &wc->w_target_to); |
1210 | } else { | 1236 | } else { |
1211 | wc->w_target_from = 0; | 1237 | wc->w_target_from = 0; |
1212 | wc->w_target_to = PAGE_CACHE_SIZE; | 1238 | wc->w_target_to = PAGE_CACHE_SIZE; |
1213 | } | 1239 | } |
1214 | } | 1240 | } |
1215 | 1241 | ||
1216 | static int ocfs2_write_begin_nolock(struct address_space *mapping, | 1242 | int ocfs2_write_begin_nolock(struct address_space *mapping, |
1217 | loff_t pos, unsigned len, unsigned flags, | 1243 | loff_t pos, unsigned len, unsigned flags, |
1218 | struct page **pagep, void **fsdata, | 1244 | struct page **pagep, void **fsdata, |
1219 | struct buffer_head *di_bh) | 1245 | struct buffer_head *di_bh, struct page *mmap_page) |
1220 | { | 1246 | { |
1221 | int ret, i, credits = OCFS2_INODE_UPDATE_CREDITS; | 1247 | int ret, i, credits = OCFS2_INODE_UPDATE_CREDITS; |
1222 | unsigned int num_clusters = 0, clusters_to_alloc = 0; | 1248 | unsigned int num_clusters = 0, clusters_to_alloc = 0; |
1223 | u32 phys = 0; | 1249 | u32 phys = 0; |
1224 | struct ocfs2_write_ctxt *wc; | 1250 | struct ocfs2_write_ctxt *wc; |
1225 | struct inode *inode = mapping->host; | 1251 | struct inode *inode = mapping->host; |
1226 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 1252 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1227 | struct ocfs2_dinode *di; | 1253 | struct ocfs2_dinode *di; |
1228 | struct ocfs2_alloc_context *data_ac = NULL; | 1254 | struct ocfs2_alloc_context *data_ac = NULL; |
1229 | struct ocfs2_alloc_context *meta_ac = NULL; | 1255 | struct ocfs2_alloc_context *meta_ac = NULL; |
1230 | handle_t *handle; | 1256 | handle_t *handle; |
1231 | struct ocfs2_write_cluster_desc *desc; | 1257 | struct ocfs2_write_cluster_desc *desc; |
1232 | 1258 | ||
1233 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); | 1259 | ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh); |
1234 | if (ret) { | 1260 | if (ret) { |
1235 | mlog_errno(ret); | 1261 | mlog_errno(ret); |
1236 | return ret; | 1262 | return ret; |
1237 | } | 1263 | } |
1238 | 1264 | ||
1239 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | 1265 | di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1240 | 1266 | ||
1241 | for (i = 0; i < wc->w_clen; i++) { | 1267 | for (i = 0; i < wc->w_clen; i++) { |
1242 | desc = &wc->w_desc[i]; | 1268 | desc = &wc->w_desc[i]; |
1243 | desc->c_cpos = wc->w_cpos + i; | 1269 | desc->c_cpos = wc->w_cpos + i; |
1244 | 1270 | ||
1245 | if (num_clusters == 0) { | 1271 | if (num_clusters == 0) { |
1246 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, | 1272 | ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys, |
1247 | &num_clusters, NULL); | 1273 | &num_clusters, NULL); |
1248 | if (ret) { | 1274 | if (ret) { |
1249 | mlog_errno(ret); | 1275 | mlog_errno(ret); |
1250 | goto out; | 1276 | goto out; |
1251 | } | 1277 | } |
1252 | } else if (phys) { | 1278 | } else if (phys) { |
1253 | /* | 1279 | /* |
1254 | * Only increment phys if it doesn't describe | 1280 | * Only increment phys if it doesn't describe |
1255 | * a hole. | 1281 | * a hole. |
1256 | */ | 1282 | */ |
1257 | phys++; | 1283 | phys++; |
1258 | } | 1284 | } |
1259 | 1285 | ||
1260 | desc->c_phys = phys; | 1286 | desc->c_phys = phys; |
1261 | if (phys == 0) { | 1287 | if (phys == 0) { |
1262 | desc->c_new = 1; | 1288 | desc->c_new = 1; |
1263 | clusters_to_alloc++; | 1289 | clusters_to_alloc++; |
1264 | } | 1290 | } |
1265 | 1291 | ||
1266 | num_clusters--; | 1292 | num_clusters--; |
1267 | } | 1293 | } |
1268 | 1294 | ||
1269 | /* | 1295 | /* |
1270 | * We set w_target_from, w_target_to here so that | 1296 | * We set w_target_from, w_target_to here so that |
1271 | * ocfs2_write_end() knows which range in the target page to | 1297 | * ocfs2_write_end() knows which range in the target page to |
1272 | * write out. An allocation requires that we write the entire | 1298 | * write out. An allocation requires that we write the entire |
1273 | * cluster range. | 1299 | * cluster range. |
1274 | */ | 1300 | */ |
1275 | if (clusters_to_alloc > 0) { | 1301 | if (clusters_to_alloc > 0) { |
1276 | /* | 1302 | /* |
1277 | * XXX: We are stretching the limits of | 1303 | * XXX: We are stretching the limits of |
1278 | * ocfs2_lock_allocators(). It greately over-estimates | 1304 | * ocfs2_lock_allocators(). It greately over-estimates |
1279 | * the work to be done. | 1305 | * the work to be done. |
1280 | */ | 1306 | */ |
1281 | ret = ocfs2_lock_allocators(inode, di, clusters_to_alloc, | 1307 | ret = ocfs2_lock_allocators(inode, di, clusters_to_alloc, |
1282 | &data_ac, &meta_ac); | 1308 | &data_ac, &meta_ac); |
1283 | if (ret) { | 1309 | if (ret) { |
1284 | mlog_errno(ret); | 1310 | mlog_errno(ret); |
1285 | goto out; | 1311 | goto out; |
1286 | } | 1312 | } |
1287 | 1313 | ||
1288 | credits = ocfs2_calc_extend_credits(inode->i_sb, di, | 1314 | credits = ocfs2_calc_extend_credits(inode->i_sb, di, |
1289 | clusters_to_alloc); | 1315 | clusters_to_alloc); |
1290 | 1316 | ||
1291 | } | 1317 | } |
1292 | 1318 | ||
1293 | ocfs2_set_target_boundaries(osb, wc, pos, len, clusters_to_alloc); | 1319 | ocfs2_set_target_boundaries(osb, wc, pos, len, clusters_to_alloc); |
1294 | 1320 | ||
1295 | handle = ocfs2_start_trans(osb, credits); | 1321 | handle = ocfs2_start_trans(osb, credits); |
1296 | if (IS_ERR(handle)) { | 1322 | if (IS_ERR(handle)) { |
1297 | ret = PTR_ERR(handle); | 1323 | ret = PTR_ERR(handle); |
1298 | mlog_errno(ret); | 1324 | mlog_errno(ret); |
1299 | goto out; | 1325 | goto out; |
1300 | } | 1326 | } |
1301 | 1327 | ||
1302 | wc->w_handle = handle; | 1328 | wc->w_handle = handle; |
1303 | 1329 | ||
1304 | /* | 1330 | /* |
1305 | * We don't want this to fail in ocfs2_write_end(), so do it | 1331 | * We don't want this to fail in ocfs2_write_end(), so do it |
1306 | * here. | 1332 | * here. |
1307 | */ | 1333 | */ |
1308 | ret = ocfs2_journal_access(handle, inode, wc->w_di_bh, | 1334 | ret = ocfs2_journal_access(handle, inode, wc->w_di_bh, |
1309 | OCFS2_JOURNAL_ACCESS_WRITE); | 1335 | OCFS2_JOURNAL_ACCESS_WRITE); |
1310 | if (ret) { | 1336 | if (ret) { |
1311 | mlog_errno(ret); | 1337 | mlog_errno(ret); |
1312 | goto out_commit; | 1338 | goto out_commit; |
1313 | } | 1339 | } |
1314 | 1340 | ||
1315 | /* | 1341 | /* |
1316 | * Fill our page array first. That way we've grabbed enough so | 1342 | * Fill our page array first. That way we've grabbed enough so |
1317 | * that we can zero and flush if we error after adding the | 1343 | * that we can zero and flush if we error after adding the |
1318 | * extent. | 1344 | * extent. |
1319 | */ | 1345 | */ |
1320 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, | 1346 | ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, |
1321 | clusters_to_alloc); | 1347 | clusters_to_alloc, mmap_page); |
1322 | if (ret) { | 1348 | if (ret) { |
1323 | mlog_errno(ret); | 1349 | mlog_errno(ret); |
1324 | goto out_commit; | 1350 | goto out_commit; |
1325 | } | 1351 | } |
1326 | 1352 | ||
1327 | for (i = 0; i < wc->w_clen; i++) { | 1353 | for (i = 0; i < wc->w_clen; i++) { |
1328 | desc = &wc->w_desc[i]; | 1354 | desc = &wc->w_desc[i]; |
1329 | 1355 | ||
1330 | ret = ocfs2_write_cluster(mapping, desc->c_phys, data_ac, | 1356 | ret = ocfs2_write_cluster(mapping, desc->c_phys, data_ac, |
1331 | meta_ac, wc, desc->c_cpos, pos, len); | 1357 | meta_ac, wc, desc->c_cpos, pos, len); |
1332 | if (ret) { | 1358 | if (ret) { |
1333 | mlog_errno(ret); | 1359 | mlog_errno(ret); |
1334 | goto out_commit; | 1360 | goto out_commit; |
1335 | } | 1361 | } |
1336 | } | 1362 | } |
1337 | 1363 | ||
1338 | if (data_ac) | 1364 | if (data_ac) |
1339 | ocfs2_free_alloc_context(data_ac); | 1365 | ocfs2_free_alloc_context(data_ac); |
1340 | if (meta_ac) | 1366 | if (meta_ac) |
1341 | ocfs2_free_alloc_context(meta_ac); | 1367 | ocfs2_free_alloc_context(meta_ac); |
1342 | 1368 | ||
1343 | *pagep = wc->w_target_page; | 1369 | *pagep = wc->w_target_page; |
1344 | *fsdata = wc; | 1370 | *fsdata = wc; |
1345 | return 0; | 1371 | return 0; |
1346 | out_commit: | 1372 | out_commit: |
1347 | ocfs2_commit_trans(osb, handle); | 1373 | ocfs2_commit_trans(osb, handle); |
1348 | 1374 | ||
1349 | out: | 1375 | out: |
1350 | ocfs2_free_write_ctxt(wc); | 1376 | ocfs2_free_write_ctxt(wc); |
1351 | 1377 | ||
1352 | if (data_ac) | 1378 | if (data_ac) |
1353 | ocfs2_free_alloc_context(data_ac); | 1379 | ocfs2_free_alloc_context(data_ac); |
1354 | if (meta_ac) | 1380 | if (meta_ac) |
1355 | ocfs2_free_alloc_context(meta_ac); | 1381 | ocfs2_free_alloc_context(meta_ac); |
1356 | return ret; | 1382 | return ret; |
1357 | } | 1383 | } |
1358 | 1384 | ||
1359 | int ocfs2_write_begin(struct file *file, struct address_space *mapping, | 1385 | int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
1360 | loff_t pos, unsigned len, unsigned flags, | 1386 | loff_t pos, unsigned len, unsigned flags, |
1361 | struct page **pagep, void **fsdata) | 1387 | struct page **pagep, void **fsdata) |
1362 | { | 1388 | { |
1363 | int ret; | 1389 | int ret; |
1364 | struct buffer_head *di_bh = NULL; | 1390 | struct buffer_head *di_bh = NULL; |
1365 | struct inode *inode = mapping->host; | 1391 | struct inode *inode = mapping->host; |
1366 | 1392 | ||
1367 | ret = ocfs2_meta_lock(inode, &di_bh, 1); | 1393 | ret = ocfs2_meta_lock(inode, &di_bh, 1); |
1368 | if (ret) { | 1394 | if (ret) { |
1369 | mlog_errno(ret); | 1395 | mlog_errno(ret); |
1370 | return ret; | 1396 | return ret; |
1371 | } | 1397 | } |
1372 | 1398 | ||
1373 | /* | 1399 | /* |
1374 | * Take alloc sem here to prevent concurrent lookups. That way | 1400 | * Take alloc sem here to prevent concurrent lookups. That way |
1375 | * the mapping, zeroing and tree manipulation within | 1401 | * the mapping, zeroing and tree manipulation within |
1376 | * ocfs2_write() will be safe against ->readpage(). This | 1402 | * ocfs2_write() will be safe against ->readpage(). This |
1377 | * should also serve to lock out allocation from a shared | 1403 | * should also serve to lock out allocation from a shared |
1378 | * writeable region. | 1404 | * writeable region. |
1379 | */ | 1405 | */ |
1380 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | 1406 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
1381 | 1407 | ||
1382 | ret = ocfs2_data_lock(inode, 1); | 1408 | ret = ocfs2_data_lock(inode, 1); |
1383 | if (ret) { | 1409 | if (ret) { |
1384 | mlog_errno(ret); | 1410 | mlog_errno(ret); |
1385 | goto out_fail; | 1411 | goto out_fail; |
1386 | } | 1412 | } |
1387 | 1413 | ||
1388 | ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep, | 1414 | ret = ocfs2_write_begin_nolock(mapping, pos, len, flags, pagep, |
1389 | fsdata, di_bh); | 1415 | fsdata, di_bh, NULL); |
1390 | if (ret) { | 1416 | if (ret) { |
1391 | mlog_errno(ret); | 1417 | mlog_errno(ret); |
1392 | goto out_fail_data; | 1418 | goto out_fail_data; |
1393 | } | 1419 | } |
1394 | 1420 | ||
1395 | brelse(di_bh); | 1421 | brelse(di_bh); |
1396 | 1422 | ||
1397 | return 0; | 1423 | return 0; |
1398 | 1424 | ||
1399 | out_fail_data: | 1425 | out_fail_data: |
1400 | ocfs2_data_unlock(inode, 1); | 1426 | ocfs2_data_unlock(inode, 1); |
1401 | out_fail: | 1427 | out_fail: |
1402 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | 1428 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
1403 | 1429 | ||
1404 | brelse(di_bh); | 1430 | brelse(di_bh); |
1405 | ocfs2_meta_unlock(inode, 1); | 1431 | ocfs2_meta_unlock(inode, 1); |
1406 | 1432 | ||
1407 | return ret; | 1433 | return ret; |
1408 | } | 1434 | } |
1409 | 1435 | ||
1410 | static int ocfs2_write_end_nolock(struct address_space *mapping, | 1436 | int ocfs2_write_end_nolock(struct address_space *mapping, |
1411 | loff_t pos, unsigned len, unsigned copied, | 1437 | loff_t pos, unsigned len, unsigned copied, |
1412 | struct page *page, void *fsdata) | 1438 | struct page *page, void *fsdata) |
1413 | { | 1439 | { |
1414 | int i; | 1440 | int i; |
1415 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); | 1441 | unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1); |
1416 | struct inode *inode = mapping->host; | 1442 | struct inode *inode = mapping->host; |
1417 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 1443 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1418 | struct ocfs2_write_ctxt *wc = fsdata; | 1444 | struct ocfs2_write_ctxt *wc = fsdata; |
1419 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; | 1445 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data; |
1420 | handle_t *handle = wc->w_handle; | 1446 | handle_t *handle = wc->w_handle; |
1421 | struct page *tmppage; | 1447 | struct page *tmppage; |
1422 | 1448 | ||
1423 | if (unlikely(copied < len)) { | 1449 | if (unlikely(copied < len)) { |
1424 | if (!PageUptodate(wc->w_target_page)) | 1450 | if (!PageUptodate(wc->w_target_page)) |
1425 | copied = 0; | 1451 | copied = 0; |
1426 | 1452 | ||
1427 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, | 1453 | ocfs2_zero_new_buffers(wc->w_target_page, start+copied, |
1428 | start+len); | 1454 | start+len); |
1429 | } | 1455 | } |
1430 | flush_dcache_page(wc->w_target_page); | 1456 | flush_dcache_page(wc->w_target_page); |
1431 | 1457 | ||
1432 | for(i = 0; i < wc->w_num_pages; i++) { | 1458 | for(i = 0; i < wc->w_num_pages; i++) { |
1433 | tmppage = wc->w_pages[i]; | 1459 | tmppage = wc->w_pages[i]; |
1434 | 1460 | ||
1435 | if (tmppage == wc->w_target_page) { | 1461 | if (tmppage == wc->w_target_page) { |
1436 | from = wc->w_target_from; | 1462 | from = wc->w_target_from; |
1437 | to = wc->w_target_to; | 1463 | to = wc->w_target_to; |
1438 | 1464 | ||
1439 | BUG_ON(from > PAGE_CACHE_SIZE || | 1465 | BUG_ON(from > PAGE_CACHE_SIZE || |
1440 | to > PAGE_CACHE_SIZE || | 1466 | to > PAGE_CACHE_SIZE || |
1441 | to < from); | 1467 | to < from); |
1442 | } else { | 1468 | } else { |
1443 | /* | 1469 | /* |
1444 | * Pages adjacent to the target (if any) imply | 1470 | * Pages adjacent to the target (if any) imply |
1445 | * a hole-filling write in which case we want | 1471 | * a hole-filling write in which case we want |
1446 | * to flush their entire range. | 1472 | * to flush their entire range. |
1447 | */ | 1473 | */ |
1448 | from = 0; | 1474 | from = 0; |
1449 | to = PAGE_CACHE_SIZE; | 1475 | to = PAGE_CACHE_SIZE; |
1450 | } | 1476 | } |
1451 | 1477 | ||
1452 | if (ocfs2_should_order_data(inode)) | 1478 | if (ocfs2_should_order_data(inode)) |
1453 | walk_page_buffers(wc->w_handle, page_buffers(tmppage), | 1479 | walk_page_buffers(wc->w_handle, page_buffers(tmppage), |
1454 | from, to, NULL, | 1480 | from, to, NULL, |
1455 | ocfs2_journal_dirty_data); | 1481 | ocfs2_journal_dirty_data); |
1456 | 1482 | ||
1457 | block_commit_write(tmppage, from, to); | 1483 | block_commit_write(tmppage, from, to); |
1458 | } | 1484 | } |
1459 | 1485 | ||
1460 | pos += copied; | 1486 | pos += copied; |
1461 | if (pos > inode->i_size) { | 1487 | if (pos > inode->i_size) { |
1462 | i_size_write(inode, pos); | 1488 | i_size_write(inode, pos); |
1463 | mark_inode_dirty(inode); | 1489 | mark_inode_dirty(inode); |
1464 | } | 1490 | } |
1465 | inode->i_blocks = ocfs2_inode_sector_count(inode); | 1491 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
1466 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | 1492 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); |
1467 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | 1493 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
1468 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | 1494 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); |
1469 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | 1495 | di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); |
1470 | 1496 | ||
1471 | ocfs2_journal_dirty(handle, wc->w_di_bh); | 1497 | ocfs2_journal_dirty(handle, wc->w_di_bh); |
1472 | 1498 | ||
1473 | ocfs2_commit_trans(osb, handle); | 1499 | ocfs2_commit_trans(osb, handle); |
1474 | ocfs2_free_write_ctxt(wc); | 1500 | ocfs2_free_write_ctxt(wc); |
1475 | 1501 | ||
1476 | return copied; | 1502 | return copied; |
1477 | } | 1503 | } |
1478 | 1504 | ||
1479 | int ocfs2_write_end(struct file *file, struct address_space *mapping, | 1505 | int ocfs2_write_end(struct file *file, struct address_space *mapping, |
1480 | loff_t pos, unsigned len, unsigned copied, | 1506 | loff_t pos, unsigned len, unsigned copied, |
1481 | struct page *page, void *fsdata) | 1507 | struct page *page, void *fsdata) |
1482 | { | 1508 | { |
1483 | int ret; | 1509 | int ret; |
1484 | struct inode *inode = mapping->host; | 1510 | struct inode *inode = mapping->host; |
1485 | 1511 | ||
1486 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); | 1512 | ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata); |
1487 | 1513 | ||
1488 | ocfs2_data_unlock(inode, 1); | 1514 | ocfs2_data_unlock(inode, 1); |
1489 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | 1515 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
1490 | ocfs2_meta_unlock(inode, 1); | 1516 | ocfs2_meta_unlock(inode, 1); |
1491 | 1517 | ||
1492 | return ret; | 1518 | return ret; |
1493 | } | 1519 | } |
1494 | 1520 | ||
1495 | const struct address_space_operations ocfs2_aops = { | 1521 | const struct address_space_operations ocfs2_aops = { |
1496 | .readpage = ocfs2_readpage, | 1522 | .readpage = ocfs2_readpage, |
1497 | .writepage = ocfs2_writepage, | 1523 | .writepage = ocfs2_writepage, |
1498 | .bmap = ocfs2_bmap, | 1524 | .bmap = ocfs2_bmap, |
1499 | .sync_page = block_sync_page, | 1525 | .sync_page = block_sync_page, |
1500 | .direct_IO = ocfs2_direct_IO, | 1526 | .direct_IO = ocfs2_direct_IO, |
1501 | .invalidatepage = ocfs2_invalidatepage, | 1527 | .invalidatepage = ocfs2_invalidatepage, |
1502 | .releasepage = ocfs2_releasepage, | 1528 | .releasepage = ocfs2_releasepage, |
1503 | .migratepage = buffer_migrate_page, | 1529 | .migratepage = buffer_migrate_page, |
1504 | }; | 1530 | }; |
1505 | 1531 |
fs/ocfs2/aops.h
1 | /* -*- mode: c; c-basic-offset: 8; -*- | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
3 | * | 3 | * |
4 | * Copyright (C) 2002, 2004, 2005 Oracle. All rights reserved. | 4 | * Copyright (C) 2002, 2004, 2005 Oracle. All rights reserved. |
5 | * | 5 | * |
