Commit e55fd011549eae01a230e3cace6f4d031b6a3453
1 parent
64fd72e0a4
Exists in
ti-lsk-linux-4.1.y
and in
12 other branches
split dentry_kill()
... into trylocks and everything else. The latter (actual killing) is __dentry_kill(). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Showing 1 changed file with 36 additions and 26 deletions Inline Diff
fs/dcache.c
1 | /* | 1 | /* |
2 | * fs/dcache.c | 2 | * fs/dcache.c |
3 | * | 3 | * |
4 | * Complete reimplementation | 4 | * Complete reimplementation |
5 | * (C) 1997 Thomas Schoebel-Theuer, | 5 | * (C) 1997 Thomas Schoebel-Theuer, |
6 | * with heavy changes by Linus Torvalds | 6 | * with heavy changes by Linus Torvalds |
7 | */ | 7 | */ |
8 | 8 | ||
9 | /* | 9 | /* |
10 | * Notes on the allocation strategy: | 10 | * Notes on the allocation strategy: |
11 | * | 11 | * |
12 | * The dcache is a master of the icache - whenever a dcache entry | 12 | * The dcache is a master of the icache - whenever a dcache entry |
13 | * exists, the inode will always exist. "iput()" is done either when | 13 | * exists, the inode will always exist. "iput()" is done either when |
14 | * the dcache entry is deleted or garbage collected. | 14 | * the dcache entry is deleted or garbage collected. |
15 | */ | 15 | */ |
16 | 16 | ||
17 | #include <linux/syscalls.h> | 17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | 18 | #include <linux/string.h> |
19 | #include <linux/mm.h> | 19 | #include <linux/mm.h> |
20 | #include <linux/fs.h> | 20 | #include <linux/fs.h> |
21 | #include <linux/fsnotify.h> | 21 | #include <linux/fsnotify.h> |
22 | #include <linux/slab.h> | 22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | 23 | #include <linux/init.h> |
24 | #include <linux/hash.h> | 24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | 25 | #include <linux/cache.h> |
26 | #include <linux/export.h> | 26 | #include <linux/export.h> |
27 | #include <linux/mount.h> | 27 | #include <linux/mount.h> |
28 | #include <linux/file.h> | 28 | #include <linux/file.h> |
29 | #include <asm/uaccess.h> | 29 | #include <asm/uaccess.h> |
30 | #include <linux/security.h> | 30 | #include <linux/security.h> |
31 | #include <linux/seqlock.h> | 31 | #include <linux/seqlock.h> |
32 | #include <linux/swap.h> | 32 | #include <linux/swap.h> |
33 | #include <linux/bootmem.h> | 33 | #include <linux/bootmem.h> |
34 | #include <linux/fs_struct.h> | 34 | #include <linux/fs_struct.h> |
35 | #include <linux/hardirq.h> | 35 | #include <linux/hardirq.h> |
36 | #include <linux/bit_spinlock.h> | 36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | 37 | #include <linux/rculist_bl.h> |
38 | #include <linux/prefetch.h> | 38 | #include <linux/prefetch.h> |
39 | #include <linux/ratelimit.h> | 39 | #include <linux/ratelimit.h> |
40 | #include <linux/list_lru.h> | 40 | #include <linux/list_lru.h> |
41 | #include "internal.h" | 41 | #include "internal.h" |
42 | #include "mount.h" | 42 | #include "mount.h" |
43 | 43 | ||
44 | /* | 44 | /* |
45 | * Usage: | 45 | * Usage: |
46 | * dcache->d_inode->i_lock protects: | 46 | * dcache->d_inode->i_lock protects: |
47 | * - i_dentry, d_alias, d_inode of aliases | 47 | * - i_dentry, d_alias, d_inode of aliases |
48 | * dcache_hash_bucket lock protects: | 48 | * dcache_hash_bucket lock protects: |
49 | * - the dcache hash table | 49 | * - the dcache hash table |
50 | * s_anon bl list spinlock protects: | 50 | * s_anon bl list spinlock protects: |
51 | * - the s_anon list (see __d_drop) | 51 | * - the s_anon list (see __d_drop) |
52 | * dentry->d_sb->s_dentry_lru_lock protects: | 52 | * dentry->d_sb->s_dentry_lru_lock protects: |
53 | * - the dcache lru lists and counters | 53 | * - the dcache lru lists and counters |
54 | * d_lock protects: | 54 | * d_lock protects: |
55 | * - d_flags | 55 | * - d_flags |
56 | * - d_name | 56 | * - d_name |
57 | * - d_lru | 57 | * - d_lru |
58 | * - d_count | 58 | * - d_count |
59 | * - d_unhashed() | 59 | * - d_unhashed() |
60 | * - d_parent and d_subdirs | 60 | * - d_parent and d_subdirs |
61 | * - childrens' d_child and d_parent | 61 | * - childrens' d_child and d_parent |
62 | * - d_alias, d_inode | 62 | * - d_alias, d_inode |
63 | * | 63 | * |
64 | * Ordering: | 64 | * Ordering: |
65 | * dentry->d_inode->i_lock | 65 | * dentry->d_inode->i_lock |
66 | * dentry->d_lock | 66 | * dentry->d_lock |
67 | * dentry->d_sb->s_dentry_lru_lock | 67 | * dentry->d_sb->s_dentry_lru_lock |
68 | * dcache_hash_bucket lock | 68 | * dcache_hash_bucket lock |
69 | * s_anon lock | 69 | * s_anon lock |
70 | * | 70 | * |
71 | * If there is an ancestor relationship: | 71 | * If there is an ancestor relationship: |
72 | * dentry->d_parent->...->d_parent->d_lock | 72 | * dentry->d_parent->...->d_parent->d_lock |
73 | * ... | 73 | * ... |
74 | * dentry->d_parent->d_lock | 74 | * dentry->d_parent->d_lock |
75 | * dentry->d_lock | 75 | * dentry->d_lock |
76 | * | 76 | * |
77 | * If no ancestor relationship: | 77 | * If no ancestor relationship: |
78 | * if (dentry1 < dentry2) | 78 | * if (dentry1 < dentry2) |
79 | * dentry1->d_lock | 79 | * dentry1->d_lock |
80 | * dentry2->d_lock | 80 | * dentry2->d_lock |
81 | */ | 81 | */ |
82 | int sysctl_vfs_cache_pressure __read_mostly = 100; | 82 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
83 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); | 83 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
84 | 84 | ||
85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
86 | 86 | ||
87 | EXPORT_SYMBOL(rename_lock); | 87 | EXPORT_SYMBOL(rename_lock); |
88 | 88 | ||
89 | static struct kmem_cache *dentry_cache __read_mostly; | 89 | static struct kmem_cache *dentry_cache __read_mostly; |
90 | 90 | ||
91 | /* | 91 | /* |
92 | * This is the single most critical data structure when it comes | 92 | * This is the single most critical data structure when it comes |
93 | * to the dcache: the hashtable for lookups. Somebody should try | 93 | * to the dcache: the hashtable for lookups. Somebody should try |
94 | * to make this good - I've just made it work. | 94 | * to make this good - I've just made it work. |
95 | * | 95 | * |
96 | * This hash-function tries to avoid losing too many bits of hash | 96 | * This hash-function tries to avoid losing too many bits of hash |
97 | * information, yet avoid using a prime hash-size or similar. | 97 | * information, yet avoid using a prime hash-size or similar. |
98 | */ | 98 | */ |
99 | 99 | ||
100 | static unsigned int d_hash_mask __read_mostly; | 100 | static unsigned int d_hash_mask __read_mostly; |
101 | static unsigned int d_hash_shift __read_mostly; | 101 | static unsigned int d_hash_shift __read_mostly; |
102 | 102 | ||
103 | static struct hlist_bl_head *dentry_hashtable __read_mostly; | 103 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
104 | 104 | ||
105 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, | 105 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
106 | unsigned int hash) | 106 | unsigned int hash) |
107 | { | 107 | { |
108 | hash += (unsigned long) parent / L1_CACHE_BYTES; | 108 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
109 | hash = hash + (hash >> d_hash_shift); | 109 | hash = hash + (hash >> d_hash_shift); |
110 | return dentry_hashtable + (hash & d_hash_mask); | 110 | return dentry_hashtable + (hash & d_hash_mask); |
111 | } | 111 | } |
112 | 112 | ||
113 | /* Statistics gathering. */ | 113 | /* Statistics gathering. */ |
114 | struct dentry_stat_t dentry_stat = { | 114 | struct dentry_stat_t dentry_stat = { |
115 | .age_limit = 45, | 115 | .age_limit = 45, |
116 | }; | 116 | }; |
117 | 117 | ||
118 | static DEFINE_PER_CPU(long, nr_dentry); | 118 | static DEFINE_PER_CPU(long, nr_dentry); |
119 | static DEFINE_PER_CPU(long, nr_dentry_unused); | 119 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
120 | 120 | ||
121 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | 121 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) |
122 | 122 | ||
123 | /* | 123 | /* |
124 | * Here we resort to our own counters instead of using generic per-cpu counters | 124 | * Here we resort to our own counters instead of using generic per-cpu counters |
125 | * for consistency with what the vfs inode code does. We are expected to harvest | 125 | * for consistency with what the vfs inode code does. We are expected to harvest |
126 | * better code and performance by having our own specialized counters. | 126 | * better code and performance by having our own specialized counters. |
127 | * | 127 | * |
128 | * Please note that the loop is done over all possible CPUs, not over all online | 128 | * Please note that the loop is done over all possible CPUs, not over all online |
129 | * CPUs. The reason for this is that we don't want to play games with CPUs going | 129 | * CPUs. The reason for this is that we don't want to play games with CPUs going |
130 | * on and off. If one of them goes off, we will just keep their counters. | 130 | * on and off. If one of them goes off, we will just keep their counters. |
131 | * | 131 | * |
132 | * glommer: See cffbc8a for details, and if you ever intend to change this, | 132 | * glommer: See cffbc8a for details, and if you ever intend to change this, |
133 | * please update all vfs counters to match. | 133 | * please update all vfs counters to match. |
134 | */ | 134 | */ |
135 | static long get_nr_dentry(void) | 135 | static long get_nr_dentry(void) |
136 | { | 136 | { |
137 | int i; | 137 | int i; |
138 | long sum = 0; | 138 | long sum = 0; |
139 | for_each_possible_cpu(i) | 139 | for_each_possible_cpu(i) |
140 | sum += per_cpu(nr_dentry, i); | 140 | sum += per_cpu(nr_dentry, i); |
141 | return sum < 0 ? 0 : sum; | 141 | return sum < 0 ? 0 : sum; |
142 | } | 142 | } |
143 | 143 | ||
144 | static long get_nr_dentry_unused(void) | 144 | static long get_nr_dentry_unused(void) |
145 | { | 145 | { |
146 | int i; | 146 | int i; |
147 | long sum = 0; | 147 | long sum = 0; |
148 | for_each_possible_cpu(i) | 148 | for_each_possible_cpu(i) |
149 | sum += per_cpu(nr_dentry_unused, i); | 149 | sum += per_cpu(nr_dentry_unused, i); |
150 | return sum < 0 ? 0 : sum; | 150 | return sum < 0 ? 0 : sum; |
151 | } | 151 | } |
152 | 152 | ||
153 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, | 153 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
154 | size_t *lenp, loff_t *ppos) | 154 | size_t *lenp, loff_t *ppos) |
155 | { | 155 | { |
156 | dentry_stat.nr_dentry = get_nr_dentry(); | 156 | dentry_stat.nr_dentry = get_nr_dentry(); |
157 | dentry_stat.nr_unused = get_nr_dentry_unused(); | 157 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
158 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); | 158 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
159 | } | 159 | } |
160 | #endif | 160 | #endif |
161 | 161 | ||
162 | /* | 162 | /* |
163 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | 163 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. |
164 | * The strings are both count bytes long, and count is non-zero. | 164 | * The strings are both count bytes long, and count is non-zero. |
165 | */ | 165 | */ |
166 | #ifdef CONFIG_DCACHE_WORD_ACCESS | 166 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
167 | 167 | ||
168 | #include <asm/word-at-a-time.h> | 168 | #include <asm/word-at-a-time.h> |
169 | /* | 169 | /* |
170 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | 170 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a |
171 | * aligned allocation for this particular component. We don't | 171 | * aligned allocation for this particular component. We don't |
172 | * strictly need the load_unaligned_zeropad() safety, but it | 172 | * strictly need the load_unaligned_zeropad() safety, but it |
173 | * doesn't hurt either. | 173 | * doesn't hurt either. |
174 | * | 174 | * |
175 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | 175 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do |
176 | * need the careful unaligned handling. | 176 | * need the careful unaligned handling. |
177 | */ | 177 | */ |
178 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) | 178 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
179 | { | 179 | { |
180 | unsigned long a,b,mask; | 180 | unsigned long a,b,mask; |
181 | 181 | ||
182 | for (;;) { | 182 | for (;;) { |
183 | a = *(unsigned long *)cs; | 183 | a = *(unsigned long *)cs; |
184 | b = load_unaligned_zeropad(ct); | 184 | b = load_unaligned_zeropad(ct); |
185 | if (tcount < sizeof(unsigned long)) | 185 | if (tcount < sizeof(unsigned long)) |
186 | break; | 186 | break; |
187 | if (unlikely(a != b)) | 187 | if (unlikely(a != b)) |
188 | return 1; | 188 | return 1; |
189 | cs += sizeof(unsigned long); | 189 | cs += sizeof(unsigned long); |
190 | ct += sizeof(unsigned long); | 190 | ct += sizeof(unsigned long); |
191 | tcount -= sizeof(unsigned long); | 191 | tcount -= sizeof(unsigned long); |
192 | if (!tcount) | 192 | if (!tcount) |
193 | return 0; | 193 | return 0; |
194 | } | 194 | } |
195 | mask = bytemask_from_count(tcount); | 195 | mask = bytemask_from_count(tcount); |
196 | return unlikely(!!((a ^ b) & mask)); | 196 | return unlikely(!!((a ^ b) & mask)); |
197 | } | 197 | } |
198 | 198 | ||
199 | #else | 199 | #else |
200 | 200 | ||
201 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) | 201 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
202 | { | 202 | { |
203 | do { | 203 | do { |
204 | if (*cs != *ct) | 204 | if (*cs != *ct) |
205 | return 1; | 205 | return 1; |
206 | cs++; | 206 | cs++; |
207 | ct++; | 207 | ct++; |
208 | tcount--; | 208 | tcount--; |
209 | } while (tcount); | 209 | } while (tcount); |
210 | return 0; | 210 | return 0; |
211 | } | 211 | } |
212 | 212 | ||
213 | #endif | 213 | #endif |
214 | 214 | ||
215 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) | 215 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
216 | { | 216 | { |
217 | const unsigned char *cs; | 217 | const unsigned char *cs; |
218 | /* | 218 | /* |
219 | * Be careful about RCU walk racing with rename: | 219 | * Be careful about RCU walk racing with rename: |
220 | * use ACCESS_ONCE to fetch the name pointer. | 220 | * use ACCESS_ONCE to fetch the name pointer. |
221 | * | 221 | * |
222 | * NOTE! Even if a rename will mean that the length | 222 | * NOTE! Even if a rename will mean that the length |
223 | * was not loaded atomically, we don't care. The | 223 | * was not loaded atomically, we don't care. The |
224 | * RCU walk will check the sequence count eventually, | 224 | * RCU walk will check the sequence count eventually, |
225 | * and catch it. And we won't overrun the buffer, | 225 | * and catch it. And we won't overrun the buffer, |
226 | * because we're reading the name pointer atomically, | 226 | * because we're reading the name pointer atomically, |
227 | * and a dentry name is guaranteed to be properly | 227 | * and a dentry name is guaranteed to be properly |
228 | * terminated with a NUL byte. | 228 | * terminated with a NUL byte. |
229 | * | 229 | * |
230 | * End result: even if 'len' is wrong, we'll exit | 230 | * End result: even if 'len' is wrong, we'll exit |
231 | * early because the data cannot match (there can | 231 | * early because the data cannot match (there can |
232 | * be no NUL in the ct/tcount data) | 232 | * be no NUL in the ct/tcount data) |
233 | */ | 233 | */ |
234 | cs = ACCESS_ONCE(dentry->d_name.name); | 234 | cs = ACCESS_ONCE(dentry->d_name.name); |
235 | smp_read_barrier_depends(); | 235 | smp_read_barrier_depends(); |
236 | return dentry_string_cmp(cs, ct, tcount); | 236 | return dentry_string_cmp(cs, ct, tcount); |
237 | } | 237 | } |
238 | 238 | ||
239 | static void __d_free(struct rcu_head *head) | 239 | static void __d_free(struct rcu_head *head) |
240 | { | 240 | { |
241 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | 241 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
242 | 242 | ||
243 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); | 243 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
244 | if (dname_external(dentry)) | 244 | if (dname_external(dentry)) |
245 | kfree(dentry->d_name.name); | 245 | kfree(dentry->d_name.name); |
246 | kmem_cache_free(dentry_cache, dentry); | 246 | kmem_cache_free(dentry_cache, dentry); |
247 | } | 247 | } |
248 | 248 | ||
249 | static void dentry_free(struct dentry *dentry) | 249 | static void dentry_free(struct dentry *dentry) |
250 | { | 250 | { |
251 | /* if dentry was never visible to RCU, immediate free is OK */ | 251 | /* if dentry was never visible to RCU, immediate free is OK */ |
252 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | 252 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) |
253 | __d_free(&dentry->d_u.d_rcu); | 253 | __d_free(&dentry->d_u.d_rcu); |
254 | else | 254 | else |
255 | call_rcu(&dentry->d_u.d_rcu, __d_free); | 255 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
256 | } | 256 | } |
257 | 257 | ||
258 | /** | 258 | /** |
259 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | 259 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups |
260 | * @dentry: the target dentry | 260 | * @dentry: the target dentry |
261 | * After this call, in-progress rcu-walk path lookup will fail. This | 261 | * After this call, in-progress rcu-walk path lookup will fail. This |
262 | * should be called after unhashing, and after changing d_inode (if | 262 | * should be called after unhashing, and after changing d_inode (if |
263 | * the dentry has not already been unhashed). | 263 | * the dentry has not already been unhashed). |
264 | */ | 264 | */ |
265 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | 265 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) |
266 | { | 266 | { |
267 | assert_spin_locked(&dentry->d_lock); | 267 | assert_spin_locked(&dentry->d_lock); |
268 | /* Go through a barrier */ | 268 | /* Go through a barrier */ |
269 | write_seqcount_barrier(&dentry->d_seq); | 269 | write_seqcount_barrier(&dentry->d_seq); |
270 | } | 270 | } |
271 | 271 | ||
272 | /* | 272 | /* |
273 | * Release the dentry's inode, using the filesystem | 273 | * Release the dentry's inode, using the filesystem |
274 | * d_iput() operation if defined. Dentry has no refcount | 274 | * d_iput() operation if defined. Dentry has no refcount |
275 | * and is unhashed. | 275 | * and is unhashed. |
276 | */ | 276 | */ |
277 | static void dentry_iput(struct dentry * dentry) | 277 | static void dentry_iput(struct dentry * dentry) |
278 | __releases(dentry->d_lock) | 278 | __releases(dentry->d_lock) |
279 | __releases(dentry->d_inode->i_lock) | 279 | __releases(dentry->d_inode->i_lock) |
280 | { | 280 | { |
281 | struct inode *inode = dentry->d_inode; | 281 | struct inode *inode = dentry->d_inode; |
282 | if (inode) { | 282 | if (inode) { |
283 | dentry->d_inode = NULL; | 283 | dentry->d_inode = NULL; |
284 | hlist_del_init(&dentry->d_alias); | 284 | hlist_del_init(&dentry->d_alias); |
285 | spin_unlock(&dentry->d_lock); | 285 | spin_unlock(&dentry->d_lock); |
286 | spin_unlock(&inode->i_lock); | 286 | spin_unlock(&inode->i_lock); |
287 | if (!inode->i_nlink) | 287 | if (!inode->i_nlink) |
288 | fsnotify_inoderemove(inode); | 288 | fsnotify_inoderemove(inode); |
289 | if (dentry->d_op && dentry->d_op->d_iput) | 289 | if (dentry->d_op && dentry->d_op->d_iput) |
290 | dentry->d_op->d_iput(dentry, inode); | 290 | dentry->d_op->d_iput(dentry, inode); |
291 | else | 291 | else |
292 | iput(inode); | 292 | iput(inode); |
293 | } else { | 293 | } else { |
294 | spin_unlock(&dentry->d_lock); | 294 | spin_unlock(&dentry->d_lock); |
295 | } | 295 | } |
296 | } | 296 | } |
297 | 297 | ||
298 | /* | 298 | /* |
299 | * Release the dentry's inode, using the filesystem | 299 | * Release the dentry's inode, using the filesystem |
300 | * d_iput() operation if defined. dentry remains in-use. | 300 | * d_iput() operation if defined. dentry remains in-use. |
301 | */ | 301 | */ |
302 | static void dentry_unlink_inode(struct dentry * dentry) | 302 | static void dentry_unlink_inode(struct dentry * dentry) |
303 | __releases(dentry->d_lock) | 303 | __releases(dentry->d_lock) |
304 | __releases(dentry->d_inode->i_lock) | 304 | __releases(dentry->d_inode->i_lock) |
305 | { | 305 | { |
306 | struct inode *inode = dentry->d_inode; | 306 | struct inode *inode = dentry->d_inode; |
307 | __d_clear_type(dentry); | 307 | __d_clear_type(dentry); |
308 | dentry->d_inode = NULL; | 308 | dentry->d_inode = NULL; |
309 | hlist_del_init(&dentry->d_alias); | 309 | hlist_del_init(&dentry->d_alias); |
310 | dentry_rcuwalk_barrier(dentry); | 310 | dentry_rcuwalk_barrier(dentry); |
311 | spin_unlock(&dentry->d_lock); | 311 | spin_unlock(&dentry->d_lock); |
312 | spin_unlock(&inode->i_lock); | 312 | spin_unlock(&inode->i_lock); |
313 | if (!inode->i_nlink) | 313 | if (!inode->i_nlink) |
314 | fsnotify_inoderemove(inode); | 314 | fsnotify_inoderemove(inode); |
315 | if (dentry->d_op && dentry->d_op->d_iput) | 315 | if (dentry->d_op && dentry->d_op->d_iput) |
316 | dentry->d_op->d_iput(dentry, inode); | 316 | dentry->d_op->d_iput(dentry, inode); |
317 | else | 317 | else |
318 | iput(inode); | 318 | iput(inode); |
319 | } | 319 | } |
320 | 320 | ||
321 | /* | 321 | /* |
322 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | 322 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry |
323 | * is in use - which includes both the "real" per-superblock | 323 | * is in use - which includes both the "real" per-superblock |
324 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | 324 | * LRU list _and_ the DCACHE_SHRINK_LIST use. |
325 | * | 325 | * |
326 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | 326 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is |
327 | * on the shrink list (ie not on the superblock LRU list). | 327 | * on the shrink list (ie not on the superblock LRU list). |
328 | * | 328 | * |
329 | * The per-cpu "nr_dentry_unused" counters are updated with | 329 | * The per-cpu "nr_dentry_unused" counters are updated with |
330 | * the DCACHE_LRU_LIST bit. | 330 | * the DCACHE_LRU_LIST bit. |
331 | * | 331 | * |
332 | * These helper functions make sure we always follow the | 332 | * These helper functions make sure we always follow the |
333 | * rules. d_lock must be held by the caller. | 333 | * rules. d_lock must be held by the caller. |
334 | */ | 334 | */ |
335 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | 335 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) |
336 | static void d_lru_add(struct dentry *dentry) | 336 | static void d_lru_add(struct dentry *dentry) |
337 | { | 337 | { |
338 | D_FLAG_VERIFY(dentry, 0); | 338 | D_FLAG_VERIFY(dentry, 0); |
339 | dentry->d_flags |= DCACHE_LRU_LIST; | 339 | dentry->d_flags |= DCACHE_LRU_LIST; |
340 | this_cpu_inc(nr_dentry_unused); | 340 | this_cpu_inc(nr_dentry_unused); |
341 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | 341 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
342 | } | 342 | } |
343 | 343 | ||
344 | static void d_lru_del(struct dentry *dentry) | 344 | static void d_lru_del(struct dentry *dentry) |
345 | { | 345 | { |
346 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | 346 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); |
347 | dentry->d_flags &= ~DCACHE_LRU_LIST; | 347 | dentry->d_flags &= ~DCACHE_LRU_LIST; |
348 | this_cpu_dec(nr_dentry_unused); | 348 | this_cpu_dec(nr_dentry_unused); |
349 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | 349 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
350 | } | 350 | } |
351 | 351 | ||
352 | static void d_shrink_del(struct dentry *dentry) | 352 | static void d_shrink_del(struct dentry *dentry) |
353 | { | 353 | { |
354 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | 354 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); |
355 | list_del_init(&dentry->d_lru); | 355 | list_del_init(&dentry->d_lru); |
356 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | 356 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); |
357 | this_cpu_dec(nr_dentry_unused); | 357 | this_cpu_dec(nr_dentry_unused); |
358 | } | 358 | } |
359 | 359 | ||
360 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | 360 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) |
361 | { | 361 | { |
362 | D_FLAG_VERIFY(dentry, 0); | 362 | D_FLAG_VERIFY(dentry, 0); |
363 | list_add(&dentry->d_lru, list); | 363 | list_add(&dentry->d_lru, list); |
364 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | 364 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; |
365 | this_cpu_inc(nr_dentry_unused); | 365 | this_cpu_inc(nr_dentry_unused); |
366 | } | 366 | } |
367 | 367 | ||
368 | /* | 368 | /* |
369 | * These can only be called under the global LRU lock, ie during the | 369 | * These can only be called under the global LRU lock, ie during the |
370 | * callback for freeing the LRU list. "isolate" removes it from the | 370 | * callback for freeing the LRU list. "isolate" removes it from the |
371 | * LRU lists entirely, while shrink_move moves it to the indicated | 371 | * LRU lists entirely, while shrink_move moves it to the indicated |
372 | * private list. | 372 | * private list. |
373 | */ | 373 | */ |
374 | static void d_lru_isolate(struct dentry *dentry) | 374 | static void d_lru_isolate(struct dentry *dentry) |