6 | * This program is free software; you can redistribute it and/or | 6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public | 7 | * modify it under the terms of the GNU General Public |
8 | * License as published by the Free Software Foundation; either | 8 | * License as published by the Free Software Foundation; either |
9 | * version 2 of the License, or (at your option) any later version. | 9 | * version 2 of the License, or (at your option) any later version. |
10 | * | 10 | * |
11 | * This program is distributed in the hope that it will be useful, | 11 | * This program is distributed in the hope that it will be useful, |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | * General Public License for more details. | 14 | * General Public License for more details. |
15 | * | 15 | * |
16 | * You should have received a copy of the GNU General Public | 16 | * You should have received a copy of the GNU General Public |
17 | * License along with this program; if not, write to the | 17 | * License along with this program; if not, write to the |
18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | 18 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
19 | * Boston, MA 021110-1307, USA. | 19 | * Boston, MA 021110-1307, USA. |
20 | */ | 20 | */ |
21 | 21 | ||
22 | #ifndef OCFS2_AOPS_H | 22 | #ifndef OCFS2_AOPS_H |
23 | #define OCFS2_AOPS_H | 23 | #define OCFS2_AOPS_H |
24 | 24 | ||
25 | int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page, | 25 | int ocfs2_prepare_write_nolock(struct inode *inode, struct page *page, |
26 | unsigned from, unsigned to); | 26 | unsigned from, unsigned to); |
27 | 27 | ||
28 | handle_t *ocfs2_start_walk_page_trans(struct inode *inode, | 28 | handle_t *ocfs2_start_walk_page_trans(struct inode *inode, |
29 | struct page *page, | 29 | struct page *page, |
30 | unsigned from, | 30 | unsigned from, |
31 | unsigned to); | 31 | unsigned to); |
32 | 32 | ||
33 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, | 33 | int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno, |
34 | struct inode *inode, unsigned int from, | 34 | struct inode *inode, unsigned int from, |
35 | unsigned int to, int new); | 35 | unsigned int to, int new); |
36 | 36 | ||
37 | int walk_page_buffers( handle_t *handle, | 37 | int walk_page_buffers( handle_t *handle, |
38 | struct buffer_head *head, | 38 | struct buffer_head *head, |
39 | unsigned from, | 39 | unsigned from, |
40 | unsigned to, | 40 | unsigned to, |
41 | int *partial, | 41 | int *partial, |
42 | int (*fn)( handle_t *handle, | 42 | int (*fn)( handle_t *handle, |
43 | struct buffer_head *bh)); | 43 | struct buffer_head *bh)); |
44 | 44 | ||
45 | int ocfs2_write_begin(struct file *file, struct address_space *mapping, | 45 | int ocfs2_write_begin(struct file *file, struct address_space *mapping, |
46 | loff_t pos, unsigned len, unsigned flags, | 46 | loff_t pos, unsigned len, unsigned flags, |
47 | struct page **pagep, void **fsdata); | 47 | struct page **pagep, void **fsdata); |
48 | 48 | ||
49 | int ocfs2_write_end(struct file *file, struct address_space *mapping, | 49 | int ocfs2_write_end(struct file *file, struct address_space *mapping, |
50 | loff_t pos, unsigned len, unsigned copied, | 50 | loff_t pos, unsigned len, unsigned copied, |
51 | struct page *page, void *fsdata); | 51 | struct page *page, void *fsdata); |
52 | 52 | ||
53 | int ocfs2_write_end_nolock(struct address_space *mapping, | ||
54 | loff_t pos, unsigned len, unsigned copied, | ||
55 | struct page *page, void *fsdata); | ||
56 | |||
57 | int ocfs2_write_begin_nolock(struct address_space *mapping, | ||
58 | loff_t pos, unsigned len, unsigned flags, | ||
59 | struct page **pagep, void **fsdata, | ||
60 | struct buffer_head *di_bh, struct page *mmap_page); | ||
61 | |||
53 | /* all ocfs2_dio_end_io()'s fault */ | 62 | /* all ocfs2_dio_end_io()'s fault */ |
54 | #define ocfs2_iocb_is_rw_locked(iocb) \ | 63 | #define ocfs2_iocb_is_rw_locked(iocb) \ |
55 | test_bit(0, (unsigned long *)&iocb->private) | 64 | test_bit(0, (unsigned long *)&iocb->private) |
56 | static inline void ocfs2_iocb_set_rw_locked(struct kiocb *iocb, int level) | 65 | static inline void ocfs2_iocb_set_rw_locked(struct kiocb *iocb, int level) |
57 | { | 66 | { |
58 | set_bit(0, (unsigned long *)&iocb->private); | 67 | set_bit(0, (unsigned long *)&iocb->private); |
59 | if (level) | 68 | if (level) |
60 | set_bit(1, (unsigned long *)&iocb->private); | 69 | set_bit(1, (unsigned long *)&iocb->private); |
61 | else | 70 | else |
62 | clear_bit(1, (unsigned long *)&iocb->private); | 71 | clear_bit(1, (unsigned long *)&iocb->private); |
63 | } | 72 | } |
64 | #define ocfs2_iocb_clear_rw_locked(iocb) \ | 73 | #define ocfs2_iocb_clear_rw_locked(iocb) \ |
65 | clear_bit(0, (unsigned long *)&iocb->private) | 74 | clear_bit(0, (unsigned long *)&iocb->private) |
66 | #define ocfs2_iocb_rw_locked_level(iocb) \ | 75 | #define ocfs2_iocb_rw_locked_level(iocb) \ |
67 | test_bit(1, (unsigned long *)&iocb->private) | 76 | test_bit(1, (unsigned long *)&iocb->private) |
68 | #endif /* OCFS2_FILE_H */ | 77 | #endif /* OCFS2_FILE_H */ |
69 | 78 |
fs/ocfs2/file.c
1 | /* -*- mode: c; c-basic-offset: 8; -*- | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
3 | * | 3 | * |
4 | * file.c | 4 | * file.c |
5 | * | 5 | * |
6 | * File open, close, extend, truncate | 6 | * File open, close, extend, truncate |
7 | * | 7 | * |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
9 | * | 9 | * |
10 | * This program is free software; you can redistribute it and/or | 10 | * This program is free software; you can redistribute it and/or |
11 | * modify it under the terms of the GNU General Public | 11 | * modify it under the terms of the GNU General Public |
12 | * License as published by the Free Software Foundation; either | 12 | * License as published by the Free Software Foundation; either |
13 | * version 2 of the License, or (at your option) any later version. | 13 | * version 2 of the License, or (at your option) any later version. |
14 | * | 14 | * |
15 | * This program is distributed in the hope that it will be useful, | 15 | * This program is distributed in the hope that it will be useful, |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 | * General Public License for more details. | 18 | * General Public License for more details. |
19 | * | 19 | * |
20 | * You should have received a copy of the GNU General Public | 20 | * You should have received a copy of the GNU General Public |
21 | * License along with this program; if not, write to the | 21 | * License along with this program; if not, write to the |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
23 | * Boston, MA 021110-1307, USA. | 23 | * Boston, MA 021110-1307, USA. |
24 | */ | 24 | */ |
25 | 25 | ||
26 | #include <linux/capability.h> | 26 | #include <linux/capability.h> |
27 | #include <linux/fs.h> | 27 | #include <linux/fs.h> |
28 | #include <linux/types.h> | 28 | #include <linux/types.h> |
29 | #include <linux/slab.h> | 29 | #include <linux/slab.h> |
30 | #include <linux/highmem.h> | 30 | #include <linux/highmem.h> |
31 | #include <linux/pagemap.h> | 31 | #include <linux/pagemap.h> |
32 | #include <linux/uio.h> | 32 | #include <linux/uio.h> |
33 | #include <linux/sched.h> | 33 | #include <linux/sched.h> |
34 | #include <linux/splice.h> | 34 | #include <linux/splice.h> |
35 | #include <linux/mount.h> | 35 | #include <linux/mount.h> |
36 | #include <linux/writeback.h> | 36 | #include <linux/writeback.h> |
37 | 37 | ||
38 | #define MLOG_MASK_PREFIX ML_INODE | 38 | #define MLOG_MASK_PREFIX ML_INODE |
39 | #include <cluster/masklog.h> | 39 | #include <cluster/masklog.h> |
40 | 40 | ||
41 | #include "ocfs2.h" | 41 | #include "ocfs2.h" |
42 | 42 | ||
43 | #include "alloc.h" | 43 | #include "alloc.h" |
44 | #include "aops.h" | 44 | #include "aops.h" |
45 | #include "dir.h" | 45 | #include "dir.h" |
46 | #include "dlmglue.h" | 46 | #include "dlmglue.h" |
47 | #include "extent_map.h" | 47 | #include "extent_map.h" |
48 | #include "file.h" | 48 | #include "file.h" |
49 | #include "sysfile.h" | 49 | #include "sysfile.h" |
50 | #include "inode.h" | 50 | #include "inode.h" |
51 | #include "ioctl.h" | 51 | #include "ioctl.h" |
52 | #include "journal.h" | 52 | #include "journal.h" |
53 | #include "mmap.h" | 53 | #include "mmap.h" |
54 | #include "suballoc.h" | 54 | #include "suballoc.h" |
55 | #include "super.h" | 55 | #include "super.h" |
56 | 56 | ||
57 | #include "buffer_head_io.h" | 57 | #include "buffer_head_io.h" |
58 | 58 | ||
59 | static int ocfs2_sync_inode(struct inode *inode) | 59 | static int ocfs2_sync_inode(struct inode *inode) |
60 | { | 60 | { |
61 | filemap_fdatawrite(inode->i_mapping); | 61 | filemap_fdatawrite(inode->i_mapping); |
62 | return sync_mapping_buffers(inode->i_mapping); | 62 | return sync_mapping_buffers(inode->i_mapping); |
63 | } | 63 | } |
64 | 64 | ||
65 | static int ocfs2_file_open(struct inode *inode, struct file *file) | 65 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
66 | { | 66 | { |
67 | int status; | 67 | int status; |
68 | int mode = file->f_flags; | 68 | int mode = file->f_flags; |
69 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | 69 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
70 | 70 | ||
71 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | 71 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
72 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); | 72 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
73 | 73 | ||
74 | spin_lock(&oi->ip_lock); | 74 | spin_lock(&oi->ip_lock); |
75 | 75 | ||
76 | /* Check that the inode hasn't been wiped from disk by another | 76 | /* Check that the inode hasn't been wiped from disk by another |
77 | * node. If it hasn't then we're safe as long as we hold the | 77 | * node. If it hasn't then we're safe as long as we hold the |
78 | * spin lock until our increment of open count. */ | 78 | * spin lock until our increment of open count. */ |
79 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | 79 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { |
80 | spin_unlock(&oi->ip_lock); | 80 | spin_unlock(&oi->ip_lock); |
81 | 81 | ||
82 | status = -ENOENT; | 82 | status = -ENOENT; |
83 | goto leave; | 83 | goto leave; |
84 | } | 84 | } |
85 | 85 | ||
86 | if (mode & O_DIRECT) | 86 | if (mode & O_DIRECT) |
87 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | 87 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; |
88 | 88 | ||
89 | oi->ip_open_count++; | 89 | oi->ip_open_count++; |
90 | spin_unlock(&oi->ip_lock); | 90 | spin_unlock(&oi->ip_lock); |
91 | status = 0; | 91 | status = 0; |
92 | leave: | 92 | leave: |
93 | mlog_exit(status); | 93 | mlog_exit(status); |
94 | return status; | 94 | return status; |
95 | } | 95 | } |
96 | 96 | ||
97 | static int ocfs2_file_release(struct inode *inode, struct file *file) | 97 | static int ocfs2_file_release(struct inode *inode, struct file *file) |
98 | { | 98 | { |
99 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | 99 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
100 | 100 | ||
101 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | 101 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, |
102 | file->f_path.dentry->d_name.len, | 102 | file->f_path.dentry->d_name.len, |
103 | file->f_path.dentry->d_name.name); | 103 | file->f_path.dentry->d_name.name); |
104 | 104 | ||
105 | spin_lock(&oi->ip_lock); | 105 | spin_lock(&oi->ip_lock); |
106 | if (!--oi->ip_open_count) | 106 | if (!--oi->ip_open_count) |
107 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | 107 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; |
108 | spin_unlock(&oi->ip_lock); | 108 | spin_unlock(&oi->ip_lock); |
109 | 109 | ||
110 | mlog_exit(0); | 110 | mlog_exit(0); |
111 | 111 | ||
112 | return 0; | 112 | return 0; |
113 | } | 113 | } |
114 | 114 | ||
115 | static int ocfs2_sync_file(struct file *file, | 115 | static int ocfs2_sync_file(struct file *file, |
116 | struct dentry *dentry, | 116 | struct dentry *dentry, |
117 | int datasync) | 117 | int datasync) |
118 | { | 118 | { |
119 | int err = 0; | 119 | int err = 0; |
120 | journal_t *journal; | 120 | journal_t *journal; |
121 | struct inode *inode = dentry->d_inode; | 121 | struct inode *inode = dentry->d_inode; |
122 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 122 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
123 | 123 | ||
124 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | 124 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, |
125 | dentry->d_name.len, dentry->d_name.name); | 125 | dentry->d_name.len, dentry->d_name.name); |
126 | 126 | ||
127 | err = ocfs2_sync_inode(dentry->d_inode); | 127 | err = ocfs2_sync_inode(dentry->d_inode); |
128 | if (err) | 128 | if (err) |
129 | goto bail; | 129 | goto bail; |
130 | 130 | ||
131 | journal = osb->journal->j_journal; | 131 | journal = osb->journal->j_journal; |
132 | err = journal_force_commit(journal); | 132 | err = journal_force_commit(journal); |
133 | 133 | ||
134 | bail: | 134 | bail: |
135 | mlog_exit(err); | 135 | mlog_exit(err); |
136 | 136 | ||
137 | return (err < 0) ? -EIO : 0; | 137 | return (err < 0) ? -EIO : 0; |
138 | } | 138 | } |
139 | 139 | ||
140 | int ocfs2_should_update_atime(struct inode *inode, | 140 | int ocfs2_should_update_atime(struct inode *inode, |
141 | struct vfsmount *vfsmnt) | 141 | struct vfsmount *vfsmnt) |
142 | { | 142 | { |
143 | struct timespec now; | 143 | struct timespec now; |
144 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 144 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
145 | 145 | ||
146 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | 146 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
147 | return 0; | 147 | return 0; |
148 | 148 | ||
149 | if ((inode->i_flags & S_NOATIME) || | 149 | if ((inode->i_flags & S_NOATIME) || |
150 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | 150 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) |
151 | return 0; | 151 | return 0; |
152 | 152 | ||
153 | /* | 153 | /* |
154 | * We can be called with no vfsmnt structure - NFSD will | 154 | * We can be called with no vfsmnt structure - NFSD will |
155 | * sometimes do this. | 155 | * sometimes do this. |
156 | * | 156 | * |
157 | * Note that our action here is different than touch_atime() - | 157 | * Note that our action here is different than touch_atime() - |
158 | * if we can't tell whether this is a noatime mount, then we | 158 | * if we can't tell whether this is a noatime mount, then we |
159 | * don't know whether to trust the value of s_atime_quantum. | 159 | * don't know whether to trust the value of s_atime_quantum. |
160 | */ | 160 | */ |
161 | if (vfsmnt == NULL) | 161 | if (vfsmnt == NULL) |
162 | return 0; | 162 | return 0; |
163 | 163 | ||
164 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || | 164 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
165 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | 165 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) |
166 | return 0; | 166 | return 0; |
167 | 167 | ||
168 | if (vfsmnt->mnt_flags & MNT_RELATIME) { | 168 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
169 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || | 169 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
170 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | 170 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) |
171 | return 1; | 171 | return 1; |
172 | 172 | ||
173 | return 0; | 173 | return 0; |
174 | } | 174 | } |
175 | 175 | ||
176 | now = CURRENT_TIME; | 176 | now = CURRENT_TIME; |
177 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | 177 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
178 | return 0; | 178 | return 0; |
179 | else | 179 | else |
180 | return 1; | 180 | return 1; |
181 | } | 181 | } |
182 | 182 | ||
183 | int ocfs2_update_inode_atime(struct inode *inode, | 183 | int ocfs2_update_inode_atime(struct inode *inode, |
184 | struct buffer_head *bh) | 184 | struct buffer_head *bh) |
185 | { | 185 | { |
186 | int ret; | 186 | int ret; |
187 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 187 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
188 | handle_t *handle; | 188 | handle_t *handle; |
189 | 189 | ||
190 | mlog_entry_void(); | 190 | mlog_entry_void(); |
191 | 191 | ||
192 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | 192 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
193 | if (handle == NULL) { | 193 | if (handle == NULL) { |
194 | ret = -ENOMEM; | 194 | ret = -ENOMEM; |
195 | mlog_errno(ret); | 195 | mlog_errno(ret); |
196 | goto out; | 196 | goto out; |
197 | } | 197 | } |
198 | 198 | ||
199 | inode->i_atime = CURRENT_TIME; | 199 | inode->i_atime = CURRENT_TIME; |
200 | ret = ocfs2_mark_inode_dirty(handle, inode, bh); | 200 | ret = ocfs2_mark_inode_dirty(handle, inode, bh); |
201 | if (ret < 0) | 201 | if (ret < 0) |
202 | mlog_errno(ret); | 202 | mlog_errno(ret); |
203 | 203 | ||
204 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | 204 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
205 | out: | 205 | out: |
206 | mlog_exit(ret); | 206 | mlog_exit(ret); |
207 | return ret; | 207 | return ret; |
208 | } | 208 | } |
209 | 209 | ||
210 | static int ocfs2_set_inode_size(handle_t *handle, | 210 | static int ocfs2_set_inode_size(handle_t *handle, |
211 | struct inode *inode, | 211 | struct inode *inode, |
212 | struct buffer_head *fe_bh, | 212 | struct buffer_head *fe_bh, |
213 | u64 new_i_size) | 213 | u64 new_i_size) |
214 | { | 214 | { |
215 | int status; | 215 | int status; |
216 | 216 | ||
217 | mlog_entry_void(); | 217 | mlog_entry_void(); |
218 | i_size_write(inode, new_i_size); | 218 | i_size_write(inode, new_i_size); |
219 | inode->i_blocks = ocfs2_inode_sector_count(inode); | 219 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
220 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | 220 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
221 | 221 | ||
222 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | 222 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); |
223 | if (status < 0) { | 223 | if (status < 0) { |
224 | mlog_errno(status); | 224 | mlog_errno(status); |
225 | goto bail; | 225 | goto bail; |
226 | } | 226 | } |
227 | 227 | ||
228 | bail: | 228 | bail: |
229 | mlog_exit(status); | 229 | mlog_exit(status); |
230 | return status; | 230 | return status; |
231 | } | 231 | } |
232 | 232 | ||
233 | static int ocfs2_simple_size_update(struct inode *inode, | 233 | static int ocfs2_simple_size_update(struct inode *inode, |
234 | struct buffer_head *di_bh, | 234 | struct buffer_head *di_bh, |
235 | u64 new_i_size) | 235 | u64 new_i_size) |
236 | { | 236 | { |
237 | int ret; | 237 | int ret; |
238 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 238 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
239 | handle_t *handle = NULL; | 239 | handle_t *handle = NULL; |
240 | 240 | ||
241 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | 241 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
242 | if (handle == NULL) { | 242 | if (handle == NULL) { |
243 | ret = -ENOMEM; | 243 | ret = -ENOMEM; |
244 | mlog_errno(ret); | 244 | mlog_errno(ret); |
245 | goto out; | 245 | goto out; |
246 | } | 246 | } |
247 | 247 | ||
248 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | 248 | ret = ocfs2_set_inode_size(handle, inode, di_bh, |
249 | new_i_size); | 249 | new_i_size); |
250 | if (ret < 0) | 250 | if (ret < 0) |
251 | mlog_errno(ret); | 251 | mlog_errno(ret); |
252 | 252 | ||
253 | ocfs2_commit_trans(osb, handle); | 253 | ocfs2_commit_trans(osb, handle); |
254 | out: | 254 | out: |
255 | return ret; | 255 | return ret; |
256 | } | 256 | } |
257 | 257 | ||
258 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, | 258 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
259 | struct inode *inode, | 259 | struct inode *inode, |
260 | struct buffer_head *fe_bh, | 260 | struct buffer_head *fe_bh, |
261 | u64 new_i_size) | 261 | u64 new_i_size) |
262 | { | 262 | { |
263 | int status; | 263 | int status; |
264 | handle_t *handle; | 264 | handle_t *handle; |
265 | struct ocfs2_dinode *di; | 265 | struct ocfs2_dinode *di; |
266 | 266 | ||
267 | mlog_entry_void(); | 267 | mlog_entry_void(); |
268 | 268 | ||
269 | /* TODO: This needs to actually orphan the inode in this | 269 | /* TODO: This needs to actually orphan the inode in this |
270 | * transaction. */ | 270 | * transaction. */ |
271 | 271 | ||
272 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | 272 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
273 | if (IS_ERR(handle)) { | 273 | if (IS_ERR(handle)) { |
274 | status = PTR_ERR(handle); | 274 | status = PTR_ERR(handle); |
275 | mlog_errno(status); | 275 | mlog_errno(status); |
276 | goto out; | 276 | goto out; |
277 | } | 277 | } |
278 | 278 | ||
279 | status = ocfs2_journal_access(handle, inode, fe_bh, | 279 | status = ocfs2_journal_access(handle, inode, fe_bh, |
280 | OCFS2_JOURNAL_ACCESS_WRITE); | 280 | OCFS2_JOURNAL_ACCESS_WRITE); |
281 | if (status < 0) { | 281 | if (status < 0) { |
282 | mlog_errno(status); | 282 | mlog_errno(status); |
283 | goto out_commit; | 283 | goto out_commit; |
284 | } | 284 | } |
285 | 285 | ||
286 | /* | 286 | /* |
287 | * Do this before setting i_size. | 287 | * Do this before setting i_size. |
288 | */ | 288 | */ |
289 | status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size); | 289 | status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size); |
290 | if (status) { | 290 | if (status) { |
291 | mlog_errno(status); | 291 | mlog_errno(status); |
292 | goto out_commit; | 292 | goto out_commit; |
293 | } | 293 | } |
294 | 294 | ||
295 | i_size_write(inode, new_i_size); | 295 | i_size_write(inode, new_i_size); |
296 | inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size); | 296 | inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size); |
297 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | 297 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
298 | 298 | ||
299 | di = (struct ocfs2_dinode *) fe_bh->b_data; | 299 | di = (struct ocfs2_dinode *) fe_bh->b_data; |
300 | di->i_size = cpu_to_le64(new_i_size); | 300 | di->i_size = cpu_to_le64(new_i_size); |
301 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | 301 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); |
302 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | 302 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); |
303 | 303 | ||
304 | status = ocfs2_journal_dirty(handle, fe_bh); | 304 | status = ocfs2_journal_dirty(handle, fe_bh); |
305 | if (status < 0) | 305 | if (status < 0) |
306 | mlog_errno(status); | 306 | mlog_errno(status); |
307 | 307 | ||
308 | out_commit: | 308 | out_commit: |
309 | ocfs2_commit_trans(osb, handle); | 309 | ocfs2_commit_trans(osb, handle); |
310 | out: | 310 | out: |
311 | 311 | ||
312 | mlog_exit(status); | 312 | mlog_exit(status); |
313 | return status; | 313 | return status; |
314 | } | 314 | } |
315 | 315 | ||
316 | static int ocfs2_truncate_file(struct inode *inode, | 316 | static int ocfs2_truncate_file(struct inode *inode, |
317 | struct buffer_head *di_bh, | 317 | struct buffer_head *di_bh, |
318 | u64 new_i_size) | 318 | u64 new_i_size) |
319 | { | 319 | { |
320 | int status = 0; | 320 | int status = 0; |
321 | struct ocfs2_dinode *fe = NULL; | 321 | struct ocfs2_dinode *fe = NULL; |
322 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 322 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
323 | struct ocfs2_truncate_context *tc = NULL; | 323 | struct ocfs2_truncate_context *tc = NULL; |
324 | 324 | ||
325 | mlog_entry("(inode = %llu, new_i_size = %llu\n", | 325 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
326 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | 326 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
327 | (unsigned long long)new_i_size); | 327 | (unsigned long long)new_i_size); |
328 | 328 | ||
329 | fe = (struct ocfs2_dinode *) di_bh->b_data; | 329 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
330 | if (!OCFS2_IS_VALID_DINODE(fe)) { | 330 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
331 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | 331 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
332 | status = -EIO; | 332 | status = -EIO; |
333 | goto bail; | 333 | goto bail; |
334 | } | 334 | } |
335 | 335 | ||
336 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | 336 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), |
337 | "Inode %llu, inode i_size = %lld != di " | 337 | "Inode %llu, inode i_size = %lld != di " |
338 | "i_size = %llu, i_flags = 0x%x\n", | 338 | "i_size = %llu, i_flags = 0x%x\n", |
339 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | 339 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
340 | i_size_read(inode), | 340 | i_size_read(inode), |
341 | (unsigned long long)le64_to_cpu(fe->i_size), | 341 | (unsigned long long)le64_to_cpu(fe->i_size), |
342 | le32_to_cpu(fe->i_flags)); | 342 | le32_to_cpu(fe->i_flags)); |
343 | 343 | ||
344 | if (new_i_size > le64_to_cpu(fe->i_size)) { | 344 | if (new_i_size > le64_to_cpu(fe->i_size)) { |
345 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", | 345 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
346 | (unsigned long long)le64_to_cpu(fe->i_size), | 346 | (unsigned long long)le64_to_cpu(fe->i_size), |
347 | (unsigned long long)new_i_size); | 347 | (unsigned long long)new_i_size); |
348 | status = -EINVAL; | 348 | status = -EINVAL; |
349 | mlog_errno(status); | 349 | mlog_errno(status); |
350 | goto bail; | 350 | goto bail; |
351 | } | 351 | } |
352 | 352 | ||
353 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", | 353 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
354 | (unsigned long long)le64_to_cpu(fe->i_blkno), | 354 | (unsigned long long)le64_to_cpu(fe->i_blkno), |
355 | (unsigned long long)le64_to_cpu(fe->i_size), | 355 | (unsigned long long)le64_to_cpu(fe->i_size), |
356 | (unsigned long long)new_i_size); | 356 | (unsigned long long)new_i_size); |
357 | 357 | ||
358 | /* lets handle the simple truncate cases before doing any more | 358 | /* lets handle the simple truncate cases before doing any more |
359 | * cluster locking. */ | 359 | * cluster locking. */ |
360 | if (new_i_size == le64_to_cpu(fe->i_size)) | 360 | if (new_i_size == le64_to_cpu(fe->i_size)) |
361 | goto bail; | 361 | goto bail; |
362 | 362 | ||
363 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | 363 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
364 | 364 | ||
365 | /* This forces other nodes to sync and drop their pages. Do | 365 | /* This forces other nodes to sync and drop their pages. Do |
366 | * this even if we have a truncate without allocation change - | 366 | * this even if we have a truncate without allocation change - |
367 | * ocfs2 cluster sizes can be much greater than page size, so | 367 | * ocfs2 cluster sizes can be much greater than page size, so |
368 | * we have to truncate them anyway. */ | 368 | * we have to truncate them anyway. */ |
369 | status = ocfs2_data_lock(inode, 1); | 369 | status = ocfs2_data_lock(inode, 1); |
370 | if (status < 0) { | 370 | if (status < 0) { |
371 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | 371 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
372 | 372 | ||
373 | mlog_errno(status); | 373 | mlog_errno(status); |
374 | goto bail; | 374 | goto bail; |
375 | } | 375 | } |
376 | 376 | ||
377 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); | 377 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
378 | truncate_inode_pages(inode->i_mapping, new_i_size); | 378 | truncate_inode_pages(inode->i_mapping, new_i_size); |
379 | 379 | ||
380 | /* alright, we're going to need to do a full blown alloc size | 380 | /* alright, we're going to need to do a full blown alloc size |
381 | * change. Orphan the inode so that recovery can complete the | 381 | * change. Orphan the inode so that recovery can complete the |
382 | * truncate if necessary. This does the task of marking | 382 | * truncate if necessary. This does the task of marking |
383 | * i_size. */ | 383 | * i_size. */ |
384 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | 384 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); |
385 | if (status < 0) { | 385 | if (status < 0) { |
386 | mlog_errno(status); | 386 | mlog_errno(status); |
387 | goto bail_unlock_data; | 387 | goto bail_unlock_data; |
388 | } | 388 | } |
389 | 389 | ||
390 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); | 390 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); |
391 | if (status < 0) { | 391 | if (status < 0) { |
392 | mlog_errno(status); | 392 | mlog_errno(status); |
393 | goto bail_unlock_data; | 393 | goto bail_unlock_data; |
394 | } | 394 | } |
395 | 395 | ||
396 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); | 396 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); |
397 | if (status < 0) { | 397 | if (status < 0) { |
398 | mlog_errno(status); | 398 | mlog_errno(status); |
399 | goto bail_unlock_data; | 399 | goto bail_unlock_data; |
400 | } | 400 | } |
401 | 401 | ||
402 | /* TODO: orphan dir cleanup here. */ | 402 | /* TODO: orphan dir cleanup here. */ |
403 | bail_unlock_data: | 403 | bail_unlock_data: |
404 | ocfs2_data_unlock(inode, 1); | 404 | ocfs2_data_unlock(inode, 1); |
405 | 405 | ||
406 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | 406 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
407 | 407 | ||
408 | bail: | 408 | bail: |
409 | 409 | ||
410 | mlog_exit(status); | 410 | mlog_exit(status); |
411 | return status; | 411 | return status; |
412 | } | 412 | } |
413 | 413 | ||
414 | /* | 414 | /* |
415 | * extend allocation only here. | 415 | * extend allocation only here. |
416 | * we'll update all the disk stuff, and oip->alloc_size | 416 | * we'll update all the disk stuff, and oip->alloc_size |
417 | * | 417 | * |
418 | * expect stuff to be locked, a transaction started and enough data / | 418 | * expect stuff to be locked, a transaction started and enough data / |
419 | * metadata reservations in the contexts. | 419 | * metadata reservations in the contexts. |
420 | * | 420 | * |
421 | * Will return -EAGAIN, and a reason if a restart is needed. | 421 | * Will return -EAGAIN, and a reason if a restart is needed. |
422 | * If passed in, *reason will always be set, even in error. | 422 | * If passed in, *reason will always be set, even in error. |
423 | */ | 423 | */ |
424 | int ocfs2_do_extend_allocation(struct ocfs2_super *osb, | 424 | int ocfs2_do_extend_allocation(struct ocfs2_super *osb, |
425 | struct inode *inode, | 425 | struct inode *inode, |
426 | u32 *logical_offset, | 426 | u32 *logical_offset, |
427 | u32 clusters_to_add, | 427 | u32 clusters_to_add, |
428 | struct buffer_head *fe_bh, | 428 | struct buffer_head *fe_bh, |
429 | handle_t *handle, | 429 | handle_t *handle, |
430 | struct ocfs2_alloc_context *data_ac, | 430 | struct ocfs2_alloc_context *data_ac, |
431 | struct ocfs2_alloc_context *meta_ac, | 431 | struct ocfs2_alloc_context *meta_ac, |
432 | enum ocfs2_alloc_restarted *reason_ret) | 432 | enum ocfs2_alloc_restarted *reason_ret) |
433 | { | 433 | { |
434 | int status = 0; | 434 | int status = 0; |
435 | int free_extents; | 435 | int free_extents; |
436 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; | 436 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; |
437 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | 437 | enum ocfs2_alloc_restarted reason = RESTART_NONE; |
438 | u32 bit_off, num_bits; | 438 | u32 bit_off, num_bits; |
439 | u64 block; | 439 | u64 block; |
440 | 440 | ||
441 | BUG_ON(!clusters_to_add); | 441 | BUG_ON(!clusters_to_add); |
442 | 442 | ||
443 | free_extents = ocfs2_num_free_extents(osb, inode, fe); | 443 | free_extents = ocfs2_num_free_extents(osb, inode, fe); |
444 | if (free_extents < 0) { | 444 | if (free_extents < 0) { |
445 | status = free_extents; | 445 | status = free_extents; |
446 | mlog_errno(status); | 446 | mlog_errno(status); |
447 | goto leave; | 447 | goto leave; |
448 | } | 448 | } |
449 | 449 | ||
450 | /* there are two cases which could cause us to EAGAIN in the | 450 | /* there are two cases which could cause us to EAGAIN in the |
451 | * we-need-more-metadata case: | 451 | * we-need-more-metadata case: |
452 | * 1) we haven't reserved *any* | 452 | * 1) we haven't reserved *any* |
453 | * 2) we are so fragmented, we've needed to add metadata too | 453 | * 2) we are so fragmented, we've needed to add metadata too |
454 | * many times. */ | 454 | * many times. */ |
455 | if (!free_extents && !meta_ac) { | 455 | if (!free_extents && !meta_ac) { |
456 | mlog(0, "we haven't reserved any metadata!\n"); | 456 | mlog(0, "we haven't reserved any metadata!\n"); |
457 | status = -EAGAIN; | 457 | status = -EAGAIN; |
458 | reason = RESTART_META; | 458 | reason = RESTART_META; |
459 | goto leave; | 459 | goto leave; |
460 | } else if ((!free_extents) | 460 | } else if ((!free_extents) |
461 | && (ocfs2_alloc_context_bits_left(meta_ac) | 461 | && (ocfs2_alloc_context_bits_left(meta_ac) |
462 | < ocfs2_extend_meta_needed(fe))) { | 462 | < ocfs2_extend_meta_needed(fe))) { |
463 | mlog(0, "filesystem is really fragmented...\n"); | 463 | mlog(0, "filesystem is really fragmented...\n"); |
464 | status = -EAGAIN; | 464 | status = -EAGAIN; |
465 | reason = RESTART_META; | 465 | reason = RESTART_META; |
466 | goto leave; | 466 | goto leave; |
467 | } | 467 | } |
468 | 468 | ||
469 | status = ocfs2_claim_clusters(osb, handle, data_ac, 1, | 469 | status = ocfs2_claim_clusters(osb, handle, data_ac, 1, |
470 | &bit_off, &num_bits); | 470 | &bit_off, &num_bits); |
471 | if (status < 0) { | 471 | if (status < 0) { |
472 | if (status != -ENOSPC) | 472 | if (status != -ENOSPC) |
473 | mlog_errno(status); | 473 | mlog_errno(status); |
474 | goto leave; | 474 | goto leave; |
475 | } | 475 | } |
476 | 476 | ||
477 | BUG_ON(num_bits > clusters_to_add); | 477 | BUG_ON(num_bits > clusters_to_add); |
478 | 478 | ||
479 | /* reserve our write early -- insert_extent may update the inode */ | 479 | /* reserve our write early -- insert_extent may update the inode */ |
480 | status = ocfs2_journal_access(handle, inode, fe_bh, | 480 | status = ocfs2_journal_access(handle, inode, fe_bh, |
481 | OCFS2_JOURNAL_ACCESS_WRITE); | 481 | OCFS2_JOURNAL_ACCESS_WRITE); |
482 | if (status < 0) { | 482 | if (status < 0) { |
483 | mlog_errno(status); | 483 | mlog_errno(status); |
484 | goto leave; | 484 | goto leave; |
485 | } | 485 | } |
486 | 486 | ||
487 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | 487 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); |
488 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", | 488 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", |
489 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | 489 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); |
490 | status = ocfs2_insert_extent(osb, handle, inode, fe_bh, | 490 | status = ocfs2_insert_extent(osb, handle, inode, fe_bh, |
491 | *logical_offset, block, num_bits, | 491 | *logical_offset, block, num_bits, |
492 | meta_ac); | 492 | meta_ac); |
493 | if (status < 0) { | 493 | if (status < 0) { |
494 | mlog_errno(status); | 494 | mlog_errno(status); |
495 | goto leave; | 495 | goto leave; |
496 | } | 496 | } |
497 | 497 | ||
498 | status = ocfs2_journal_dirty(handle, fe_bh); | 498 | status = ocfs2_journal_dirty(handle, fe_bh); |
499 | if (status < 0) { | 499 | if (status < 0) { |
500 | mlog_errno(status); | 500 | mlog_errno(status); |
501 | goto leave; | 501 | goto leave; |
502 | } | 502 | } |
503 | 503 | ||
504 | clusters_to_add -= num_bits; | 504 | clusters_to_add -= num_bits; |
505 | *logical_offset += num_bits; | 505 | *logical_offset += num_bits; |
506 | 506 | ||
507 | if (clusters_to_add) { | 507 | if (clusters_to_add) { |
508 | mlog(0, "need to alloc once more, clusters = %u, wanted = " | 508 | mlog(0, "need to alloc once more, clusters = %u, wanted = " |
509 | "%u\n", fe->i_clusters, clusters_to_add); | 509 | "%u\n", fe->i_clusters, clusters_to_add); |
510 | status = -EAGAIN; | 510 | status = -EAGAIN; |
511 | reason = RESTART_TRANS; | 511 | reason = RESTART_TRANS; |
512 | } | 512 | } |
513 | 513 | ||
514 | leave: | 514 | leave: |
515 | mlog_exit(status); | 515 | mlog_exit(status); |
516 | if (reason_ret) | 516 | if (reason_ret) |
517 | *reason_ret = reason; | 517 | *reason_ret = reason; |
518 | return status; | 518 | return status; |