375 | { | 375 | { |
376 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | 376 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); |
377 | dentry->d_flags &= ~DCACHE_LRU_LIST; | 377 | dentry->d_flags &= ~DCACHE_LRU_LIST; |
378 | this_cpu_dec(nr_dentry_unused); | 378 | this_cpu_dec(nr_dentry_unused); |
379 | list_del_init(&dentry->d_lru); | 379 | list_del_init(&dentry->d_lru); |
380 | } | 380 | } |
381 | 381 | ||
382 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) | 382 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) |
383 | { | 383 | { |
384 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | 384 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); |
385 | dentry->d_flags |= DCACHE_SHRINK_LIST; | 385 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
386 | list_move_tail(&dentry->d_lru, list); | 386 | list_move_tail(&dentry->d_lru, list); |
387 | } | 387 | } |
388 | 388 | ||
389 | /* | 389 | /* |
390 | * dentry_lru_(add|del)_list) must be called with d_lock held. | 390 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
391 | */ | 391 | */ |
392 | static void dentry_lru_add(struct dentry *dentry) | 392 | static void dentry_lru_add(struct dentry *dentry) |
393 | { | 393 | { |
394 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | 394 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
395 | d_lru_add(dentry); | 395 | d_lru_add(dentry); |
396 | } | 396 | } |
397 | 397 | ||
398 | /** | 398 | /** |
399 | * d_drop - drop a dentry | 399 | * d_drop - drop a dentry |
400 | * @dentry: dentry to drop | 400 | * @dentry: dentry to drop |
401 | * | 401 | * |
402 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | 402 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't |
403 | * be found through a VFS lookup any more. Note that this is different from | 403 | * be found through a VFS lookup any more. Note that this is different from |
404 | * deleting the dentry - d_delete will try to mark the dentry negative if | 404 | * deleting the dentry - d_delete will try to mark the dentry negative if |
405 | * possible, giving a successful _negative_ lookup, while d_drop will | 405 | * possible, giving a successful _negative_ lookup, while d_drop will |
406 | * just make the cache lookup fail. | 406 | * just make the cache lookup fail. |
407 | * | 407 | * |
408 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | 408 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some |
409 | * reason (NFS timeouts or autofs deletes). | 409 | * reason (NFS timeouts or autofs deletes). |
410 | * | 410 | * |
411 | * __d_drop requires dentry->d_lock. | 411 | * __d_drop requires dentry->d_lock. |
412 | */ | 412 | */ |
413 | void __d_drop(struct dentry *dentry) | 413 | void __d_drop(struct dentry *dentry) |
414 | { | 414 | { |
415 | if (!d_unhashed(dentry)) { | 415 | if (!d_unhashed(dentry)) { |
416 | struct hlist_bl_head *b; | 416 | struct hlist_bl_head *b; |
417 | /* | 417 | /* |
418 | * Hashed dentries are normally on the dentry hashtable, | 418 | * Hashed dentries are normally on the dentry hashtable, |
419 | * with the exception of those newly allocated by | 419 | * with the exception of those newly allocated by |
420 | * d_obtain_alias, which are always IS_ROOT: | 420 | * d_obtain_alias, which are always IS_ROOT: |
421 | */ | 421 | */ |
422 | if (unlikely(IS_ROOT(dentry))) | 422 | if (unlikely(IS_ROOT(dentry))) |
423 | b = &dentry->d_sb->s_anon; | 423 | b = &dentry->d_sb->s_anon; |
424 | else | 424 | else |
425 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | 425 | b = d_hash(dentry->d_parent, dentry->d_name.hash); |
426 | 426 | ||
427 | hlist_bl_lock(b); | 427 | hlist_bl_lock(b); |
428 | __hlist_bl_del(&dentry->d_hash); | 428 | __hlist_bl_del(&dentry->d_hash); |
429 | dentry->d_hash.pprev = NULL; | 429 | dentry->d_hash.pprev = NULL; |
430 | hlist_bl_unlock(b); | 430 | hlist_bl_unlock(b); |
431 | dentry_rcuwalk_barrier(dentry); | 431 | dentry_rcuwalk_barrier(dentry); |
432 | } | 432 | } |
433 | } | 433 | } |
434 | EXPORT_SYMBOL(__d_drop); | 434 | EXPORT_SYMBOL(__d_drop); |
435 | 435 | ||
436 | void d_drop(struct dentry *dentry) | 436 | void d_drop(struct dentry *dentry) |
437 | { | 437 | { |
438 | spin_lock(&dentry->d_lock); | 438 | spin_lock(&dentry->d_lock); |
439 | __d_drop(dentry); | 439 | __d_drop(dentry); |
440 | spin_unlock(&dentry->d_lock); | 440 | spin_unlock(&dentry->d_lock); |
441 | } | 441 | } |
442 | EXPORT_SYMBOL(d_drop); | 442 | EXPORT_SYMBOL(d_drop); |
443 | 443 | ||
444 | /* | 444 | static void __dentry_kill(struct dentry *dentry) |
445 | * Finish off a dentry we've decided to kill. | ||
446 | * dentry->d_lock must be held, returns with it unlocked. | ||
447 | * If ref is non-zero, then decrement the refcount too. | ||
448 | * Returns dentry requiring refcount drop, or NULL if we're done. | ||
449 | */ | ||
450 | static struct dentry * | ||
451 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | ||
452 | __releases(dentry->d_lock) | ||
453 | { | 445 | { |
454 | struct inode *inode; | ||
455 | struct dentry *parent = NULL; | 446 | struct dentry *parent = NULL; |
456 | bool can_free = true; | 447 | bool can_free = true; |
457 | |||
458 | inode = dentry->d_inode; | ||
459 | if (inode && !spin_trylock(&inode->i_lock)) { | ||
460 | relock: | ||
461 | if (unlock_on_failure) { | ||
462 | spin_unlock(&dentry->d_lock); | ||
463 | cpu_relax(); | ||
464 | } | ||
465 | return dentry; /* try again with same dentry */ | ||
466 | } | ||
467 | if (!IS_ROOT(dentry)) | 448 | if (!IS_ROOT(dentry)) |
468 | parent = dentry->d_parent; | 449 | parent = dentry->d_parent; |
469 | if (parent && !spin_trylock(&parent->d_lock)) { | ||
470 | if (inode) | ||
471 | spin_unlock(&inode->i_lock); | ||
472 | goto relock; | ||
473 | } | ||
474 | 450 | ||
475 | /* | 451 | /* |
476 | * The dentry is now unrecoverably dead to the world. | 452 | * The dentry is now unrecoverably dead to the world. |
477 | */ | 453 | */ |
478 | lockref_mark_dead(&dentry->d_lockref); | 454 | lockref_mark_dead(&dentry->d_lockref); |
479 | 455 | ||
480 | /* | 456 | /* |
481 | * inform the fs via d_prune that this dentry is about to be | 457 | * inform the fs via d_prune that this dentry is about to be |
482 | * unhashed and destroyed. | 458 | * unhashed and destroyed. |
483 | */ | 459 | */ |
484 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) | 460 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
485 | dentry->d_op->d_prune(dentry); | 461 | dentry->d_op->d_prune(dentry); |
486 | 462 | ||
487 | if (dentry->d_flags & DCACHE_LRU_LIST) { | 463 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
488 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | 464 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) |
489 | d_lru_del(dentry); | 465 | d_lru_del(dentry); |
490 | } | 466 | } |
491 | /* if it was on the hash then remove it */ | 467 | /* if it was on the hash then remove it */ |
492 | __d_drop(dentry); | 468 | __d_drop(dentry); |
493 | list_del(&dentry->d_u.d_child); | 469 | list_del(&dentry->d_u.d_child); |
494 | /* | 470 | /* |
495 | * Inform d_walk() that we are no longer attached to the | 471 | * Inform d_walk() that we are no longer attached to the |
496 | * dentry tree | 472 | * dentry tree |
497 | */ | 473 | */ |
498 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | 474 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
499 | if (parent) | 475 | if (parent) |
500 | spin_unlock(&parent->d_lock); | 476 | spin_unlock(&parent->d_lock); |
501 | dentry_iput(dentry); | 477 | dentry_iput(dentry); |
502 | /* | 478 | /* |
503 | * dentry_iput drops the locks, at which point nobody (except | 479 | * dentry_iput drops the locks, at which point nobody (except |
504 | * transient RCU lookups) can reach this dentry. | 480 | * transient RCU lookups) can reach this dentry. |
505 | */ | 481 | */ |
506 | BUG_ON((int)dentry->d_lockref.count > 0); | 482 | BUG_ON((int)dentry->d_lockref.count > 0); |
507 | this_cpu_dec(nr_dentry); | 483 | this_cpu_dec(nr_dentry); |
508 | if (dentry->d_op && dentry->d_op->d_release) | 484 | if (dentry->d_op && dentry->d_op->d_release) |
509 | dentry->d_op->d_release(dentry); | 485 | dentry->d_op->d_release(dentry); |
510 | 486 | ||
511 | spin_lock(&dentry->d_lock); | 487 | spin_lock(&dentry->d_lock); |
512 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | 488 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
513 | dentry->d_flags |= DCACHE_MAY_FREE; | 489 | dentry->d_flags |= DCACHE_MAY_FREE; |
514 | can_free = false; | 490 | can_free = false; |
515 | } | 491 | } |
516 | spin_unlock(&dentry->d_lock); | 492 | spin_unlock(&dentry->d_lock); |
517 | out: | ||
518 | if (likely(can_free)) | 493 | if (likely(can_free)) |
519 | dentry_free(dentry); | 494 | dentry_free(dentry); |
495 | } | ||
496 | |||
497 | /* | ||
498 | * Finish off a dentry we've decided to kill. | ||
499 | * dentry->d_lock must be held, returns with it unlocked. | ||
500 | * If ref is non-zero, then decrement the refcount too. | ||
501 | * Returns dentry requiring refcount drop, or NULL if we're done. | ||
502 | */ | ||
503 | static struct dentry * | ||
504 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | ||
505 | __releases(dentry->d_lock) | ||
506 | { | ||
507 | struct inode *inode = dentry->d_inode; | ||
508 | struct dentry *parent = NULL; | ||
509 | |||
510 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | ||
511 | goto failed; | ||
512 | |||
513 | if (!IS_ROOT(dentry)) { | ||
514 | parent = dentry->d_parent; | ||
515 | if (unlikely(!spin_trylock(&parent->d_lock))) { | ||
516 | if (inode) | ||
517 | spin_unlock(&inode->i_lock); | ||
518 | goto failed; | ||
519 | } | ||
520 | } | ||
521 | |||
522 | __dentry_kill(dentry); | ||
520 | return parent; | 523 | return parent; |
524 | |||
525 | failed: | ||
526 | if (unlock_on_failure) { | ||
527 | spin_unlock(&dentry->d_lock); | ||
528 | cpu_relax(); | ||
529 | } | ||
530 | return dentry; /* try again with same dentry */ | ||
521 | } | 531 | } |
522 | 532 | ||
523 | /* | 533 | /* |
524 | * This is dput | 534 | * This is dput |
525 | * | 535 | * |
526 | * This is complicated by the fact that we do not want to put | 536 | * This is complicated by the fact that we do not want to put |
527 | * dentries that are no longer on any hash chain on the unused | 537 | * dentries that are no longer on any hash chain on the unused |
528 | * list: we'd much rather just get rid of them immediately. | 538 | * list: we'd much rather just get rid of them immediately. |
529 | * | 539 | * |
530 | * However, that implies that we have to traverse the dentry | 540 | * However, that implies that we have to traverse the dentry |
531 | * tree upwards to the parents which might _also_ now be | 541 | * tree upwards to the parents which might _also_ now be |
532 | * scheduled for deletion (it may have been only waiting for | 542 | * scheduled for deletion (it may have been only waiting for |
533 | * its last child to go away). | 543 | * its last child to go away). |
534 | * | 544 | * |
535 | * This tail recursion is done by hand as we don't want to depend | 545 | * This tail recursion is done by hand as we don't want to depend |
536 | * on the compiler to always get this right (gcc generally doesn't). | 546 | * on the compiler to always get this right (gcc generally doesn't). |
537 | * Real recursion would eat up our stack space. | 547 | * Real recursion would eat up our stack space. |
538 | */ | 548 | */ |
539 | 549 | ||
540 | /* | 550 | /* |
541 | * dput - release a dentry | 551 | * dput - release a dentry |
542 | * @dentry: dentry to release | 552 | * @dentry: dentry to release |
543 | * | 553 | * |
544 | * Release a dentry. This will drop the usage count and if appropriate | 554 | * Release a dentry. This will drop the usage count and if appropriate |
545 | * call the dentry unlink method as well as removing it from the queues and | 555 | * call the dentry unlink method as well as removing it from the queues and |
546 | * releasing its resources. If the parent dentries were scheduled for release | 556 | * releasing its resources. If the parent dentries were scheduled for release |
547 | * they too may now get deleted. | 557 | * they too may now get deleted. |
548 | */ | 558 | */ |
549 | void dput(struct dentry *dentry) | 559 | void dput(struct dentry *dentry) |
550 | { | 560 | { |
551 | if (unlikely(!dentry)) | 561 | if (unlikely(!dentry)) |
552 | return; | 562 | return; |
553 | 563 | ||
554 | repeat: | 564 | repeat: |
555 | if (lockref_put_or_lock(&dentry->d_lockref)) | 565 | if (lockref_put_or_lock(&dentry->d_lockref)) |
556 | return; | 566 | return; |
557 | 567 | ||
558 | /* Unreachable? Get rid of it */ | 568 | /* Unreachable? Get rid of it */ |
559 | if (unlikely(d_unhashed(dentry))) | 569 | if (unlikely(d_unhashed(dentry))) |
560 | goto kill_it; | 570 | goto kill_it; |
561 | 571 | ||
562 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | 572 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { |
563 | if (dentry->d_op->d_delete(dentry)) | 573 | if (dentry->d_op->d_delete(dentry)) |
564 | goto kill_it; | 574 | goto kill_it; |
565 | } | 575 | } |
566 | 576 | ||
567 | if (!(dentry->d_flags & DCACHE_REFERENCED)) | 577 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
568 | dentry->d_flags |= DCACHE_REFERENCED; | 578 | dentry->d_flags |= DCACHE_REFERENCED; |
569 | dentry_lru_add(dentry); | 579 | dentry_lru_add(dentry); |
570 | 580 | ||
571 | dentry->d_lockref.count--; | 581 | dentry->d_lockref.count--; |
572 | spin_unlock(&dentry->d_lock); | 582 | spin_unlock(&dentry->d_lock); |
573 | return; | 583 | return; |
574 | 584 | ||
575 | kill_it: | 585 | kill_it: |
576 | dentry = dentry_kill(dentry, 1); | 586 | dentry = dentry_kill(dentry, 1); |
577 | if (dentry) | 587 | if (dentry) |
578 | goto repeat; | 588 | goto repeat; |
579 | } | 589 | } |
580 | EXPORT_SYMBOL(dput); | 590 | EXPORT_SYMBOL(dput); |
581 | 591 | ||
582 | /** | 592 | /** |
583 | * d_invalidate - invalidate a dentry | 593 | * d_invalidate - invalidate a dentry |
584 | * @dentry: dentry to invalidate | 594 | * @dentry: dentry to invalidate |
585 | * | 595 | * |
586 | * Try to invalidate the dentry if it turns out to be | 596 | * Try to invalidate the dentry if it turns out to be |
587 | * possible. If there are other dentries that can be | 597 | * possible. If there are other dentries that can be |
588 | * reached through this one we can't delete it and we | 598 | * reached through this one we can't delete it and we |
589 | * return -EBUSY. On success we return 0. | 599 | * return -EBUSY. On success we return 0. |
590 | * | 600 | * |
591 | * no dcache lock. | 601 | * no dcache lock. |
592 | */ | 602 | */ |
593 | 603 | ||
594 | int d_invalidate(struct dentry * dentry) | 604 | int d_invalidate(struct dentry * dentry) |
595 | { | 605 | { |
596 | /* | 606 | /* |
597 | * If it's already been dropped, return OK. | 607 | * If it's already been dropped, return OK. |
598 | */ | 608 | */ |
599 | spin_lock(&dentry->d_lock); | 609 | spin_lock(&dentry->d_lock); |
600 | if (d_unhashed(dentry)) { | 610 | if (d_unhashed(dentry)) { |
601 | spin_unlock(&dentry->d_lock); | 611 | spin_unlock(&dentry->d_lock); |
602 | return 0; | 612 | return 0; |
603 | } | 613 | } |
604 | /* | 614 | /* |
605 | * Check whether to do a partial shrink_dcache | 615 | * Check whether to do a partial shrink_dcache |
606 | * to get rid of unused child entries. | 616 | * to get rid of unused child entries. |
607 | */ | 617 | */ |
608 | if (!list_empty(&dentry->d_subdirs)) { | 618 | if (!list_empty(&dentry->d_subdirs)) { |
609 | spin_unlock(&dentry->d_lock); | 619 | spin_unlock(&dentry->d_lock); |
610 | shrink_dcache_parent(dentry); | 620 | shrink_dcache_parent(dentry); |
611 | spin_lock(&dentry->d_lock); | 621 | spin_lock(&dentry->d_lock); |
612 | } | 622 | } |
613 | 623 | ||
614 | /* | 624 | /* |
615 | * Somebody else still using it? | 625 | * Somebody else still using it? |
616 | * | 626 | * |
617 | * If it's a directory, we can't drop it | 627 | * If it's a directory, we can't drop it |
618 | * for fear of somebody re-populating it | 628 | * for fear of somebody re-populating it |
619 | * with children (even though dropping it | 629 | * with children (even though dropping it |
620 | * would make it unreachable from the root, | 630 | * would make it unreachable from the root, |
621 | * we might still populate it if it was a | 631 | * we might still populate it if it was a |
622 | * working directory or similar). | 632 | * working directory or similar). |
623 | * We also need to leave mountpoints alone, | 633 | * We also need to leave mountpoints alone, |
624 | * directory or not. | 634 | * directory or not. |
625 | */ | 635 | */ |
626 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { | 636 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
627 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { | 637 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
628 | spin_unlock(&dentry->d_lock); | 638 | spin_unlock(&dentry->d_lock); |
629 | return -EBUSY; | 639 | return -EBUSY; |
630 | } | 640 | } |
631 | } | 641 | } |
632 | 642 | ||
633 | __d_drop(dentry); | 643 | __d_drop(dentry); |
634 | spin_unlock(&dentry->d_lock); | 644 | spin_unlock(&dentry->d_lock); |
635 | return 0; | 645 | return 0; |
636 | } | 646 | } |
637 | EXPORT_SYMBOL(d_invalidate); | 647 | EXPORT_SYMBOL(d_invalidate); |
638 | 648 | ||
639 | /* This must be called with d_lock held */ | 649 | /* This must be called with d_lock held */ |
640 | static inline void __dget_dlock(struct dentry *dentry) | 650 | static inline void __dget_dlock(struct dentry *dentry) |
641 | { | 651 | { |
642 | dentry->d_lockref.count++; | 652 | dentry->d_lockref.count++; |
643 | } | 653 | } |
644 | 654 | ||
645 | static inline void __dget(struct dentry *dentry) | 655 | static inline void __dget(struct dentry *dentry) |
646 | { | 656 | { |
647 | lockref_get(&dentry->d_lockref); | 657 | lockref_get(&dentry->d_lockref); |
648 | } | 658 | } |
649 | 659 | ||
650 | struct dentry *dget_parent(struct dentry *dentry) | 660 | struct dentry *dget_parent(struct dentry *dentry) |
651 | { | 661 | { |
652 | int gotref; | 662 | int gotref; |
653 | struct dentry *ret; | 663 | struct dentry *ret; |
654 | 664 | ||
655 | /* | 665 | /* |
656 | * Do optimistic parent lookup without any | 666 | * Do optimistic parent lookup without any |
657 | * locking. | 667 | * locking. |
658 | */ | 668 | */ |
659 | rcu_read_lock(); | 669 | rcu_read_lock(); |
660 | ret = ACCESS_ONCE(dentry->d_parent); | 670 | ret = ACCESS_ONCE(dentry->d_parent); |
661 | gotref = lockref_get_not_zero(&ret->d_lockref); | 671 | gotref = lockref_get_not_zero(&ret->d_lockref); |
662 | rcu_read_unlock(); | 672 | rcu_read_unlock(); |
663 | if (likely(gotref)) { | 673 | if (likely(gotref)) { |
664 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | 674 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) |
665 | return ret; | 675 | return ret; |
666 | dput(ret); | 676 | dput(ret); |
667 | } | 677 | } |
668 | 678 | ||
669 | repeat: | 679 | repeat: |
670 | /* | 680 | /* |
671 | * Don't need rcu_dereference because we re-check it was correct under | 681 | * Don't need rcu_dereference because we re-check it was correct under |
672 | * the lock. | 682 | * the lock. |
673 | */ | 683 | */ |
674 | rcu_read_lock(); | 684 | rcu_read_lock(); |
675 | ret = dentry->d_parent; | 685 | ret = dentry->d_parent; |
676 | spin_lock(&ret->d_lock); | 686 | spin_lock(&ret->d_lock); |
677 | if (unlikely(ret != dentry->d_parent)) { | 687 | if (unlikely(ret != dentry->d_parent)) { |
678 | spin_unlock(&ret->d_lock); | 688 | spin_unlock(&ret->d_lock); |
679 | rcu_read_unlock(); | 689 | rcu_read_unlock(); |
680 | goto repeat; | 690 | goto repeat; |
681 | } | 691 | } |
682 | rcu_read_unlock(); | 692 | rcu_read_unlock(); |
683 | BUG_ON(!ret->d_lockref.count); | 693 | BUG_ON(!ret->d_lockref.count); |
684 | ret->d_lockref.count++; | 694 | ret->d_lockref.count++; |
685 | spin_unlock(&ret->d_lock); | 695 | spin_unlock(&ret->d_lock); |
686 | return ret; | 696 | return ret; |
687 | } | 697 | } |
688 | EXPORT_SYMBOL(dget_parent); | 698 | EXPORT_SYMBOL(dget_parent); |
689 | 699 | ||
690 | /** | 700 | /** |
691 | * d_find_alias - grab a hashed alias of inode | 701 | * d_find_alias - grab a hashed alias of inode |
692 | * @inode: inode in question | 702 | * @inode: inode in question |
693 | * @want_discon: flag, used by d_splice_alias, to request | 703 | * @want_discon: flag, used by d_splice_alias, to request |
694 | * that only a DISCONNECTED alias be returned. | 704 | * that only a DISCONNECTED alias be returned. |
695 | * | 705 | * |
696 | * If inode has a hashed alias, or is a directory and has any alias, | 706 | * If inode has a hashed alias, or is a directory and has any alias, |
697 | * acquire the reference to alias and return it. Otherwise return NULL. | 707 | * acquire the reference to alias and return it. Otherwise return NULL. |
698 | * Notice that if inode is a directory there can be only one alias and | 708 | * Notice that if inode is a directory there can be only one alias and |
699 | * it can be unhashed only if it has no children, or if it is the root | 709 | * it can be unhashed only if it has no children, or if it is the root |
700 | * of a filesystem. | 710 | * of a filesystem. |
701 | * | 711 | * |
702 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer | 712 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
703 | * any other hashed alias over that one unless @want_discon is set, | 713 | * any other hashed alias over that one unless @want_discon is set, |
704 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | 714 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
705 | */ | 715 | */ |
706 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) | 716 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
707 | { | 717 | { |
708 | struct dentry *alias, *discon_alias; | 718 | struct dentry *alias, *discon_alias; |
709 | 719 | ||
710 | again: | 720 | again: |
711 | discon_alias = NULL; | 721 | discon_alias = NULL; |
712 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { | 722 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
713 | spin_lock(&alias->d_lock); | 723 | spin_lock(&alias->d_lock); |
714 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | 724 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
715 | if (IS_ROOT(alias) && | 725 | if (IS_ROOT(alias) && |
716 | (alias->d_flags & DCACHE_DISCONNECTED)) { | 726 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
717 | discon_alias = alias; | 727 | discon_alias = alias; |
718 | } else if (!want_discon) { | 728 | } else if (!want_discon) { |
719 | __dget_dlock(alias); | 729 | __dget_dlock(alias); |
720 | spin_unlock(&alias->d_lock); | 730 | spin_unlock(&alias->d_lock); |
721 | return alias; | 731 | return alias; |
722 | } | 732 | } |
723 | } | 733 | } |
724 | spin_unlock(&alias->d_lock); | 734 | spin_unlock(&alias->d_lock); |
725 | } | 735 | } |
726 | if (discon_alias) { | 736 | if (discon_alias) { |
727 | alias = discon_alias; | 737 | alias = discon_alias; |
728 | spin_lock(&alias->d_lock); | 738 | spin_lock(&alias->d_lock); |
729 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | 739 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
730 | if (IS_ROOT(alias) && | 740 | if (IS_ROOT(alias) && |
731 | (alias->d_flags & DCACHE_DISCONNECTED)) { | 741 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
732 | __dget_dlock(alias); | 742 | __dget_dlock(alias); |
733 | spin_unlock(&alias->d_lock); | 743 | spin_unlock(&alias->d_lock); |
734 | return alias; | 744 | return alias; |
735 | } | 745 | } |
736 | } | 746 | } |
737 | spin_unlock(&alias->d_lock); | 747 | spin_unlock(&alias->d_lock); |
738 | goto again; | 748 | goto again; |
739 | } | 749 | } |
740 | return NULL; | 750 | return NULL; |
741 | } | 751 | } |
742 | 752 | ||
743 | struct dentry *d_find_alias(struct inode *inode) | 753 | struct dentry *d_find_alias(struct inode *inode) |
744 | { | 754 | { |
745 | struct dentry *de = NULL; | 755 | struct dentry *de = NULL; |
746 | 756 | ||
747 | if (!hlist_empty(&inode->i_dentry)) { | 757 | if (!hlist_empty(&inode->i_dentry)) { |
748 | spin_lock(&inode->i_lock); | 758 | spin_lock(&inode->i_lock); |
749 | de = __d_find_alias(inode, 0); | 759 | de = __d_find_alias(inode, 0); |
750 | spin_unlock(&inode->i_lock); | 760 | spin_unlock(&inode->i_lock); |
751 | } | 761 | } |
752 | return de; | 762 | return de; |
753 | } | 763 | } |
754 | EXPORT_SYMBOL(d_find_alias); | 764 | EXPORT_SYMBOL(d_find_alias); |
755 | 765 | ||
756 | /* | 766 | /* |
757 | * Try to kill dentries associated with this inode. | 767 | * Try to kill dentries associated with this inode. |
758 | * WARNING: you must own a reference to inode. | 768 | * WARNING: you must own a reference to inode. |
759 | */ | 769 | */ |
760 | void d_prune_aliases(struct inode *inode) | 770 | void d_prune_aliases(struct inode *inode) |
761 | { | 771 | { |
762 | struct dentry *dentry; | 772 | struct dentry *dentry; |
763 | restart: | 773 | restart: |
764 | spin_lock(&inode->i_lock); | 774 | spin_lock(&inode->i_lock); |
765 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { | 775 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
766 | spin_lock(&dentry->d_lock); | 776 | spin_lock(&dentry->d_lock); |