519 | } | 519 | } |
520 | 520 | ||
521 | /* | 521 | /* |
522 | * For a given allocation, determine which allocators will need to be | 522 | * For a given allocation, determine which allocators will need to be |
523 | * accessed, and lock them, reserving the appropriate number of bits. | 523 | * accessed, and lock them, reserving the appropriate number of bits. |
524 | * | 524 | * |
525 | * Called from ocfs2_extend_allocation() for file systems which don't | 525 | * Called from ocfs2_extend_allocation() for file systems which don't |
526 | * support holes, and from ocfs2_write() for file systems which | 526 | * support holes, and from ocfs2_write() for file systems which |
527 | * understand sparse inodes. | 527 | * understand sparse inodes. |
528 | */ | 528 | */ |
529 | int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di, | 529 | int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di, |
530 | u32 clusters_to_add, | 530 | u32 clusters_to_add, |
531 | struct ocfs2_alloc_context **data_ac, | 531 | struct ocfs2_alloc_context **data_ac, |
532 | struct ocfs2_alloc_context **meta_ac) | 532 | struct ocfs2_alloc_context **meta_ac) |
533 | { | 533 | { |
534 | int ret, num_free_extents; | 534 | int ret, num_free_extents; |
535 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 535 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
536 | 536 | ||
537 | *meta_ac = NULL; | 537 | *meta_ac = NULL; |
538 | *data_ac = NULL; | 538 | *data_ac = NULL; |
539 | 539 | ||
540 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " | 540 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
541 | "clusters_to_add = %u\n", | 541 | "clusters_to_add = %u\n", |
542 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode), | 542 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode), |
543 | le32_to_cpu(di->i_clusters), clusters_to_add); | 543 | le32_to_cpu(di->i_clusters), clusters_to_add); |
544 | 544 | ||
545 | num_free_extents = ocfs2_num_free_extents(osb, inode, di); | 545 | num_free_extents = ocfs2_num_free_extents(osb, inode, di); |
546 | if (num_free_extents < 0) { | 546 | if (num_free_extents < 0) { |
547 | ret = num_free_extents; | 547 | ret = num_free_extents; |
548 | mlog_errno(ret); | 548 | mlog_errno(ret); |
549 | goto out; | 549 | goto out; |
550 | } | 550 | } |
551 | 551 | ||
552 | /* | 552 | /* |
553 | * Sparse allocation file systems need to be more conservative | 553 | * Sparse allocation file systems need to be more conservative |
554 | * with reserving room for expansion - the actual allocation | 554 | * with reserving room for expansion - the actual allocation |
555 | * happens while we've got a journal handle open so re-taking | 555 | * happens while we've got a journal handle open so re-taking |
556 | * a cluster lock (because we ran out of room for another | 556 | * a cluster lock (because we ran out of room for another |
557 | * extent) will violate ordering rules. | 557 | * extent) will violate ordering rules. |
558 | * | 558 | * |
559 | * Most of the time we'll only be seeing this 1 cluster at a time | 559 | * Most of the time we'll only be seeing this 1 cluster at a time |
560 | * anyway. | 560 | * anyway. |
561 | */ | 561 | */ |
562 | if (!num_free_extents || | 562 | if (!num_free_extents || |
563 | (ocfs2_sparse_alloc(osb) && num_free_extents < clusters_to_add)) { | 563 | (ocfs2_sparse_alloc(osb) && num_free_extents < clusters_to_add)) { |
564 | ret = ocfs2_reserve_new_metadata(osb, di, meta_ac); | 564 | ret = ocfs2_reserve_new_metadata(osb, di, meta_ac); |
565 | if (ret < 0) { | 565 | if (ret < 0) { |
566 | if (ret != -ENOSPC) | 566 | if (ret != -ENOSPC) |
567 | mlog_errno(ret); | 567 | mlog_errno(ret); |
568 | goto out; | 568 | goto out; |
569 | } | 569 | } |
570 | } | 570 | } |
571 | 571 | ||
572 | ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac); | 572 | ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac); |
573 | if (ret < 0) { | 573 | if (ret < 0) { |
574 | if (ret != -ENOSPC) | 574 | if (ret != -ENOSPC) |
575 | mlog_errno(ret); | 575 | mlog_errno(ret); |
576 | goto out; | 576 | goto out; |
577 | } | 577 | } |
578 | 578 | ||
579 | out: | 579 | out: |
580 | if (ret) { | 580 | if (ret) { |
581 | if (*meta_ac) { | 581 | if (*meta_ac) { |
582 | ocfs2_free_alloc_context(*meta_ac); | 582 | ocfs2_free_alloc_context(*meta_ac); |
583 | *meta_ac = NULL; | 583 | *meta_ac = NULL; |
584 | } | 584 | } |
585 | 585 | ||
586 | /* | 586 | /* |
587 | * We cannot have an error and a non null *data_ac. | 587 | * We cannot have an error and a non null *data_ac. |
588 | */ | 588 | */ |
589 | } | 589 | } |
590 | 590 | ||
591 | return ret; | 591 | return ret; |
592 | } | 592 | } |
593 | 593 | ||
594 | static int ocfs2_extend_allocation(struct inode *inode, | 594 | static int ocfs2_extend_allocation(struct inode *inode, |
595 | u32 clusters_to_add) | 595 | u32 clusters_to_add) |
596 | { | 596 | { |
597 | int status = 0; | 597 | int status = 0; |
598 | int restart_func = 0; | 598 | int restart_func = 0; |
599 | int drop_alloc_sem = 0; | 599 | int drop_alloc_sem = 0; |
600 | int credits; | 600 | int credits; |
601 | u32 prev_clusters, logical_start; | 601 | u32 prev_clusters, logical_start; |
602 | struct buffer_head *bh = NULL; | 602 | struct buffer_head *bh = NULL; |
603 | struct ocfs2_dinode *fe = NULL; | 603 | struct ocfs2_dinode *fe = NULL; |
604 | handle_t *handle = NULL; | 604 | handle_t *handle = NULL; |
605 | struct ocfs2_alloc_context *data_ac = NULL; | 605 | struct ocfs2_alloc_context *data_ac = NULL; |
606 | struct ocfs2_alloc_context *meta_ac = NULL; | 606 | struct ocfs2_alloc_context *meta_ac = NULL; |
607 | enum ocfs2_alloc_restarted why; | 607 | enum ocfs2_alloc_restarted why; |
608 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 608 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
609 | 609 | ||
610 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | 610 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); |
611 | 611 | ||
612 | /* | 612 | /* |
613 | * This function only exists for file systems which don't | 613 | * This function only exists for file systems which don't |
614 | * support holes. | 614 | * support holes. |
615 | */ | 615 | */ |
616 | BUG_ON(ocfs2_sparse_alloc(osb)); | 616 | BUG_ON(ocfs2_sparse_alloc(osb)); |
617 | 617 | ||
618 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, | 618 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, |
619 | OCFS2_BH_CACHED, inode); | 619 | OCFS2_BH_CACHED, inode); |
620 | if (status < 0) { | 620 | if (status < 0) { |
621 | mlog_errno(status); | 621 | mlog_errno(status); |
622 | goto leave; | 622 | goto leave; |
623 | } | 623 | } |
624 | 624 | ||
625 | fe = (struct ocfs2_dinode *) bh->b_data; | 625 | fe = (struct ocfs2_dinode *) bh->b_data; |
626 | if (!OCFS2_IS_VALID_DINODE(fe)) { | 626 | if (!OCFS2_IS_VALID_DINODE(fe)) { |
627 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | 627 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); |
628 | status = -EIO; | 628 | status = -EIO; |
629 | goto leave; | 629 | goto leave; |
630 | } | 630 | } |
631 | 631 | ||
632 | logical_start = OCFS2_I(inode)->ip_clusters; | 632 | logical_start = OCFS2_I(inode)->ip_clusters; |
633 | 633 | ||
634 | restart_all: | 634 | restart_all: |
635 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | 635 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); |
636 | 636 | ||
637 | /* blocks peope in read/write from reading our allocation | 637 | /* blocks peope in read/write from reading our allocation |
638 | * until we're done changing it. We depend on i_mutex to block | 638 | * until we're done changing it. We depend on i_mutex to block |
639 | * other extend/truncate calls while we're here. Ordering wrt | 639 | * other extend/truncate calls while we're here. Ordering wrt |
640 | * start_trans is important here -- always do it before! */ | 640 | * start_trans is important here -- always do it before! */ |
641 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | 641 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
642 | drop_alloc_sem = 1; | 642 | drop_alloc_sem = 1; |
643 | 643 | ||
644 | status = ocfs2_lock_allocators(inode, fe, clusters_to_add, &data_ac, | 644 | status = ocfs2_lock_allocators(inode, fe, clusters_to_add, &data_ac, |
645 | &meta_ac); | 645 | &meta_ac); |
646 | if (status) { | 646 | if (status) { |
647 | mlog_errno(status); | 647 | mlog_errno(status); |
648 | goto leave; | 648 | goto leave; |
649 | } | 649 | } |
650 | 650 | ||
651 | credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add); | 651 | credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add); |
652 | handle = ocfs2_start_trans(osb, credits); | 652 | handle = ocfs2_start_trans(osb, credits); |
653 | if (IS_ERR(handle)) { | 653 | if (IS_ERR(handle)) { |
654 | status = PTR_ERR(handle); | 654 | status = PTR_ERR(handle); |
655 | handle = NULL; | 655 | handle = NULL; |
656 | mlog_errno(status); | 656 | mlog_errno(status); |
657 | goto leave; | 657 | goto leave; |
658 | } | 658 | } |
659 | 659 | ||
660 | restarted_transaction: | 660 | restarted_transaction: |
661 | /* reserve a write to the file entry early on - that we if we | 661 | /* reserve a write to the file entry early on - that we if we |
662 | * run out of credits in the allocation path, we can still | 662 | * run out of credits in the allocation path, we can still |
663 | * update i_size. */ | 663 | * update i_size. */ |
664 | status = ocfs2_journal_access(handle, inode, bh, | 664 | status = ocfs2_journal_access(handle, inode, bh, |
665 | OCFS2_JOURNAL_ACCESS_WRITE); | 665 | OCFS2_JOURNAL_ACCESS_WRITE); |
666 | if (status < 0) { | 666 | if (status < 0) { |
667 | mlog_errno(status); | 667 | mlog_errno(status); |
668 | goto leave; | 668 | goto leave; |
669 | } | 669 | } |
670 | 670 | ||
671 | prev_clusters = OCFS2_I(inode)->ip_clusters; | 671 | prev_clusters = OCFS2_I(inode)->ip_clusters; |
672 | 672 | ||
673 | status = ocfs2_do_extend_allocation(osb, | 673 | status = ocfs2_do_extend_allocation(osb, |
674 | inode, | 674 | inode, |
675 | &logical_start, | 675 | &logical_start, |
676 | clusters_to_add, | 676 | clusters_to_add, |
677 | bh, | 677 | bh, |
678 | handle, | 678 | handle, |
679 | data_ac, | 679 | data_ac, |
680 | meta_ac, | 680 | meta_ac, |
681 | &why); | 681 | &why); |
682 | if ((status < 0) && (status != -EAGAIN)) { | 682 | if ((status < 0) && (status != -EAGAIN)) { |
683 | if (status != -ENOSPC) | 683 | if (status != -ENOSPC) |
684 | mlog_errno(status); | 684 | mlog_errno(status); |
685 | goto leave; | 685 | goto leave; |
686 | } | 686 | } |
687 | 687 | ||
688 | status = ocfs2_journal_dirty(handle, bh); | 688 | status = ocfs2_journal_dirty(handle, bh); |
689 | if (status < 0) { | 689 | if (status < 0) { |
690 | mlog_errno(status); | 690 | mlog_errno(status); |
691 | goto leave; | 691 | goto leave; |
692 | } | 692 | } |
693 | 693 | ||
694 | spin_lock(&OCFS2_I(inode)->ip_lock); | 694 | spin_lock(&OCFS2_I(inode)->ip_lock); |
695 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | 695 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); |
696 | spin_unlock(&OCFS2_I(inode)->ip_lock); | 696 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
697 | 697 | ||
698 | if (why != RESTART_NONE && clusters_to_add) { | 698 | if (why != RESTART_NONE && clusters_to_add) { |
699 | if (why == RESTART_META) { | 699 | if (why == RESTART_META) { |
700 | mlog(0, "restarting function.\n"); | 700 | mlog(0, "restarting function.\n"); |
701 | restart_func = 1; | 701 | restart_func = 1; |
702 | } else { | 702 | } else { |
703 | BUG_ON(why != RESTART_TRANS); | 703 | BUG_ON(why != RESTART_TRANS); |
704 | 704 | ||
705 | mlog(0, "restarting transaction.\n"); | 705 | mlog(0, "restarting transaction.\n"); |
706 | /* TODO: This can be more intelligent. */ | 706 | /* TODO: This can be more intelligent. */ |
707 | credits = ocfs2_calc_extend_credits(osb->sb, | 707 | credits = ocfs2_calc_extend_credits(osb->sb, |
708 | fe, | 708 | fe, |
709 | clusters_to_add); | 709 | clusters_to_add); |
710 | status = ocfs2_extend_trans(handle, credits); | 710 | status = ocfs2_extend_trans(handle, credits); |
711 | if (status < 0) { | 711 | if (status < 0) { |
712 | /* handle still has to be committed at | 712 | /* handle still has to be committed at |
713 | * this point. */ | 713 | * this point. */ |
714 | status = -ENOMEM; | 714 | status = -ENOMEM; |
715 | mlog_errno(status); | 715 | mlog_errno(status); |
716 | goto leave; | 716 | goto leave; |
717 | } | 717 | } |
718 | goto restarted_transaction; | 718 | goto restarted_transaction; |
719 | } | 719 | } |
720 | } | 720 | } |
721 | 721 | ||
722 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", | 722 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
723 | le32_to_cpu(fe->i_clusters), | 723 | le32_to_cpu(fe->i_clusters), |
724 | (unsigned long long)le64_to_cpu(fe->i_size)); | 724 | (unsigned long long)le64_to_cpu(fe->i_size)); |
725 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", | 725 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
726 | OCFS2_I(inode)->ip_clusters, i_size_read(inode)); | 726 | OCFS2_I(inode)->ip_clusters, i_size_read(inode)); |
727 | 727 | ||
728 | leave: | 728 | leave: |
729 | if (drop_alloc_sem) { | 729 | if (drop_alloc_sem) { |
730 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | 730 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
731 | drop_alloc_sem = 0; | 731 | drop_alloc_sem = 0; |
732 | } | 732 | } |
733 | if (handle) { | 733 | if (handle) { |
734 | ocfs2_commit_trans(osb, handle); | 734 | ocfs2_commit_trans(osb, handle); |
735 | handle = NULL; | 735 | handle = NULL; |
736 | } | 736 | } |
737 | if (data_ac) { | 737 | if (data_ac) { |
738 | ocfs2_free_alloc_context(data_ac); | 738 | ocfs2_free_alloc_context(data_ac); |
739 | data_ac = NULL; | 739 | data_ac = NULL; |
740 | } | 740 | } |
741 | if (meta_ac) { | 741 | if (meta_ac) { |
742 | ocfs2_free_alloc_context(meta_ac); | 742 | ocfs2_free_alloc_context(meta_ac); |
743 | meta_ac = NULL; | 743 | meta_ac = NULL; |
744 | } | 744 | } |
745 | if ((!status) && restart_func) { | 745 | if ((!status) && restart_func) { |
746 | restart_func = 0; | 746 | restart_func = 0; |
747 | goto restart_all; | 747 | goto restart_all; |
748 | } | 748 | } |
749 | if (bh) { | 749 | if (bh) { |
750 | brelse(bh); | 750 | brelse(bh); |
751 | bh = NULL; | 751 | bh = NULL; |
752 | } | 752 | } |
753 | 753 | ||
754 | mlog_exit(status); | 754 | mlog_exit(status); |
755 | return status; | 755 | return status; |
756 | } | 756 | } |
757 | 757 | ||
758 | /* Some parts of this taken from generic_cont_expand, which turned out | 758 | /* Some parts of this taken from generic_cont_expand, which turned out |
759 | * to be too fragile to do exactly what we need without us having to | 759 | * to be too fragile to do exactly what we need without us having to |
760 | * worry about recursive locking in ->prepare_write() and | 760 | * worry about recursive locking in ->prepare_write() and |
761 | * ->commit_write(). */ | 761 | * ->commit_write(). */ |
762 | static int ocfs2_write_zero_page(struct inode *inode, | 762 | static int ocfs2_write_zero_page(struct inode *inode, |
763 | u64 size) | 763 | u64 size) |
764 | { | 764 | { |
765 | struct address_space *mapping = inode->i_mapping; | 765 | struct address_space *mapping = inode->i_mapping; |
766 | struct page *page; | 766 | struct page *page; |
767 | unsigned long index; | 767 | unsigned long index; |
768 | unsigned int offset; | 768 | unsigned int offset; |
769 | handle_t *handle = NULL; | 769 | handle_t *handle = NULL; |
770 | int ret; | 770 | int ret; |
771 | 771 | ||
772 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ | 772 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ |
773 | /* ugh. in prepare/commit_write, if from==to==start of block, we | 773 | /* ugh. in prepare/commit_write, if from==to==start of block, we |
774 | ** skip the prepare. make sure we never send an offset for the start | 774 | ** skip the prepare. make sure we never send an offset for the start |
775 | ** of a block | 775 | ** of a block |
776 | */ | 776 | */ |
777 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { | 777 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { |
778 | offset++; | 778 | offset++; |
779 | } | 779 | } |
780 | index = size >> PAGE_CACHE_SHIFT; | 780 | index = size >> PAGE_CACHE_SHIFT; |
781 | 781 | ||
782 | page = grab_cache_page(mapping, index); | 782 | page = grab_cache_page(mapping, index); |
783 | if (!page) { | 783 | if (!page) { |
784 | ret = -ENOMEM; | 784 | ret = -ENOMEM; |
785 | mlog_errno(ret); | 785 | mlog_errno(ret); |
786 | goto out; | 786 | goto out; |
787 | } | 787 | } |
788 | 788 | ||
789 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); | 789 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
790 | if (ret < 0) { | 790 | if (ret < 0) { |
791 | mlog_errno(ret); | 791 | mlog_errno(ret); |
792 | goto out_unlock; | 792 | goto out_unlock; |
793 | } | 793 | } |
794 | 794 | ||
795 | if (ocfs2_should_order_data(inode)) { | 795 | if (ocfs2_should_order_data(inode)) { |
796 | handle = ocfs2_start_walk_page_trans(inode, page, offset, | 796 | handle = ocfs2_start_walk_page_trans(inode, page, offset, |
797 | offset); | 797 | offset); |
798 | if (IS_ERR(handle)) { | 798 | if (IS_ERR(handle)) { |
799 | ret = PTR_ERR(handle); | 799 | ret = PTR_ERR(handle); |
800 | handle = NULL; | 800 | handle = NULL; |
801 | goto out_unlock; | 801 | goto out_unlock; |
802 | } | 802 | } |
803 | } | 803 | } |
804 | 804 | ||
805 | /* must not update i_size! */ | 805 | /* must not update i_size! */ |
806 | ret = block_commit_write(page, offset, offset); | 806 | ret = block_commit_write(page, offset, offset); |
807 | if (ret < 0) | 807 | if (ret < 0) |
808 | mlog_errno(ret); | 808 | mlog_errno(ret); |
809 | else | 809 | else |
810 | ret = 0; | 810 | ret = 0; |
811 | 811 | ||
812 | if (handle) | 812 | if (handle) |
813 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | 813 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
814 | out_unlock: | 814 | out_unlock: |
815 | unlock_page(page); | 815 | unlock_page(page); |
816 | page_cache_release(page); | 816 | page_cache_release(page); |
817 | out: | 817 | out: |
818 | return ret; | 818 | return ret; |