767 | if (!dentry->d_lockref.count) { | 777 | if (!dentry->d_lockref.count) { |
768 | /* | 778 | /* |
769 | * inform the fs via d_prune that this dentry | 779 | * inform the fs via d_prune that this dentry |
770 | * is about to be unhashed and destroyed. | 780 | * is about to be unhashed and destroyed. |
771 | */ | 781 | */ |
772 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | 782 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && |
773 | !d_unhashed(dentry)) | 783 | !d_unhashed(dentry)) |
774 | dentry->d_op->d_prune(dentry); | 784 | dentry->d_op->d_prune(dentry); |
775 | 785 | ||
776 | __dget_dlock(dentry); | 786 | __dget_dlock(dentry); |
777 | __d_drop(dentry); | 787 | __d_drop(dentry); |
778 | spin_unlock(&dentry->d_lock); | 788 | spin_unlock(&dentry->d_lock); |
779 | spin_unlock(&inode->i_lock); | 789 | spin_unlock(&inode->i_lock); |
780 | dput(dentry); | 790 | dput(dentry); |
781 | goto restart; | 791 | goto restart; |
782 | } | 792 | } |
783 | spin_unlock(&dentry->d_lock); | 793 | spin_unlock(&dentry->d_lock); |
784 | } | 794 | } |
785 | spin_unlock(&inode->i_lock); | 795 | spin_unlock(&inode->i_lock); |
786 | } | 796 | } |
787 | EXPORT_SYMBOL(d_prune_aliases); | 797 | EXPORT_SYMBOL(d_prune_aliases); |
788 | 798 | ||
789 | static void shrink_dentry_list(struct list_head *list) | 799 | static void shrink_dentry_list(struct list_head *list) |
790 | { | 800 | { |
791 | struct dentry *dentry, *parent; | 801 | struct dentry *dentry, *parent; |
792 | 802 | ||
793 | while (!list_empty(list)) { | 803 | while (!list_empty(list)) { |
794 | dentry = list_entry(list->prev, struct dentry, d_lru); | 804 | dentry = list_entry(list->prev, struct dentry, d_lru); |
795 | spin_lock(&dentry->d_lock); | 805 | spin_lock(&dentry->d_lock); |
796 | /* | 806 | /* |
797 | * The dispose list is isolated and dentries are not accounted | 807 | * The dispose list is isolated and dentries are not accounted |
798 | * to the LRU here, so we can simply remove it from the list | 808 | * to the LRU here, so we can simply remove it from the list |
799 | * here regardless of whether it is referenced or not. | 809 | * here regardless of whether it is referenced or not. |
800 | */ | 810 | */ |
801 | d_shrink_del(dentry); | 811 | d_shrink_del(dentry); |
802 | 812 | ||
803 | /* | 813 | /* |
804 | * We found an inuse dentry which was not removed from | 814 | * We found an inuse dentry which was not removed from |
805 | * the LRU because of laziness during lookup. Do not free it. | 815 | * the LRU because of laziness during lookup. Do not free it. |
806 | */ | 816 | */ |
807 | if ((int)dentry->d_lockref.count > 0) { | 817 | if ((int)dentry->d_lockref.count > 0) { |
808 | spin_unlock(&dentry->d_lock); | 818 | spin_unlock(&dentry->d_lock); |
809 | continue; | 819 | continue; |
810 | } | 820 | } |
811 | 821 | ||
812 | 822 | ||
813 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | 823 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { |
814 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | 824 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; |
815 | spin_unlock(&dentry->d_lock); | 825 | spin_unlock(&dentry->d_lock); |
816 | if (can_free) | 826 | if (can_free) |
817 | dentry_free(dentry); | 827 | dentry_free(dentry); |
818 | continue; | 828 | continue; |
819 | } | 829 | } |
820 | 830 | ||
821 | parent = dentry_kill(dentry, 0); | 831 | parent = dentry_kill(dentry, 0); |
822 | /* | 832 | /* |
823 | * If dentry_kill returns NULL, we have nothing more to do. | 833 | * If dentry_kill returns NULL, we have nothing more to do. |
824 | */ | 834 | */ |
825 | if (!parent) | 835 | if (!parent) |
826 | continue; | 836 | continue; |
827 | 837 | ||
828 | if (unlikely(parent == dentry)) { | 838 | if (unlikely(parent == dentry)) { |
829 | /* | 839 | /* |
830 | * trylocks have failed and d_lock has been held the | 840 | * trylocks have failed and d_lock has been held the |
831 | * whole time, so it could not have been added to any | 841 | * whole time, so it could not have been added to any |
832 | * other lists. Just add it back to the shrink list. | 842 | * other lists. Just add it back to the shrink list. |
833 | */ | 843 | */ |
834 | d_shrink_add(dentry, list); | 844 | d_shrink_add(dentry, list); |
835 | spin_unlock(&dentry->d_lock); | 845 | spin_unlock(&dentry->d_lock); |
836 | continue; | 846 | continue; |
837 | } | 847 | } |
838 | /* | 848 | /* |
839 | * We need to prune ancestors too. This is necessary to prevent | 849 | * We need to prune ancestors too. This is necessary to prevent |
840 | * quadratic behavior of shrink_dcache_parent(), but is also | 850 | * quadratic behavior of shrink_dcache_parent(), but is also |
841 | * expected to be beneficial in reducing dentry cache | 851 | * expected to be beneficial in reducing dentry cache |
842 | * fragmentation. | 852 | * fragmentation. |
843 | */ | 853 | */ |
844 | dentry = parent; | 854 | dentry = parent; |
845 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) | 855 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) |
846 | dentry = dentry_kill(dentry, 1); | 856 | dentry = dentry_kill(dentry, 1); |
847 | } | 857 | } |
848 | } | 858 | } |
849 | 859 | ||
850 | static enum lru_status | 860 | static enum lru_status |
851 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | 861 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) |
852 | { | 862 | { |
853 | struct list_head *freeable = arg; | 863 | struct list_head *freeable = arg; |
854 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | 864 | struct dentry *dentry = container_of(item, struct dentry, d_lru); |
855 | 865 | ||
856 | 866 | ||
857 | /* | 867 | /* |
858 | * we are inverting the lru lock/dentry->d_lock here, | 868 | * we are inverting the lru lock/dentry->d_lock here, |
859 | * so use a trylock. If we fail to get the lock, just skip | 869 | * so use a trylock. If we fail to get the lock, just skip |
860 | * it | 870 | * it |
861 | */ | 871 | */ |
862 | if (!spin_trylock(&dentry->d_lock)) | 872 | if (!spin_trylock(&dentry->d_lock)) |
863 | return LRU_SKIP; | 873 | return LRU_SKIP; |
864 | 874 | ||
865 | /* | 875 | /* |
866 | * Referenced dentries are still in use. If they have active | 876 | * Referenced dentries are still in use. If they have active |
867 | * counts, just remove them from the LRU. Otherwise give them | 877 | * counts, just remove them from the LRU. Otherwise give them |
868 | * another pass through the LRU. | 878 | * another pass through the LRU. |
869 | */ | 879 | */ |
870 | if (dentry->d_lockref.count) { | 880 | if (dentry->d_lockref.count) { |
871 | d_lru_isolate(dentry); | 881 | d_lru_isolate(dentry); |
872 | spin_unlock(&dentry->d_lock); | 882 | spin_unlock(&dentry->d_lock); |
873 | return LRU_REMOVED; | 883 | return LRU_REMOVED; |
874 | } | 884 | } |
875 | 885 | ||
876 | if (dentry->d_flags & DCACHE_REFERENCED) { | 886 | if (dentry->d_flags & DCACHE_REFERENCED) { |
877 | dentry->d_flags &= ~DCACHE_REFERENCED; | 887 | dentry->d_flags &= ~DCACHE_REFERENCED; |
878 | spin_unlock(&dentry->d_lock); | 888 | spin_unlock(&dentry->d_lock); |
879 | 889 | ||
880 | /* | 890 | /* |
881 | * The list move itself will be made by the common LRU code. At | 891 | * The list move itself will be made by the common LRU code. At |
882 | * this point, we've dropped the dentry->d_lock but keep the | 892 | * this point, we've dropped the dentry->d_lock but keep the |
883 | * lru lock. This is safe to do, since every list movement is | 893 | * lru lock. This is safe to do, since every list movement is |
884 | * protected by the lru lock even if both locks are held. | 894 | * protected by the lru lock even if both locks are held. |
885 | * | 895 | * |
886 | * This is guaranteed by the fact that all LRU management | 896 | * This is guaranteed by the fact that all LRU management |
887 | * functions are intermediated by the LRU API calls like | 897 | * functions are intermediated by the LRU API calls like |
888 | * list_lru_add and list_lru_del. List movement in this file | 898 | * list_lru_add and list_lru_del. List movement in this file |
889 | * only ever occur through this functions or through callbacks | 899 | * only ever occur through this functions or through callbacks |
890 | * like this one, that are called from the LRU API. | 900 | * like this one, that are called from the LRU API. |
891 | * | 901 | * |
892 | * The only exceptions to this are functions like | 902 | * The only exceptions to this are functions like |
893 | * shrink_dentry_list, and code that first checks for the | 903 | * shrink_dentry_list, and code that first checks for the |
894 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | 904 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be |
895 | * operating only with stack provided lists after they are | 905 | * operating only with stack provided lists after they are |
896 | * properly isolated from the main list. It is thus, always a | 906 | * properly isolated from the main list. It is thus, always a |
897 | * local access. | 907 | * local access. |
898 | */ | 908 | */ |
899 | return LRU_ROTATE; | 909 | return LRU_ROTATE; |
900 | } | 910 | } |
901 | 911 | ||
902 | d_lru_shrink_move(dentry, freeable); | 912 | d_lru_shrink_move(dentry, freeable); |
903 | spin_unlock(&dentry->d_lock); | 913 | spin_unlock(&dentry->d_lock); |
904 | 914 | ||
905 | return LRU_REMOVED; | 915 | return LRU_REMOVED; |
906 | } | 916 | } |
907 | 917 | ||
908 | /** | 918 | /** |
909 | * prune_dcache_sb - shrink the dcache | 919 | * prune_dcache_sb - shrink the dcache |
910 | * @sb: superblock | 920 | * @sb: superblock |
911 | * @nr_to_scan : number of entries to try to free | 921 | * @nr_to_scan : number of entries to try to free |
912 | * @nid: which node to scan for freeable entities | 922 | * @nid: which node to scan for freeable entities |
913 | * | 923 | * |
914 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is | 924 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
915 | * done when we need more memory an called from the superblock shrinker | 925 | * done when we need more memory an called from the superblock shrinker |
916 | * function. | 926 | * function. |
917 | * | 927 | * |
918 | * This function may fail to free any resources if all the dentries are in | 928 | * This function may fail to free any resources if all the dentries are in |
919 | * use. | 929 | * use. |
920 | */ | 930 | */ |
921 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, | 931 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, |
922 | int nid) | 932 | int nid) |
923 | { | 933 | { |
924 | LIST_HEAD(dispose); | 934 | LIST_HEAD(dispose); |
925 | long freed; | 935 | long freed; |
926 | 936 | ||
927 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, | 937 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, |
928 | &dispose, &nr_to_scan); | 938 | &dispose, &nr_to_scan); |
929 | shrink_dentry_list(&dispose); | 939 | shrink_dentry_list(&dispose); |
930 | return freed; | 940 | return freed; |
931 | } | 941 | } |
932 | 942 | ||
933 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, | 943 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
934 | spinlock_t *lru_lock, void *arg) | 944 | spinlock_t *lru_lock, void *arg) |
935 | { | 945 | { |
936 | struct list_head *freeable = arg; | 946 | struct list_head *freeable = arg; |
937 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | 947 | struct dentry *dentry = container_of(item, struct dentry, d_lru); |
938 | 948 | ||
939 | /* | 949 | /* |
940 | * we are inverting the lru lock/dentry->d_lock here, | 950 | * we are inverting the lru lock/dentry->d_lock here, |
941 | * so use a trylock. If we fail to get the lock, just skip | 951 | * so use a trylock. If we fail to get the lock, just skip |
942 | * it | 952 | * it |
943 | */ | 953 | */ |
944 | if (!spin_trylock(&dentry->d_lock)) | 954 | if (!spin_trylock(&dentry->d_lock)) |
945 | return LRU_SKIP; | 955 | return LRU_SKIP; |
946 | 956 | ||
947 | d_lru_shrink_move(dentry, freeable); | 957 | d_lru_shrink_move(dentry, freeable); |
948 | spin_unlock(&dentry->d_lock); | 958 | spin_unlock(&dentry->d_lock); |
949 | 959 | ||
950 | return LRU_REMOVED; | 960 | return LRU_REMOVED; |
951 | } | 961 | } |
952 | 962 | ||
953 | 963 | ||
954 | /** | 964 | /** |
955 | * shrink_dcache_sb - shrink dcache for a superblock | 965 | * shrink_dcache_sb - shrink dcache for a superblock |
956 | * @sb: superblock | 966 | * @sb: superblock |
957 | * | 967 | * |
958 | * Shrink the dcache for the specified super block. This is used to free | 968 | * Shrink the dcache for the specified super block. This is used to free |
959 | * the dcache before unmounting a file system. | 969 | * the dcache before unmounting a file system. |
960 | */ | 970 | */ |
961 | void shrink_dcache_sb(struct super_block *sb) | 971 | void shrink_dcache_sb(struct super_block *sb) |
962 | { | 972 | { |
963 | long freed; | 973 | long freed; |
964 | 974 | ||
965 | do { | 975 | do { |
966 | LIST_HEAD(dispose); | 976 | LIST_HEAD(dispose); |
967 | 977 | ||
968 | freed = list_lru_walk(&sb->s_dentry_lru, | 978 | freed = list_lru_walk(&sb->s_dentry_lru, |
969 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | 979 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); |
970 | 980 | ||
971 | this_cpu_sub(nr_dentry_unused, freed); | 981 | this_cpu_sub(nr_dentry_unused, freed); |
972 | shrink_dentry_list(&dispose); | 982 | shrink_dentry_list(&dispose); |
973 | } while (freed > 0); | 983 | } while (freed > 0); |
974 | } | 984 | } |
975 | EXPORT_SYMBOL(shrink_dcache_sb); | 985 | EXPORT_SYMBOL(shrink_dcache_sb); |
976 | 986 | ||
977 | /** | 987 | /** |
978 | * enum d_walk_ret - action to talke during tree walk | 988 | * enum d_walk_ret - action to talke during tree walk |
979 | * @D_WALK_CONTINUE: contrinue walk | 989 | * @D_WALK_CONTINUE: contrinue walk |
980 | * @D_WALK_QUIT: quit walk | 990 | * @D_WALK_QUIT: quit walk |
981 | * @D_WALK_NORETRY: quit when retry is needed | 991 | * @D_WALK_NORETRY: quit when retry is needed |
982 | * @D_WALK_SKIP: skip this dentry and its children | 992 | * @D_WALK_SKIP: skip this dentry and its children |
983 | */ | 993 | */ |
984 | enum d_walk_ret { | 994 | enum d_walk_ret { |
985 | D_WALK_CONTINUE, | 995 | D_WALK_CONTINUE, |
986 | D_WALK_QUIT, | 996 | D_WALK_QUIT, |
987 | D_WALK_NORETRY, | 997 | D_WALK_NORETRY, |
988 | D_WALK_SKIP, | 998 | D_WALK_SKIP, |
989 | }; | 999 | }; |
990 | 1000 | ||
991 | /** | 1001 | /** |
992 | * d_walk - walk the dentry tree | 1002 | * d_walk - walk the dentry tree |
993 | * @parent: start of walk | 1003 | * @parent: start of walk |
994 | * @data: data passed to @enter() and @finish() | 1004 | * @data: data passed to @enter() and @finish() |
995 | * @enter: callback when first entering the dentry | 1005 | * @enter: callback when first entering the dentry |
996 | * @finish: callback when successfully finished the walk | 1006 | * @finish: callback when successfully finished the walk |
997 | * | 1007 | * |
998 | * The @enter() and @finish() callbacks are called with d_lock held. | 1008 | * The @enter() and @finish() callbacks are called with d_lock held. |
999 | */ | 1009 | */ |
1000 | static void d_walk(struct dentry *parent, void *data, | 1010 | static void d_walk(struct dentry *parent, void *data, |
1001 | enum d_walk_ret (*enter)(void *, struct dentry *), | 1011 | enum d_walk_ret (*enter)(void *, struct dentry *), |
1002 | void (*finish)(void *)) | 1012 | void (*finish)(void *)) |
1003 | { | 1013 | { |
1004 | struct dentry *this_parent; | 1014 | struct dentry *this_parent; |
1005 | struct list_head *next; | 1015 | struct list_head *next; |
1006 | unsigned seq = 0; | 1016 | unsigned seq = 0; |
1007 | enum d_walk_ret ret; | 1017 | enum d_walk_ret ret; |
1008 | bool retry = true; | 1018 | bool retry = true; |
1009 | 1019 | ||
1010 | again: | 1020 | again: |
1011 | read_seqbegin_or_lock(&rename_lock, &seq); | 1021 | read_seqbegin_or_lock(&rename_lock, &seq); |
1012 | this_parent = parent; | 1022 | this_parent = parent; |
1013 | spin_lock(&this_parent->d_lock); | 1023 | spin_lock(&this_parent->d_lock); |
1014 | 1024 | ||
1015 | ret = enter(data, this_parent); | 1025 | ret = enter(data, this_parent); |
1016 | switch (ret) { | 1026 | switch (ret) { |
1017 | case D_WALK_CONTINUE: | 1027 | case D_WALK_CONTINUE: |
1018 | break; | 1028 | break; |
1019 | case D_WALK_QUIT: | 1029 | case D_WALK_QUIT: |
1020 | case D_WALK_SKIP: | 1030 | case D_WALK_SKIP: |
1021 | goto out_unlock; | 1031 | goto out_unlock; |
1022 | case D_WALK_NORETRY: | 1032 | case D_WALK_NORETRY: |
1023 | retry = false; | 1033 | retry = false; |
1024 | break; | 1034 | break; |
1025 | } | 1035 | } |
1026 | repeat: | 1036 | repeat: |
1027 | next = this_parent->d_subdirs.next; | 1037 | next = this_parent->d_subdirs.next; |
1028 | resume: | 1038 | resume: |
1029 | while (next != &this_parent->d_subdirs) { | 1039 | while (next != &this_parent->d_subdirs) { |
1030 | struct list_head *tmp = next; | 1040 | struct list_head *tmp = next; |
1031 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); | 1041 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1032 | next = tmp->next; | 1042 | next = tmp->next; |
1033 | 1043 | ||
1034 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | 1044 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1035 | 1045 | ||
1036 | ret = enter(data, dentry); | 1046 | ret = enter(data, dentry); |
1037 | switch (ret) { | 1047 | switch (ret) { |
1038 | case D_WALK_CONTINUE: | 1048 | case D_WALK_CONTINUE: |
1039 | break; | 1049 | break; |
1040 | case D_WALK_QUIT: | 1050 | case D_WALK_QUIT: |
1041 | spin_unlock(&dentry->d_lock); | 1051 | spin_unlock(&dentry->d_lock); |
1042 | goto out_unlock; | 1052 | goto out_unlock; |
1043 | case D_WALK_NORETRY: | 1053 | case D_WALK_NORETRY: |
1044 | retry = false; | 1054 | retry = false; |
1045 | break; | 1055 | break; |
1046 | case D_WALK_SKIP: | 1056 | case D_WALK_SKIP: |
1047 | spin_unlock(&dentry->d_lock); | 1057 | spin_unlock(&dentry->d_lock); |
1048 | continue; | 1058 | continue; |
1049 | } | 1059 | } |
1050 | 1060 | ||
1051 | if (!list_empty(&dentry->d_subdirs)) { | 1061 | if (!list_empty(&dentry->d_subdirs)) { |
1052 | spin_unlock(&this_parent->d_lock); | 1062 | spin_unlock(&this_parent->d_lock); |
1053 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | 1063 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); |
1054 | this_parent = dentry; | 1064 | this_parent = dentry; |
1055 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); | 1065 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1056 | goto repeat; | 1066 | goto repeat; |
1057 | } | 1067 | } |
1058 | spin_unlock(&dentry->d_lock); | 1068 | spin_unlock(&dentry->d_lock); |
1059 | } | 1069 | } |
1060 | /* | 1070 | /* |
1061 | * All done at this level ... ascend and resume the search. | 1071 | * All done at this level ... ascend and resume the search. |
1062 | */ | 1072 | */ |
1063 | if (this_parent != parent) { | 1073 | if (this_parent != parent) { |
1064 | struct dentry *child = this_parent; | 1074 | struct dentry *child = this_parent; |
1065 | this_parent = child->d_parent; | 1075 | this_parent = child->d_parent; |
1066 | 1076 | ||
1067 | rcu_read_lock(); | 1077 | rcu_read_lock(); |
1068 | spin_unlock(&child->d_lock); | 1078 | spin_unlock(&child->d_lock); |
1069 | spin_lock(&this_parent->d_lock); | 1079 | spin_lock(&this_parent->d_lock); |
1070 | 1080 | ||
1071 | /* | 1081 | /* |
1072 | * might go back up the wrong parent if we have had a rename | 1082 | * might go back up the wrong parent if we have had a rename |
1073 | * or deletion | 1083 | * or deletion |
1074 | */ | 1084 | */ |
1075 | if (this_parent != child->d_parent || | 1085 | if (this_parent != child->d_parent || |
1076 | (child->d_flags & DCACHE_DENTRY_KILLED) || | 1086 | (child->d_flags & DCACHE_DENTRY_KILLED) || |
1077 | need_seqretry(&rename_lock, seq)) { | 1087 | need_seqretry(&rename_lock, seq)) { |
1078 | spin_unlock(&this_parent->d_lock); | 1088 | spin_unlock(&this_parent->d_lock); |
1079 | rcu_read_unlock(); | 1089 | rcu_read_unlock(); |
1080 | goto rename_retry; | 1090 | goto rename_retry; |
1081 | } | 1091 | } |
1082 | rcu_read_unlock(); | 1092 | rcu_read_unlock(); |
1083 | next = child->d_u.d_child.next; | 1093 | next = child->d_u.d_child.next; |
1084 | goto resume; | 1094 | goto resume; |
1085 | } | 1095 | } |
1086 | if (need_seqretry(&rename_lock, seq)) { | 1096 | if (need_seqretry(&rename_lock, seq)) { |
1087 | spin_unlock(&this_parent->d_lock); | 1097 | spin_unlock(&this_parent->d_lock); |
1088 | goto rename_retry; | 1098 | goto rename_retry; |
1089 | } | 1099 | } |
1090 | if (finish) | 1100 | if (finish) |
1091 | finish(data); | 1101 | finish(data); |
1092 | 1102 | ||
1093 | out_unlock: | 1103 | out_unlock: |
1094 | spin_unlock(&this_parent->d_lock); | 1104 | spin_unlock(&this_parent->d_lock); |
1095 | done_seqretry(&rename_lock, seq); | 1105 | done_seqretry(&rename_lock, seq); |
1096 | return; | 1106 | return; |
1097 | 1107 | ||
1098 | rename_retry: | 1108 | rename_retry: |
1099 | if (!retry) | 1109 | if (!retry) |
1100 | return; | 1110 | return; |
1101 | seq = 1; | 1111 | seq = 1; |
1102 | goto again; | 1112 | goto again; |
1103 | } | 1113 | } |
1104 | 1114 | ||
1105 | /* | 1115 | /* |
1106 | * Search for at least 1 mount point in the dentry's subdirs. | 1116 | * Search for at least 1 mount point in the dentry's subdirs. |
1107 | * We descend to the next level whenever the d_subdirs | 1117 | * We descend to the next level whenever the d_subdirs |
1108 | * list is non-empty and continue searching. | 1118 | * list is non-empty and continue searching. |
1109 | */ | 1119 | */ |
1110 | 1120 | ||
1111 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | 1121 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1112 | { | 1122 | { |
1113 | int *ret = data; | 1123 | int *ret = data; |
1114 | if (d_mountpoint(dentry)) { | 1124 | if (d_mountpoint(dentry)) { |
1115 | *ret = 1; | 1125 | *ret = 1; |
1116 | return D_WALK_QUIT; | 1126 | return D_WALK_QUIT; |
1117 | } | 1127 | } |
1118 | return D_WALK_CONTINUE; | 1128 | return D_WALK_CONTINUE; |
1119 | } | 1129 | } |
1120 | 1130 | ||
1121 | /** | 1131 | /** |
1122 | * have_submounts - check for mounts over a dentry | 1132 | * have_submounts - check for mounts over a dentry |
1123 | * @parent: dentry to check. | 1133 | * @parent: dentry to check. |
1124 | * | 1134 | * |
1125 | * Return true if the parent or its subdirectories contain | 1135 | * Return true if the parent or its subdirectories contain |
1126 | * a mount point | 1136 | * a mount point |
1127 | */ | 1137 | */ |
1128 | int have_submounts(struct dentry *parent) | 1138 | int have_submounts(struct dentry *parent) |
1129 | { | 1139 | { |
1130 | int ret = 0; | 1140 | int ret = 0; |
1131 | 1141 | ||
1132 | d_walk(parent, &ret, check_mount, NULL); | 1142 | d_walk(parent, &ret, check_mount, NULL); |
1133 | 1143 | ||
1134 | return ret; | 1144 | return ret; |
1135 | } | 1145 | } |
1136 | EXPORT_SYMBOL(have_submounts); | 1146 | EXPORT_SYMBOL(have_submounts); |
1137 | 1147 | ||
1138 | /* | 1148 | /* |
1139 | * Called by mount code to set a mountpoint and check if the mountpoint is | 1149 | * Called by mount code to set a mountpoint and check if the mountpoint is |
1140 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | 1150 | * reachable (e.g. NFS can unhash a directory dentry and then the complete |
1141 | * subtree can become unreachable). | 1151 | * subtree can become unreachable). |
1142 | * | 1152 | * |
1143 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | 1153 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For |
1144 | * this reason take rename_lock and d_lock on dentry and ancestors. | 1154 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1145 | */ | 1155 | */ |
1146 | int d_set_mounted(struct dentry *dentry) | 1156 | int d_set_mounted(struct dentry *dentry) |
1147 | { | 1157 | { |
1148 | struct dentry *p; | 1158 | struct dentry *p; |
1149 | int ret = -ENOENT; | 1159 | int ret = -ENOENT; |
1150 | write_seqlock(&rename_lock); | 1160 | write_seqlock(&rename_lock); |
1151 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | 1161 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { |
1152 | /* Need exclusion wrt. check_submounts_and_drop() */ | 1162 | /* Need exclusion wrt. check_submounts_and_drop() */ |
1153 | spin_lock(&p->d_lock); | 1163 | spin_lock(&p->d_lock); |
1154 | if (unlikely(d_unhashed(p))) { | 1164 | if (unlikely(d_unhashed(p))) { |
1155 | spin_unlock(&p->d_lock); | 1165 | spin_unlock(&p->d_lock); |
1156 | goto out; | 1166 | goto out; |
1157 | } | 1167 | } |
1158 | spin_unlock(&p->d_lock); | 1168 | spin_unlock(&p->d_lock); |
1159 | } | 1169 | } |
1160 | spin_lock(&dentry->d_lock); | 1170 | spin_lock(&dentry->d_lock); |
1161 | if (!d_unlinked(dentry)) { | 1171 | if (!d_unlinked(dentry)) { |
1162 | dentry->d_flags |= DCACHE_MOUNTED; | 1172 | dentry->d_flags |= DCACHE_MOUNTED; |
1163 | ret = 0; | 1173 | ret = 0; |
1164 | } | 1174 | } |
1165 | spin_unlock(&dentry->d_lock); | 1175 | spin_unlock(&dentry->d_lock); |
1166 | out: | 1176 | out: |
1167 | write_sequnlock(&rename_lock); | 1177 | write_sequnlock(&rename_lock); |
1168 | return ret; | 1178 | return ret; |
1169 | } | 1179 | } |
1170 | 1180 | ||
1171 | /* | 1181 | /* |
1172 | * Search the dentry child list of the specified parent, | 1182 | * Search the dentry child list of the specified parent, |
1173 | * and move any unused dentries to the end of the unused | 1183 | * and move any unused dentries to the end of the unused |
1174 | * list for prune_dcache(). We descend to the next level | 1184 | * list for prune_dcache(). We descend to the next level |
1175 | * whenever the d_subdirs list is non-empty and continue | 1185 | * whenever the d_subdirs list is non-empty and continue |
1176 | * searching. | 1186 | * searching. |
1177 | * | 1187 | * |
1178 | * It returns zero iff there are no unused children, | 1188 | * It returns zero iff there are no unused children, |
1179 | * otherwise it returns the number of children moved to | 1189 | * otherwise it returns the number of children moved to |
1180 | * the end of the unused list. This may not be the total | 1190 | * the end of the unused list. This may not be the total |
1181 | * number of unused children, because select_parent can | 1191 | * number of unused children, because select_parent can |
1182 | * drop the lock and return early due to latency | 1192 | * drop the lock and return early due to latency |
1183 | * constraints. | 1193 | * constraints. |
1184 | */ | 1194 | */ |
1185 | 1195 | ||
1186 | struct select_data { | 1196 | struct select_data { |
1187 | struct dentry *start; | 1197 | struct dentry *start; |
1188 | struct list_head dispose; | 1198 | struct list_head dispose; |
1189 | int found; | 1199 | int found; |
1190 | }; | 1200 | }; |
1191 | 1201 | ||
1192 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) | 1202 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1193 | { | 1203 | { |
1194 | struct select_data *data = _data; | 1204 | struct select_data *data = _data; |
1195 | enum d_walk_ret ret = D_WALK_CONTINUE; | 1205 | enum d_walk_ret ret = D_WALK_CONTINUE; |
1196 | 1206 | ||
1197 | if (data->start == dentry) | 1207 | if (data->start == dentry) |
1198 | goto out; | 1208 | goto out; |
1199 | 1209 | ||
1200 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | 1210 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
1201 | data->found++; | 1211 | data->found++; |
1202 | } else { | 1212 | } else { |
1203 | if (dentry->d_flags & DCACHE_LRU_LIST) | 1213 | if (dentry->d_flags & DCACHE_LRU_LIST) |
1204 | d_lru_del(dentry); | 1214 | d_lru_del(dentry); |
1205 | if (!dentry->d_lockref.count) { | 1215 | if (!dentry->d_lockref.count) { |
1206 | d_shrink_add(dentry, &data->dispose); | 1216 | d_shrink_add(dentry, &data->dispose); |
1207 | data->found++; | 1217 | data->found++; |
1208 | } | 1218 | } |
1209 | } | 1219 | } |
1210 | /* | 1220 | /* |
1211 | * We can return to the caller if we have found some (this | 1221 | * We can return to the caller if we have found some (this |
1212 | * ensures forward progress). We'll be coming back to find | 1222 | * ensures forward progress). We'll be coming back to find |
1213 | * the rest. | 1223 | * the rest. |
1214 | */ | 1224 | */ |
1215 | if (!list_empty(&data->dispose)) | 1225 | if (!list_empty(&data->dispose)) |
1216 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | 1226 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; |
1217 | out: | 1227 | out: |
1218 | return ret; | 1228 | return ret; |
1219 | } | 1229 | } |
1220 | 1230 | ||
1221 | /** | 1231 | /** |
1222 | * shrink_dcache_parent - prune dcache | 1232 | * shrink_dcache_parent - prune dcache |
1223 | * @parent: parent of entries to prune | 1233 | * @parent: parent of entries to prune |
1224 | * | 1234 | * |
1225 | * Prune the dcache to remove unused children of the parent dentry. | 1235 | * Prune the dcache to remove unused children of the parent dentry. |
1226 | */ | 1236 | */ |
1227 | void shrink_dcache_parent(struct dentry *parent) | 1237 | void shrink_dcache_parent(struct dentry *parent) |
1228 | { | 1238 | { |
1229 | for (;;) { | 1239 | for (;;) { |
1230 | struct select_data data; | 1240 | struct select_data data; |
1231 | 1241 | ||
1232 | INIT_LIST_HEAD(&data.dispose); | 1242 | INIT_LIST_HEAD(&data.dispose); |
1233 | data.start = parent; | 1243 | data.start = parent; |
1234 | data.found = 0; | 1244 | data.found = 0; |
1235 | 1245 | ||
1236 | d_walk(parent, &data, select_collect, NULL); | 1246 | d_walk(parent, &data, select_collect, NULL); |
1237 | if (!data.found) | 1247 | if (!data.found) |
1238 | break; | 1248 | break; |
1239 | 1249 | ||
1240 | shrink_dentry_list(&data.dispose); | 1250 | shrink_dentry_list(&data.dispose); |
1241 | cond_resched(); | 1251 | cond_resched(); |
1242 | } | 1252 | } |
1243 | } | 1253 | } |
1244 | EXPORT_SYMBOL(shrink_dcache_parent); | 1254 | EXPORT_SYMBOL(shrink_dcache_parent); |
1245 | 1255 | ||
1246 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) | 1256 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
1247 | { | 1257 | { |
1248 | /* it has busy descendents; complain about those instead */ | 1258 | /* it has busy descendents; complain about those instead */ |
1249 | if (!list_empty(&dentry->d_subdirs)) | 1259 | if (!list_empty(&dentry->d_subdirs)) |
1250 | return D_WALK_CONTINUE; | 1260 | return D_WALK_CONTINUE; |
1251 | 1261 | ||
1252 | /* root with refcount 1 is fine */ | 1262 | /* root with refcount 1 is fine */ |
1253 | if (dentry == _data && dentry->d_lockref.count == 1) | 1263 | if (dentry == _data && dentry->d_lockref.count == 1) |
1254 | return D_WALK_CONTINUE; | 1264 | return D_WALK_CONTINUE; |
1255 | 1265 | ||
1256 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | 1266 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " |
1257 | " still in use (%d) [unmount of %s %s]\n", | 1267 | " still in use (%d) [unmount of %s %s]\n", |
1258 | dentry, | 1268 | dentry, |
1259 | dentry->d_inode ? | 1269 | dentry->d_inode ? |
1260 | dentry->d_inode->i_ino : 0UL, | 1270 | dentry->d_inode->i_ino : 0UL, |
1261 | dentry, | 1271 | dentry, |
1262 | dentry->d_lockref.count, | 1272 | dentry->d_lockref.count, |
1263 | dentry->d_sb->s_type->name, | 1273 | dentry->d_sb->s_type->name, |
1264 | dentry->d_sb->s_id); | 1274 | dentry->d_sb->s_id); |
1265 | WARN_ON(1); | 1275 | WARN_ON(1); |
1266 | return D_WALK_CONTINUE; | 1276 | return D_WALK_CONTINUE; |
1267 | } | 1277 | } |
1268 | 1278 | ||
1269 | static void do_one_tree(struct dentry *dentry) | 1279 | static void do_one_tree(struct dentry *dentry) |
1270 | { | 1280 | { |
1271 | shrink_dcache_parent(dentry); | 1281 | shrink_dcache_parent(dentry); |
1272 | d_walk(dentry, dentry, umount_check, NULL); | 1282 | d_walk(dentry, dentry, umount_check, NULL); |
1273 | d_drop(dentry); | 1283 | d_drop(dentry); |
1274 | dput(dentry); | 1284 | dput(dentry); |
1275 | } | 1285 | } |
1276 | 1286 | ||
1277 | /* | 1287 | /* |
1278 | * destroy the dentries attached to a superblock on unmounting | 1288 | * destroy the dentries attached to a superblock on unmounting |
1279 | */ | 1289 | */ |
1280 | void shrink_dcache_for_umount(struct super_block *sb) | 1290 | void shrink_dcache_for_umount(struct super_block *sb) |
1281 | { | 1291 | { |
1282 | struct dentry *dentry; | 1292 | struct dentry *dentry; |
1283 | 1293 | ||
1284 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); | 1294 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
1285 | 1295 | ||
1286 | dentry = sb->s_root; | 1296 | dentry = sb->s_root; |
1287 | sb->s_root = NULL; | 1297 | sb->s_root = NULL; |
1288 | do_one_tree(dentry); | 1298 | do_one_tree(dentry); |
1289 | 1299 | ||
1290 | while (!hlist_bl_empty(&sb->s_anon)) { | 1300 | while (!hlist_bl_empty(&sb->s_anon)) { |
1291 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); | 1301 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1292 | do_one_tree(dentry); | 1302 | do_one_tree(dentry); |
1293 | } | 1303 | } |
1294 | } | 1304 | } |
1295 | 1305 | ||
1296 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) | 1306 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1297 | { | 1307 | { |
1298 | struct select_data *data = _data; | 1308 | struct select_data *data = _data; |
1299 | 1309 | ||
1300 | if (d_mountpoint(dentry)) { | 1310 | if (d_mountpoint(dentry)) { |
1301 | data->found = -EBUSY; | 1311 | data->found = -EBUSY; |
1302 | return D_WALK_QUIT; | 1312 | return D_WALK_QUIT; |
1303 | } | 1313 | } |
1304 | 1314 | ||
1305 | return select_collect(_data, dentry); | 1315 | return select_collect(_data, dentry); |
1306 | } | 1316 | } |
1307 | 1317 | ||
1308 | static void check_and_drop(void *_data) | 1318 | static void check_and_drop(void *_data) |
1309 | { | 1319 | { |
1310 | struct select_data *data = _data; | 1320 | struct select_data *data = _data; |
1311 | 1321 | ||
1312 | if (d_mountpoint(data->start)) | 1322 | if (d_mountpoint(data->start)) |
1313 | data->found = -EBUSY; | 1323 | data->found = -EBUSY; |
1314 | if (!data->found) | 1324 | if (!data->found) |
1315 | __d_drop(data->start); | 1325 | __d_drop(data->start); |
1316 | } | 1326 | } |
1317 | 1327 | ||
1318 | /** | 1328 | /** |
1319 | * check_submounts_and_drop - prune dcache, check for submounts and drop | 1329 | * check_submounts_and_drop - prune dcache, check for submounts and drop |
1320 | * | 1330 | * |
1321 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | 1331 | * All done as a single atomic operation relative to has_unlinked_ancestor(). |
1322 | * Returns 0 if successfully unhashed @parent. If there were submounts then | 1332 | * Returns 0 if successfully unhashed @parent. If there were submounts then |
1323 | * return -EBUSY. | 1333 | * return -EBUSY. |
1324 | * | 1334 | * |
1325 | * @dentry: dentry to prune and drop | 1335 | * @dentry: dentry to prune and drop |
1326 | */ | 1336 | */ |
1327 | int check_submounts_and_drop(struct dentry *dentry) | 1337 | int check_submounts_and_drop(struct dentry *dentry) |
1328 | { | 1338 | { |
1329 | int ret = 0; | 1339 | int ret = 0; |
1330 | 1340 | ||
1331 | /* Negative dentries can be dropped without further checks */ | 1341 | /* Negative dentries can be dropped without further checks */ |
1332 | if (!dentry->d_inode) { | 1342 | if (!dentry->d_inode) { |
1333 | d_drop(dentry); | 1343 | d_drop(dentry); |
1334 | goto out; | 1344 | goto out; |
1335 | } | 1345 | } |
1336 | 1346 | ||
1337 | for (;;) { | 1347 | for (;;) { |
1338 | struct select_data data; | 1348 | struct select_data data; |
1339 | 1349 | ||
1340 | INIT_LIST_HEAD(&data.dispose); | 1350 | INIT_LIST_HEAD(&data.dispose); |
1341 | data.start = dentry; | 1351 | data.start = dentry; |
1342 | data.found = 0; | 1352 | data.found = 0; |
1343 | 1353 | ||
1344 | d_walk(dentry, &data, check_and_collect, check_and_drop); | 1354 | d_walk(dentry, &data, check_and_collect, check_and_drop); |
1345 | ret = data.found; | 1355 | ret = data.found; |
1346 | 1356 | ||
1347 | if (!list_empty(&data.dispose)) | 1357 | if (!list_empty(&data.dispose)) |
1348 | shrink_dentry_list(&data.dispose); | 1358 | shrink_dentry_list(&data.dispose); |
1349 | 1359 | ||
1350 | if (ret <= 0) | 1360 | if (ret <= 0) |
1351 | break; | 1361 | break; |
1352 | 1362 | ||
1353 | cond_resched(); | 1363 | cond_resched(); |
1354 | } | 1364 | } |
1355 | 1365 | ||
1356 | out: | 1366 | out: |
1357 | return ret; | 1367 | return ret; |
1358 | } | 1368 | } |
1359 | EXPORT_SYMBOL(check_submounts_and_drop); | 1369 | EXPORT_SYMBOL(check_submounts_and_drop); |
1360 | 1370 | ||
1361 | /** | 1371 | /** |
1362 | * __d_alloc - allocate a dcache entry | 1372 | * __d_alloc - allocate a dcache entry |
1363 | * @sb: filesystem it will belong to | 1373 | * @sb: filesystem it will belong to |
1364 | * @name: qstr of the name | 1374 | * @name: qstr of the name |
1365 | * | 1375 | * |
1366 | * Allocates a dentry. It returns %NULL if there is insufficient memory | 1376 | * Allocates a dentry. It returns %NULL if there is insufficient memory |
1367 | * available. On a success the dentry is returned. The name passed in is | 1377 | * available. On a success the dentry is returned. The name passed in is |
1368 | * copied and the copy passed in may be reused after this call. | 1378 | * copied and the copy passed in may be reused after this call. |
1369 | */ | 1379 | */ |
1370 | 1380 | ||
1371 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) | 1381 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1372 | { | 1382 | { |
1373 | struct dentry *dentry; | 1383 | struct dentry *dentry; |
1374 | char *dname; | 1384 | char *dname; |
1375 | 1385 | ||
1376 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); | 1386 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1377 | if (!dentry) | 1387 | if (!dentry) |
1378 | return NULL; | 1388 | return NULL; |
1379 | 1389 | ||
1380 | /* | 1390 | /* |
1381 | * We guarantee that the inline name is always NUL-terminated. | 1391 | * We guarantee that the inline name is always NUL-terminated. |
1382 | * This way the memcpy() done by the name switching in rename | 1392 | * This way the memcpy() done by the name switching in rename |
1383 | * will still always have a NUL at the end, even if we might | 1393 | * will still always have a NUL at the end, even if we might |
1384 | * be overwriting an internal NUL character | 1394 | * be overwriting an internal NUL character |
1385 | */ | 1395 | */ |
1386 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | 1396 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; |
1387 | if (name->len > DNAME_INLINE_LEN-1) { | 1397 | if (name->len > DNAME_INLINE_LEN-1) { |
1388 | dname = kmalloc(name->len + 1, GFP_KERNEL); | 1398 | dname = kmalloc(name->len + 1, GFP_KERNEL); |
1389 | if (!dname) { | 1399 | if (!dname) { |
1390 | kmem_cache_free(dentry_cache, dentry); | 1400 | kmem_cache_free(dentry_cache, dentry); |
1391 | return NULL; | 1401 | return NULL; |
1392 | } | 1402 | } |
1393 | } else { | 1403 | } else { |
1394 | dname = dentry->d_iname; | 1404 | dname = dentry->d_iname; |
1395 | } | 1405 | } |
1396 | 1406 | ||
1397 | dentry->d_name.len = name->len; | 1407 | dentry->d_name.len = name->len; |
1398 | dentry->d_name.hash = name->hash; | 1408 | dentry->d_name.hash = name->hash; |
1399 | memcpy(dname, name->name, name->len); | 1409 | memcpy(dname, name->name, name->len); |
1400 | dname[name->len] = 0; | 1410 | dname[name->len] = 0; |
1401 | 1411 | ||
1402 | /* Make sure we always see the terminating NUL character */ | 1412 | /* Make sure we always see the terminating NUL character */ |
1403 | smp_wmb(); | 1413 | smp_wmb(); |
1404 | dentry->d_name.name = dname; | 1414 | dentry->d_name.name = dname; |
1405 | 1415 | ||
1406 | dentry->d_lockref.count = 1; | 1416 | dentry->d_lockref.count = 1; |
1407 | dentry->d_flags = 0; | 1417 | dentry->d_flags = 0; |
1408 | spin_lock_init(&dentry->d_lock); | 1418 | spin_lock_init(&dentry->d_lock); |
1409 | seqcount_init(&dentry->d_seq); | 1419 | seqcount_init(&dentry->d_seq); |
1410 | dentry->d_inode = NULL; | 1420 | dentry->d_inode = NULL; |
1411 | dentry->d_parent = dentry; | 1421 | dentry->d_parent = dentry; |
1412 | dentry->d_sb = sb; | 1422 | dentry->d_sb = sb; |
1413 | dentry->d_op = NULL; | 1423 | dentry->d_op = NULL; |
1414 | dentry->d_fsdata = NULL; | 1424 | dentry->d_fsdata = NULL; |
1415 | INIT_HLIST_BL_NODE(&dentry->d_hash); | 1425 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1416 | INIT_LIST_HEAD(&dentry->d_lru); | 1426 | INIT_LIST_HEAD(&dentry->d_lru); |
1417 | INIT_LIST_HEAD(&dentry->d_subdirs); | 1427 | INIT_LIST_HEAD(&dentry->d_subdirs); |
1418 | INIT_HLIST_NODE(&dentry->d_alias); | 1428 | INIT_HLIST_NODE(&dentry->d_alias); |
1419 | INIT_LIST_HEAD(&dentry->d_u.d_child); | 1429 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1420 | d_set_d_op(dentry, dentry->d_sb->s_d_op); | 1430 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1421 | 1431 | ||
1422 | this_cpu_inc(nr_dentry); | 1432 | this_cpu_inc(nr_dentry); |
1423 | 1433 | ||
1424 | return dentry; | 1434 | return dentry; |
1425 | } | 1435 | } |
1426 | 1436 | ||
1427 | /** | 1437 | /** |
1428 | * d_alloc - allocate a dcache entry | 1438 | * d_alloc - allocate a dcache entry |
1429 | * @parent: parent of entry to allocate | 1439 | * @parent: parent of entry to allocate |
1430 | * @name: qstr of the name | 1440 | * @name: qstr of the name |
1431 | * | 1441 | * |
1432 | * Allocates a dentry. It returns %NULL if there is insufficient memory | 1442 | * Allocates a dentry. It returns %NULL if there is insufficient memory |
1433 | * available. On a success the dentry is returned. The name passed in is | 1443 | * available. On a success the dentry is returned. The name passed in is |
1434 | * copied and the copy passed in may be reused after this call. | 1444 | * copied and the copy passed in may be reused after this call. |
1435 | */ | 1445 | */ |
1436 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | 1446 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) |
1437 | { | 1447 | { |
1438 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | 1448 | struct dentry *dentry = __d_alloc(parent->d_sb, name); |
1439 | if (!dentry) | 1449 | if (!dentry) |
1440 | return NULL; | 1450 | return NULL; |
1441 | 1451 | ||
1442 | spin_lock(&parent->d_lock); | 1452 | spin_lock(&parent->d_lock); |
1443 | /* | 1453 | /* |
1444 | * don't need child lock because it is not subject | 1454 | * don't need child lock because it is not subject |
1445 | * to concurrency here | 1455 | * to concurrency here |
1446 | */ | 1456 | */ |
1447 | __dget_dlock(parent); | 1457 | __dget_dlock(parent); |
1448 | dentry->d_parent = parent; | 1458 | dentry->d_parent = parent; |
1449 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | 1459 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1450 | spin_unlock(&parent->d_lock); | 1460 | spin_unlock(&parent->d_lock); |
1451 | 1461 | ||
1452 | return dentry; | 1462 | return dentry; |
1453 | } | 1463 | } |
1454 | EXPORT_SYMBOL(d_alloc); | 1464 | EXPORT_SYMBOL(d_alloc); |
1455 | 1465 | ||
1456 | /** | 1466 | /** |
1457 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | 1467 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) |
1458 | * @sb: the superblock | 1468 | * @sb: the superblock |
1459 | * @name: qstr of the name | 1469 | * @name: qstr of the name |
1460 | * | 1470 | * |
1461 | * For a filesystem that just pins its dentries in memory and never | 1471 | * For a filesystem that just pins its dentries in memory and never |
1462 | * performs lookups at all, return an unhashed IS_ROOT dentry. | 1472 | * performs lookups at all, return an unhashed IS_ROOT dentry. |
1463 | */ | 1473 | */ |
1464 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) | 1474 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1465 | { | 1475 | { |
1466 | return __d_alloc(sb, name); | 1476 | return __d_alloc(sb, name); |
1467 | } | 1477 | } |
1468 | EXPORT_SYMBOL(d_alloc_pseudo); | 1478 | EXPORT_SYMBOL(d_alloc_pseudo); |
1469 | 1479 | ||
1470 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | 1480 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1471 | { | 1481 | { |
1472 | struct qstr q; | 1482 | struct qstr q; |
1473 | 1483 | ||
1474 | q.name = name; | 1484 | q.name = name; |
1475 | q.len = strlen(name); | 1485 | q.len = strlen(name); |
1476 | q.hash = full_name_hash(q.name, q.len); | 1486 | q.hash = full_name_hash(q.name, q.len); |
1477 | return d_alloc(parent, &q); | 1487 | return d_alloc(parent, &q); |
1478 | } | 1488 | } |
1479 | EXPORT_SYMBOL(d_alloc_name); | 1489 | EXPORT_SYMBOL(d_alloc_name); |
1480 | 1490 | ||
1481 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) | 1491 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1482 | { | 1492 | { |
1483 | WARN_ON_ONCE(dentry->d_op); | 1493 | WARN_ON_ONCE(dentry->d_op); |
1484 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | 1494 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | |
1485 | DCACHE_OP_COMPARE | | 1495 | DCACHE_OP_COMPARE | |
1486 | DCACHE_OP_REVALIDATE | | 1496 | DCACHE_OP_REVALIDATE | |
1487 | DCACHE_OP_WEAK_REVALIDATE | | 1497 | DCACHE_OP_WEAK_REVALIDATE | |
1488 | DCACHE_OP_DELETE )); | 1498 | DCACHE_OP_DELETE )); |
1489 | dentry->d_op = op; | 1499 | dentry->d_op = op; |
1490 | if (!op) | 1500 | if (!op) |
1491 | return; | 1501 | return; |
1492 | if (op->d_hash) | 1502 | if (op->d_hash) |
1493 | dentry->d_flags |= DCACHE_OP_HASH; | 1503 | dentry->d_flags |= DCACHE_OP_HASH; |
1494 | if (op->d_compare) | 1504 | if (op->d_compare) |
1495 | dentry->d_flags |= DCACHE_OP_COMPARE; | 1505 | dentry->d_flags |= DCACHE_OP_COMPARE; |
1496 | if (op->d_revalidate) | 1506 | if (op->d_revalidate) |
1497 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | 1507 | dentry->d_flags |= DCACHE_OP_REVALIDATE; |
1498 | if (op->d_weak_revalidate) | 1508 | if (op->d_weak_revalidate) |
1499 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | 1509 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; |
1500 | if (op->d_delete) | 1510 | if (op->d_delete) |
1501 | dentry->d_flags |= DCACHE_OP_DELETE; | 1511 | dentry->d_flags |= DCACHE_OP_DELETE; |
1502 | if (op->d_prune) | 1512 | if (op->d_prune) |
1503 | dentry->d_flags |= DCACHE_OP_PRUNE; | 1513 | dentry->d_flags |= DCACHE_OP_PRUNE; |
1504 | 1514 | ||
1505 | } | 1515 | } |
1506 | EXPORT_SYMBOL(d_set_d_op); | 1516 | EXPORT_SYMBOL(d_set_d_op); |
1507 | 1517 | ||
1508 | static unsigned d_flags_for_inode(struct inode *inode) | 1518 | static unsigned d_flags_for_inode(struct inode *inode) |
1509 | { | 1519 | { |
1510 | unsigned add_flags = DCACHE_FILE_TYPE; | 1520 | unsigned add_flags = DCACHE_FILE_TYPE; |
1511 | 1521 | ||
1512 | if (!inode) | 1522 | if (!inode) |
1513 | return DCACHE_MISS_TYPE; | 1523 | return DCACHE_MISS_TYPE; |
1514 | 1524 | ||
1515 | if (S_ISDIR(inode->i_mode)) { | 1525 | if (S_ISDIR(inode->i_mode)) { |
1516 | add_flags = DCACHE_DIRECTORY_TYPE; | 1526 | add_flags = DCACHE_DIRECTORY_TYPE; |
1517 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | 1527 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { |
1518 | if (unlikely(!inode->i_op->lookup)) | 1528 | if (unlikely(!inode->i_op->lookup)) |
1519 | add_flags = DCACHE_AUTODIR_TYPE; | 1529 | add_flags = DCACHE_AUTODIR_TYPE; |
1520 | else | 1530 | else |
1521 | inode->i_opflags |= IOP_LOOKUP; | 1531 | inode->i_opflags |= IOP_LOOKUP; |
1522 | } | 1532 | } |
1523 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | 1533 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { |
1524 | if (unlikely(inode->i_op->follow_link)) | 1534 | if (unlikely(inode->i_op->follow_link)) |
1525 | add_flags = DCACHE_SYMLINK_TYPE; | 1535 | add_flags = DCACHE_SYMLINK_TYPE; |
1526 | else | 1536 | else |
1527 | inode->i_opflags |= IOP_NOFOLLOW; | 1537 | inode->i_opflags |= IOP_NOFOLLOW; |
1528 | } | 1538 | } |
1529 | 1539 | ||
1530 | if (unlikely(IS_AUTOMOUNT(inode))) | 1540 | if (unlikely(IS_AUTOMOUNT(inode))) |
1531 | add_flags |= DCACHE_NEED_AUTOMOUNT; | 1541 | add_flags |= DCACHE_NEED_AUTOMOUNT; |
1532 | return add_flags; | 1542 | return add_flags; |
1533 | } | 1543 | } |
1534 | 1544 | ||
1535 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | 1545 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1536 | { | 1546 | { |
1537 | unsigned add_flags = d_flags_for_inode(inode); | 1547 | unsigned add_flags = d_flags_for_inode(inode); |
1538 | 1548 | ||
1539 | spin_lock(&dentry->d_lock); | 1549 | spin_lock(&dentry->d_lock); |
1540 | __d_set_type(dentry, add_flags); | 1550 | __d_set_type(dentry, add_flags); |
1541 | if (inode) | 1551 | if (inode) |
1542 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); | 1552 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
1543 | dentry->d_inode = inode; | 1553 | dentry->d_inode = inode; |
1544 | dentry_rcuwalk_barrier(dentry); | 1554 | dentry_rcuwalk_barrier(dentry); |
1545 | spin_unlock(&dentry->d_lock); | 1555 | spin_unlock(&dentry->d_lock); |
1546 | fsnotify_d_instantiate(dentry, inode); | 1556 | fsnotify_d_instantiate(dentry, inode); |
1547 | } | 1557 | } |
1548 | 1558 | ||
1549 | /** | 1559 | /** |
1550 | * d_instantiate - fill in inode information for a dentry | 1560 | * d_instantiate - fill in inode information for a dentry |
1551 | * @entry: dentry to complete | 1561 | * @entry: dentry to complete |
1552 | * @inode: inode to attach to this dentry | 1562 | * @inode: inode to attach to this dentry |
1553 | * | 1563 | * |
1554 | * Fill in inode information in the entry. | 1564 | * Fill in inode information in the entry. |
1555 | * | 1565 | * |
1556 | * This turns negative dentries into productive full members | 1566 | * This turns negative dentries into productive full members |
1557 | * of society. | 1567 | * of society. |
1558 | * | 1568 | * |
1559 | * NOTE! This assumes that the inode count has been incremented | 1569 | * NOTE! This assumes that the inode count has been incremented |
1560 | * (or otherwise set) by the caller to indicate that it is now | 1570 | * (or otherwise set) by the caller to indicate that it is now |
1561 | * in use by the dcache. | 1571 | * in use by the dcache. |
1562 | */ | 1572 | */ |
1563 | 1573 | ||
1564 | void d_instantiate(struct dentry *entry, struct inode * inode) | 1574 | void d_instantiate(struct dentry *entry, struct inode * inode) |
1565 | { | 1575 | { |
1566 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | 1576 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
1567 | if (inode) | 1577 | if (inode) |
1568 | spin_lock(&inode->i_lock); | 1578 | spin_lock(&inode->i_lock); |