819 | } | 819 | } |
820 | 820 | ||
821 | static int ocfs2_zero_extend(struct inode *inode, | 821 | static int ocfs2_zero_extend(struct inode *inode, |
822 | u64 zero_to_size) | 822 | u64 zero_to_size) |
823 | { | 823 | { |
824 | int ret = 0; | 824 | int ret = 0; |
825 | u64 start_off; | 825 | u64 start_off; |
826 | struct super_block *sb = inode->i_sb; | 826 | struct super_block *sb = inode->i_sb; |
827 | 827 | ||
828 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | 828 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); |
829 | while (start_off < zero_to_size) { | 829 | while (start_off < zero_to_size) { |
830 | ret = ocfs2_write_zero_page(inode, start_off); | 830 | ret = ocfs2_write_zero_page(inode, start_off); |
831 | if (ret < 0) { | 831 | if (ret < 0) { |
832 | mlog_errno(ret); | 832 | mlog_errno(ret); |
833 | goto out; | 833 | goto out; |
834 | } | 834 | } |
835 | 835 | ||
836 | start_off += sb->s_blocksize; | 836 | start_off += sb->s_blocksize; |
837 | 837 | ||
838 | /* | 838 | /* |
839 | * Very large extends have the potential to lock up | 839 | * Very large extends have the potential to lock up |
840 | * the cpu for extended periods of time. | 840 | * the cpu for extended periods of time. |
841 | */ | 841 | */ |
842 | cond_resched(); | 842 | cond_resched(); |
843 | } | 843 | } |
844 | 844 | ||
845 | out: | 845 | out: |
846 | return ret; | 846 | return ret; |
847 | } | 847 | } |
848 | 848 | ||
849 | /* | 849 | /* |
850 | * A tail_to_skip value > 0 indicates that we're being called from | 850 | * A tail_to_skip value > 0 indicates that we're being called from |
851 | * ocfs2_file_aio_write(). This has the following implications: | 851 | * ocfs2_file_aio_write(). This has the following implications: |
852 | * | 852 | * |
853 | * - we don't want to update i_size | 853 | * - we don't want to update i_size |
854 | * - di_bh will be NULL, which is fine because it's only used in the | 854 | * - di_bh will be NULL, which is fine because it's only used in the |
855 | * case where we want to update i_size. | 855 | * case where we want to update i_size. |
856 | * - ocfs2_zero_extend() will then only be filling the hole created | 856 | * - ocfs2_zero_extend() will then only be filling the hole created |
857 | * between i_size and the start of the write. | 857 | * between i_size and the start of the write. |
858 | */ | 858 | */ |
859 | static int ocfs2_extend_file(struct inode *inode, | 859 | static int ocfs2_extend_file(struct inode *inode, |
860 | struct buffer_head *di_bh, | 860 | struct buffer_head *di_bh, |
861 | u64 new_i_size, | 861 | u64 new_i_size, |
862 | size_t tail_to_skip) | 862 | size_t tail_to_skip) |
863 | { | 863 | { |
864 | int ret = 0; | 864 | int ret = 0; |
865 | u32 clusters_to_add = 0; | 865 | u32 clusters_to_add = 0; |
866 | 866 | ||
867 | BUG_ON(!tail_to_skip && !di_bh); | 867 | BUG_ON(!tail_to_skip && !di_bh); |
868 | 868 | ||
869 | /* setattr sometimes calls us like this. */ | 869 | /* setattr sometimes calls us like this. */ |
870 | if (new_i_size == 0) | 870 | if (new_i_size == 0) |
871 | goto out; | 871 | goto out; |
872 | 872 | ||
873 | if (i_size_read(inode) == new_i_size) | 873 | if (i_size_read(inode) == new_i_size) |
874 | goto out; | 874 | goto out; |
875 | BUG_ON(new_i_size < i_size_read(inode)); | 875 | BUG_ON(new_i_size < i_size_read(inode)); |
876 | 876 | ||
877 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { | 877 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { |
878 | BUG_ON(tail_to_skip != 0); | 878 | BUG_ON(tail_to_skip != 0); |
879 | goto out_update_size; | 879 | goto out_update_size; |
880 | } | 880 | } |
881 | 881 | ||
882 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - | 882 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - |
883 | OCFS2_I(inode)->ip_clusters; | 883 | OCFS2_I(inode)->ip_clusters; |
884 | 884 | ||
885 | /* | 885 | /* |
886 | * protect the pages that ocfs2_zero_extend is going to be | 886 | * protect the pages that ocfs2_zero_extend is going to be |
887 | * pulling into the page cache.. we do this before the | 887 | * pulling into the page cache.. we do this before the |
888 | * metadata extend so that we don't get into the situation | 888 | * metadata extend so that we don't get into the situation |
889 | * where we've extended the metadata but can't get the data | 889 | * where we've extended the metadata but can't get the data |
890 | * lock to zero. | 890 | * lock to zero. |
891 | */ | 891 | */ |
892 | ret = ocfs2_data_lock(inode, 1); | 892 | ret = ocfs2_data_lock(inode, 1); |
893 | if (ret < 0) { | 893 | if (ret < 0) { |
894 | mlog_errno(ret); | 894 | mlog_errno(ret); |
895 | goto out; | 895 | goto out; |
896 | } | 896 | } |
897 | 897 | ||
898 | if (clusters_to_add) { | 898 | if (clusters_to_add) { |
899 | ret = ocfs2_extend_allocation(inode, clusters_to_add); | 899 | ret = ocfs2_extend_allocation(inode, clusters_to_add); |
900 | if (ret < 0) { | 900 | if (ret < 0) { |
901 | mlog_errno(ret); | 901 | mlog_errno(ret); |
902 | goto out_unlock; | 902 | goto out_unlock; |
903 | } | 903 | } |
904 | } | 904 | } |
905 | 905 | ||
906 | /* | 906 | /* |
907 | * Call this even if we don't add any clusters to the tree. We | 907 | * Call this even if we don't add any clusters to the tree. We |
908 | * still need to zero the area between the old i_size and the | 908 | * still need to zero the area between the old i_size and the |
909 | * new i_size. | 909 | * new i_size. |
910 | */ | 910 | */ |
911 | ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip); | 911 | ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip); |
912 | if (ret < 0) { | 912 | if (ret < 0) { |
913 | mlog_errno(ret); | 913 | mlog_errno(ret); |
914 | goto out_unlock; | 914 | goto out_unlock; |
915 | } | 915 | } |
916 | 916 | ||
917 | out_update_size: | 917 | out_update_size: |
918 | if (!tail_to_skip) { | 918 | if (!tail_to_skip) { |
919 | /* We're being called from ocfs2_setattr() which wants | 919 | /* We're being called from ocfs2_setattr() which wants |
920 | * us to update i_size */ | 920 | * us to update i_size */ |
921 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); | 921 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
922 | if (ret < 0) | 922 | if (ret < 0) |
923 | mlog_errno(ret); | 923 | mlog_errno(ret); |
924 | } | 924 | } |
925 | 925 | ||
926 | out_unlock: | 926 | out_unlock: |
927 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | 927 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
928 | ocfs2_data_unlock(inode, 1); | 928 | ocfs2_data_unlock(inode, 1); |
929 | 929 | ||
930 | out: | 930 | out: |
931 | return ret; | 931 | return ret; |
932 | } | 932 | } |
933 | 933 | ||
934 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | 934 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) |
935 | { | 935 | { |
936 | int status = 0, size_change; | 936 | int status = 0, size_change; |
937 | struct inode *inode = dentry->d_inode; | 937 | struct inode *inode = dentry->d_inode; |
938 | struct super_block *sb = inode->i_sb; | 938 | struct super_block *sb = inode->i_sb; |
939 | struct ocfs2_super *osb = OCFS2_SB(sb); | 939 | struct ocfs2_super *osb = OCFS2_SB(sb); |
940 | struct buffer_head *bh = NULL; | 940 | struct buffer_head *bh = NULL; |
941 | handle_t *handle = NULL; | 941 | handle_t *handle = NULL; |
942 | 942 | ||
943 | mlog_entry("(0x%p, '%.*s')\n", dentry, | 943 | mlog_entry("(0x%p, '%.*s')\n", dentry, |
944 | dentry->d_name.len, dentry->d_name.name); | 944 | dentry->d_name.len, dentry->d_name.name); |
945 | 945 | ||
946 | if (attr->ia_valid & ATTR_MODE) | 946 | if (attr->ia_valid & ATTR_MODE) |
947 | mlog(0, "mode change: %d\n", attr->ia_mode); | 947 | mlog(0, "mode change: %d\n", attr->ia_mode); |
948 | if (attr->ia_valid & ATTR_UID) | 948 | if (attr->ia_valid & ATTR_UID) |
949 | mlog(0, "uid change: %d\n", attr->ia_uid); | 949 | mlog(0, "uid change: %d\n", attr->ia_uid); |
950 | if (attr->ia_valid & ATTR_GID) | 950 | if (attr->ia_valid & ATTR_GID) |
951 | mlog(0, "gid change: %d\n", attr->ia_gid); | 951 | mlog(0, "gid change: %d\n", attr->ia_gid); |
952 | if (attr->ia_valid & ATTR_SIZE) | 952 | if (attr->ia_valid & ATTR_SIZE) |
953 | mlog(0, "size change...\n"); | 953 | mlog(0, "size change...\n"); |
954 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | 954 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) |
955 | mlog(0, "time change...\n"); | 955 | mlog(0, "time change...\n"); |
956 | 956 | ||
957 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | 957 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
958 | | ATTR_GID | ATTR_UID | ATTR_MODE) | 958 | | ATTR_GID | ATTR_UID | ATTR_MODE) |
959 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | 959 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { |
960 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | 960 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); |
961 | return 0; | 961 | return 0; |
962 | } | 962 | } |
963 | 963 | ||
964 | status = inode_change_ok(inode, attr); | 964 | status = inode_change_ok(inode, attr); |
965 | if (status) | 965 | if (status) |
966 | return status; | 966 | return status; |
967 | 967 | ||
968 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; | 968 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
969 | if (size_change) { | 969 | if (size_change) { |
970 | status = ocfs2_rw_lock(inode, 1); | 970 | status = ocfs2_rw_lock(inode, 1); |
971 | if (status < 0) { | 971 | if (status < 0) { |
972 | mlog_errno(status); | 972 | mlog_errno(status); |
973 | goto bail; | 973 | goto bail; |
974 | } | 974 | } |
975 | } | 975 | } |
976 | 976 | ||
977 | status = ocfs2_meta_lock(inode, &bh, 1); | 977 | status = ocfs2_meta_lock(inode, &bh, 1); |
978 | if (status < 0) { | 978 | if (status < 0) { |
979 | if (status != -ENOENT) | 979 | if (status != -ENOENT) |
980 | mlog_errno(status); | 980 | mlog_errno(status); |
981 | goto bail_unlock_rw; | 981 | goto bail_unlock_rw; |
982 | } | 982 | } |
983 | 983 | ||
984 | if (size_change && attr->ia_size != i_size_read(inode)) { | 984 | if (size_change && attr->ia_size != i_size_read(inode)) { |
985 | if (i_size_read(inode) > attr->ia_size) | 985 | if (i_size_read(inode) > attr->ia_size) |
986 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | 986 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
987 | else | 987 | else |
988 | status = ocfs2_extend_file(inode, bh, attr->ia_size, 0); | 988 | status = ocfs2_extend_file(inode, bh, attr->ia_size, 0); |
989 | if (status < 0) { | 989 | if (status < 0) { |
990 | if (status != -ENOSPC) | 990 | if (status != -ENOSPC) |
991 | mlog_errno(status); | 991 | mlog_errno(status); |
992 | status = -ENOSPC; | 992 | status = -ENOSPC; |
993 | goto bail_unlock; | 993 | goto bail_unlock; |
994 | } | 994 | } |
995 | } | 995 | } |
996 | 996 | ||
997 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | 997 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
998 | if (IS_ERR(handle)) { | 998 | if (IS_ERR(handle)) { |
999 | status = PTR_ERR(handle); | 999 | status = PTR_ERR(handle); |
1000 | mlog_errno(status); | 1000 | mlog_errno(status); |
1001 | goto bail_unlock; | 1001 | goto bail_unlock; |
1002 | } | 1002 | } |
1003 | 1003 | ||
1004 | /* | ||
1005 | * This will intentionally not wind up calling vmtruncate(), | ||
1006 | * since all the work for a size change has been done above. | ||
1007 | * Otherwise, we could get into problems with truncate as | ||
1008 | * ip_alloc_sem is used there to protect against i_size | ||
1009 | * changes. | ||
1010 | */ | ||
1004 | status = inode_setattr(inode, attr); | 1011 | status = inode_setattr(inode, attr); |
1005 | if (status < 0) { | 1012 | if (status < 0) { |
1006 | mlog_errno(status); | 1013 | mlog_errno(status); |
1007 | goto bail_commit; | 1014 | goto bail_commit; |
1008 | } | 1015 | } |
1009 | 1016 | ||
1010 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | 1017 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
1011 | if (status < 0) | 1018 | if (status < 0) |
1012 | mlog_errno(status); | 1019 | mlog_errno(status); |
1013 | 1020 | ||
1014 | bail_commit: | 1021 | bail_commit: |
1015 | ocfs2_commit_trans(osb, handle); | 1022 | ocfs2_commit_trans(osb, handle); |
1016 | bail_unlock: | 1023 | bail_unlock: |
1017 | ocfs2_meta_unlock(inode, 1); | 1024 | ocfs2_meta_unlock(inode, 1); |
1018 | bail_unlock_rw: | 1025 | bail_unlock_rw: |
1019 | if (size_change) | 1026 | if (size_change) |
1020 | ocfs2_rw_unlock(inode, 1); | 1027 | ocfs2_rw_unlock(inode, 1); |
1021 | bail: | 1028 | bail: |
1022 | if (bh) | 1029 | if (bh) |
1023 | brelse(bh); | 1030 | brelse(bh); |
1024 | 1031 | ||
1025 | mlog_exit(status); | 1032 | mlog_exit(status); |
1026 | return status; | 1033 | return status; |
1027 | } | 1034 | } |
1028 | 1035 | ||
1029 | int ocfs2_getattr(struct vfsmount *mnt, | 1036 | int ocfs2_getattr(struct vfsmount *mnt, |
1030 | struct dentry *dentry, | 1037 | struct dentry *dentry, |
1031 | struct kstat *stat) | 1038 | struct kstat *stat) |
1032 | { | 1039 | { |
1033 | struct inode *inode = dentry->d_inode; | 1040 | struct inode *inode = dentry->d_inode; |
1034 | struct super_block *sb = dentry->d_inode->i_sb; | 1041 | struct super_block *sb = dentry->d_inode->i_sb; |
1035 | struct ocfs2_super *osb = sb->s_fs_info; | 1042 | struct ocfs2_super *osb = sb->s_fs_info; |
1036 | int err; | 1043 | int err; |
1037 | 1044 | ||
1038 | mlog_entry_void(); | 1045 | mlog_entry_void(); |
1039 | 1046 | ||
1040 | err = ocfs2_inode_revalidate(dentry); | 1047 | err = ocfs2_inode_revalidate(dentry); |
1041 | if (err) { | 1048 | if (err) { |
1042 | if (err != -ENOENT) | 1049 | if (err != -ENOENT) |
1043 | mlog_errno(err); | 1050 | mlog_errno(err); |
1044 | goto bail; | 1051 | goto bail; |
1045 | } | 1052 | } |
1046 | 1053 | ||
1047 | generic_fillattr(inode, stat); | 1054 | generic_fillattr(inode, stat); |
1048 | 1055 | ||
1049 | /* We set the blksize from the cluster size for performance */ | 1056 | /* We set the blksize from the cluster size for performance */ |
1050 | stat->blksize = osb->s_clustersize; | 1057 | stat->blksize = osb->s_clustersize; |
1051 | 1058 | ||
1052 | bail: | 1059 | bail: |
1053 | mlog_exit(err); | 1060 | mlog_exit(err); |
1054 | 1061 | ||
1055 | return err; | 1062 | return err; |
1056 | } | 1063 | } |
1057 | 1064 | ||
1058 | int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd) | 1065 | int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd) |
1059 | { | 1066 | { |
1060 | int ret; | 1067 | int ret; |
1061 | 1068 | ||
1062 | mlog_entry_void(); | 1069 | mlog_entry_void(); |
1063 | 1070 | ||
1064 | ret = ocfs2_meta_lock(inode, NULL, 0); | 1071 | ret = ocfs2_meta_lock(inode, NULL, 0); |
1065 | if (ret) { | 1072 | if (ret) { |
1066 | if (ret != -ENOENT) | 1073 | if (ret != -ENOENT) |
1067 | mlog_errno(ret); | 1074 | mlog_errno(ret); |
1068 | goto out; | 1075 | goto out; |
1069 | } | 1076 | } |
1070 | 1077 | ||
1071 | ret = generic_permission(inode, mask, NULL); | 1078 | ret = generic_permission(inode, mask, NULL); |
1072 | 1079 | ||
1073 | ocfs2_meta_unlock(inode, 0); | 1080 | ocfs2_meta_unlock(inode, 0); |
1074 | out: | 1081 | out: |
1075 | mlog_exit(ret); | 1082 | mlog_exit(ret); |
1076 | return ret; | 1083 | return ret; |
1077 | } | 1084 | } |
1078 | 1085 | ||
1079 | static int ocfs2_write_remove_suid(struct inode *inode) | 1086 | static int ocfs2_write_remove_suid(struct inode *inode) |
1080 | { | 1087 | { |
1081 | int ret; | 1088 | int ret; |
1082 | struct buffer_head *bh = NULL; | 1089 | struct buffer_head *bh = NULL; |
1083 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | 1090 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
1084 | handle_t *handle; | 1091 | handle_t *handle; |
1085 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | 1092 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1086 | struct ocfs2_dinode *di; | 1093 | struct ocfs2_dinode *di; |
1087 | 1094 | ||
1088 | mlog_entry("(Inode %llu, mode 0%o)\n", | 1095 | mlog_entry("(Inode %llu, mode 0%o)\n", |
1089 | (unsigned long long)oi->ip_blkno, inode->i_mode); | 1096 | (unsigned long long)oi->ip_blkno, inode->i_mode); |
1090 | 1097 | ||
1091 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | 1098 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
1092 | if (handle == NULL) { | 1099 | if (handle == NULL) { |
1093 | ret = -ENOMEM; | 1100 | ret = -ENOMEM; |
1094 | mlog_errno(ret); | 1101 | mlog_errno(ret); |
1095 | goto out; | 1102 | goto out; |
1096 | } | 1103 | } |
1097 | 1104 | ||
1098 | ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); | 1105 | ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); |
1099 | if (ret < 0) { | 1106 | if (ret < 0) { |
1100 | mlog_errno(ret); | 1107 | mlog_errno(ret); |
1101 | goto out_trans; | 1108 | goto out_trans; |
1102 | } | 1109 | } |
1103 | 1110 | ||
1104 | ret = ocfs2_journal_access(handle, inode, bh, | 1111 | ret = ocfs2_journal_access(handle, inode, bh, |
1105 | OCFS2_JOURNAL_ACCESS_WRITE); | 1112 | OCFS2_JOURNAL_ACCESS_WRITE); |
1106 | if (ret < 0) { | 1113 | if (ret < 0) { |
1107 | mlog_errno(ret); | 1114 | mlog_errno(ret); |
1108 | goto out_bh; | 1115 | goto out_bh; |
1109 | } | 1116 | } |
1110 | 1117 | ||
1111 | inode->i_mode &= ~S_ISUID; | 1118 | inode->i_mode &= ~S_ISUID; |
1112 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | 1119 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) |
1113 | inode->i_mode &= ~S_ISGID; | 1120 | inode->i_mode &= ~S_ISGID; |
1114 | 1121 | ||
1115 | di = (struct ocfs2_dinode *) bh->b_data; | 1122 | di = (struct ocfs2_dinode *) bh->b_data; |
1116 | di->i_mode = cpu_to_le16(inode->i_mode); | 1123 | di->i_mode = cpu_to_le16(inode->i_mode); |
1117 | 1124 | ||
1118 | ret = ocfs2_journal_dirty(handle, bh); | 1125 | ret = ocfs2_journal_dirty(handle, bh); |
1119 | if (ret < 0) | 1126 | if (ret < 0) |
1120 | mlog_errno(ret); | 1127 | mlog_errno(ret); |
1121 | out_bh: | 1128 | out_bh: |
1122 | brelse(bh); | 1129 | brelse(bh); |
1123 | out_trans: | 1130 | out_trans: |
1124 | ocfs2_commit_trans(osb, handle); | 1131 | ocfs2_commit_trans(osb, handle); |
1125 | out: | 1132 | out: |
1126 | mlog_exit(ret); | 1133 | mlog_exit(ret); |
1127 | return ret; | 1134 | return ret; |
1128 | } | 1135 | } |
1129 | 1136 | ||
1130 | /* | 1137 | /* |
1131 | * Will look for holes and unwritten extents in the range starting at | 1138 | * Will look for holes and unwritten extents in the range starting at |
1132 | * pos for count bytes (inclusive). | 1139 | * pos for count bytes (inclusive). |
1133 | */ | 1140 | */ |
1134 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, | 1141 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, |
1135 | size_t count) | 1142 | size_t count) |
1136 | { | 1143 | { |
1137 | int ret = 0; | 1144 | int ret = 0; |
1138 | unsigned int extent_flags; | 1145 | unsigned int extent_flags; |
1139 | u32 cpos, clusters, extent_len, phys_cpos; | 1146 | u32 cpos, clusters, extent_len, phys_cpos; |
1140 | struct super_block *sb = inode->i_sb; | 1147 | struct super_block *sb = inode->i_sb; |
1141 | 1148 | ||
1142 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | 1149 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; |
1143 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | 1150 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; |
1144 | 1151 | ||
1145 | while (clusters) { | 1152 | while (clusters) { |
1146 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, | 1153 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
1147 | &extent_flags); | 1154 | &extent_flags); |
1148 | if (ret < 0) { | 1155 | if (ret < 0) { |
1149 | mlog_errno(ret); | 1156 | mlog_errno(ret); |
1150 | goto out; | 1157 | goto out; |
1151 | } | 1158 | } |
1152 | 1159 | ||
1153 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { | 1160 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
1154 | ret = 1; | 1161 | ret = 1; |
1155 | break; | 1162 | break; |
1156 | } | 1163 | } |
1157 | 1164 | ||
1158 | if (extent_len > clusters) | 1165 | if (extent_len > clusters) |
1159 | extent_len = clusters; | 1166 | extent_len = clusters; |
1160 | 1167 | ||
1161 | clusters -= extent_len; | 1168 | clusters -= extent_len; |
1162 | cpos += extent_len; | 1169 | cpos += extent_len; |
1163 | } | 1170 | } |
1164 | out: | 1171 | out: |
1165 | return ret; | 1172 | return ret; |
1166 | } | 1173 | } |
1167 | 1174 | ||
1168 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, | 1175 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
1169 | loff_t *ppos, | 1176 | loff_t *ppos, |
1170 | size_t count, | 1177 | size_t count, |
1171 | int appending, | 1178 | int appending, |
1172 | int *direct_io) | 1179 | int *direct_io) |
1173 | { | 1180 | { |
1174 | int ret = 0, meta_level = appending; | 1181 | int ret = 0, meta_level = appending; |
1175 | struct inode *inode = dentry->d_inode; | 1182 | struct inode *inode = dentry->d_inode; |
1176 | u32 clusters; | 1183 | u32 clusters; |
1177 | loff_t newsize, saved_pos; | 1184 | loff_t newsize, saved_pos; |
1178 | 1185 | ||
1179 | /* | 1186 | /* |
1180 | * We sample i_size under a read level meta lock to see if our write | 1187 | * We sample i_size under a read level meta lock to see if our write |
1181 | * is extending the file, if it is we back off and get a write level | 1188 | * is extending the file, if it is we back off and get a write level |
1182 | * meta lock. | 1189 | * meta lock. |
1183 | */ | 1190 | */ |
1184 | for(;;) { | 1191 | for(;;) { |
1185 | ret = ocfs2_meta_lock(inode, NULL, meta_level); | 1192 | ret = ocfs2_meta_lock(inode, NULL, meta_level); |
1186 | if (ret < 0) { | 1193 | if (ret < 0) { |
1187 | meta_level = -1; | 1194 | meta_level = -1; |
1188 | mlog_errno(ret); | 1195 | mlog_errno(ret); |
1189 | goto out; | 1196 | goto out; |
1190 | } | 1197 | } |
1191 | 1198 | ||
1192 | /* Clear suid / sgid if necessary. We do this here | 1199 | /* Clear suid / sgid if necessary. We do this here |
1193 | * instead of later in the write path because | 1200 | * instead of later in the write path because |
1194 | * remove_suid() calls ->setattr without any hint that | 1201 | * remove_suid() calls ->setattr without any hint that |
1195 | * we may have already done our cluster locking. Since | 1202 | * we may have already done our cluster locking. Since |
1196 | * ocfs2_setattr() *must* take cluster locks to | 1203 | * ocfs2_setattr() *must* take cluster locks to |
1197 | * proceeed, this will lead us to recursively lock the | 1204 | * proceeed, this will lead us to recursively lock the |
1198 | * inode. There's also the dinode i_size state which | 1205 | * inode. There's also the dinode i_size state which |
1199 | * can be lost via setattr during extending writes (we | 1206 | * can be lost via setattr during extending writes (we |
1200 | * set inode->i_size at the end of a write. */ | 1207 | * set inode->i_size at the end of a write. */ |
1201 | if (should_remove_suid(dentry)) { | 1208 | if (should_remove_suid(dentry)) { |
1202 | if (meta_level == 0) { | 1209 | if (meta_level == 0) { |
1203 | ocfs2_meta_unlock(inode, meta_level); | 1210 | ocfs2_meta_unlock(inode, meta_level); |
1204 | meta_level = 1; | 1211 | meta_level = 1; |
1205 | continue; | 1212 | continue; |
1206 | } | 1213 | } |
1207 | 1214 | ||
1208 | ret = ocfs2_write_remove_suid(inode); | 1215 | ret = ocfs2_write_remove_suid(inode); |
1209 | if (ret < 0) { | 1216 | if (ret < 0) { |
1210 | mlog_errno(ret); | 1217 | mlog_errno(ret); |
1211 | goto out_unlock; | 1218 | goto out_unlock; |
1212 | } | 1219 | } |
1213 | } | 1220 | } |
1214 | 1221 | ||
1215 | /* work on a copy of ppos until we're sure that we won't have | 1222 | /* work on a copy of ppos until we're sure that we won't have |
1216 | * to recalculate it due to relocking. */ | 1223 | * to recalculate it due to relocking. */ |
1217 | if (appending) { | 1224 | if (appending) { |
1218 | saved_pos = i_size_read(inode); | 1225 | saved_pos = i_size_read(inode); |
1219 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | 1226 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); |
1220 | } else { | 1227 | } else { |
1221 | saved_pos = *ppos; | 1228 | saved_pos = *ppos; |
1222 | } | 1229 | } |
1223 | 1230 | ||
1224 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { | 1231 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { |
1225 | loff_t end = saved_pos + count; | 1232 | loff_t end = saved_pos + count; |
1226 | 1233 | ||
1227 | /* | 1234 | /* |
1228 | * Skip the O_DIRECT checks if we don't need | 1235 | * Skip the O_DIRECT checks if we don't need |
1229 | * them. | 1236 | * them. |
1230 | */ | 1237 | */ |
1231 | if (!direct_io || !(*direct_io)) | 1238 | if (!direct_io || !(*direct_io)) |
1232 | break; | 1239 | break; |
1233 | 1240 | ||
1234 | /* | 1241 | /* |
1235 | * Allowing concurrent direct writes means | 1242 | * Allowing concurrent direct writes means |
1236 | * i_size changes wouldn't be synchronized, so | 1243 | * i_size changes wouldn't be synchronized, so |
1237 | * one node could wind up truncating another | 1244 | * one node could wind up truncating another |
1238 | * nodes writes. | 1245 | * nodes writes. |
1239 | */ | 1246 | */ |
1240 | if (end > i_size_read(inode)) { | 1247 | if (end > i_size_read(inode)) { |
1241 | *direct_io = 0; | 1248 | *direct_io = 0; |
1242 | break; | 1249 | break; |
1243 | } | 1250 | } |
1244 | 1251 | ||
1245 | /* | 1252 | /* |
1246 | * We don't fill holes during direct io, so | 1253 | * We don't fill holes during direct io, so |
1247 | * check for them here. If any are found, the | 1254 | * check for them here. If any are found, the |
1248 | * caller will have to retake some cluster | 1255 | * caller will have to retake some cluster |
1249 | * locks and initiate the io as buffered. | 1256 | * locks and initiate the io as buffered. |
1250 | */ | 1257 | */ |
1251 | ret = ocfs2_check_range_for_holes(inode, saved_pos, | 1258 | ret = ocfs2_check_range_for_holes(inode, saved_pos, |
1252 | count); | 1259 | count); |
1253 | if (ret == 1) { | 1260 | if (ret == 1) { |
1254 | *direct_io = 0; | 1261 | *direct_io = 0; |
1255 | ret = 0; | 1262 | ret = 0; |
1256 | } else if (ret < 0) | 1263 | } else if (ret < 0) |
1257 | mlog_errno(ret); | 1264 | mlog_errno(ret); |
1258 | break; | 1265 | break; |
1259 | } | 1266 | } |
1260 | 1267 | ||
1261 | /* | 1268 | /* |
1262 | * The rest of this loop is concerned with legacy file | 1269 | * The rest of this loop is concerned with legacy file |
1263 | * systems which don't support sparse files. | 1270 | * systems which don't support sparse files. |
1264 | */ | 1271 | */ |
1265 | 1272 | ||
1266 | newsize = count + saved_pos; | 1273 | newsize = count + saved_pos; |
1267 | 1274 | ||
1268 | mlog(0, "pos=%lld newsize=%lld cursize=%lld\n", | 1275 | mlog(0, "pos=%lld newsize=%lld cursize=%lld\n", |
1269 | (long long) saved_pos, (long long) newsize, | 1276 | (long long) saved_pos, (long long) newsize, |
1270 | (long long) i_size_read(inode)); | 1277 | (long long) i_size_read(inode)); |
1271 | 1278 | ||
1272 | /* No need for a higher level metadata lock if we're | 1279 | /* No need for a higher level metadata lock if we're |
1273 | * never going past i_size. */ | 1280 | * never going past i_size. */ |
1274 | if (newsize <= i_size_read(inode)) | 1281 | if (newsize <= i_size_read(inode)) |
1275 | break; | 1282 | break; |
1276 | 1283 | ||
1277 | if (meta_level == 0) { | 1284 | if (meta_level == 0) { |
1278 | ocfs2_meta_unlock(inode, meta_level); | 1285 | ocfs2_meta_unlock(inode, meta_level); |
1279 | meta_level = 1; | 1286 | meta_level = 1; |
1280 | continue; | 1287 | continue; |
1281 | } | 1288 | } |
1282 | 1289 | ||
1283 | spin_lock(&OCFS2_I(inode)->ip_lock); | 1290 | spin_lock(&OCFS2_I(inode)->ip_lock); |
1284 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) - | 1291 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) - |
1285 | OCFS2_I(inode)->ip_clusters; | 1292 | OCFS2_I(inode)->ip_clusters; |
1286 | spin_unlock(&OCFS2_I(inode)->ip_lock); | 1293 | spin_unlock(&OCFS2_I(inode)->ip_lock); |
1287 | 1294 | ||
1288 | mlog(0, "Writing at EOF, may need more allocation: " | 1295 | mlog(0, "Writing at EOF, may need more allocation: " |
1289 | "i_size = %lld, newsize = %lld, need %u clusters\n", | 1296 | "i_size = %lld, newsize = %lld, need %u clusters\n", |
1290 | (long long) i_size_read(inode), (long long) newsize, | 1297 | (long long) i_size_read(inode), (long long) newsize, |
1291 | clusters); | 1298 | clusters); |
1292 | 1299 | ||
1293 | /* We only want to continue the rest of this loop if | 1300 | /* We only want to continue the rest of this loop if |
1294 | * our extend will actually require more | 1301 | * our extend will actually require more |
1295 | * allocation. */ | 1302 | * allocation. */ |
1296 | if (!clusters) | 1303 | if (!clusters) |
1297 | break; | 1304 | break; |
1298 | 1305 | ||
1299 | ret = ocfs2_extend_file(inode, NULL, newsize, count); | 1306 | ret = ocfs2_extend_file(inode, NULL, newsize, count); |
1300 | if (ret < 0) { | 1307 | if (ret < 0) { |
1301 | if (ret != -ENOSPC) | 1308 | if (ret != -ENOSPC) |
1302 | mlog_errno(ret); | 1309 | mlog_errno(ret); |
1303 | goto out_unlock; | 1310 | goto out_unlock; |
1304 | } | 1311 | } |
1305 | break; | 1312 | break; |
1306 | } | 1313 | } |
1307 | 1314 | ||
1308 | if (appending) | 1315 | if (appending) |
1309 | *ppos = saved_pos; | 1316 | *ppos = saved_pos; |
1310 | 1317 | ||
1311 | out_unlock: | 1318 | out_unlock: |
1312 | ocfs2_meta_unlock(inode, meta_level); | 1319 | ocfs2_meta_unlock(inode, meta_level); |
1313 | 1320 | ||
1314 | out: | 1321 | out: |
1315 | return ret; | 1322 | return ret; |
1316 | } | 1323 | } |
1317 | 1324 | ||
1318 | static inline void | 1325 | static inline void |
1319 | ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes) | 1326 | ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes) |
1320 | { | 1327 | { |
1321 | const struct iovec *iov = *iovp; | 1328 | const struct iovec *iov = *iovp; |
1322 | size_t base = *basep; | 1329 | size_t base = *basep; |
1323 | 1330 | ||
1324 | do { | 1331 | do { |
1325 | int copy = min(bytes, iov->iov_len - base); | 1332 | int copy = min(bytes, iov->iov_len - base); |
1326 | 1333 | ||
1327 | bytes -= copy; | 1334 | bytes -= copy; |
1328 | base += copy; | 1335 | base += copy; |
1329 | if (iov->iov_len == base) { | 1336 | if (iov->iov_len == base) { |
1330 | iov++; | 1337 | iov++; |
1331 | base = 0; | 1338 | base = 0; |
1332 | } | 1339 | } |
1333 | } while (bytes); | 1340 | } while (bytes); |
1334 | *iovp = iov; | 1341 | *iovp = iov; |
1335 | *basep = base; | 1342 | *basep = base; |
1336 | } | 1343 | } |
1337 | 1344 | ||
1338 | static struct page * ocfs2_get_write_source(char **ret_src_buf, | 1345 | static struct page * ocfs2_get_write_source(char **ret_src_buf, |
1339 | const struct iovec *cur_iov, | 1346 | const struct iovec *cur_iov, |
1340 | size_t iov_offset) | 1347 | size_t iov_offset) |
1341 | { | 1348 | { |
1342 | int ret; | 1349 | int ret; |
1343 | char *buf = cur_iov->iov_base + iov_offset; | 1350 | char *buf = cur_iov->iov_base + iov_offset; |
1344 | struct page *src_page = NULL; | 1351 | struct page *src_page = NULL; |
1345 | unsigned long off; | 1352 | unsigned long off; |
1346 | 1353 | ||
1347 | off = (unsigned long)(buf) & ~PAGE_CACHE_MASK; | 1354 | off = (unsigned long)(buf) & ~PAGE_CACHE_MASK; |
1348 | 1355 | ||
1349 | if (!segment_eq(get_fs(), KERNEL_DS)) { | 1356 | if (!segment_eq(get_fs(), KERNEL_DS)) { |
1350 | /* | 1357 | /* |
1351 | * Pull in the user page. We want to do this outside | 1358 | * Pull in the user page. We want to do this outside |
1352 | * of the meta data locks in order to preserve locking | 1359 | * of the meta data locks in order to preserve locking |
1353 | * order in case of page fault. | 1360 | * order in case of page fault. |
1354 | */ | 1361 | */ |
1355 | ret = get_user_pages(current, current->mm, | 1362 | ret = get_user_pages(current, current->mm, |
1356 | (unsigned long)buf & PAGE_CACHE_MASK, 1, | 1363 | (unsigned long)buf & PAGE_CACHE_MASK, 1, |
1357 | 0, 0, &src_page, NULL); | 1364 | 0, 0, &src_page, NULL); |
1358 | if (ret == 1) | 1365 | if (ret == 1) |
1359 | *ret_src_buf = kmap(src_page) + off; | 1366 | *ret_src_buf = kmap(src_page) + off; |
1360 | else | 1367 | else |
1361 | src_page = ERR_PTR(-EFAULT); | 1368 | src_page = ERR_PTR(-EFAULT); |
1362 | } else { | 1369 | } else { |
1363 | *ret_src_buf = buf; | 1370 | *ret_src_buf = buf; |
1364 | } | 1371 | } |
1365 | 1372 | ||
1366 | return src_page; | 1373 | return src_page; |
1367 | } | 1374 | } |
1368 | 1375 | ||
1369 | static void ocfs2_put_write_source(struct page *page) | 1376 | static void ocfs2_put_write_source(struct page *page) |
1370 | { | 1377 | { |
1371 | if (page) { | 1378 | if (page) { |
1372 | kunmap(page); | 1379 | kunmap(page); |
1373 | page_cache_release(page); | 1380 | page_cache_release(page); |
1374 | } | 1381 | } |
1375 | } | 1382 | } |
1376 | 1383 | ||
1377 | static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos, | 1384 | static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos, |
1378 | const struct iovec *iov, | 1385 | const struct iovec *iov, |
1379 | unsigned long nr_segs, | 1386 | unsigned long nr_segs, |
1380 | size_t count, | 1387 | size_t count, |
1381 | ssize_t o_direct_written) | 1388 | ssize_t o_direct_written) |
1382 | { | 1389 | { |
1383 | int ret = 0; | 1390 | int ret = 0; |
1384 | ssize_t copied, total = 0; | 1391 | ssize_t copied, total = 0; |
1385 | size_t iov_offset = 0, bytes; | 1392 | size_t iov_offset = 0, bytes; |
1386 | loff_t pos; | 1393 | loff_t pos; |
1387 | const struct iovec *cur_iov = iov; | 1394 | const struct iovec *cur_iov = iov; |
1388 | struct page *user_page, *page; | 1395 | struct page *user_page, *page; |
1389 | char *buf, *dst; | 1396 | char *buf, *dst; |
1390 | void *fsdata; | 1397 | void *fsdata; |
1391 | 1398 | ||
1392 | /* | 1399 | /* |
1393 | * handle partial DIO write. Adjust cur_iov if needed. | 1400 | * handle partial DIO write. Adjust cur_iov if needed. |
1394 | */ | 1401 | */ |
1395 | ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written); | 1402 | ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written); |
1396 | 1403 | ||
1397 | do { | 1404 | do { |
1398 | pos = *ppos; | 1405 | pos = *ppos; |
1399 | 1406 | ||
1400 | user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset); | 1407 | user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset); |
1401 | if (IS_ERR(user_page)) { | 1408 | if (IS_ERR(user_page)) { |
1402 | ret = PTR_ERR(user_page); | 1409 | ret = PTR_ERR(user_page); |
1403 | goto out; | 1410 | goto out; |
1404 | } | 1411 | } |
1405 | 1412 | ||
1406 | /* Stay within our page boundaries */ | 1413 | /* Stay within our page boundaries */ |
1407 | bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)), | 1414 | bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)), |
1408 | (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK))); | 1415 | (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK))); |
1409 | /* Stay within the vector boundary */ | 1416 | /* Stay within the vector boundary */ |
1410 | bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset); | 1417 | bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset); |
1411 | /* Stay within count */ | 1418 | /* Stay within count */ |
1412 | bytes = min(bytes, count); | 1419 | bytes = min(bytes, count); |
1413 | 1420 | ||
1414 | page = NULL; | 1421 | page = NULL; |
1415 | ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0, | 1422 | ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0, |
1416 | &page, &fsdata); | 1423 | &page, &fsdata); |
1417 | if (ret) { | 1424 | if (ret) { |
1418 | mlog_errno(ret); | 1425 | mlog_errno(ret); |
1419 | goto out; | 1426 | goto out; |
1420 | } | 1427 | } |
1421 | 1428 | ||
1422 | dst = kmap_atomic(page, KM_USER0); | 1429 | dst = kmap_atomic(page, KM_USER0); |
1423 | memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes); | 1430 | memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes); |
1424 | kunmap_atomic(dst, KM_USER0); | 1431 | kunmap_atomic(dst, KM_USER0); |
1425 | flush_dcache_page(page); | 1432 | flush_dcache_page(page); |
1426 | ocfs2_put_write_source(user_page); | 1433 | ocfs2_put_write_source(user_page); |
1427 | 1434 | ||
1428 | copied = ocfs2_write_end(file, file->f_mapping, pos, bytes, | 1435 | copied = ocfs2_write_end(file, file->f_mapping, pos, bytes, |
1429 | bytes, page, fsdata); | 1436 | bytes, page, fsdata); |
1430 | if (copied < 0) { | 1437 | if (copied < 0) { |
1431 | mlog_errno(copied); | 1438 | mlog_errno(copied); |
1432 | ret = copied; | 1439 | ret = copied; |