1569 | __d_instantiate(entry, inode); | 1579 | __d_instantiate(entry, inode); |
1570 | if (inode) | 1580 | if (inode) |
1571 | spin_unlock(&inode->i_lock); | 1581 | spin_unlock(&inode->i_lock); |
1572 | security_d_instantiate(entry, inode); | 1582 | security_d_instantiate(entry, inode); |
1573 | } | 1583 | } |
1574 | EXPORT_SYMBOL(d_instantiate); | 1584 | EXPORT_SYMBOL(d_instantiate); |
1575 | 1585 | ||
1576 | /** | 1586 | /** |
1577 | * d_instantiate_unique - instantiate a non-aliased dentry | 1587 | * d_instantiate_unique - instantiate a non-aliased dentry |
1578 | * @entry: dentry to instantiate | 1588 | * @entry: dentry to instantiate |
1579 | * @inode: inode to attach to this dentry | 1589 | * @inode: inode to attach to this dentry |
1580 | * | 1590 | * |
1581 | * Fill in inode information in the entry. On success, it returns NULL. | 1591 | * Fill in inode information in the entry. On success, it returns NULL. |
1582 | * If an unhashed alias of "entry" already exists, then we return the | 1592 | * If an unhashed alias of "entry" already exists, then we return the |
1583 | * aliased dentry instead and drop one reference to inode. | 1593 | * aliased dentry instead and drop one reference to inode. |
1584 | * | 1594 | * |
1585 | * Note that in order to avoid conflicts with rename() etc, the caller | 1595 | * Note that in order to avoid conflicts with rename() etc, the caller |
1586 | * had better be holding the parent directory semaphore. | 1596 | * had better be holding the parent directory semaphore. |
1587 | * | 1597 | * |
1588 | * This also assumes that the inode count has been incremented | 1598 | * This also assumes that the inode count has been incremented |
1589 | * (or otherwise set) by the caller to indicate that it is now | 1599 | * (or otherwise set) by the caller to indicate that it is now |
1590 | * in use by the dcache. | 1600 | * in use by the dcache. |
1591 | */ | 1601 | */ |
1592 | static struct dentry *__d_instantiate_unique(struct dentry *entry, | 1602 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1593 | struct inode *inode) | 1603 | struct inode *inode) |
1594 | { | 1604 | { |
1595 | struct dentry *alias; | 1605 | struct dentry *alias; |
1596 | int len = entry->d_name.len; | 1606 | int len = entry->d_name.len; |
1597 | const char *name = entry->d_name.name; | 1607 | const char *name = entry->d_name.name; |
1598 | unsigned int hash = entry->d_name.hash; | 1608 | unsigned int hash = entry->d_name.hash; |
1599 | 1609 | ||
1600 | if (!inode) { | 1610 | if (!inode) { |
1601 | __d_instantiate(entry, NULL); | 1611 | __d_instantiate(entry, NULL); |
1602 | return NULL; | 1612 | return NULL; |
1603 | } | 1613 | } |
1604 | 1614 | ||
1605 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { | 1615 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1606 | /* | 1616 | /* |
1607 | * Don't need alias->d_lock here, because aliases with | 1617 | * Don't need alias->d_lock here, because aliases with |
1608 | * d_parent == entry->d_parent are not subject to name or | 1618 | * d_parent == entry->d_parent are not subject to name or |
1609 | * parent changes, because the parent inode i_mutex is held. | 1619 | * parent changes, because the parent inode i_mutex is held. |
1610 | */ | 1620 | */ |
1611 | if (alias->d_name.hash != hash) | 1621 | if (alias->d_name.hash != hash) |
1612 | continue; | 1622 | continue; |
1613 | if (alias->d_parent != entry->d_parent) | 1623 | if (alias->d_parent != entry->d_parent) |
1614 | continue; | 1624 | continue; |
1615 | if (alias->d_name.len != len) | 1625 | if (alias->d_name.len != len) |
1616 | continue; | 1626 | continue; |
1617 | if (dentry_cmp(alias, name, len)) | 1627 | if (dentry_cmp(alias, name, len)) |
1618 | continue; | 1628 | continue; |
1619 | __dget(alias); | 1629 | __dget(alias); |
1620 | return alias; | 1630 | return alias; |
1621 | } | 1631 | } |
1622 | 1632 | ||
1623 | __d_instantiate(entry, inode); | 1633 | __d_instantiate(entry, inode); |
1624 | return NULL; | 1634 | return NULL; |
1625 | } | 1635 | } |
1626 | 1636 | ||
1627 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | 1637 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) |
1628 | { | 1638 | { |
1629 | struct dentry *result; | 1639 | struct dentry *result; |
1630 | 1640 | ||
1631 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | 1641 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
1632 | 1642 | ||
1633 | if (inode) | 1643 | if (inode) |
1634 | spin_lock(&inode->i_lock); | 1644 | spin_lock(&inode->i_lock); |
1635 | result = __d_instantiate_unique(entry, inode); | 1645 | result = __d_instantiate_unique(entry, inode); |
1636 | if (inode) | 1646 | if (inode) |
1637 | spin_unlock(&inode->i_lock); | 1647 | spin_unlock(&inode->i_lock); |
1638 | 1648 | ||
1639 | if (!result) { | 1649 | if (!result) { |
1640 | security_d_instantiate(entry, inode); | 1650 | security_d_instantiate(entry, inode); |
1641 | return NULL; | 1651 | return NULL; |
1642 | } | 1652 | } |
1643 | 1653 | ||
1644 | BUG_ON(!d_unhashed(result)); | 1654 | BUG_ON(!d_unhashed(result)); |
1645 | iput(inode); | 1655 | iput(inode); |
1646 | return result; | 1656 | return result; |
1647 | } | 1657 | } |
1648 | 1658 | ||
1649 | EXPORT_SYMBOL(d_instantiate_unique); | 1659 | EXPORT_SYMBOL(d_instantiate_unique); |
1650 | 1660 | ||
1651 | /** | 1661 | /** |
1652 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | 1662 | * d_instantiate_no_diralias - instantiate a non-aliased dentry |
1653 | * @entry: dentry to complete | 1663 | * @entry: dentry to complete |
1654 | * @inode: inode to attach to this dentry | 1664 | * @inode: inode to attach to this dentry |
1655 | * | 1665 | * |
1656 | * Fill in inode information in the entry. If a directory alias is found, then | 1666 | * Fill in inode information in the entry. If a directory alias is found, then |
1657 | * return an error (and drop inode). Together with d_materialise_unique() this | 1667 | * return an error (and drop inode). Together with d_materialise_unique() this |
1658 | * guarantees that a directory inode may never have more than one alias. | 1668 | * guarantees that a directory inode may never have more than one alias. |
1659 | */ | 1669 | */ |
1660 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | 1670 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) |
1661 | { | 1671 | { |
1662 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | 1672 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
1663 | 1673 | ||
1664 | spin_lock(&inode->i_lock); | 1674 | spin_lock(&inode->i_lock); |
1665 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | 1675 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { |
1666 | spin_unlock(&inode->i_lock); | 1676 | spin_unlock(&inode->i_lock); |
1667 | iput(inode); | 1677 | iput(inode); |
1668 | return -EBUSY; | 1678 | return -EBUSY; |
1669 | } | 1679 | } |
1670 | __d_instantiate(entry, inode); | 1680 | __d_instantiate(entry, inode); |
1671 | spin_unlock(&inode->i_lock); | 1681 | spin_unlock(&inode->i_lock); |
1672 | security_d_instantiate(entry, inode); | 1682 | security_d_instantiate(entry, inode); |
1673 | 1683 | ||
1674 | return 0; | 1684 | return 0; |
1675 | } | 1685 | } |
1676 | EXPORT_SYMBOL(d_instantiate_no_diralias); | 1686 | EXPORT_SYMBOL(d_instantiate_no_diralias); |
1677 | 1687 | ||
1678 | struct dentry *d_make_root(struct inode *root_inode) | 1688 | struct dentry *d_make_root(struct inode *root_inode) |
1679 | { | 1689 | { |
1680 | struct dentry *res = NULL; | 1690 | struct dentry *res = NULL; |
1681 | 1691 | ||
1682 | if (root_inode) { | 1692 | if (root_inode) { |
1683 | static const struct qstr name = QSTR_INIT("/", 1); | 1693 | static const struct qstr name = QSTR_INIT("/", 1); |
1684 | 1694 | ||
1685 | res = __d_alloc(root_inode->i_sb, &name); | 1695 | res = __d_alloc(root_inode->i_sb, &name); |
1686 | if (res) | 1696 | if (res) |
1687 | d_instantiate(res, root_inode); | 1697 | d_instantiate(res, root_inode); |
1688 | else | 1698 | else |
1689 | iput(root_inode); | 1699 | iput(root_inode); |
1690 | } | 1700 | } |
1691 | return res; | 1701 | return res; |
1692 | } | 1702 | } |
1693 | EXPORT_SYMBOL(d_make_root); | 1703 | EXPORT_SYMBOL(d_make_root); |
1694 | 1704 | ||
1695 | static struct dentry * __d_find_any_alias(struct inode *inode) | 1705 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1696 | { | 1706 | { |
1697 | struct dentry *alias; | 1707 | struct dentry *alias; |
1698 | 1708 | ||
1699 | if (hlist_empty(&inode->i_dentry)) | 1709 | if (hlist_empty(&inode->i_dentry)) |
1700 | return NULL; | 1710 | return NULL; |
1701 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); | 1711 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
1702 | __dget(alias); | 1712 | __dget(alias); |
1703 | return alias; | 1713 | return alias; |
1704 | } | 1714 | } |
1705 | 1715 | ||
1706 | /** | 1716 | /** |
1707 | * d_find_any_alias - find any alias for a given inode | 1717 | * d_find_any_alias - find any alias for a given inode |
1708 | * @inode: inode to find an alias for | 1718 | * @inode: inode to find an alias for |
1709 | * | 1719 | * |
1710 | * If any aliases exist for the given inode, take and return a | 1720 | * If any aliases exist for the given inode, take and return a |
1711 | * reference for one of them. If no aliases exist, return %NULL. | 1721 | * reference for one of them. If no aliases exist, return %NULL. |
1712 | */ | 1722 | */ |
1713 | struct dentry *d_find_any_alias(struct inode *inode) | 1723 | struct dentry *d_find_any_alias(struct inode *inode) |
1714 | { | 1724 | { |
1715 | struct dentry *de; | 1725 | struct dentry *de; |
1716 | 1726 | ||
1717 | spin_lock(&inode->i_lock); | 1727 | spin_lock(&inode->i_lock); |
1718 | de = __d_find_any_alias(inode); | 1728 | de = __d_find_any_alias(inode); |
1719 | spin_unlock(&inode->i_lock); | 1729 | spin_unlock(&inode->i_lock); |
1720 | return de; | 1730 | return de; |
1721 | } | 1731 | } |
1722 | EXPORT_SYMBOL(d_find_any_alias); | 1732 | EXPORT_SYMBOL(d_find_any_alias); |
1723 | 1733 | ||
1724 | /** | 1734 | /** |
1725 | * d_obtain_alias - find or allocate a dentry for a given inode | 1735 | * d_obtain_alias - find or allocate a dentry for a given inode |
1726 | * @inode: inode to allocate the dentry for | 1736 | * @inode: inode to allocate the dentry for |
1727 | * | 1737 | * |
1728 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | 1738 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or |
1729 | * similar open by handle operations. The returned dentry may be anonymous, | 1739 | * similar open by handle operations. The returned dentry may be anonymous, |
1730 | * or may have a full name (if the inode was already in the cache). | 1740 | * or may have a full name (if the inode was already in the cache). |
1731 | * | 1741 | * |
1732 | * When called on a directory inode, we must ensure that the inode only ever | 1742 | * When called on a directory inode, we must ensure that the inode only ever |
1733 | * has one dentry. If a dentry is found, that is returned instead of | 1743 | * has one dentry. If a dentry is found, that is returned instead of |
1734 | * allocating a new one. | 1744 | * allocating a new one. |
1735 | * | 1745 | * |
1736 | * On successful return, the reference to the inode has been transferred | 1746 | * On successful return, the reference to the inode has been transferred |
1737 | * to the dentry. In case of an error the reference on the inode is released. | 1747 | * to the dentry. In case of an error the reference on the inode is released. |
1738 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | 1748 | * To make it easier to use in export operations a %NULL or IS_ERR inode may |
1739 | * be passed in and will be the error will be propagate to the return value, | 1749 | * be passed in and will be the error will be propagate to the return value, |
1740 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | 1750 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). |
1741 | */ | 1751 | */ |
1742 | struct dentry *d_obtain_alias(struct inode *inode) | 1752 | struct dentry *d_obtain_alias(struct inode *inode) |
1743 | { | 1753 | { |
1744 | static const struct qstr anonstring = QSTR_INIT("/", 1); | 1754 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
1745 | struct dentry *tmp; | 1755 | struct dentry *tmp; |
1746 | struct dentry *res; | 1756 | struct dentry *res; |
1747 | unsigned add_flags; | 1757 | unsigned add_flags; |
1748 | 1758 | ||
1749 | if (!inode) | 1759 | if (!inode) |
1750 | return ERR_PTR(-ESTALE); | 1760 | return ERR_PTR(-ESTALE); |
1751 | if (IS_ERR(inode)) | 1761 | if (IS_ERR(inode)) |
1752 | return ERR_CAST(inode); | 1762 | return ERR_CAST(inode); |
1753 | 1763 | ||
1754 | res = d_find_any_alias(inode); | 1764 | res = d_find_any_alias(inode); |
1755 | if (res) | 1765 | if (res) |
1756 | goto out_iput; | 1766 | goto out_iput; |
1757 | 1767 | ||
1758 | tmp = __d_alloc(inode->i_sb, &anonstring); | 1768 | tmp = __d_alloc(inode->i_sb, &anonstring); |
1759 | if (!tmp) { | 1769 | if (!tmp) { |
1760 | res = ERR_PTR(-ENOMEM); | 1770 | res = ERR_PTR(-ENOMEM); |
1761 | goto out_iput; | 1771 | goto out_iput; |
1762 | } | 1772 | } |
1763 | 1773 | ||
1764 | spin_lock(&inode->i_lock); | 1774 | spin_lock(&inode->i_lock); |
1765 | res = __d_find_any_alias(inode); | 1775 | res = __d_find_any_alias(inode); |
1766 | if (res) { | 1776 | if (res) { |
1767 | spin_unlock(&inode->i_lock); | 1777 | spin_unlock(&inode->i_lock); |
1768 | dput(tmp); | 1778 | dput(tmp); |
1769 | goto out_iput; | 1779 | goto out_iput; |
1770 | } | 1780 | } |
1771 | 1781 | ||
1772 | /* attach a disconnected dentry */ | 1782 | /* attach a disconnected dentry */ |
1773 | add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED; | 1783 | add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED; |
1774 | 1784 | ||
1775 | spin_lock(&tmp->d_lock); | 1785 | spin_lock(&tmp->d_lock); |
1776 | tmp->d_inode = inode; | 1786 | tmp->d_inode = inode; |
1777 | tmp->d_flags |= add_flags; | 1787 | tmp->d_flags |= add_flags; |
1778 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); | 1788 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1779 | hlist_bl_lock(&tmp->d_sb->s_anon); | 1789 | hlist_bl_lock(&tmp->d_sb->s_anon); |
1780 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); | 1790 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1781 | hlist_bl_unlock(&tmp->d_sb->s_anon); | 1791 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
1782 | spin_unlock(&tmp->d_lock); | 1792 | spin_unlock(&tmp->d_lock); |
1783 | spin_unlock(&inode->i_lock); | 1793 | spin_unlock(&inode->i_lock); |
1784 | security_d_instantiate(tmp, inode); | 1794 | security_d_instantiate(tmp, inode); |
1785 | 1795 | ||
1786 | return tmp; | 1796 | return tmp; |
1787 | 1797 | ||
1788 | out_iput: | 1798 | out_iput: |
1789 | if (res && !IS_ERR(res)) | 1799 | if (res && !IS_ERR(res)) |
1790 | security_d_instantiate(res, inode); | 1800 | security_d_instantiate(res, inode); |
1791 | iput(inode); | 1801 | iput(inode); |
1792 | return res; | 1802 | return res; |
1793 | } | 1803 | } |
1794 | EXPORT_SYMBOL(d_obtain_alias); | 1804 | EXPORT_SYMBOL(d_obtain_alias); |
1795 | 1805 | ||
1796 | /** | 1806 | /** |
1797 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | 1807 | * d_splice_alias - splice a disconnected dentry into the tree if one exists |
1798 | * @inode: the inode which may have a disconnected dentry | 1808 | * @inode: the inode which may have a disconnected dentry |
1799 | * @dentry: a negative dentry which we want to point to the inode. | 1809 | * @dentry: a negative dentry which we want to point to the inode. |
1800 | * | 1810 | * |
1801 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | 1811 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and |
1802 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | 1812 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry |
1803 | * and return it, else simply d_add the inode to the dentry and return NULL. | 1813 | * and return it, else simply d_add the inode to the dentry and return NULL. |
1804 | * | 1814 | * |
1805 | * This is needed in the lookup routine of any filesystem that is exportable | 1815 | * This is needed in the lookup routine of any filesystem that is exportable |
1806 | * (via knfsd) so that we can build dcache paths to directories effectively. | 1816 | * (via knfsd) so that we can build dcache paths to directories effectively. |
1807 | * | 1817 | * |
1808 | * If a dentry was found and moved, then it is returned. Otherwise NULL | 1818 | * If a dentry was found and moved, then it is returned. Otherwise NULL |
1809 | * is returned. This matches the expected return value of ->lookup. | 1819 | * is returned. This matches the expected return value of ->lookup. |
1810 | * | 1820 | * |
1811 | * Cluster filesystems may call this function with a negative, hashed dentry. | 1821 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1812 | * In that case, we know that the inode will be a regular file, and also this | 1822 | * In that case, we know that the inode will be a regular file, and also this |
1813 | * will only occur during atomic_open. So we need to check for the dentry | 1823 | * will only occur during atomic_open. So we need to check for the dentry |
1814 | * being already hashed only in the final case. | 1824 | * being already hashed only in the final case. |
1815 | */ | 1825 | */ |
1816 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | 1826 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) |
1817 | { | 1827 | { |
1818 | struct dentry *new = NULL; | 1828 | struct dentry *new = NULL; |
1819 | 1829 | ||
1820 | if (IS_ERR(inode)) | 1830 | if (IS_ERR(inode)) |
1821 | return ERR_CAST(inode); | 1831 | return ERR_CAST(inode); |
1822 | 1832 | ||
1823 | if (inode && S_ISDIR(inode->i_mode)) { | 1833 | if (inode && S_ISDIR(inode->i_mode)) { |
1824 | spin_lock(&inode->i_lock); | 1834 | spin_lock(&inode->i_lock); |
1825 | new = __d_find_alias(inode, 1); | 1835 | new = __d_find_alias(inode, 1); |
1826 | if (new) { | 1836 | if (new) { |
1827 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | 1837 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
1828 | spin_unlock(&inode->i_lock); | 1838 | spin_unlock(&inode->i_lock); |
1829 | security_d_instantiate(new, inode); | 1839 | security_d_instantiate(new, inode); |
1830 | d_move(new, dentry); | 1840 | d_move(new, dentry); |
1831 | iput(inode); | 1841 | iput(inode); |
1832 | } else { | 1842 | } else { |
1833 | /* already taking inode->i_lock, so d_add() by hand */ | 1843 | /* already taking inode->i_lock, so d_add() by hand */ |
1834 | __d_instantiate(dentry, inode); | 1844 | __d_instantiate(dentry, inode); |
1835 | spin_unlock(&inode->i_lock); | 1845 | spin_unlock(&inode->i_lock); |
1836 | security_d_instantiate(dentry, inode); | 1846 | security_d_instantiate(dentry, inode); |
1837 | d_rehash(dentry); | 1847 | d_rehash(dentry); |
1838 | } | 1848 | } |
1839 | } else { | 1849 | } else { |
1840 | d_instantiate(dentry, inode); | 1850 | d_instantiate(dentry, inode); |
1841 | if (d_unhashed(dentry)) | 1851 | if (d_unhashed(dentry)) |
1842 | d_rehash(dentry); | 1852 | d_rehash(dentry); |
1843 | } | 1853 | } |
1844 | return new; | 1854 | return new; |
1845 | } | 1855 | } |
1846 | EXPORT_SYMBOL(d_splice_alias); | 1856 | EXPORT_SYMBOL(d_splice_alias); |
1847 | 1857 | ||
1848 | /** | 1858 | /** |
1849 | * d_add_ci - lookup or allocate new dentry with case-exact name | 1859 | * d_add_ci - lookup or allocate new dentry with case-exact name |
1850 | * @inode: the inode case-insensitive lookup has found | 1860 | * @inode: the inode case-insensitive lookup has found |
1851 | * @dentry: the negative dentry that was passed to the parent's lookup func | 1861 | * @dentry: the negative dentry that was passed to the parent's lookup func |
1852 | * @name: the case-exact name to be associated with the returned dentry | 1862 | * @name: the case-exact name to be associated with the returned dentry |
1853 | * | 1863 | * |
1854 | * This is to avoid filling the dcache with case-insensitive names to the | 1864 | * This is to avoid filling the dcache with case-insensitive names to the |
1855 | * same inode, only the actual correct case is stored in the dcache for | 1865 | * same inode, only the actual correct case is stored in the dcache for |
1856 | * case-insensitive filesystems. | 1866 | * case-insensitive filesystems. |
1857 | * | 1867 | * |
1858 | * For a case-insensitive lookup match and if the the case-exact dentry | 1868 | * For a case-insensitive lookup match and if the the case-exact dentry |
1859 | * already exists in in the dcache, use it and return it. | 1869 | * already exists in in the dcache, use it and return it. |
1860 | * | 1870 | * |
1861 | * If no entry exists with the exact case name, allocate new dentry with | 1871 | * If no entry exists with the exact case name, allocate new dentry with |
1862 | * the exact case, and return the spliced entry. | 1872 | * the exact case, and return the spliced entry. |
1863 | */ | 1873 | */ |
1864 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, | 1874 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
1865 | struct qstr *name) | 1875 | struct qstr *name) |
1866 | { | 1876 | { |
1867 | struct dentry *found; | 1877 | struct dentry *found; |
1868 | struct dentry *new; | 1878 | struct dentry *new; |
1869 | 1879 | ||
1870 | /* | 1880 | /* |
1871 | * First check if a dentry matching the name already exists, | 1881 | * First check if a dentry matching the name already exists, |
1872 | * if not go ahead and create it now. | 1882 | * if not go ahead and create it now. |
1873 | */ | 1883 | */ |
1874 | found = d_hash_and_lookup(dentry->d_parent, name); | 1884 | found = d_hash_and_lookup(dentry->d_parent, name); |
1875 | if (unlikely(IS_ERR(found))) | 1885 | if (unlikely(IS_ERR(found))) |
1876 | goto err_out; | 1886 | goto err_out; |
1877 | if (!found) { | 1887 | if (!found) { |
1878 | new = d_alloc(dentry->d_parent, name); | 1888 | new = d_alloc(dentry->d_parent, name); |
1879 | if (!new) { | 1889 | if (!new) { |
1880 | found = ERR_PTR(-ENOMEM); | 1890 | found = ERR_PTR(-ENOMEM); |
1881 | goto err_out; | 1891 | goto err_out; |
1882 | } | 1892 | } |
1883 | 1893 | ||
1884 | found = d_splice_alias(inode, new); | 1894 | found = d_splice_alias(inode, new); |
1885 | if (found) { | 1895 | if (found) { |
1886 | dput(new); | 1896 | dput(new); |
1887 | return found; | 1897 | return found; |
1888 | } | 1898 | } |
1889 | return new; | 1899 | return new; |
1890 | } | 1900 | } |
1891 | 1901 | ||
1892 | /* | 1902 | /* |
1893 | * If a matching dentry exists, and it's not negative use it. | 1903 | * If a matching dentry exists, and it's not negative use it. |
1894 | * | 1904 | * |
1895 | * Decrement the reference count to balance the iget() done | 1905 | * Decrement the reference count to balance the iget() done |
1896 | * earlier on. | 1906 | * earlier on. |
1897 | */ | 1907 | */ |
1898 | if (found->d_inode) { | 1908 | if (found->d_inode) { |
1899 | if (unlikely(found->d_inode != inode)) { | 1909 | if (unlikely(found->d_inode != inode)) { |
1900 | /* This can't happen because bad inodes are unhashed. */ | 1910 | /* This can't happen because bad inodes are unhashed. */ |
1901 | BUG_ON(!is_bad_inode(inode)); | 1911 | BUG_ON(!is_bad_inode(inode)); |
1902 | BUG_ON(!is_bad_inode(found->d_inode)); | 1912 | BUG_ON(!is_bad_inode(found->d_inode)); |
1903 | } | 1913 | } |
1904 | iput(inode); | 1914 | iput(inode); |
1905 | return found; | 1915 | return found; |
1906 | } | 1916 | } |
1907 | 1917 | ||
1908 | /* | 1918 | /* |
1909 | * Negative dentry: instantiate it unless the inode is a directory and | 1919 | * Negative dentry: instantiate it unless the inode is a directory and |
1910 | * already has a dentry. | 1920 | * already has a dentry. |
1911 | */ | 1921 | */ |
1912 | new = d_splice_alias(inode, found); | 1922 | new = d_splice_alias(inode, found); |
1913 | if (new) { | 1923 | if (new) { |
1914 | dput(found); | 1924 | dput(found); |
1915 | found = new; | 1925 | found = new; |
1916 | } | 1926 | } |
1917 | return found; | 1927 | return found; |
1918 | 1928 | ||
1919 | err_out: | 1929 | err_out: |
1920 | iput(inode); | 1930 | iput(inode); |
1921 | return found; | 1931 | return found; |
1922 | } | 1932 | } |
1923 | EXPORT_SYMBOL(d_add_ci); | 1933 | EXPORT_SYMBOL(d_add_ci); |
1924 | 1934 | ||
1925 | /* | 1935 | /* |
1926 | * Do the slow-case of the dentry name compare. | 1936 | * Do the slow-case of the dentry name compare. |
1927 | * | 1937 | * |
1928 | * Unlike the dentry_cmp() function, we need to atomically | 1938 | * Unlike the dentry_cmp() function, we need to atomically |
1929 | * load the name and length information, so that the | 1939 | * load the name and length information, so that the |
1930 | * filesystem can rely on them, and can use the 'name' and | 1940 | * filesystem can rely on them, and can use the 'name' and |
1931 | * 'len' information without worrying about walking off the | 1941 | * 'len' information without worrying about walking off the |
1932 | * end of memory etc. | 1942 | * end of memory etc. |
1933 | * | 1943 | * |
1934 | * Thus the read_seqcount_retry() and the "duplicate" info | 1944 | * Thus the read_seqcount_retry() and the "duplicate" info |