1433 | goto out; | 1440 | goto out; |
1434 | } | 1441 | } |
1435 | 1442 | ||
1436 | total += copied; | 1443 | total += copied; |
1437 | *ppos = pos + copied; | 1444 | *ppos = pos + copied; |
1438 | count -= copied; | 1445 | count -= copied; |
1439 | 1446 | ||
1440 | ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied); | 1447 | ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied); |
1441 | } while(count); | 1448 | } while(count); |
1442 | 1449 | ||
1443 | out: | 1450 | out: |
1444 | return total ? total : ret; | 1451 | return total ? total : ret; |
1445 | } | 1452 | } |
1446 | 1453 | ||
1447 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | 1454 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
1448 | const struct iovec *iov, | 1455 | const struct iovec *iov, |
1449 | unsigned long nr_segs, | 1456 | unsigned long nr_segs, |
1450 | loff_t pos) | 1457 | loff_t pos) |
1451 | { | 1458 | { |
1452 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; | 1459 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
1453 | int can_do_direct, sync = 0; | 1460 | int can_do_direct, sync = 0; |
1454 | ssize_t written = 0; | 1461 | ssize_t written = 0; |
1455 | size_t ocount; /* original count */ | 1462 | size_t ocount; /* original count */ |
1456 | size_t count; /* after file limit checks */ | 1463 | size_t count; /* after file limit checks */ |
1457 | loff_t *ppos = &iocb->ki_pos; | 1464 | loff_t *ppos = &iocb->ki_pos; |
1458 | struct file *file = iocb->ki_filp; | 1465 | struct file *file = iocb->ki_filp; |
1459 | struct inode *inode = file->f_path.dentry->d_inode; | 1466 | struct inode *inode = file->f_path.dentry->d_inode; |
1460 | 1467 | ||
1461 | mlog_entry("(0x%p, %u, '%.*s')\n", file, | 1468 | mlog_entry("(0x%p, %u, '%.*s')\n", file, |
1462 | (unsigned int)nr_segs, | 1469 | (unsigned int)nr_segs, |
1463 | file->f_path.dentry->d_name.len, | 1470 | file->f_path.dentry->d_name.len, |
1464 | file->f_path.dentry->d_name.name); | 1471 | file->f_path.dentry->d_name.name); |
1465 | 1472 | ||
1466 | if (iocb->ki_left == 0) | 1473 | if (iocb->ki_left == 0) |
1467 | return 0; | 1474 | return 0; |
1468 | 1475 | ||
1469 | ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | 1476 | ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
1470 | if (ret) | 1477 | if (ret) |
1471 | return ret; | 1478 | return ret; |
1472 | 1479 | ||
1473 | count = ocount; | 1480 | count = ocount; |
1474 | 1481 | ||
1475 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | 1482 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
1476 | 1483 | ||
1477 | appending = file->f_flags & O_APPEND ? 1 : 0; | 1484 | appending = file->f_flags & O_APPEND ? 1 : 0; |
1478 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; | 1485 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; |
1479 | 1486 | ||
1480 | mutex_lock(&inode->i_mutex); | 1487 | mutex_lock(&inode->i_mutex); |
1481 | 1488 | ||
1482 | relock: | 1489 | relock: |
1483 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ | 1490 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
1484 | if (direct_io) { | 1491 | if (direct_io) { |
1485 | down_read(&inode->i_alloc_sem); | 1492 | down_read(&inode->i_alloc_sem); |
1486 | have_alloc_sem = 1; | 1493 | have_alloc_sem = 1; |
1487 | } | 1494 | } |
1488 | 1495 | ||
1489 | /* concurrent O_DIRECT writes are allowed */ | 1496 | /* concurrent O_DIRECT writes are allowed */ |
1490 | rw_level = !direct_io; | 1497 | rw_level = !direct_io; |
1491 | ret = ocfs2_rw_lock(inode, rw_level); | 1498 | ret = ocfs2_rw_lock(inode, rw_level); |
1492 | if (ret < 0) { | 1499 | if (ret < 0) { |
1493 | mlog_errno(ret); | 1500 | mlog_errno(ret); |
1494 | goto out_sems; | 1501 | goto out_sems; |
1495 | } | 1502 | } |
1496 | 1503 | ||
1497 | can_do_direct = direct_io; | 1504 | can_do_direct = direct_io; |
1498 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, | 1505 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, |
1499 | iocb->ki_left, appending, | 1506 | iocb->ki_left, appending, |
1500 | &can_do_direct); | 1507 | &can_do_direct); |
1501 | if (ret < 0) { | 1508 | if (ret < 0) { |
1502 | mlog_errno(ret); | 1509 | mlog_errno(ret); |
1503 | goto out; | 1510 | goto out; |
1504 | } | 1511 | } |
1505 | 1512 | ||
1506 | /* | 1513 | /* |
1507 | * We can't complete the direct I/O as requested, fall back to | 1514 | * We can't complete the direct I/O as requested, fall back to |
1508 | * buffered I/O. | 1515 | * buffered I/O. |
1509 | */ | 1516 | */ |
1510 | if (direct_io && !can_do_direct) { | 1517 | if (direct_io && !can_do_direct) { |
1511 | ocfs2_rw_unlock(inode, rw_level); | 1518 | ocfs2_rw_unlock(inode, rw_level); |
1512 | up_read(&inode->i_alloc_sem); | 1519 | up_read(&inode->i_alloc_sem); |
1513 | 1520 | ||
1514 | have_alloc_sem = 0; | 1521 | have_alloc_sem = 0; |
1515 | rw_level = -1; | 1522 | rw_level = -1; |
1516 | 1523 | ||
1517 | direct_io = 0; | 1524 | direct_io = 0; |
1518 | sync = 1; | 1525 | sync = 1; |
1519 | goto relock; | 1526 | goto relock; |
1520 | } | 1527 | } |
1521 | 1528 | ||
1522 | if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) | 1529 | if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) |
1523 | sync = 1; | 1530 | sync = 1; |
1524 | 1531 | ||
1525 | /* | 1532 | /* |
1526 | * XXX: Is it ok to execute these checks a second time? | 1533 | * XXX: Is it ok to execute these checks a second time? |
1527 | */ | 1534 | */ |
1528 | ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode)); | 1535 | ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode)); |
1529 | if (ret) | 1536 | if (ret) |
1530 | goto out; | 1537 | goto out; |
1531 | 1538 | ||
1532 | /* | 1539 | /* |
1533 | * Set pos so that sync_page_range_nolock() below understands | 1540 | * Set pos so that sync_page_range_nolock() below understands |
1534 | * where to start from. We might've moved it around via the | 1541 | * where to start from. We might've moved it around via the |
1535 | * calls above. The range we want to actually sync starts from | 1542 | * calls above. The range we want to actually sync starts from |
1536 | * *ppos here. | 1543 | * *ppos here. |
1537 | * | 1544 | * |
1538 | */ | 1545 | */ |
1539 | pos = *ppos; | 1546 | pos = *ppos; |
1540 | 1547 | ||
1541 | /* communicate with ocfs2_dio_end_io */ | 1548 | /* communicate with ocfs2_dio_end_io */ |
1542 | ocfs2_iocb_set_rw_locked(iocb, rw_level); | 1549 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
1543 | 1550 | ||
1544 | if (direct_io) { | 1551 | if (direct_io) { |
1545 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, | 1552 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
1546 | ppos, count, ocount); | 1553 | ppos, count, ocount); |
1547 | if (written < 0) { | 1554 | if (written < 0) { |
1548 | ret = written; | 1555 | ret = written; |
1549 | goto out_dio; | 1556 | goto out_dio; |
1550 | } | 1557 | } |
1551 | } else { | 1558 | } else { |
1552 | written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs, | 1559 | written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs, |
1553 | count, written); | 1560 | count, written); |
1554 | if (written < 0) { | 1561 | if (written < 0) { |
1555 | ret = written; | 1562 | ret = written; |
1556 | if (ret != -EFAULT || ret != -ENOSPC) | 1563 | if (ret != -EFAULT || ret != -ENOSPC) |
1557 | mlog_errno(ret); | 1564 | mlog_errno(ret); |
1558 | goto out; | 1565 | goto out; |
1559 | } | 1566 | } |
1560 | } | 1567 | } |
1561 | 1568 | ||
1562 | out_dio: | 1569 | out_dio: |
1563 | /* buffered aio wouldn't have proper lock coverage today */ | 1570 | /* buffered aio wouldn't have proper lock coverage today */ |
1564 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); | 1571 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
1565 | 1572 | ||
1566 | /* | 1573 | /* |
1567 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | 1574 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
1568 | * function pointer which is called when o_direct io completes so that | 1575 | * function pointer which is called when o_direct io completes so that |
1569 | * it can unlock our rw lock. (it's the clustered equivalent of | 1576 | * it can unlock our rw lock. (it's the clustered equivalent of |
1570 | * i_alloc_sem; protects truncate from racing with pending ios). | 1577 | * i_alloc_sem; protects truncate from racing with pending ios). |
1571 | * Unfortunately there are error cases which call end_io and others | 1578 | * Unfortunately there are error cases which call end_io and others |
1572 | * that don't. so we don't have to unlock the rw_lock if either an | 1579 | * that don't. so we don't have to unlock the rw_lock if either an |
1573 | * async dio is going to do it in the future or an end_io after an | 1580 | * async dio is going to do it in the future or an end_io after an |
1574 | * error has already done it. | 1581 | * error has already done it. |
1575 | */ | 1582 | */ |
1576 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | 1583 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
1577 | rw_level = -1; | 1584 | rw_level = -1; |
1578 | have_alloc_sem = 0; | 1585 | have_alloc_sem = 0; |
1579 | } | 1586 | } |
1580 | 1587 | ||
1581 | out: | 1588 | out: |
1582 | if (rw_level != -1) | 1589 | if (rw_level != -1) |
1583 | ocfs2_rw_unlock(inode, rw_level); | 1590 | ocfs2_rw_unlock(inode, rw_level); |
1584 | 1591 | ||
1585 | out_sems: | 1592 | out_sems: |
1586 | if (have_alloc_sem) | 1593 | if (have_alloc_sem) |
1587 | up_read(&inode->i_alloc_sem); | 1594 | up_read(&inode->i_alloc_sem); |
1588 | 1595 | ||
1589 | if (written > 0 && sync) { | 1596 | if (written > 0 && sync) { |
1590 | ssize_t err; | 1597 | ssize_t err; |
1591 | 1598 | ||
1592 | err = sync_page_range_nolock(inode, file->f_mapping, pos, count); | 1599 | err = sync_page_range_nolock(inode, file->f_mapping, pos, count); |
1593 | if (err < 0) | 1600 | if (err < 0) |
1594 | written = err; | 1601 | written = err; |
1595 | } | 1602 | } |
1596 | 1603 | ||
1597 | mutex_unlock(&inode->i_mutex); | 1604 | mutex_unlock(&inode->i_mutex); |
1598 | 1605 | ||
1599 | mlog_exit(ret); | 1606 | mlog_exit(ret); |
1600 | return written ? written : ret; | 1607 | return written ? written : ret; |
1601 | } | 1608 | } |
1602 | 1609 | ||
1603 | static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe, | 1610 | static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe, |
1604 | struct pipe_buffer *buf, | 1611 | struct pipe_buffer *buf, |
1605 | struct splice_desc *sd) | 1612 | struct splice_desc *sd) |
1606 | { | 1613 | { |
1607 | int ret, count; | 1614 | int ret, count; |
1608 | ssize_t copied = 0; | 1615 | ssize_t copied = 0; |
1609 | struct file *file = sd->u.file; | 1616 | struct file *file = sd->u.file; |
1610 | unsigned int offset; | 1617 | unsigned int offset; |
1611 | struct page *page = NULL; | 1618 | struct page *page = NULL; |
1612 | void *fsdata; | 1619 | void *fsdata; |
1613 | char *src, *dst; | 1620 | char *src, *dst; |
1614 | 1621 | ||
1615 | ret = buf->ops->confirm(pipe, buf); | 1622 | ret = buf->ops->confirm(pipe, buf); |
1616 | if (ret) | 1623 | if (ret) |
1617 | goto out; | 1624 | goto out; |
1618 | 1625 | ||
1619 | offset = sd->pos & ~PAGE_CACHE_MASK; | 1626 | offset = sd->pos & ~PAGE_CACHE_MASK; |
1620 | count = sd->len; | 1627 | count = sd->len; |
1621 | if (count + offset > PAGE_CACHE_SIZE) | 1628 | if (count + offset > PAGE_CACHE_SIZE) |
1622 | count = PAGE_CACHE_SIZE - offset; | 1629 | count = PAGE_CACHE_SIZE - offset; |
1623 | 1630 | ||
1624 | ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0, | 1631 | ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0, |
1625 | &page, &fsdata); | 1632 | &page, &fsdata); |
1626 | if (ret) { | 1633 | if (ret) { |
1627 | mlog_errno(ret); | 1634 | mlog_errno(ret); |
1628 | goto out; | 1635 | goto out; |
1629 | } | 1636 | } |
1630 | 1637 | ||
1631 | src = buf->ops->map(pipe, buf, 1); | 1638 | src = buf->ops->map(pipe, buf, 1); |
1632 | dst = kmap_atomic(page, KM_USER1); | 1639 | dst = kmap_atomic(page, KM_USER1); |
1633 | memcpy(dst + offset, src + buf->offset, count); | 1640 | memcpy(dst + offset, src + buf->offset, count); |
1634 | kunmap_atomic(page, KM_USER1); | 1641 | kunmap_atomic(page, KM_USER1); |
1635 | buf->ops->unmap(pipe, buf, src); | 1642 | buf->ops->unmap(pipe, buf, src); |
1636 | 1643 | ||
1637 | copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count, | 1644 | copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count, |
1638 | page, fsdata); | 1645 | page, fsdata); |
1639 | if (copied < 0) { | 1646 | if (copied < 0) { |
1640 | mlog_errno(copied); | 1647 | mlog_errno(copied); |
1641 | ret = copied; | 1648 | ret = copied; |
1642 | goto out; | 1649 | goto out; |
1643 | } | 1650 | } |
1644 | out: | 1651 | out: |
1645 | 1652 | ||
1646 | return copied ? copied : ret; | 1653 | return copied ? copied : ret; |
1647 | } | 1654 | } |
1648 | 1655 | ||
1649 | static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe, | 1656 | static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
1650 | struct file *out, | 1657 | struct file *out, |
1651 | loff_t *ppos, | 1658 | loff_t *ppos, |
1652 | size_t len, | 1659 | size_t len, |
1653 | unsigned int flags) | 1660 | unsigned int flags) |
1654 | { | 1661 | { |
1655 | int ret, err; | 1662 | int ret, err; |
1656 | struct address_space *mapping = out->f_mapping; | 1663 | struct address_space *mapping = out->f_mapping; |
1657 | struct inode *inode = mapping->host; | 1664 | struct inode *inode = mapping->host; |
1658 | struct splice_desc sd = { | 1665 | struct splice_desc sd = { |
1659 | .total_len = len, | 1666 | .total_len = len, |
1660 | .flags = flags, | 1667 | .flags = flags, |
1661 | .pos = *ppos, | 1668 | .pos = *ppos, |
1662 | .u.file = out, | 1669 | .u.file = out, |
1663 | }; | 1670 | }; |
1664 | 1671 | ||
1665 | ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor); | 1672 | ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor); |
1666 | if (ret > 0) { | 1673 | if (ret > 0) { |
1667 | *ppos += ret; | 1674 | *ppos += ret; |
1668 | 1675 | ||
1669 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | 1676 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { |
1670 | err = generic_osync_inode(inode, mapping, | 1677 | err = generic_osync_inode(inode, mapping, |
1671 | OSYNC_METADATA|OSYNC_DATA); | 1678 | OSYNC_METADATA|OSYNC_DATA); |
1672 | if (err) | 1679 | if (err) |
1673 | ret = err; | 1680 | ret = err; |
1674 | } | 1681 | } |
1675 | } | 1682 | } |
1676 | 1683 | ||
1677 | return ret; | 1684 | return ret; |
1678 | } | 1685 | } |
1679 | 1686 | ||
1680 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, | 1687 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
1681 | struct file *out, | 1688 | struct file *out, |
1682 | loff_t *ppos, | 1689 | loff_t *ppos, |
1683 | size_t len, | 1690 | size_t len, |
1684 | unsigned int flags) | 1691 | unsigned int flags) |
1685 | { | 1692 | { |
1686 | int ret; | 1693 | int ret; |
1687 | struct inode *inode = out->f_path.dentry->d_inode; | 1694 | struct inode *inode = out->f_path.dentry->d_inode; |
1688 | 1695 | ||
1689 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, | 1696 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, |
1690 | (unsigned int)len, | 1697 | (unsigned int)len, |
1691 | out->f_path.dentry->d_name.len, | 1698 | out->f_path.dentry->d_name.len, |
1692 | out->f_path.dentry->d_name.name); | 1699 | out->f_path.dentry->d_name.name); |
1693 | 1700 | ||
1694 | inode_double_lock(inode, pipe->inode); | 1701 | inode_double_lock(inode, pipe->inode); |
1695 | 1702 | ||
1696 | ret = ocfs2_rw_lock(inode, 1); | 1703 | ret = ocfs2_rw_lock(inode, 1); |
1697 | if (ret < 0) { | 1704 | if (ret < 0) { |
1698 | mlog_errno(ret); | 1705 | mlog_errno(ret); |
1699 | goto out; | 1706 | goto out; |
1700 | } | 1707 | } |
1701 | 1708 | ||
1702 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, | 1709 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
1703 | NULL); | 1710 | NULL); |
1704 | if (ret < 0) { | 1711 | if (ret < 0) { |
1705 | mlog_errno(ret); | 1712 | mlog_errno(ret); |
1706 | goto out_unlock; | 1713 | goto out_unlock; |
1707 | } | 1714 | } |
1708 | 1715 | ||
1709 | /* ok, we're done with i_size and alloc work */ | 1716 | /* ok, we're done with i_size and alloc work */ |
1710 | ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags); | 1717 | ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags); |
1711 | 1718 | ||
1712 | out_unlock: | 1719 | out_unlock: |
1713 | ocfs2_rw_unlock(inode, 1); | 1720 | ocfs2_rw_unlock(inode, 1); |
1714 | out: | 1721 | out: |
1715 | inode_double_unlock(inode, pipe->inode); | 1722 | inode_double_unlock(inode, pipe->inode); |
1716 | 1723 | ||
1717 | mlog_exit(ret); | 1724 | mlog_exit(ret); |
1718 | return ret; | 1725 | return ret; |
1719 | } | 1726 | } |
1720 | 1727 | ||
1721 | static ssize_t ocfs2_file_splice_read(struct file *in, | 1728 | static ssize_t ocfs2_file_splice_read(struct file *in, |
1722 | loff_t *ppos, | 1729 | loff_t *ppos, |
1723 | struct pipe_inode_info *pipe, | 1730 | struct pipe_inode_info *pipe, |
1724 | size_t len, | 1731 | size_t len, |
1725 | unsigned int flags) | 1732 | unsigned int flags) |
1726 | { | 1733 | { |
1727 | int ret = 0; | 1734 | int ret = 0; |
1728 | struct inode *inode = in->f_path.dentry->d_inode; | 1735 | struct inode *inode = in->f_path.dentry->d_inode; |
1729 | 1736 | ||
1730 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, | 1737 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, |
1731 | (unsigned int)len, | 1738 | (unsigned int)len, |
1732 | in->f_path.dentry->d_name.len, | 1739 | in->f_path.dentry->d_name.len, |
1733 | in->f_path.dentry->d_name.name); | 1740 | in->f_path.dentry->d_name.name); |
1734 | 1741 | ||
1735 | /* | 1742 | /* |
1736 | * See the comment in ocfs2_file_aio_read() | 1743 | * See the comment in ocfs2_file_aio_read() |
1737 | */ | 1744 | */ |
1738 | ret = ocfs2_meta_lock(inode, NULL, 0); | 1745 | ret = ocfs2_meta_lock(inode, NULL, 0); |
1739 | if (ret < 0) { | 1746 | if (ret < 0) { |
1740 | mlog_errno(ret); | 1747 | mlog_errno(ret); |
1741 | goto bail; | 1748 | goto bail; |
1742 | } | 1749 | } |
1743 | ocfs2_meta_unlock(inode, 0); | 1750 | ocfs2_meta_unlock(inode, 0); |
1744 | 1751 | ||