1935 | * in arguments (the low-level filesystem should not look | 1945 | * in arguments (the low-level filesystem should not look |
1936 | * at the dentry inode or name contents directly, since | 1946 | * at the dentry inode or name contents directly, since |
1937 | * rename can change them while we're in RCU mode). | 1947 | * rename can change them while we're in RCU mode). |
1938 | */ | 1948 | */ |
1939 | enum slow_d_compare { | 1949 | enum slow_d_compare { |
1940 | D_COMP_OK, | 1950 | D_COMP_OK, |
1941 | D_COMP_NOMATCH, | 1951 | D_COMP_NOMATCH, |
1942 | D_COMP_SEQRETRY, | 1952 | D_COMP_SEQRETRY, |
1943 | }; | 1953 | }; |
1944 | 1954 | ||
1945 | static noinline enum slow_d_compare slow_dentry_cmp( | 1955 | static noinline enum slow_d_compare slow_dentry_cmp( |
1946 | const struct dentry *parent, | 1956 | const struct dentry *parent, |
1947 | struct dentry *dentry, | 1957 | struct dentry *dentry, |
1948 | unsigned int seq, | 1958 | unsigned int seq, |
1949 | const struct qstr *name) | 1959 | const struct qstr *name) |
1950 | { | 1960 | { |
1951 | int tlen = dentry->d_name.len; | 1961 | int tlen = dentry->d_name.len; |
1952 | const char *tname = dentry->d_name.name; | 1962 | const char *tname = dentry->d_name.name; |
1953 | 1963 | ||
1954 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | 1964 | if (read_seqcount_retry(&dentry->d_seq, seq)) { |
1955 | cpu_relax(); | 1965 | cpu_relax(); |
1956 | return D_COMP_SEQRETRY; | 1966 | return D_COMP_SEQRETRY; |
1957 | } | 1967 | } |
1958 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) | 1968 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1959 | return D_COMP_NOMATCH; | 1969 | return D_COMP_NOMATCH; |
1960 | return D_COMP_OK; | 1970 | return D_COMP_OK; |
1961 | } | 1971 | } |
1962 | 1972 | ||
1963 | /** | 1973 | /** |
1964 | * __d_lookup_rcu - search for a dentry (racy, store-free) | 1974 | * __d_lookup_rcu - search for a dentry (racy, store-free) |
1965 | * @parent: parent dentry | 1975 | * @parent: parent dentry |
1966 | * @name: qstr of name we wish to find | 1976 | * @name: qstr of name we wish to find |
1967 | * @seqp: returns d_seq value at the point where the dentry was found | 1977 | * @seqp: returns d_seq value at the point where the dentry was found |
1968 | * Returns: dentry, or NULL | 1978 | * Returns: dentry, or NULL |
1969 | * | 1979 | * |
1970 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | 1980 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name |
1971 | * resolution (store-free path walking) design described in | 1981 | * resolution (store-free path walking) design described in |
1972 | * Documentation/filesystems/path-lookup.txt. | 1982 | * Documentation/filesystems/path-lookup.txt. |
1973 | * | 1983 | * |
1974 | * This is not to be used outside core vfs. | 1984 | * This is not to be used outside core vfs. |
1975 | * | 1985 | * |
1976 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | 1986 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock |
1977 | * held, and rcu_read_lock held. The returned dentry must not be stored into | 1987 | * held, and rcu_read_lock held. The returned dentry must not be stored into |
1978 | * without taking d_lock and checking d_seq sequence count against @seq | 1988 | * without taking d_lock and checking d_seq sequence count against @seq |
1979 | * returned here. | 1989 | * returned here. |
1980 | * | 1990 | * |
1981 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount | 1991 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
1982 | * function. | 1992 | * function. |
1983 | * | 1993 | * |
1984 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | 1994 | * Alternatively, __d_lookup_rcu may be called again to look up the child of |
1985 | * the returned dentry, so long as its parent's seqlock is checked after the | 1995 | * the returned dentry, so long as its parent's seqlock is checked after the |
1986 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | 1996 | * child is looked up. Thus, an interlocking stepping of sequence lock checks |
1987 | * is formed, giving integrity down the path walk. | 1997 | * is formed, giving integrity down the path walk. |
1988 | * | 1998 | * |
1989 | * NOTE! The caller *has* to check the resulting dentry against the sequence | 1999 | * NOTE! The caller *has* to check the resulting dentry against the sequence |
1990 | * number we've returned before using any of the resulting dentry state! | 2000 | * number we've returned before using any of the resulting dentry state! |
1991 | */ | 2001 | */ |
1992 | struct dentry *__d_lookup_rcu(const struct dentry *parent, | 2002 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1993 | const struct qstr *name, | 2003 | const struct qstr *name, |
1994 | unsigned *seqp) | 2004 | unsigned *seqp) |
1995 | { | 2005 | { |
1996 | u64 hashlen = name->hash_len; | 2006 | u64 hashlen = name->hash_len; |
1997 | const unsigned char *str = name->name; | 2007 | const unsigned char *str = name->name; |
1998 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); | 2008 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
1999 | struct hlist_bl_node *node; | 2009 | struct hlist_bl_node *node; |
2000 | struct dentry *dentry; | 2010 | struct dentry *dentry; |
2001 | 2011 | ||
2002 | /* | 2012 | /* |
2003 | * Note: There is significant duplication with __d_lookup_rcu which is | 2013 | * Note: There is significant duplication with __d_lookup_rcu which is |
2004 | * required to prevent single threaded performance regressions | 2014 | * required to prevent single threaded performance regressions |
2005 | * especially on architectures where smp_rmb (in seqcounts) are costly. | 2015 | * especially on architectures where smp_rmb (in seqcounts) are costly. |
2006 | * Keep the two functions in sync. | 2016 | * Keep the two functions in sync. |
2007 | */ | 2017 | */ |
2008 | 2018 | ||
2009 | /* | 2019 | /* |
2010 | * The hash list is protected using RCU. | 2020 | * The hash list is protected using RCU. |
2011 | * | 2021 | * |
2012 | * Carefully use d_seq when comparing a candidate dentry, to avoid | 2022 | * Carefully use d_seq when comparing a candidate dentry, to avoid |
2013 | * races with d_move(). | 2023 | * races with d_move(). |
2014 | * | 2024 | * |
2015 | * It is possible that concurrent renames can mess up our list | 2025 | * It is possible that concurrent renames can mess up our list |
2016 | * walk here and result in missing our dentry, resulting in the | 2026 | * walk here and result in missing our dentry, resulting in the |
2017 | * false-negative result. d_lookup() protects against concurrent | 2027 | * false-negative result. d_lookup() protects against concurrent |
2018 | * renames using rename_lock seqlock. | 2028 | * renames using rename_lock seqlock. |
2019 | * | 2029 | * |
2020 | * See Documentation/filesystems/path-lookup.txt for more details. | 2030 | * See Documentation/filesystems/path-lookup.txt for more details. |
2021 | */ | 2031 | */ |
2022 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | 2032 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
2023 | unsigned seq; | 2033 | unsigned seq; |
2024 | 2034 | ||
2025 | seqretry: | 2035 | seqretry: |
2026 | /* | 2036 | /* |
2027 | * The dentry sequence count protects us from concurrent | 2037 | * The dentry sequence count protects us from concurrent |
2028 | * renames, and thus protects parent and name fields. | 2038 | * renames, and thus protects parent and name fields. |
2029 | * | 2039 | * |
2030 | * The caller must perform a seqcount check in order | 2040 | * The caller must perform a seqcount check in order |
2031 | * to do anything useful with the returned dentry. | 2041 | * to do anything useful with the returned dentry. |
2032 | * | 2042 | * |
2033 | * NOTE! We do a "raw" seqcount_begin here. That means that | 2043 | * NOTE! We do a "raw" seqcount_begin here. That means that |
2034 | * we don't wait for the sequence count to stabilize if it | 2044 | * we don't wait for the sequence count to stabilize if it |
2035 | * is in the middle of a sequence change. If we do the slow | 2045 | * is in the middle of a sequence change. If we do the slow |
2036 | * dentry compare, we will do seqretries until it is stable, | 2046 | * dentry compare, we will do seqretries until it is stable, |
2037 | * and if we end up with a successful lookup, we actually | 2047 | * and if we end up with a successful lookup, we actually |
2038 | * want to exit RCU lookup anyway. | 2048 | * want to exit RCU lookup anyway. |
2039 | */ | 2049 | */ |
2040 | seq = raw_seqcount_begin(&dentry->d_seq); | 2050 | seq = raw_seqcount_begin(&dentry->d_seq); |
2041 | if (dentry->d_parent != parent) | 2051 | if (dentry->d_parent != parent) |
2042 | continue; | 2052 | continue; |
2043 | if (d_unhashed(dentry)) | 2053 | if (d_unhashed(dentry)) |
2044 | continue; | 2054 | continue; |
2045 | 2055 | ||
2046 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { | 2056 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
2047 | if (dentry->d_name.hash != hashlen_hash(hashlen)) | 2057 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2048 | continue; | 2058 | continue; |
2049 | *seqp = seq; | 2059 | *seqp = seq; |
2050 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | 2060 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { |
2051 | case D_COMP_OK: | 2061 | case D_COMP_OK: |
2052 | return dentry; | 2062 | return dentry; |
2053 | case D_COMP_NOMATCH: | 2063 | case D_COMP_NOMATCH: |
2054 | continue; | 2064 | continue; |
2055 | default: | 2065 | default: |
2056 | goto seqretry; | 2066 | goto seqretry; |
2057 | } | 2067 | } |
2058 | } | 2068 | } |
2059 | 2069 | ||
2060 | if (dentry->d_name.hash_len != hashlen) | 2070 | if (dentry->d_name.hash_len != hashlen) |
2061 | continue; | 2071 | continue; |
2062 | *seqp = seq; | 2072 | *seqp = seq; |
2063 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) | 2073 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
2064 | return dentry; | 2074 | return dentry; |
2065 | } | 2075 | } |
2066 | return NULL; | 2076 | return NULL; |
2067 | } | 2077 | } |
2068 | 2078 | ||
2069 | /** | 2079 | /** |
2070 | * d_lookup - search for a dentry | 2080 | * d_lookup - search for a dentry |
2071 | * @parent: parent dentry | 2081 | * @parent: parent dentry |
2072 | * @name: qstr of name we wish to find | 2082 | * @name: qstr of name we wish to find |
2073 | * Returns: dentry, or NULL | 2083 | * Returns: dentry, or NULL |
2074 | * | 2084 | * |
2075 | * d_lookup searches the children of the parent dentry for the name in | 2085 | * d_lookup searches the children of the parent dentry for the name in |
2076 | * question. If the dentry is found its reference count is incremented and the | 2086 | * question. If the dentry is found its reference count is incremented and the |
2077 | * dentry is returned. The caller must use dput to free the entry when it has | 2087 | * dentry is returned. The caller must use dput to free the entry when it has |
2078 | * finished using it. %NULL is returned if the dentry does not exist. | 2088 | * finished using it. %NULL is returned if the dentry does not exist. |
2079 | */ | 2089 | */ |
2080 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) | 2090 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
2081 | { | 2091 | { |
2082 | struct dentry *dentry; | 2092 | struct dentry *dentry; |
2083 | unsigned seq; | 2093 | unsigned seq; |
2084 | 2094 | ||
2085 | do { | 2095 | do { |
2086 | seq = read_seqbegin(&rename_lock); | 2096 | seq = read_seqbegin(&rename_lock); |
2087 | dentry = __d_lookup(parent, name); | 2097 | dentry = __d_lookup(parent, name); |
2088 | if (dentry) | 2098 | if (dentry) |
2089 | break; | 2099 | break; |
2090 | } while (read_seqretry(&rename_lock, seq)); | 2100 | } while (read_seqretry(&rename_lock, seq)); |
2091 | return dentry; | 2101 | return dentry; |
2092 | } | 2102 | } |
2093 | EXPORT_SYMBOL(d_lookup); | 2103 | EXPORT_SYMBOL(d_lookup); |
2094 | 2104 | ||
2095 | /** | 2105 | /** |
2096 | * __d_lookup - search for a dentry (racy) | 2106 | * __d_lookup - search for a dentry (racy) |
2097 | * @parent: parent dentry | 2107 | * @parent: parent dentry |
2098 | * @name: qstr of name we wish to find | 2108 | * @name: qstr of name we wish to find |
2099 | * Returns: dentry, or NULL | 2109 | * Returns: dentry, or NULL |
2100 | * | 2110 | * |
2101 | * __d_lookup is like d_lookup, however it may (rarely) return a | 2111 | * __d_lookup is like d_lookup, however it may (rarely) return a |
2102 | * false-negative result due to unrelated rename activity. | 2112 | * false-negative result due to unrelated rename activity. |
2103 | * | 2113 | * |
2104 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | 2114 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, |
2105 | * however it must be used carefully, eg. with a following d_lookup in | 2115 | * however it must be used carefully, eg. with a following d_lookup in |
2106 | * the case of failure. | 2116 | * the case of failure. |
2107 | * | 2117 | * |
2108 | * __d_lookup callers must be commented. | 2118 | * __d_lookup callers must be commented. |
2109 | */ | 2119 | */ |
2110 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) | 2120 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
2111 | { | 2121 | { |
2112 | unsigned int len = name->len; | 2122 | unsigned int len = name->len; |
2113 | unsigned int hash = name->hash; | 2123 | unsigned int hash = name->hash; |
2114 | const unsigned char *str = name->name; | 2124 | const unsigned char *str = name->name; |
2115 | struct hlist_bl_head *b = d_hash(parent, hash); | 2125 | struct hlist_bl_head *b = d_hash(parent, hash); |
2116 | struct hlist_bl_node *node; | 2126 | struct hlist_bl_node *node; |
2117 | struct dentry *found = NULL; | 2127 | struct dentry *found = NULL; |
2118 | struct dentry *dentry; | 2128 | struct dentry *dentry; |
2119 | 2129 | ||
2120 | /* | 2130 | /* |
2121 | * Note: There is significant duplication with __d_lookup_rcu which is | 2131 | * Note: There is significant duplication with __d_lookup_rcu which is |
2122 | * required to prevent single threaded performance regressions | 2132 | * required to prevent single threaded performance regressions |
2123 | * especially on architectures where smp_rmb (in seqcounts) are costly. | 2133 | * especially on architectures where smp_rmb (in seqcounts) are costly. |
2124 | * Keep the two functions in sync. | 2134 | * Keep the two functions in sync. |
2125 | */ | 2135 | */ |
2126 | 2136 | ||
2127 | /* | 2137 | /* |
2128 | * The hash list is protected using RCU. | 2138 | * The hash list is protected using RCU. |
2129 | * | 2139 | * |
2130 | * Take d_lock when comparing a candidate dentry, to avoid races | 2140 | * Take d_lock when comparing a candidate dentry, to avoid races |
2131 | * with d_move(). | 2141 | * with d_move(). |
2132 | * | 2142 | * |
2133 | * It is possible that concurrent renames can mess up our list | 2143 | * It is possible that concurrent renames can mess up our list |
2134 | * walk here and result in missing our dentry, resulting in the | 2144 | * walk here and result in missing our dentry, resulting in the |
2135 | * false-negative result. d_lookup() protects against concurrent | 2145 | * false-negative result. d_lookup() protects against concurrent |
2136 | * renames using rename_lock seqlock. | 2146 | * renames using rename_lock seqlock. |
2137 | * | 2147 | * |
2138 | * See Documentation/filesystems/path-lookup.txt for more details. | 2148 | * See Documentation/filesystems/path-lookup.txt for more details. |
2139 | */ | 2149 | */ |
2140 | rcu_read_lock(); | 2150 | rcu_read_lock(); |
2141 | 2151 | ||
2142 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | 2152 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
2143 | 2153 | ||
2144 | if (dentry->d_name.hash != hash) | 2154 | if (dentry->d_name.hash != hash) |
2145 | continue; | 2155 | continue; |
2146 | 2156 | ||
2147 | spin_lock(&dentry->d_lock); | 2157 | spin_lock(&dentry->d_lock); |
2148 | if (dentry->d_parent != parent) | 2158 | if (dentry->d_parent != parent) |
2149 | goto next; | 2159 | goto next; |
2150 | if (d_unhashed(dentry)) | 2160 | if (d_unhashed(dentry)) |
2151 | goto next; | 2161 | goto next; |
2152 | 2162 | ||
2153 | /* | 2163 | /* |
2154 | * It is safe to compare names since d_move() cannot | 2164 | * It is safe to compare names since d_move() cannot |
2155 | * change the qstr (protected by d_lock). | 2165 | * change the qstr (protected by d_lock). |
2156 | */ | 2166 | */ |
2157 | if (parent->d_flags & DCACHE_OP_COMPARE) { | 2167 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
2158 | int tlen = dentry->d_name.len; | 2168 | int tlen = dentry->d_name.len; |
2159 | const char *tname = dentry->d_name.name; | 2169 | const char *tname = dentry->d_name.name; |
2160 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) | 2170 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
2161 | goto next; | 2171 | goto next; |
2162 | } else { | 2172 | } else { |
2163 | if (dentry->d_name.len != len) | 2173 | if (dentry->d_name.len != len) |
2164 | goto next; | 2174 | goto next; |
2165 | if (dentry_cmp(dentry, str, len)) | 2175 | if (dentry_cmp(dentry, str, len)) |
2166 | goto next; | 2176 | goto next; |
2167 | } | 2177 | } |
2168 | 2178 | ||
2169 | dentry->d_lockref.count++; | 2179 | dentry->d_lockref.count++; |
2170 | found = dentry; | 2180 | found = dentry; |
2171 | spin_unlock(&dentry->d_lock); | 2181 | spin_unlock(&dentry->d_lock); |
2172 | break; | 2182 | break; |
2173 | next: | 2183 | next: |
2174 | spin_unlock(&dentry->d_lock); | 2184 | spin_unlock(&dentry->d_lock); |
2175 | } | 2185 | } |
2176 | rcu_read_unlock(); | 2186 | rcu_read_unlock(); |
2177 | 2187 | ||
2178 | return found; | 2188 | return found; |
2179 | } | 2189 | } |
2180 | 2190 | ||
2181 | /** | 2191 | /** |
2182 | * d_hash_and_lookup - hash the qstr then search for a dentry | 2192 | * d_hash_and_lookup - hash the qstr then search for a dentry |
2183 | * @dir: Directory to search in | 2193 | * @dir: Directory to search in |
2184 | * @name: qstr of name we wish to find | 2194 | * @name: qstr of name we wish to find |
2185 | * | 2195 | * |
2186 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) | 2196 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
2187 | */ | 2197 | */ |
2188 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | 2198 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) |
2189 | { | 2199 | { |
2190 | /* | 2200 | /* |
2191 | * Check for a fs-specific hash function. Note that we must | 2201 | * Check for a fs-specific hash function. Note that we must |
2192 | * calculate the standard hash first, as the d_op->d_hash() | 2202 | * calculate the standard hash first, as the d_op->d_hash() |
2193 | * routine may choose to leave the hash value unchanged. | 2203 | * routine may choose to leave the hash value unchanged. |
2194 | */ | 2204 | */ |
2195 | name->hash = full_name_hash(name->name, name->len); | 2205 | name->hash = full_name_hash(name->name, name->len); |
2196 | if (dir->d_flags & DCACHE_OP_HASH) { | 2206 | if (dir->d_flags & DCACHE_OP_HASH) { |
2197 | int err = dir->d_op->d_hash(dir, name); | 2207 | int err = dir->d_op->d_hash(dir, name); |
2198 | if (unlikely(err < 0)) | 2208 | if (unlikely(err < 0)) |
2199 | return ERR_PTR(err); | 2209 | return ERR_PTR(err); |
2200 | } | 2210 | } |
2201 | return d_lookup(dir, name); | 2211 | return d_lookup(dir, name); |
2202 | } | 2212 | } |
2203 | EXPORT_SYMBOL(d_hash_and_lookup); | 2213 | EXPORT_SYMBOL(d_hash_and_lookup); |
2204 | 2214 | ||
2205 | /** | 2215 | /** |
2206 | * d_validate - verify dentry provided from insecure source (deprecated) | 2216 | * d_validate - verify dentry provided from insecure source (deprecated) |
2207 | * @dentry: The dentry alleged to be valid child of @dparent | 2217 | * @dentry: The dentry alleged to be valid child of @dparent |
2208 | * @dparent: The parent dentry (known to be valid) | 2218 | * @dparent: The parent dentry (known to be valid) |
2209 | * | 2219 | * |
2210 | * An insecure source has sent us a dentry, here we verify it and dget() it. | 2220 | * An insecure source has sent us a dentry, here we verify it and dget() it. |
2211 | * This is used by ncpfs in its readdir implementation. | 2221 | * This is used by ncpfs in its readdir implementation. |
2212 | * Zero is returned in the dentry is invalid. | 2222 | * Zero is returned in the dentry is invalid. |
2213 | * | 2223 | * |
2214 | * This function is slow for big directories, and deprecated, do not use it. | 2224 | * This function is slow for big directories, and deprecated, do not use it. |
2215 | */ | 2225 | */ |
2216 | int d_validate(struct dentry *dentry, struct dentry *dparent) | 2226 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
2217 | { | 2227 | { |
2218 | struct dentry *child; | 2228 | struct dentry *child; |
2219 | 2229 | ||
2220 | spin_lock(&dparent->d_lock); | 2230 | spin_lock(&dparent->d_lock); |
2221 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { | 2231 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2222 | if (dentry == child) { | 2232 | if (dentry == child) { |
2223 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | 2233 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2224 | __dget_dlock(dentry); | 2234 | __dget_dlock(dentry); |
2225 | spin_unlock(&dentry->d_lock); | 2235 | spin_unlock(&dentry->d_lock); |
2226 | spin_unlock(&dparent->d_lock); | 2236 | spin_unlock(&dparent->d_lock); |
2227 | return 1; | 2237 | return 1; |
2228 | } | 2238 | } |
2229 | } | 2239 | } |
2230 | spin_unlock(&dparent->d_lock); | 2240 | spin_unlock(&dparent->d_lock); |
2231 | 2241 | ||
2232 | return 0; | 2242 | return 0; |
2233 | } | 2243 | } |
2234 | EXPORT_SYMBOL(d_validate); | 2244 | EXPORT_SYMBOL(d_validate); |
2235 | 2245 | ||
2236 | /* | 2246 | /* |
2237 | * When a file is deleted, we have two options: | 2247 | * When a file is deleted, we have two options: |
2238 | * - turn this dentry into a negative dentry | 2248 | * - turn this dentry into a negative dentry |
2239 | * - unhash this dentry and free it. | 2249 | * - unhash this dentry and free it. |
2240 | * | 2250 | * |
2241 | * Usually, we want to just turn this into | 2251 | * Usually, we want to just turn this into |
2242 | * a negative dentry, but if anybody else is | 2252 | * a negative dentry, but if anybody else is |
2243 | * currently using the dentry or the inode | 2253 | * currently using the dentry or the inode |
2244 | * we can't do that and we fall back on removing | 2254 | * we can't do that and we fall back on removing |
2245 | * it from the hash queues and waiting for | 2255 | * it from the hash queues and waiting for |
2246 | * it to be deleted later when it has no users | 2256 | * it to be deleted later when it has no users |
2247 | */ | 2257 | */ |
2248 | 2258 | ||
2249 | /** | 2259 | /** |
2250 | * d_delete - delete a dentry | 2260 | * d_delete - delete a dentry |
2251 | * @dentry: The dentry to delete | 2261 | * @dentry: The dentry to delete |
2252 | * | 2262 | * |
2253 | * Turn the dentry into a negative dentry if possible, otherwise | 2263 | * Turn the dentry into a negative dentry if possible, otherwise |
2254 | * remove it from the hash queues so it can be deleted later | 2264 | * remove it from the hash queues so it can be deleted later |
2255 | */ | 2265 | */ |
2256 | 2266 | ||
2257 | void d_delete(struct dentry * dentry) | 2267 | void d_delete(struct dentry * dentry) |
2258 | { | 2268 | { |
2259 | struct inode *inode; | 2269 | struct inode *inode; |
2260 | int isdir = 0; | 2270 | int isdir = 0; |
2261 | /* | 2271 | /* |
2262 | * Are we the only user? | 2272 | * Are we the only user? |
2263 | */ | 2273 | */ |
2264 | again: | 2274 | again: |
2265 | spin_lock(&dentry->d_lock); | 2275 | spin_lock(&dentry->d_lock); |
2266 | inode = dentry->d_inode; | 2276 | inode = dentry->d_inode; |
2267 | isdir = S_ISDIR(inode->i_mode); | 2277 | isdir = S_ISDIR(inode->i_mode); |
2268 | if (dentry->d_lockref.count == 1) { | 2278 | if (dentry->d_lockref.count == 1) { |
2269 | if (!spin_trylock(&inode->i_lock)) { | 2279 | if (!spin_trylock(&inode->i_lock)) { |
2270 | spin_unlock(&dentry->d_lock); | 2280 | spin_unlock(&dentry->d_lock); |
2271 | cpu_relax(); | 2281 | cpu_relax(); |
2272 | goto again; | 2282 | goto again; |
2273 | } | 2283 | } |
2274 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; | 2284 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
2275 | dentry_unlink_inode(dentry); | 2285 | dentry_unlink_inode(dentry); |
2276 | fsnotify_nameremove(dentry, isdir); | 2286 | fsnotify_nameremove(dentry, isdir); |
2277 | return; | 2287 | return; |
2278 | } | 2288 | } |
2279 | 2289 | ||
2280 | if (!d_unhashed(dentry)) | 2290 | if (!d_unhashed(dentry)) |
2281 | __d_drop(dentry); | 2291 | __d_drop(dentry); |
2282 | 2292 | ||
2283 | spin_unlock(&dentry->d_lock); | 2293 | spin_unlock(&dentry->d_lock); |
2284 | 2294 | ||
2285 | fsnotify_nameremove(dentry, isdir); | 2295 | fsnotify_nameremove(dentry, isdir); |
2286 | } | 2296 | } |
2287 | EXPORT_SYMBOL(d_delete); | 2297 | EXPORT_SYMBOL(d_delete); |
2288 | 2298 | ||
2289 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) | 2299 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
2290 | { | 2300 | { |
2291 | BUG_ON(!d_unhashed(entry)); | 2301 | BUG_ON(!d_unhashed(entry)); |
2292 | hlist_bl_lock(b); | 2302 | hlist_bl_lock(b); |
2293 | entry->d_flags |= DCACHE_RCUACCESS; | 2303 | entry->d_flags |= DCACHE_RCUACCESS; |
2294 | hlist_bl_add_head_rcu(&entry->d_hash, b); | 2304 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
2295 | hlist_bl_unlock(b); | 2305 | hlist_bl_unlock(b); |
2296 | } | 2306 | } |
2297 | 2307 | ||
2298 | static void _d_rehash(struct dentry * entry) | 2308 | static void _d_rehash(struct dentry * entry) |
2299 | { | 2309 | { |
2300 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | 2310 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); |
2301 | } | 2311 | } |
2302 | 2312 | ||
2303 | /** | 2313 | /** |
2304 | * d_rehash - add an entry back to the hash | 2314 | * d_rehash - add an entry back to the hash |
2305 | * @entry: dentry to add to the hash | 2315 | * @entry: dentry to add to the hash |
2306 | * | 2316 | * |
2307 | * Adds a dentry to the hash according to its name. | 2317 | * Adds a dentry to the hash according to its name. |
2308 | */ | 2318 | */ |
2309 | 2319 | ||
2310 | void d_rehash(struct dentry * entry) | 2320 | void d_rehash(struct dentry * entry) |
2311 | { | 2321 | { |
2312 | spin_lock(&entry->d_lock); | 2322 | spin_lock(&entry->d_lock); |
2313 | _d_rehash(entry); | 2323 | _d_rehash(entry); |
2314 | spin_unlock(&entry->d_lock); | 2324 | spin_unlock(&entry->d_lock); |
2315 | } | 2325 | } |
2316 | EXPORT_SYMBOL(d_rehash); | 2326 | EXPORT_SYMBOL(d_rehash); |
2317 | 2327 | ||
2318 | /** | 2328 | /** |
2319 | * dentry_update_name_case - update case insensitive dentry with a new name | 2329 | * dentry_update_name_case - update case insensitive dentry with a new name |
2320 | * @dentry: dentry to be updated | 2330 | * @dentry: dentry to be updated |
2321 | * @name: new name | 2331 | * @name: new name |
2322 | * | 2332 | * |
2323 | * Update a case insensitive dentry with new case of name. | 2333 | * Update a case insensitive dentry with new case of name. |
2324 | * | 2334 | * |
2325 | * dentry must have been returned by d_lookup with name @name. Old and new | 2335 | * dentry must have been returned by d_lookup with name @name. Old and new |
2326 | * name lengths must match (ie. no d_compare which allows mismatched name | 2336 | * name lengths must match (ie. no d_compare which allows mismatched name |
2327 | * lengths). | 2337 | * lengths). |
2328 | * | 2338 | * |
2329 | * Parent inode i_mutex must be held over d_lookup and into this call (to | 2339 | * Parent inode i_mutex must be held over d_lookup and into this call (to |
2330 | * keep renames and concurrent inserts, and readdir(2) away). | 2340 | * keep renames and concurrent inserts, and readdir(2) away). |
2331 | */ | 2341 | */ |
2332 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | 2342 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) |
2333 | { | 2343 | { |
2334 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); | 2344 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
2335 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | 2345 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2336 | 2346 | ||
2337 | spin_lock(&dentry->d_lock); | 2347 | spin_lock(&dentry->d_lock); |
2338 | write_seqcount_begin(&dentry->d_seq); | 2348 | write_seqcount_begin(&dentry->d_seq); |
2339 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); | 2349 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
2340 | write_seqcount_end(&dentry->d_seq); | 2350 | write_seqcount_end(&dentry->d_seq); |
2341 | spin_unlock(&dentry->d_lock); | 2351 | spin_unlock(&dentry->d_lock); |
2342 | } | 2352 | } |
2343 | EXPORT_SYMBOL(dentry_update_name_case); | 2353 | EXPORT_SYMBOL(dentry_update_name_case); |
2344 | 2354 | ||
2345 | static void switch_names(struct dentry *dentry, struct dentry *target) | 2355 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2346 | { | 2356 | { |
2347 | if (dname_external(target)) { | 2357 | if (dname_external(target)) { |
2348 | if (dname_external(dentry)) { | 2358 | if (dname_external(dentry)) { |
2349 | /* | 2359 | /* |
2350 | * Both external: swap the pointers | 2360 | * Both external: swap the pointers |
2351 | */ | 2361 | */ |
2352 | swap(target->d_name.name, dentry->d_name.name); | 2362 | swap(target->d_name.name, dentry->d_name.name); |
2353 | } else { | 2363 | } else { |
2354 | /* | 2364 | /* |
2355 | * dentry:internal, target:external. Steal target's | 2365 | * dentry:internal, target:external. Steal target's |
2356 | * storage and make target internal. | 2366 | * storage and make target internal. |
2357 | */ | 2367 | */ |
2358 | memcpy(target->d_iname, dentry->d_name.name, | 2368 | memcpy(target->d_iname, dentry->d_name.name, |
2359 | dentry->d_name.len + 1); | 2369 | dentry->d_name.len + 1); |
2360 | dentry->d_name.name = target->d_name.name; | 2370 | dentry->d_name.name = target->d_name.name; |
2361 | target->d_name.name = target->d_iname; | 2371 | target->d_name.name = target->d_iname; |
2362 | } | 2372 | } |
2363 | } else { | 2373 | } else { |
2364 | if (dname_external(dentry)) { | 2374 | if (dname_external(dentry)) { |
2365 | /* | 2375 | /* |
2366 | * dentry:external, target:internal. Give dentry's | 2376 | * dentry:external, target:internal. Give dentry's |
2367 | * storage to target and make dentry internal | 2377 | * storage to target and make dentry internal |
2368 | */ | 2378 | */ |
2369 | memcpy(dentry->d_iname, target->d_name.name, | 2379 | memcpy(dentry->d_iname, target->d_name.name, |
2370 | target->d_name.len + 1); | 2380 | target->d_name.len + 1); |
2371 | target->d_name.name = dentry->d_name.name; | 2381 | target->d_name.name = dentry->d_name.name; |
2372 | dentry->d_name.name = dentry->d_iname; | 2382 | dentry->d_name.name = dentry->d_iname; |
2373 | } else { | 2383 | } else { |
2374 | /* | 2384 | /* |
2375 | * Both are internal. | 2385 | * Both are internal. |
2376 | */ | 2386 | */ |
2377 | unsigned int i; | 2387 | unsigned int i; |
2378 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | 2388 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); |
2379 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | 2389 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { |
2380 | swap(((long *) &dentry->d_iname)[i], | 2390 | swap(((long *) &dentry->d_iname)[i], |
2381 | ((long *) &target->d_iname)[i]); | 2391 | ((long *) &target->d_iname)[i]); |
2382 | } | 2392 | } |
2383 | } | 2393 | } |
2384 | } | 2394 | } |
2385 | swap(dentry->d_name.len, target->d_name.len); | 2395 | swap(dentry->d_name.len, target->d_name.len); |
2386 | } | 2396 | } |
2387 | 2397 | ||
2388 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) | 2398 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2389 | { | 2399 | { |
2390 | /* | 2400 | /* |
2391 | * XXXX: do we really need to take target->d_lock? | 2401 | * XXXX: do we really need to take target->d_lock? |
2392 | */ | 2402 | */ |
2393 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | 2403 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) |
2394 | spin_lock(&target->d_parent->d_lock); | 2404 | spin_lock(&target->d_parent->d_lock); |
2395 | else { | 2405 | else { |
2396 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | 2406 | if (d_ancestor(dentry->d_parent, target->d_parent)) { |
2397 | spin_lock(&dentry->d_parent->d_lock); | 2407 | spin_lock(&dentry->d_parent->d_lock); |
2398 | spin_lock_nested(&target->d_parent->d_lock, | 2408 | spin_lock_nested(&target->d_parent->d_lock, |
2399 | DENTRY_D_LOCK_NESTED); | 2409 | DENTRY_D_LOCK_NESTED); |
2400 | } else { | 2410 | } else { |
2401 | spin_lock(&target->d_parent->d_lock); | 2411 | spin_lock(&target->d_parent->d_lock); |
2402 | spin_lock_nested(&dentry->d_parent->d_lock, | 2412 | spin_lock_nested(&dentry->d_parent->d_lock, |
2403 | DENTRY_D_LOCK_NESTED); | 2413 | DENTRY_D_LOCK_NESTED); |
2404 | } | 2414 | } |
2405 | } | 2415 | } |
2406 | if (target < dentry) { | 2416 | if (target < dentry) { |
2407 | spin_lock_nested(&target->d_lock, 2); | 2417 | spin_lock_nested(&target->d_lock, 2); |
2408 | spin_lock_nested(&dentry->d_lock, 3); | 2418 | spin_lock_nested(&dentry->d_lock, 3); |
2409 | } else { | 2419 | } else { |
2410 | spin_lock_nested(&dentry->d_lock, 2); | 2420 | spin_lock_nested(&dentry->d_lock, 2); |
2411 | spin_lock_nested(&target->d_lock, 3); | 2421 | spin_lock_nested(&target->d_lock, 3); |
2412 | } | 2422 | } |
2413 | } | 2423 | } |
2414 | 2424 | ||
2415 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | 2425 | static void dentry_unlock_parents_for_move(struct dentry *dentry, |
2416 | struct dentry *target) | 2426 | struct dentry *target) |
2417 | { | 2427 | { |
2418 | if (target->d_parent != dentry->d_parent) | 2428 | if (target->d_parent != dentry->d_parent) |
2419 | spin_unlock(&dentry->d_parent->d_lock); | 2429 | spin_unlock(&dentry->d_parent->d_lock); |
2420 | if (target->d_parent != target) | 2430 | if (target->d_parent != target) |
2421 | spin_unlock(&target->d_parent->d_lock); | 2431 | spin_unlock(&target->d_parent->d_lock); |
2422 | } | 2432 | } |
2423 | 2433 | ||
2424 | /* | 2434 | /* |
2425 | * When switching names, the actual string doesn't strictly have to | 2435 | * When switching names, the actual string doesn't strictly have to |
2426 | * be preserved in the target - because we're dropping the target | 2436 | * be preserved in the target - because we're dropping the target |
2427 | * anyway. As such, we can just do a simple memcpy() to copy over | 2437 | * anyway. As such, we can just do a simple memcpy() to copy over |
2428 | * the new name before we switch. | 2438 | * the new name before we switch. |
2429 | * | 2439 | * |
2430 | * Note that we have to be a lot more careful about getting the hash | 2440 | * Note that we have to be a lot more careful about getting the hash |
2431 | * switched - we have to switch the hash value properly even if it | 2441 | * switched - we have to switch the hash value properly even if it |
2432 | * then no longer matches the actual (corrupted) string of the target. | 2442 | * then no longer matches the actual (corrupted) string of the target. |
2433 | * The hash value has to match the hash queue that the dentry is on.. | 2443 | * The hash value has to match the hash queue that the dentry is on.. |
2434 | */ | 2444 | */ |
2435 | /* | 2445 | /* |
2436 | * __d_move - move a dentry | 2446 | * __d_move - move a dentry |
2437 | * @dentry: entry to move | 2447 | * @dentry: entry to move |
2438 | * @target: new dentry | 2448 | * @target: new dentry |
2439 | * @exchange: exchange the two dentries | 2449 | * @exchange: exchange the two dentries |
2440 | * | 2450 | * |
2441 | * Update the dcache to reflect the move of a file name. Negative | 2451 | * Update the dcache to reflect the move of a file name. Negative |
2442 | * dcache entries should not be moved in this way. Caller must hold | 2452 | * dcache entries should not be moved in this way. Caller must hold |
2443 | * rename_lock, the i_mutex of the source and target directories, | 2453 | * rename_lock, the i_mutex of the source and target directories, |
2444 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | 2454 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). |
2445 | */ | 2455 | */ |
2446 | static void __d_move(struct dentry *dentry, struct dentry *target, | 2456 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2447 | bool exchange) | 2457 | bool exchange) |
2448 | { | 2458 | { |
2449 | if (!dentry->d_inode) | 2459 | if (!dentry->d_inode) |
2450 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | 2460 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); |
2451 | 2461 | ||
2452 | BUG_ON(d_ancestor(dentry, target)); | 2462 | BUG_ON(d_ancestor(dentry, target)); |
2453 | BUG_ON(d_ancestor(target, dentry)); | 2463 | BUG_ON(d_ancestor(target, dentry)); |
2454 | 2464 | ||
2455 | dentry_lock_for_move(dentry, target); | 2465 | dentry_lock_for_move(dentry, target); |
2456 | 2466 | ||
2457 | write_seqcount_begin(&dentry->d_seq); | 2467 | write_seqcount_begin(&dentry->d_seq); |
2458 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); | 2468 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
2459 | 2469 | ||
2460 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ | 2470 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2461 | 2471 | ||
2462 | /* | 2472 | /* |
2463 | * Move the dentry to the target hash queue. Don't bother checking | 2473 | * Move the dentry to the target hash queue. Don't bother checking |
2464 | * for the same hash queue because of how unlikely it is. | 2474 | * for the same hash queue because of how unlikely it is. |
2465 | */ | 2475 | */ |
2466 | __d_drop(dentry); | 2476 | __d_drop(dentry); |
2467 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | 2477 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
2468 | 2478 | ||
2469 | /* | 2479 | /* |
2470 | * Unhash the target (d_delete() is not usable here). If exchanging | 2480 | * Unhash the target (d_delete() is not usable here). If exchanging |
2471 | * the two dentries, then rehash onto the other's hash queue. | 2481 | * the two dentries, then rehash onto the other's hash queue. |
2472 | */ | 2482 | */ |
2473 | __d_drop(target); | 2483 | __d_drop(target); |
2474 | if (exchange) { | 2484 | if (exchange) { |
2475 | __d_rehash(target, | 2485 | __d_rehash(target, |
2476 | d_hash(dentry->d_parent, dentry->d_name.hash)); | 2486 | d_hash(dentry->d_parent, dentry->d_name.hash)); |
2477 | } | 2487 | } |
2478 | 2488 | ||
2479 | list_del(&dentry->d_u.d_child); | 2489 | list_del(&dentry->d_u.d_child); |
2480 | list_del(&target->d_u.d_child); | 2490 | list_del(&target->d_u.d_child); |
2481 | 2491 | ||
2482 | /* Switch the names.. */ | 2492 | /* Switch the names.. */ |
2483 | switch_names(dentry, target); | 2493 | switch_names(dentry, target); |
2484 | swap(dentry->d_name.hash, target->d_name.hash); | 2494 | swap(dentry->d_name.hash, target->d_name.hash); |
2485 | 2495 | ||
2486 | /* ... and switch the parents */ | 2496 | /* ... and switch the parents */ |
2487 | if (IS_ROOT(dentry)) { | 2497 | if (IS_ROOT(dentry)) { |
2488 | dentry->d_parent = target->d_parent; | 2498 | dentry->d_parent = target->d_parent; |
2489 | target->d_parent = target; | 2499 | target->d_parent = target; |
2490 | INIT_LIST_HEAD(&target->d_u.d_child); | 2500 | INIT_LIST_HEAD(&target->d_u.d_child); |
2491 | } else { | 2501 | } else { |
2492 | swap(dentry->d_parent, target->d_parent); | 2502 | swap(dentry->d_parent, target->d_parent); |
2493 | 2503 | ||
2494 | /* And add them back to the (new) parent lists */ | 2504 | /* And add them back to the (new) parent lists */ |
2495 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); | 2505 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
2496 | } | 2506 | } |
2497 | 2507 | ||
2498 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | 2508 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2499 | 2509 | ||
2500 | write_seqcount_end(&target->d_seq); | 2510 | write_seqcount_end(&target->d_seq); |
2501 | write_seqcount_end(&dentry->d_seq); | 2511 | write_seqcount_end(&dentry->d_seq); |
2502 | 2512 | ||
2503 | dentry_unlock_parents_for_move(dentry, target); | 2513 | dentry_unlock_parents_for_move(dentry, target); |
2504 | if (exchange) | 2514 | if (exchange) |
2505 | fsnotify_d_move(target); | 2515 | fsnotify_d_move(target); |
2506 | spin_unlock(&target->d_lock); | 2516 | spin_unlock(&target->d_lock); |
2507 | fsnotify_d_move(dentry); | 2517 | fsnotify_d_move(dentry); |
2508 | spin_unlock(&dentry->d_lock); | 2518 | spin_unlock(&dentry->d_lock); |
2509 | } | 2519 | } |
2510 | 2520 | ||
2511 | /* | 2521 | /* |
2512 | * d_move - move a dentry | 2522 | * d_move - move a dentry |
2513 | * @dentry: entry to move | 2523 | * @dentry: entry to move |
2514 | * @target: new dentry | 2524 | * @target: new dentry |
2515 | * | 2525 | * |
2516 | * Update the dcache to reflect the move of a file name. Negative | 2526 | * Update the dcache to reflect the move of a file name. Negative |
2517 | * dcache entries should not be moved in this way. See the locking | 2527 | * dcache entries should not be moved in this way. See the locking |
2518 | * requirements for __d_move. | 2528 | * requirements for __d_move. |
2519 | */ | 2529 | */ |
2520 | void d_move(struct dentry *dentry, struct dentry *target) | 2530 | void d_move(struct dentry *dentry, struct dentry *target) |
2521 | { | 2531 | { |
2522 | write_seqlock(&rename_lock); | 2532 | write_seqlock(&rename_lock); |
2523 | __d_move(dentry, target, false); | 2533 | __d_move(dentry, target, false); |
2524 | write_sequnlock(&rename_lock); | 2534 | write_sequnlock(&rename_lock); |
2525 | } | 2535 | } |
2526 | EXPORT_SYMBOL(d_move); | 2536 | EXPORT_SYMBOL(d_move); |
2527 | 2537 | ||
2528 | /* | 2538 | /* |
2529 | * d_exchange - exchange two dentries | 2539 | * d_exchange - exchange two dentries |
2530 | * @dentry1: first dentry | 2540 | * @dentry1: first dentry |
2531 | * @dentry2: second dentry | 2541 | * @dentry2: second dentry |
2532 | */ | 2542 | */ |
2533 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | 2543 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) |
2534 | { | 2544 | { |
2535 | write_seqlock(&rename_lock); | 2545 | write_seqlock(&rename_lock); |
2536 | 2546 | ||
2537 | WARN_ON(!dentry1->d_inode); | 2547 | WARN_ON(!dentry1->d_inode); |
2538 | WARN_ON(!dentry2->d_inode); | 2548 | WARN_ON(!dentry2->d_inode); |
2539 | WARN_ON(IS_ROOT(dentry1)); | 2549 | WARN_ON(IS_ROOT(dentry1)); |
2540 | WARN_ON(IS_ROOT(dentry2)); | 2550 | WARN_ON(IS_ROOT(dentry2)); |
2541 | 2551 | ||
2542 | __d_move(dentry1, dentry2, true); | 2552 | __d_move(dentry1, dentry2, true); |
2543 | 2553 | ||
2544 | write_sequnlock(&rename_lock); | 2554 | write_sequnlock(&rename_lock); |
2545 | } | 2555 | } |
2546 | 2556 | ||
2547 | /** | 2557 | /** |
2548 | * d_ancestor - search for an ancestor | 2558 | * d_ancestor - search for an ancestor |
2549 | * @p1: ancestor dentry | 2559 | * @p1: ancestor dentry |
2550 | * @p2: child dentry | 2560 | * @p2: child dentry |
2551 | * | 2561 | * |
2552 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | 2562 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is |
2553 | * an ancestor of p2, else NULL. | 2563 | * an ancestor of p2, else NULL. |
2554 | */ | 2564 | */ |
2555 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) | 2565 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
2556 | { | 2566 | { |
2557 | struct dentry *p; | 2567 | struct dentry *p; |
2558 | 2568 | ||
2559 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { | 2569 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
2560 | if (p->d_parent == p1) | 2570 | if (p->d_parent == p1) |
2561 | return p; | 2571 | return p; |
2562 | } | 2572 | } |
2563 | return NULL; | 2573 | return NULL; |
2564 | } | 2574 | } |
2565 | 2575 | ||
2566 | /* | 2576 | /* |
2567 | * This helper attempts to cope with remotely renamed directories | 2577 | * This helper attempts to cope with remotely renamed directories |
2568 | * | 2578 | * |
2569 | * It assumes that the caller is already holding | 2579 | * It assumes that the caller is already holding |
2570 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock | 2580 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
2571 | * | 2581 | * |
2572 | * Note: If ever the locking in lock_rename() changes, then please | 2582 | * Note: If ever the locking in lock_rename() changes, then please |
2573 | * remember to update this too... | 2583 | * remember to update this too... |
2574 | */ | 2584 | */ |
2575 | static struct dentry *__d_unalias(struct inode *inode, | 2585 | static struct dentry *__d_unalias(struct inode *inode, |
2576 | struct dentry *dentry, struct dentry *alias) | 2586 | struct dentry *dentry, struct dentry *alias) |
2577 | { | 2587 | { |
2578 | struct mutex *m1 = NULL, *m2 = NULL; | 2588 | struct mutex *m1 = NULL, *m2 = NULL; |
2579 | struct dentry *ret = ERR_PTR(-EBUSY); | 2589 | struct dentry *ret = ERR_PTR(-EBUSY); |
2580 | 2590 | ||
2581 | /* If alias and dentry share a parent, then no extra locks required */ | 2591 | /* If alias and dentry share a parent, then no extra locks required */ |
2582 | if (alias->d_parent == dentry->d_parent) | 2592 | if (alias->d_parent == dentry->d_parent) |
2583 | goto out_unalias; | 2593 | goto out_unalias; |
2584 | 2594 | ||
2585 | /* See lock_rename() */ | 2595 | /* See lock_rename() */ |
2586 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | 2596 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2587 | goto out_err; | 2597 | goto out_err; |
2588 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | 2598 | m1 = &dentry->d_sb->s_vfs_rename_mutex; |
2589 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | 2599 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) |
2590 | goto out_err; | 2600 | goto out_err; |
2591 | m2 = &alias->d_parent->d_inode->i_mutex; | 2601 | m2 = &alias->d_parent->d_inode->i_mutex; |
2592 | out_unalias: | 2602 | out_unalias: |
2593 | if (likely(!d_mountpoint(alias))) { | 2603 | if (likely(!d_mountpoint(alias))) { |
2594 | __d_move(alias, dentry, false); | 2604 | __d_move(alias, dentry, false); |
2595 | ret = alias; | 2605 | ret = alias; |
2596 | } | 2606 | } |
2597 | out_err: | 2607 | out_err: |
2598 | spin_unlock(&inode->i_lock); | 2608 | spin_unlock(&inode->i_lock); |
2599 | if (m2) | 2609 | if (m2) |
2600 | mutex_unlock(m2); | 2610 | mutex_unlock(m2); |
2601 | if (m1) | 2611 | if (m1) |
2602 | mutex_unlock(m1); | 2612 | mutex_unlock(m1); |
2603 | return ret; | 2613 | return ret; |
2604 | } | 2614 | } |
2605 | 2615 | ||
2606 | /* | 2616 | /* |
2607 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | 2617 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a |
2608 | * named dentry in place of the dentry to be replaced. | 2618 | * named dentry in place of the dentry to be replaced. |
2609 | * returns with anon->d_lock held! | 2619 | * returns with anon->d_lock held! |
2610 | */ | 2620 | */ |
2611 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | 2621 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) |
2612 | { | 2622 | { |
2613 | struct dentry *dparent; | 2623 | struct dentry *dparent; |
2614 | 2624 | ||
2615 | dentry_lock_for_move(anon, dentry); | 2625 | dentry_lock_for_move(anon, dentry); |
2616 | 2626 | ||
2617 | write_seqcount_begin(&dentry->d_seq); | 2627 | write_seqcount_begin(&dentry->d_seq); |
2618 | write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED); | 2628 | write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED); |
2619 | 2629 | ||
2620 | dparent = dentry->d_parent; | 2630 | dparent = dentry->d_parent; |
2621 | 2631 | ||
2622 | switch_names(dentry, anon); | 2632 | switch_names(dentry, anon); |
2623 | swap(dentry->d_name.hash, anon->d_name.hash); | 2633 | swap(dentry->d_name.hash, anon->d_name.hash); |
2624 | 2634 | ||
2625 | dentry->d_parent = dentry; | 2635 | dentry->d_parent = dentry; |
2626 | list_del_init(&dentry->d_u.d_child); | 2636 | list_del_init(&dentry->d_u.d_child); |
2627 | anon->d_parent = dparent; | 2637 | anon->d_parent = dparent; |
2628 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); | 2638 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
2629 | 2639 | ||
2630 | write_seqcount_end(&dentry->d_seq); | 2640 | write_seqcount_end(&dentry->d_seq); |
2631 | write_seqcount_end(&anon->d_seq); | 2641 | write_seqcount_end(&anon->d_seq); |
2632 | 2642 | ||
2633 | dentry_unlock_parents_for_move(anon, dentry); | 2643 | dentry_unlock_parents_for_move(anon, dentry); |
2634 | spin_unlock(&dentry->d_lock); | 2644 | spin_unlock(&dentry->d_lock); |
2635 | 2645 | ||
2636 | /* anon->d_lock still locked, returns locked */ | 2646 | /* anon->d_lock still locked, returns locked */ |
2637 | } | 2647 | } |
2638 | 2648 | ||
2639 | /** | 2649 | /** |
2640 | * d_materialise_unique - introduce an inode into the tree | 2650 | * d_materialise_unique - introduce an inode into the tree |
2641 | * @dentry: candidate dentry | 2651 | * @dentry: candidate dentry |
2642 | * @inode: inode to bind to the dentry, to which aliases may be attached | 2652 | * @inode: inode to bind to the dentry, to which aliases may be attached |
2643 | * | 2653 | * |
2644 | * Introduces an dentry into the tree, substituting an extant disconnected | 2654 | * Introduces an dentry into the tree, substituting an extant disconnected |
2645 | * root directory alias in its place if there is one. Caller must hold the | 2655 | * root directory alias in its place if there is one. Caller must hold the |
2646 | * i_mutex of the parent directory. | 2656 | * i_mutex of the parent directory. |
2647 | */ | 2657 | */ |
2648 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | 2658 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) |
2649 | { | 2659 | { |
2650 | struct dentry *actual; | 2660 | struct dentry *actual; |
2651 | 2661 | ||
2652 | BUG_ON(!d_unhashed(dentry)); | 2662 | BUG_ON(!d_unhashed(dentry)); |
2653 | 2663 | ||
2654 | if (!inode) { | 2664 | if (!inode) { |
2655 | actual = dentry; | 2665 | actual = dentry; |
2656 | __d_instantiate(dentry, NULL); | 2666 | __d_instantiate(dentry, NULL); |
2657 | d_rehash(actual); | 2667 | d_rehash(actual); |
2658 | goto out_nolock; | 2668 | goto out_nolock; |
2659 | } | 2669 | } |
2660 | 2670 | ||
2661 | spin_lock(&inode->i_lock); | 2671 | spin_lock(&inode->i_lock); |
2662 | 2672 | ||