1745 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | 1752 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); |
1746 | 1753 | ||
1747 | bail: | 1754 | bail: |
1748 | mlog_exit(ret); | 1755 | mlog_exit(ret); |
1749 | return ret; | 1756 | return ret; |
1750 | } | 1757 | } |
1751 | 1758 | ||
1752 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, | 1759 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
1753 | const struct iovec *iov, | 1760 | const struct iovec *iov, |
1754 | unsigned long nr_segs, | 1761 | unsigned long nr_segs, |
1755 | loff_t pos) | 1762 | loff_t pos) |
1756 | { | 1763 | { |
1757 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; | 1764 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
1758 | struct file *filp = iocb->ki_filp; | 1765 | struct file *filp = iocb->ki_filp; |
1759 | struct inode *inode = filp->f_path.dentry->d_inode; | 1766 | struct inode *inode = filp->f_path.dentry->d_inode; |
1760 | 1767 | ||
1761 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, | 1768 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
1762 | (unsigned int)nr_segs, | 1769 | (unsigned int)nr_segs, |
1763 | filp->f_path.dentry->d_name.len, | 1770 | filp->f_path.dentry->d_name.len, |
1764 | filp->f_path.dentry->d_name.name); | 1771 | filp->f_path.dentry->d_name.name); |
1765 | 1772 | ||
1766 | if (!inode) { | 1773 | if (!inode) { |
1767 | ret = -EINVAL; | 1774 | ret = -EINVAL; |
1768 | mlog_errno(ret); | 1775 | mlog_errno(ret); |
1769 | goto bail; | 1776 | goto bail; |
1770 | } | 1777 | } |
1771 | 1778 | ||
1772 | /* | 1779 | /* |
1773 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads | 1780 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
1774 | * need locks to protect pending reads from racing with truncate. | 1781 | * need locks to protect pending reads from racing with truncate. |
1775 | */ | 1782 | */ |
1776 | if (filp->f_flags & O_DIRECT) { | 1783 | if (filp->f_flags & O_DIRECT) { |
1777 | down_read(&inode->i_alloc_sem); | 1784 | down_read(&inode->i_alloc_sem); |
1778 | have_alloc_sem = 1; | 1785 | have_alloc_sem = 1; |
1779 | 1786 | ||
1780 | ret = ocfs2_rw_lock(inode, 0); | 1787 | ret = ocfs2_rw_lock(inode, 0); |
1781 | if (ret < 0) { | 1788 | if (ret < 0) { |
1782 | mlog_errno(ret); | 1789 | mlog_errno(ret); |
1783 | goto bail; | 1790 | goto bail; |
1784 | } | 1791 | } |
1785 | rw_level = 0; | 1792 | rw_level = 0; |
1786 | /* communicate with ocfs2_dio_end_io */ | 1793 | /* communicate with ocfs2_dio_end_io */ |
1787 | ocfs2_iocb_set_rw_locked(iocb, rw_level); | 1794 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
1788 | } | 1795 | } |
1789 | 1796 | ||
1790 | /* | 1797 | /* |
1791 | * We're fine letting folks race truncates and extending | 1798 | * We're fine letting folks race truncates and extending |
1792 | * writes with read across the cluster, just like they can | 1799 | * writes with read across the cluster, just like they can |
1793 | * locally. Hence no rw_lock during read. | 1800 | * locally. Hence no rw_lock during read. |
1794 | * | 1801 | * |
1795 | * Take and drop the meta data lock to update inode fields | 1802 | * Take and drop the meta data lock to update inode fields |
1796 | * like i_size. This allows the checks down below | 1803 | * like i_size. This allows the checks down below |
1797 | * generic_file_aio_read() a chance of actually working. | 1804 | * generic_file_aio_read() a chance of actually working. |
1798 | */ | 1805 | */ |
1799 | ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level); | 1806 | ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
1800 | if (ret < 0) { | 1807 | if (ret < 0) { |
1801 | mlog_errno(ret); | 1808 | mlog_errno(ret); |
1802 | goto bail; | 1809 | goto bail; |
1803 | } | 1810 | } |
1804 | ocfs2_meta_unlock(inode, lock_level); | 1811 | ocfs2_meta_unlock(inode, lock_level); |
1805 | 1812 | ||
1806 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); | 1813 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
1807 | if (ret == -EINVAL) | 1814 | if (ret == -EINVAL) |
1808 | mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n"); | 1815 | mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n"); |
1809 | 1816 | ||
1810 | /* buffered aio wouldn't have proper lock coverage today */ | 1817 | /* buffered aio wouldn't have proper lock coverage today */ |
1811 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | 1818 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); |
1812 | 1819 | ||
1813 | /* see ocfs2_file_aio_write */ | 1820 | /* see ocfs2_file_aio_write */ |
1814 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | 1821 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
1815 | rw_level = -1; | 1822 | rw_level = -1; |
1816 | have_alloc_sem = 0; | 1823 | have_alloc_sem = 0; |
1817 | } | 1824 | } |
1818 | 1825 | ||
1819 | bail: | 1826 | bail: |
1820 | if (have_alloc_sem) | 1827 | if (have_alloc_sem) |
1821 | up_read(&inode->i_alloc_sem); | 1828 | up_read(&inode->i_alloc_sem); |
1822 | if (rw_level != -1) | 1829 | if (rw_level != -1) |
1823 | ocfs2_rw_unlock(inode, rw_level); | 1830 | ocfs2_rw_unlock(inode, rw_level); |
1824 | mlog_exit(ret); | 1831 | mlog_exit(ret); |
1825 | 1832 | ||
1826 | return ret; | 1833 | return ret; |
1827 | } | 1834 | } |
1828 | 1835 | ||
1829 | const struct inode_operations ocfs2_file_iops = { | 1836 | const struct inode_operations ocfs2_file_iops = { |
1830 | .setattr = ocfs2_setattr, | 1837 | .setattr = ocfs2_setattr, |
1831 | .getattr = ocfs2_getattr, | 1838 | .getattr = ocfs2_getattr, |
1832 | .permission = ocfs2_permission, | 1839 | .permission = ocfs2_permission, |
1833 | }; | 1840 | }; |
1834 | 1841 | ||
1835 | const struct inode_operations ocfs2_special_file_iops = { | 1842 | const struct inode_operations ocfs2_special_file_iops = { |
1836 | .setattr = ocfs2_setattr, | 1843 | .setattr = ocfs2_setattr, |
1837 | .getattr = ocfs2_getattr, | 1844 | .getattr = ocfs2_getattr, |
1838 | .permission = ocfs2_permission, | 1845 | .permission = ocfs2_permission, |
1839 | }; | 1846 | }; |
1840 | 1847 | ||
1841 | const struct file_operations ocfs2_fops = { | 1848 | const struct file_operations ocfs2_fops = { |
1842 | .read = do_sync_read, | 1849 | .read = do_sync_read, |
1843 | .write = do_sync_write, | 1850 | .write = do_sync_write, |
1844 | .mmap = ocfs2_mmap, | 1851 | .mmap = ocfs2_mmap, |
1845 | .fsync = ocfs2_sync_file, | 1852 | .fsync = ocfs2_sync_file, |
1846 | .release = ocfs2_file_release, | 1853 | .release = ocfs2_file_release, |
1847 | .open = ocfs2_file_open, | 1854 | .open = ocfs2_file_open, |
1848 | .aio_read = ocfs2_file_aio_read, | 1855 | .aio_read = ocfs2_file_aio_read, |
1849 | .aio_write = ocfs2_file_aio_write, | 1856 | .aio_write = ocfs2_file_aio_write, |
1850 | .ioctl = ocfs2_ioctl, | 1857 | .ioctl = ocfs2_ioctl, |
1851 | #ifdef CONFIG_COMPAT | 1858 | #ifdef CONFIG_COMPAT |
1852 | .compat_ioctl = ocfs2_compat_ioctl, | 1859 | .compat_ioctl = ocfs2_compat_ioctl, |
1853 | #endif | 1860 | #endif |
1854 | .splice_read = ocfs2_file_splice_read, | 1861 | .splice_read = ocfs2_file_splice_read, |
1855 | .splice_write = ocfs2_file_splice_write, | 1862 | .splice_write = ocfs2_file_splice_write, |
1856 | }; | 1863 | }; |
1857 | 1864 | ||
1858 | const struct file_operations ocfs2_dops = { | 1865 | const struct file_operations ocfs2_dops = { |
1859 | .read = generic_read_dir, | 1866 | .read = generic_read_dir, |
1860 | .readdir = ocfs2_readdir, | 1867 | .readdir = ocfs2_readdir, |
1861 | .fsync = ocfs2_sync_file, | 1868 | .fsync = ocfs2_sync_file, |
1862 | .ioctl = ocfs2_ioctl, | 1869 | .ioctl = ocfs2_ioctl, |
1863 | #ifdef CONFIG_COMPAT | 1870 | #ifdef CONFIG_COMPAT |
1864 | .compat_ioctl = ocfs2_compat_ioctl, | 1871 | .compat_ioctl = ocfs2_compat_ioctl, |
1865 | #endif | 1872 | #endif |
1866 | }; | 1873 | }; |
1867 | 1874 |
fs/ocfs2/mmap.c
1 | /* -*- mode: c; c-basic-offset: 8; -*- | 1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | 2 | * vim: noexpandtab sw=8 ts=8 sts=0: |
3 | * | 3 | * |
4 | * mmap.c | 4 | * mmap.c |
5 | * | 5 | * |
6 | * Code to deal with the mess that is clustered mmap. | 6 | * Code to deal with the mess that is clustered mmap. |
7 | * | 7 | * |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. |
9 | * | 9 | * |
10 | * This program is free software; you can redistribute it and/or | 10 | * This program is free software; you can redistribute it and/or |
11 | * modify it under the terms of the GNU General Public | 11 | * modify it under the terms of the GNU General Public |
12 | * License as published by the Free Software Foundation; either | 12 | * License as published by the Free Software Foundation; either |
13 | * version 2 of the License, or (at your option) any later version. | 13 | * version 2 of the License, or (at your option) any later version. |
14 | * | 14 | * |
15 | * This program is distributed in the hope that it will be useful, | 15 | * This program is distributed in the hope that it will be useful, |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
18 | * General Public License for more details. | 18 | * General Public License for more details. |
19 | * | 19 | * |
20 | * You should have received a copy of the GNU General Public | 20 | * You should have received a copy of the GNU General Public |
21 | * License along with this program; if not, write to the | 21 | * License along with this program; if not, write to the |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | 22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
23 | * Boston, MA 021110-1307, USA. | 23 | * Boston, MA 021110-1307, USA. |
24 | */ | 24 | */ |
25 | 25 | ||
26 | #include <linux/fs.h> | 26 | #include <linux/fs.h> |
27 | #include <linux/types.h> | 27 | #include <linux/types.h> |
28 | #include <linux/slab.h> | 28 | #include <linux/slab.h> |
29 | #include <linux/highmem.h> | 29 | #include <linux/highmem.h> |
30 | #include <linux/pagemap.h> | 30 | #include <linux/pagemap.h> |
31 | #include <linux/uio.h> | 31 | #include <linux/uio.h> |
32 | #include <linux/signal.h> | 32 | #include <linux/signal.h> |
33 | #include <linux/rbtree.h> | 33 | #include <linux/rbtree.h> |
34 | 34 | ||
35 | #define MLOG_MASK_PREFIX ML_FILE_IO | 35 | #define MLOG_MASK_PREFIX ML_FILE_IO |
36 | #include <cluster/masklog.h> | 36 | #include <cluster/masklog.h> |
37 | 37 | ||
38 | #include "ocfs2.h" | 38 | #include "ocfs2.h" |
39 | 39 | ||
40 | #include "aops.h" | ||
40 | #include "dlmglue.h" | 41 | #include "dlmglue.h" |
41 | #include "file.h" | 42 | #include "file.h" |
42 | #include "inode.h" | 43 | #include "inode.h" |
43 | #include "mmap.h" | 44 | #include "mmap.h" |
44 | 45 | ||
46 | static inline int ocfs2_vm_op_block_sigs(sigset_t *blocked, sigset_t *oldset) | ||
47 | { | ||
48 | /* The best way to deal with signals in the vm path is | ||
49 | * to block them upfront, rather than allowing the | ||
50 | * locking paths to return -ERESTARTSYS. */ | ||
51 | sigfillset(blocked); | ||
52 | |||
53 | /* We should technically never get a bad return value | ||
54 | * from sigprocmask */ | ||
55 | return sigprocmask(SIG_BLOCK, blocked, oldset); | ||
56 | } | ||
57 | |||
58 | static inline int ocfs2_vm_op_unblock_sigs(sigset_t *oldset) | ||
59 | { | ||
60 | return sigprocmask(SIG_SETMASK, oldset, NULL); | ||
61 | } | ||
62 | |||
45 | static struct page *ocfs2_nopage(struct vm_area_struct * area, | 63 | static struct page *ocfs2_nopage(struct vm_area_struct * area, |
46 | unsigned long address, | 64 | unsigned long address, |
47 | int *type) | 65 | int *type) |
48 | { | 66 | { |
49 | struct page *page = NOPAGE_SIGBUS; | 67 | struct page *page = NOPAGE_SIGBUS; |
50 | sigset_t blocked, oldset; | 68 | sigset_t blocked, oldset; |
51 | int ret; | 69 | int ret; |
52 | 70 | ||
53 | mlog_entry("(area=%p, address=%lu, type=%p)\n", area, address, | 71 | mlog_entry("(area=%p, address=%lu, type=%p)\n", area, address, |
54 | type); | 72 | type); |
55 | 73 | ||
56 | /* The best way to deal with signals in this path is | 74 | ret = ocfs2_vm_op_block_sigs(&blocked, &oldset); |
57 | * to block them upfront, rather than allowing the | ||
58 | * locking paths to return -ERESTARTSYS. */ | ||
59 | sigfillset(&blocked); | ||
60 | |||
61 | /* We should technically never get a bad ret return | ||
62 | * from sigprocmask */ | ||
63 | ret = sigprocmask(SIG_BLOCK, &blocked, &oldset); | ||
64 | if (ret < 0) { | 75 | if (ret < 0) { |
65 | mlog_errno(ret); | 76 | mlog_errno(ret); |
66 | goto out; | 77 | goto out; |
67 | } | 78 | } |
68 | 79 | ||
69 | page = filemap_nopage(area, address, type); | 80 | page = filemap_nopage(area, address, type); |
70 | 81 | ||
71 | ret = sigprocmask(SIG_SETMASK, &oldset, NULL); | 82 | ret = ocfs2_vm_op_unblock_sigs(&oldset); |
72 | if (ret < 0) | 83 | if (ret < 0) |
73 | mlog_errno(ret); | 84 | mlog_errno(ret); |
74 | out: | 85 | out: |
75 | mlog_exit_ptr(page); | 86 | mlog_exit_ptr(page); |
76 | return page; | 87 | return page; |
77 | } | 88 | } |
78 | 89 | ||
79 | static struct vm_operations_struct ocfs2_file_vm_ops = { | 90 | static int __ocfs2_page_mkwrite(struct inode *inode, struct buffer_head *di_bh, |
80 | .nopage = ocfs2_nopage, | 91 | struct page *page) |
81 | }; | 92 | { |
93 | int ret; | ||
94 | struct address_space *mapping = inode->i_mapping; | ||
95 | loff_t pos = page->index << PAGE_CACHE_SHIFT; | ||
96 | unsigned int len = PAGE_CACHE_SIZE; | ||
97 | pgoff_t last_index; | ||
98 | struct page *locked_page = NULL; | ||
99 | void *fsdata; | ||
100 | loff_t size = i_size_read(inode); | ||
82 | 101 | ||
83 | int ocfs2_mmap(struct file *file, struct vm_area_struct *vma) | 102 | /* |
103 | * Another node might have truncated while we were waiting on | ||
104 | * cluster locks. | ||
105 | */ | ||
106 | last_index = size >> PAGE_CACHE_SHIFT; | ||
107 | if (page->index > last_index) { | ||
108 | ret = -EINVAL; | ||
109 | goto out; | ||
110 | } | ||
111 | |||
112 | /* | ||
113 | * The i_size check above doesn't catch the case where nodes | ||
114 | * truncated and then re-extended the file. We'll re-check the | ||
115 | * page mapping after taking the page lock inside of | ||
116 | * ocfs2_write_begin_nolock(). | ||
117 | */ | ||
118 | if (!PageUptodate(page) || page->mapping != inode->i_mapping) { | ||
119 | ret = -EINVAL; | ||
120 | goto out; | ||
121 | } | ||
122 | |||
123 | /* | ||
124 | * Call ocfs2_write_begin() and ocfs2_write_end() to take | ||
125 | * advantage of the allocation code there. We pass a write | ||
126 | * length of the whole page (chopped to i_size) to make sure | ||
127 | * the whole thing is allocated. | ||
128 | * | ||
129 | * Since we know the page is up to date, we don't have to | ||
130 | * worry about ocfs2_write_begin() skipping some buffer reads | ||
131 | * because the "write" would invalidate their data. | ||
132 | */ | ||
133 | if (page->index == last_index) | ||
134 | len = size & ~PAGE_CACHE_MASK; | ||
135 | |||
136 | ret = ocfs2_write_begin_nolock(mapping, pos, len, 0, &locked_page, | ||
137 | &fsdata, di_bh, page); | ||
138 | if (ret) { | ||
139 | if (ret != -ENOSPC) | ||
140 | mlog_errno(ret); | ||
141 | goto out; | ||
142 | } | ||
143 | |||
144 | ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page, | ||
145 | fsdata); | ||
146 | if (ret < 0) { | ||
147 | mlog_errno(ret); | ||
148 | goto out; | ||
149 | } | ||
150 | BUG_ON(ret != len); | ||
151 | ret = 0; | ||
152 | out: | ||
153 | return ret; | ||
154 | } | ||
155 | |||
156 | static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct page *page) | ||
84 | { | 157 | { |
85 | int ret = 0, lock_level = 0; | 158 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
86 | struct ocfs2_super *osb = OCFS2_SB(file->f_dentry->d_inode->i_sb); | 159 | struct buffer_head *di_bh = NULL; |
160 | sigset_t blocked, oldset; | ||
161 | int ret, ret2; | ||
87 | 162 | ||
163 | ret = ocfs2_vm_op_block_sigs(&blocked, &oldset); | ||
164 | if (ret < 0) { | ||
165 | mlog_errno(ret); | ||
166 | return ret; | ||
167 | } | ||
168 | |||
88 | /* | 169 | /* |
89 | * Only support shared writeable mmap for local mounts which | 170 | * The cluster locks taken will block a truncate from another |
90 | * don't know about holes. | 171 | * node. Taking the data lock will also ensure that we don't |
172 | * attempt page truncation as part of a downconvert. | ||
91 | */ | 173 | */ |
92 | if ((!ocfs2_mount_local(osb) || ocfs2_sparse_alloc(osb)) && | 174 | ret = ocfs2_meta_lock(inode, &di_bh, 1); |
93 | ((vma->vm_flags & VM_SHARED) || (vma->vm_flags & VM_MAYSHARE)) && | 175 | if (ret < 0) { |
94 | ((vma->vm_flags & VM_WRITE) || (vma->vm_flags & VM_MAYWRITE))) { | 176 | mlog_errno(ret); |
95 | mlog(0, "disallow shared writable mmaps %lx\n", vma->vm_flags); | 177 | goto out; |
96 | /* This is -EINVAL because generic_file_readonly_mmap | ||
97 | * returns it in a similar situation. */ | ||
98 | return -EINVAL; | ||
99 | } | 178 | } |
179 | |||
180 | /* | ||
181 | * The alloc sem should be enough to serialize with | ||
182 | * ocfs2_truncate_file() changing i_size as well as any thread | ||
183 | * modifying the inode btree. | ||
184 | */ | ||
185 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | ||
186 | |||
187 | ret = ocfs2_data_lock(inode, 1); | ||
188 | if (ret < 0) { | ||
189 | mlog_errno(ret); | ||
190 | goto out_meta_unlock; | ||
191 | } | ||
192 | |||
193 | ret = __ocfs2_page_mkwrite(inode, di_bh, page); | ||
194 | |||
195 | ocfs2_data_unlock(inode, 1); | ||
196 | |||
197 | out_meta_unlock: | ||
198 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | ||
199 | |||
200 | brelse(di_bh); | ||
201 | ocfs2_meta_unlock(inode, 1); | ||
202 | |||
203 | out: | ||
204 | ret2 = ocfs2_vm_op_unblock_sigs(&oldset); | ||
205 | if (ret2 < 0) | ||
206 | mlog_errno(ret2); | ||
207 | |||
208 | return ret; | ||
209 | } | ||
210 | |||
211 | static struct vm_operations_struct ocfs2_file_vm_ops = { | ||
212 | .nopage = ocfs2_nopage, | ||
213 | .page_mkwrite = ocfs2_page_mkwrite, | ||
214 | }; | ||
215 | |||
216 | int ocfs2_mmap(struct file *file, struct vm_area_struct *vma) | ||
217 | { | ||
218 | int ret = 0, lock_level = 0; | ||
100 | 219 | ||
101 | ret = ocfs2_meta_lock_atime(file->f_dentry->d_inode, | 220 | ret = ocfs2_meta_lock_atime(file->f_dentry->d_inode, |
102 | file->f_vfsmnt, &lock_level); | 221 | file->f_vfsmnt, &lock_level); |
103 | if (ret < 0) { | 222 | if (ret < 0) { |