2663 | if (S_ISDIR(inode->i_mode)) { | 2673 | if (S_ISDIR(inode->i_mode)) { |
2664 | struct dentry *alias; | 2674 | struct dentry *alias; |
2665 | 2675 | ||
2666 | /* Does an aliased dentry already exist? */ | 2676 | /* Does an aliased dentry already exist? */ |
2667 | alias = __d_find_alias(inode, 0); | 2677 | alias = __d_find_alias(inode, 0); |
2668 | if (alias) { | 2678 | if (alias) { |
2669 | actual = alias; | 2679 | actual = alias; |
2670 | write_seqlock(&rename_lock); | 2680 | write_seqlock(&rename_lock); |
2671 | 2681 | ||
2672 | if (d_ancestor(alias, dentry)) { | 2682 | if (d_ancestor(alias, dentry)) { |
2673 | /* Check for loops */ | 2683 | /* Check for loops */ |
2674 | actual = ERR_PTR(-ELOOP); | 2684 | actual = ERR_PTR(-ELOOP); |
2675 | spin_unlock(&inode->i_lock); | 2685 | spin_unlock(&inode->i_lock); |
2676 | } else if (IS_ROOT(alias)) { | 2686 | } else if (IS_ROOT(alias)) { |
2677 | /* Is this an anonymous mountpoint that we | 2687 | /* Is this an anonymous mountpoint that we |
2678 | * could splice into our tree? */ | 2688 | * could splice into our tree? */ |
2679 | __d_materialise_dentry(dentry, alias); | 2689 | __d_materialise_dentry(dentry, alias); |
2680 | write_sequnlock(&rename_lock); | 2690 | write_sequnlock(&rename_lock); |
2681 | __d_drop(alias); | 2691 | __d_drop(alias); |
2682 | goto found; | 2692 | goto found; |
2683 | } else { | 2693 | } else { |
2684 | /* Nope, but we must(!) avoid directory | 2694 | /* Nope, but we must(!) avoid directory |
2685 | * aliasing. This drops inode->i_lock */ | 2695 | * aliasing. This drops inode->i_lock */ |
2686 | actual = __d_unalias(inode, dentry, alias); | 2696 | actual = __d_unalias(inode, dentry, alias); |
2687 | } | 2697 | } |
2688 | write_sequnlock(&rename_lock); | 2698 | write_sequnlock(&rename_lock); |
2689 | if (IS_ERR(actual)) { | 2699 | if (IS_ERR(actual)) { |
2690 | if (PTR_ERR(actual) == -ELOOP) | 2700 | if (PTR_ERR(actual) == -ELOOP) |
2691 | pr_warn_ratelimited( | 2701 | pr_warn_ratelimited( |
2692 | "VFS: Lookup of '%s' in %s %s" | 2702 | "VFS: Lookup of '%s' in %s %s" |
2693 | " would have caused loop\n", | 2703 | " would have caused loop\n", |
2694 | dentry->d_name.name, | 2704 | dentry->d_name.name, |
2695 | inode->i_sb->s_type->name, | 2705 | inode->i_sb->s_type->name, |
2696 | inode->i_sb->s_id); | 2706 | inode->i_sb->s_id); |
2697 | dput(alias); | 2707 | dput(alias); |
2698 | } | 2708 | } |
2699 | goto out_nolock; | 2709 | goto out_nolock; |
2700 | } | 2710 | } |
2701 | } | 2711 | } |
2702 | 2712 | ||
2703 | /* Add a unique reference */ | 2713 | /* Add a unique reference */ |
2704 | actual = __d_instantiate_unique(dentry, inode); | 2714 | actual = __d_instantiate_unique(dentry, inode); |
2705 | if (!actual) | 2715 | if (!actual) |
2706 | actual = dentry; | 2716 | actual = dentry; |
2707 | else | 2717 | else |
2708 | BUG_ON(!d_unhashed(actual)); | 2718 | BUG_ON(!d_unhashed(actual)); |
2709 | 2719 | ||
2710 | spin_lock(&actual->d_lock); | 2720 | spin_lock(&actual->d_lock); |
2711 | found: | 2721 | found: |
2712 | _d_rehash(actual); | 2722 | _d_rehash(actual); |
2713 | spin_unlock(&actual->d_lock); | 2723 | spin_unlock(&actual->d_lock); |
2714 | spin_unlock(&inode->i_lock); | 2724 | spin_unlock(&inode->i_lock); |
2715 | out_nolock: | 2725 | out_nolock: |
2716 | if (actual == dentry) { | 2726 | if (actual == dentry) { |
2717 | security_d_instantiate(dentry, inode); | 2727 | security_d_instantiate(dentry, inode); |
2718 | return NULL; | 2728 | return NULL; |
2719 | } | 2729 | } |
2720 | 2730 | ||
2721 | iput(inode); | 2731 | iput(inode); |
2722 | return actual; | 2732 | return actual; |
2723 | } | 2733 | } |
2724 | EXPORT_SYMBOL_GPL(d_materialise_unique); | 2734 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
2725 | 2735 | ||
2726 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) | 2736 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
2727 | { | 2737 | { |
2728 | *buflen -= namelen; | 2738 | *buflen -= namelen; |
2729 | if (*buflen < 0) | 2739 | if (*buflen < 0) |
2730 | return -ENAMETOOLONG; | 2740 | return -ENAMETOOLONG; |
2731 | *buffer -= namelen; | 2741 | *buffer -= namelen; |
2732 | memcpy(*buffer, str, namelen); | 2742 | memcpy(*buffer, str, namelen); |
2733 | return 0; | 2743 | return 0; |
2734 | } | 2744 | } |
2735 | 2745 | ||
2736 | /** | 2746 | /** |
2737 | * prepend_name - prepend a pathname in front of current buffer pointer | 2747 | * prepend_name - prepend a pathname in front of current buffer pointer |
2738 | * @buffer: buffer pointer | 2748 | * @buffer: buffer pointer |
2739 | * @buflen: allocated length of the buffer | 2749 | * @buflen: allocated length of the buffer |
2740 | * @name: name string and length qstr structure | 2750 | * @name: name string and length qstr structure |
2741 | * | 2751 | * |
2742 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | 2752 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to |
2743 | * make sure that either the old or the new name pointer and length are | 2753 | * make sure that either the old or the new name pointer and length are |
2744 | * fetched. However, there may be mismatch between length and pointer. | 2754 | * fetched. However, there may be mismatch between length and pointer. |
2745 | * The length cannot be trusted, we need to copy it byte-by-byte until | 2755 | * The length cannot be trusted, we need to copy it byte-by-byte until |
2746 | * the length is reached or a null byte is found. It also prepends "/" at | 2756 | * the length is reached or a null byte is found. It also prepends "/" at |
2747 | * the beginning of the name. The sequence number check at the caller will | 2757 | * the beginning of the name. The sequence number check at the caller will |
2748 | * retry it again when a d_move() does happen. So any garbage in the buffer | 2758 | * retry it again when a d_move() does happen. So any garbage in the buffer |
2749 | * due to mismatched pointer and length will be discarded. | 2759 | * due to mismatched pointer and length will be discarded. |
2750 | */ | 2760 | */ |
2751 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) | 2761 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2752 | { | 2762 | { |
2753 | const char *dname = ACCESS_ONCE(name->name); | 2763 | const char *dname = ACCESS_ONCE(name->name); |
2754 | u32 dlen = ACCESS_ONCE(name->len); | 2764 | u32 dlen = ACCESS_ONCE(name->len); |
2755 | char *p; | 2765 | char *p; |
2756 | 2766 | ||
2757 | *buflen -= dlen + 1; | 2767 | *buflen -= dlen + 1; |
2758 | if (*buflen < 0) | 2768 | if (*buflen < 0) |
2759 | return -ENAMETOOLONG; | 2769 | return -ENAMETOOLONG; |
2760 | p = *buffer -= dlen + 1; | 2770 | p = *buffer -= dlen + 1; |
2761 | *p++ = '/'; | 2771 | *p++ = '/'; |
2762 | while (dlen--) { | 2772 | while (dlen--) { |
2763 | char c = *dname++; | 2773 | char c = *dname++; |
2764 | if (!c) | 2774 | if (!c) |
2765 | break; | 2775 | break; |
2766 | *p++ = c; | 2776 | *p++ = c; |
2767 | } | 2777 | } |
2768 | return 0; | 2778 | return 0; |
2769 | } | 2779 | } |
2770 | 2780 | ||
2771 | /** | 2781 | /** |
2772 | * prepend_path - Prepend path string to a buffer | 2782 | * prepend_path - Prepend path string to a buffer |
2773 | * @path: the dentry/vfsmount to report | 2783 | * @path: the dentry/vfsmount to report |
2774 | * @root: root vfsmnt/dentry | 2784 | * @root: root vfsmnt/dentry |
2775 | * @buffer: pointer to the end of the buffer | 2785 | * @buffer: pointer to the end of the buffer |
2776 | * @buflen: pointer to buffer length | 2786 | * @buflen: pointer to buffer length |
2777 | * | 2787 | * |
2778 | * The function will first try to write out the pathname without taking any | 2788 | * The function will first try to write out the pathname without taking any |
2779 | * lock other than the RCU read lock to make sure that dentries won't go away. | 2789 | * lock other than the RCU read lock to make sure that dentries won't go away. |
2780 | * It only checks the sequence number of the global rename_lock as any change | 2790 | * It only checks the sequence number of the global rename_lock as any change |
2781 | * in the dentry's d_seq will be preceded by changes in the rename_lock | 2791 | * in the dentry's d_seq will be preceded by changes in the rename_lock |
2782 | * sequence number. If the sequence number had been changed, it will restart | 2792 | * sequence number. If the sequence number had been changed, it will restart |
2783 | * the whole pathname back-tracing sequence again by taking the rename_lock. | 2793 | * the whole pathname back-tracing sequence again by taking the rename_lock. |
2784 | * In this case, there is no need to take the RCU read lock as the recursive | 2794 | * In this case, there is no need to take the RCU read lock as the recursive |
2785 | * parent pointer references will keep the dentry chain alive as long as no | 2795 | * parent pointer references will keep the dentry chain alive as long as no |
2786 | * rename operation is performed. | 2796 | * rename operation is performed. |
2787 | */ | 2797 | */ |
2788 | static int prepend_path(const struct path *path, | 2798 | static int prepend_path(const struct path *path, |
2789 | const struct path *root, | 2799 | const struct path *root, |
2790 | char **buffer, int *buflen) | 2800 | char **buffer, int *buflen) |
2791 | { | 2801 | { |
2792 | struct dentry *dentry; | 2802 | struct dentry *dentry; |
2793 | struct vfsmount *vfsmnt; | 2803 | struct vfsmount *vfsmnt; |
2794 | struct mount *mnt; | 2804 | struct mount *mnt; |
2795 | int error = 0; | 2805 | int error = 0; |
2796 | unsigned seq, m_seq = 0; | 2806 | unsigned seq, m_seq = 0; |
2797 | char *bptr; | 2807 | char *bptr; |
2798 | int blen; | 2808 | int blen; |
2799 | 2809 | ||
2800 | rcu_read_lock(); | 2810 | rcu_read_lock(); |
2801 | restart_mnt: | 2811 | restart_mnt: |
2802 | read_seqbegin_or_lock(&mount_lock, &m_seq); | 2812 | read_seqbegin_or_lock(&mount_lock, &m_seq); |
2803 | seq = 0; | 2813 | seq = 0; |
2804 | rcu_read_lock(); | 2814 | rcu_read_lock(); |
2805 | restart: | 2815 | restart: |
2806 | bptr = *buffer; | 2816 | bptr = *buffer; |
2807 | blen = *buflen; | 2817 | blen = *buflen; |
2808 | error = 0; | 2818 | error = 0; |
2809 | dentry = path->dentry; | 2819 | dentry = path->dentry; |
2810 | vfsmnt = path->mnt; | 2820 | vfsmnt = path->mnt; |
2811 | mnt = real_mount(vfsmnt); | 2821 | mnt = real_mount(vfsmnt); |
2812 | read_seqbegin_or_lock(&rename_lock, &seq); | 2822 | read_seqbegin_or_lock(&rename_lock, &seq); |
2813 | while (dentry != root->dentry || vfsmnt != root->mnt) { | 2823 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
2814 | struct dentry * parent; | 2824 | struct dentry * parent; |
2815 | 2825 | ||
2816 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { | 2826 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
2817 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); | 2827 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
2818 | /* Global root? */ | 2828 | /* Global root? */ |
2819 | if (mnt != parent) { | 2829 | if (mnt != parent) { |
2820 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | 2830 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); |
2821 | mnt = parent; | 2831 | mnt = parent; |
2822 | vfsmnt = &mnt->mnt; | 2832 | vfsmnt = &mnt->mnt; |
2823 | continue; | 2833 | continue; |
2824 | } | 2834 | } |
2825 | /* | 2835 | /* |
2826 | * Filesystems needing to implement special "root names" | 2836 | * Filesystems needing to implement special "root names" |
2827 | * should do so with ->d_dname() | 2837 | * should do so with ->d_dname() |
2828 | */ | 2838 | */ |
2829 | if (IS_ROOT(dentry) && | 2839 | if (IS_ROOT(dentry) && |
2830 | (dentry->d_name.len != 1 || | 2840 | (dentry->d_name.len != 1 || |
2831 | dentry->d_name.name[0] != '/')) { | 2841 | dentry->d_name.name[0] != '/')) { |
2832 | WARN(1, "Root dentry has weird name <%.*s>\n", | 2842 | WARN(1, "Root dentry has weird name <%.*s>\n", |
2833 | (int) dentry->d_name.len, | 2843 | (int) dentry->d_name.len, |
2834 | dentry->d_name.name); | 2844 | dentry->d_name.name); |
2835 | } | 2845 | } |
2836 | if (!error) | 2846 | if (!error) |
2837 | error = is_mounted(vfsmnt) ? 1 : 2; | 2847 | error = is_mounted(vfsmnt) ? 1 : 2; |
2838 | break; | 2848 | break; |
2839 | } | 2849 | } |
2840 | parent = dentry->d_parent; | 2850 | parent = dentry->d_parent; |
2841 | prefetch(parent); | 2851 | prefetch(parent); |
2842 | error = prepend_name(&bptr, &blen, &dentry->d_name); | 2852 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
2843 | if (error) | 2853 | if (error) |
2844 | break; | 2854 | break; |
2845 | 2855 | ||
2846 | dentry = parent; | 2856 | dentry = parent; |
2847 | } | 2857 | } |
2848 | if (!(seq & 1)) | 2858 | if (!(seq & 1)) |
2849 | rcu_read_unlock(); | 2859 | rcu_read_unlock(); |
2850 | if (need_seqretry(&rename_lock, seq)) { | 2860 | if (need_seqretry(&rename_lock, seq)) { |
2851 | seq = 1; | 2861 | seq = 1; |
2852 | goto restart; | 2862 | goto restart; |
2853 | } | 2863 | } |
2854 | done_seqretry(&rename_lock, seq); | 2864 | done_seqretry(&rename_lock, seq); |
2855 | 2865 | ||
2856 | if (!(m_seq & 1)) | 2866 | if (!(m_seq & 1)) |
2857 | rcu_read_unlock(); | 2867 | rcu_read_unlock(); |
2858 | if (need_seqretry(&mount_lock, m_seq)) { | 2868 | if (need_seqretry(&mount_lock, m_seq)) { |
2859 | m_seq = 1; | 2869 | m_seq = 1; |
2860 | goto restart_mnt; | 2870 | goto restart_mnt; |
2861 | } | 2871 | } |
2862 | done_seqretry(&mount_lock, m_seq); | 2872 | done_seqretry(&mount_lock, m_seq); |
2863 | 2873 | ||
2864 | if (error >= 0 && bptr == *buffer) { | 2874 | if (error >= 0 && bptr == *buffer) { |
2865 | if (--blen < 0) | 2875 | if (--blen < 0) |
2866 | error = -ENAMETOOLONG; | 2876 | error = -ENAMETOOLONG; |
2867 | else | 2877 | else |
2868 | *--bptr = '/'; | 2878 | *--bptr = '/'; |
2869 | } | 2879 | } |
2870 | *buffer = bptr; | 2880 | *buffer = bptr; |
2871 | *buflen = blen; | 2881 | *buflen = blen; |
2872 | return error; | 2882 | return error; |
2873 | } | 2883 | } |
2874 | 2884 | ||
2875 | /** | 2885 | /** |
2876 | * __d_path - return the path of a dentry | 2886 | * __d_path - return the path of a dentry |
2877 | * @path: the dentry/vfsmount to report | 2887 | * @path: the dentry/vfsmount to report |
2878 | * @root: root vfsmnt/dentry | 2888 | * @root: root vfsmnt/dentry |
2879 | * @buf: buffer to return value in | 2889 | * @buf: buffer to return value in |
2880 | * @buflen: buffer length | 2890 | * @buflen: buffer length |
2881 | * | 2891 | * |
2882 | * Convert a dentry into an ASCII path name. | 2892 | * Convert a dentry into an ASCII path name. |
2883 | * | 2893 | * |
2884 | * Returns a pointer into the buffer or an error code if the | 2894 | * Returns a pointer into the buffer or an error code if the |
2885 | * path was too long. | 2895 | * path was too long. |
2886 | * | 2896 | * |
2887 | * "buflen" should be positive. | 2897 | * "buflen" should be positive. |
2888 | * | 2898 | * |
2889 | * If the path is not reachable from the supplied root, return %NULL. | 2899 | * If the path is not reachable from the supplied root, return %NULL. |
2890 | */ | 2900 | */ |
2891 | char *__d_path(const struct path *path, | 2901 | char *__d_path(const struct path *path, |
2892 | const struct path *root, | 2902 | const struct path *root, |
2893 | char *buf, int buflen) | 2903 | char *buf, int buflen) |
2894 | { | 2904 | { |
2895 | char *res = buf + buflen; | 2905 | char *res = buf + buflen; |
2896 | int error; | 2906 | int error; |
2897 | 2907 | ||
2898 | prepend(&res, &buflen, "\0", 1); | 2908 | prepend(&res, &buflen, "\0", 1); |
2899 | error = prepend_path(path, root, &res, &buflen); | 2909 | error = prepend_path(path, root, &res, &buflen); |
2900 | 2910 | ||
2901 | if (error < 0) | 2911 | if (error < 0) |
2902 | return ERR_PTR(error); | 2912 | return ERR_PTR(error); |
2903 | if (error > 0) | 2913 | if (error > 0) |
2904 | return NULL; | 2914 | return NULL; |
2905 | return res; | 2915 | return res; |
2906 | } | 2916 | } |
2907 | 2917 | ||
2908 | char *d_absolute_path(const struct path *path, | 2918 | char *d_absolute_path(const struct path *path, |
2909 | char *buf, int buflen) | 2919 | char *buf, int buflen) |
2910 | { | 2920 | { |
2911 | struct path root = {}; | 2921 | struct path root = {}; |
2912 | char *res = buf + buflen; | 2922 | char *res = buf + buflen; |
2913 | int error; | 2923 | int error; |
2914 | 2924 | ||
2915 | prepend(&res, &buflen, "\0", 1); | 2925 | prepend(&res, &buflen, "\0", 1); |
2916 | error = prepend_path(path, &root, &res, &buflen); | 2926 | error = prepend_path(path, &root, &res, &buflen); |
2917 | 2927 | ||
2918 | if (error > 1) | 2928 | if (error > 1) |
2919 | error = -EINVAL; | 2929 | error = -EINVAL; |
2920 | if (error < 0) | 2930 | if (error < 0) |
2921 | return ERR_PTR(error); | 2931 | return ERR_PTR(error); |
2922 | return res; | 2932 | return res; |
2923 | } | 2933 | } |
2924 | 2934 | ||
2925 | /* | 2935 | /* |
2926 | * same as __d_path but appends "(deleted)" for unlinked files. | 2936 | * same as __d_path but appends "(deleted)" for unlinked files. |
2927 | */ | 2937 | */ |
2928 | static int path_with_deleted(const struct path *path, | 2938 | static int path_with_deleted(const struct path *path, |
2929 | const struct path *root, | 2939 | const struct path *root, |
2930 | char **buf, int *buflen) | 2940 | char **buf, int *buflen) |
2931 | { | 2941 | { |
2932 | prepend(buf, buflen, "\0", 1); | 2942 | prepend(buf, buflen, "\0", 1); |
2933 | if (d_unlinked(path->dentry)) { | 2943 | if (d_unlinked(path->dentry)) { |
2934 | int error = prepend(buf, buflen, " (deleted)", 10); | 2944 | int error = prepend(buf, buflen, " (deleted)", 10); |
2935 | if (error) | 2945 | if (error) |
2936 | return error; | 2946 | return error; |
2937 | } | 2947 | } |
2938 | 2948 | ||
2939 | return prepend_path(path, root, buf, buflen); | 2949 | return prepend_path(path, root, buf, buflen); |
2940 | } | 2950 | } |
2941 | 2951 | ||
2942 | static int prepend_unreachable(char **buffer, int *buflen) | 2952 | static int prepend_unreachable(char **buffer, int *buflen) |
2943 | { | 2953 | { |
2944 | return prepend(buffer, buflen, "(unreachable)", 13); | 2954 | return prepend(buffer, buflen, "(unreachable)", 13); |
2945 | } | 2955 | } |
2946 | 2956 | ||
2947 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) | 2957 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
2948 | { | 2958 | { |
2949 | unsigned seq; | 2959 | unsigned seq; |
2950 | 2960 | ||
2951 | do { | 2961 | do { |
2952 | seq = read_seqcount_begin(&fs->seq); | 2962 | seq = read_seqcount_begin(&fs->seq); |
2953 | *root = fs->root; | 2963 | *root = fs->root; |
2954 | } while (read_seqcount_retry(&fs->seq, seq)); | 2964 | } while (read_seqcount_retry(&fs->seq, seq)); |
2955 | } | 2965 | } |
2956 | 2966 | ||
2957 | /** | 2967 | /** |
2958 | * d_path - return the path of a dentry | 2968 | * d_path - return the path of a dentry |
2959 | * @path: path to report | 2969 | * @path: path to report |
2960 | * @buf: buffer to return value in | 2970 | * @buf: buffer to return value in |
2961 | * @buflen: buffer length | 2971 | * @buflen: buffer length |
2962 | * | 2972 | * |
2963 | * Convert a dentry into an ASCII path name. If the entry has been deleted | 2973 | * Convert a dentry into an ASCII path name. If the entry has been deleted |
2964 | * the string " (deleted)" is appended. Note that this is ambiguous. | 2974 | * the string " (deleted)" is appended. Note that this is ambiguous. |
2965 | * | 2975 | * |
2966 | * Returns a pointer into the buffer or an error code if the path was | 2976 | * Returns a pointer into the buffer or an error code if the path was |
2967 | * too long. Note: Callers should use the returned pointer, not the passed | 2977 | * too long. Note: Callers should use the returned pointer, not the passed |
2968 | * in buffer, to use the name! The implementation often starts at an offset | 2978 | * in buffer, to use the name! The implementation often starts at an offset |
2969 | * into the buffer, and may leave 0 bytes at the start. | 2979 | * into the buffer, and may leave 0 bytes at the start. |
2970 | * | 2980 | * |
2971 | * "buflen" should be positive. | 2981 | * "buflen" should be positive. |
2972 | */ | 2982 | */ |
2973 | char *d_path(const struct path *path, char *buf, int buflen) | 2983 | char *d_path(const struct path *path, char *buf, int buflen) |
2974 | { | 2984 | { |
2975 | char *res = buf + buflen; | 2985 | char *res = buf + buflen; |
2976 | struct path root; | 2986 | struct path root; |
2977 | int error; | 2987 | int error; |
2978 | 2988 | ||
2979 | /* | 2989 | /* |
2980 | * We have various synthetic filesystems that never get mounted. On | 2990 | * We have various synthetic filesystems that never get mounted. On |
2981 | * these filesystems dentries are never used for lookup purposes, and | 2991 | * these filesystems dentries are never used for lookup purposes, and |
2982 | * thus don't need to be hashed. They also don't need a name until a | 2992 | * thus don't need to be hashed. They also don't need a name until a |
2983 | * user wants to identify the object in /proc/pid/fd/. The little hack | 2993 | * user wants to identify the object in /proc/pid/fd/. The little hack |
2984 | * below allows us to generate a name for these objects on demand: | 2994 | * below allows us to generate a name for these objects on demand: |
2985 | * | 2995 | * |
2986 | * Some pseudo inodes are mountable. When they are mounted | 2996 | * Some pseudo inodes are mountable. When they are mounted |
2987 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | 2997 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname |
2988 | * and instead have d_path return the mounted path. | 2998 | * and instead have d_path return the mounted path. |
2989 | */ | 2999 | */ |
2990 | if (path->dentry->d_op && path->dentry->d_op->d_dname && | 3000 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
2991 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | 3001 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) |
2992 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | 3002 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
2993 | 3003 | ||
2994 | rcu_read_lock(); | 3004 | rcu_read_lock(); |
2995 | get_fs_root_rcu(current->fs, &root); | 3005 | get_fs_root_rcu(current->fs, &root); |
2996 | error = path_with_deleted(path, &root, &res, &buflen); | 3006 | error = path_with_deleted(path, &root, &res, &buflen); |
2997 | rcu_read_unlock(); | 3007 | rcu_read_unlock(); |
2998 | 3008 | ||
2999 | if (error < 0) | 3009 | if (error < 0) |
3000 | res = ERR_PTR(error); | 3010 | res = ERR_PTR(error); |
3001 | return res; | 3011 | return res; |
3002 | } | 3012 | } |
3003 | EXPORT_SYMBOL(d_path); | 3013 | EXPORT_SYMBOL(d_path); |
3004 | 3014 | ||
3005 | /* | 3015 | /* |
3006 | * Helper function for dentry_operations.d_dname() members | 3016 | * Helper function for dentry_operations.d_dname() members |
3007 | */ | 3017 | */ |
3008 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | 3018 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, |
3009 | const char *fmt, ...) | 3019 | const char *fmt, ...) |
3010 | { | 3020 | { |
3011 | va_list args; | 3021 | va_list args; |
3012 | char temp[64]; | 3022 | char temp[64]; |
3013 | int sz; | 3023 | int sz; |
3014 | 3024 | ||
3015 | va_start(args, fmt); | 3025 | va_start(args, fmt); |
3016 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | 3026 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; |
3017 | va_end(args); | 3027 | va_end(args); |
3018 | 3028 | ||
3019 | if (sz > sizeof(temp) || sz > buflen) | 3029 | if (sz > sizeof(temp) || sz > buflen) |
3020 | return ERR_PTR(-ENAMETOOLONG); | 3030 | return ERR_PTR(-ENAMETOOLONG); |
3021 | 3031 | ||
3022 | buffer += buflen - sz; | 3032 | buffer += buflen - sz; |
3023 | return memcpy(buffer, temp, sz); | 3033 | return memcpy(buffer, temp, sz); |
3024 | } | 3034 | } |
3025 | 3035 | ||
3026 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) | 3036 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3027 | { | 3037 | { |
3028 | char *end = buffer + buflen; | 3038 | char *end = buffer + buflen; |
3029 | /* these dentries are never renamed, so d_lock is not needed */ | 3039 | /* these dentries are never renamed, so d_lock is not needed */ |
3030 | if (prepend(&end, &buflen, " (deleted)", 11) || | 3040 | if (prepend(&end, &buflen, " (deleted)", 11) || |
3031 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || | 3041 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
3032 | prepend(&end, &buflen, "/", 1)) | 3042 | prepend(&end, &buflen, "/", 1)) |
3033 | end = ERR_PTR(-ENAMETOOLONG); | 3043 | end = ERR_PTR(-ENAMETOOLONG); |
3034 | return end; | 3044 | return end; |
3035 | } | 3045 | } |
3036 | EXPORT_SYMBOL(simple_dname); | 3046 | EXPORT_SYMBOL(simple_dname); |
3037 | 3047 | ||
3038 | /* | 3048 | /* |
3039 | * Write full pathname from the root of the filesystem into the buffer. | 3049 | * Write full pathname from the root of the filesystem into the buffer. |
3040 | */ | 3050 | */ |
3041 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) | 3051 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
3042 | { | 3052 | { |
3043 | struct dentry *dentry; | 3053 | struct dentry *dentry; |
3044 | char *end, *retval; | 3054 | char *end, *retval; |
3045 | int len, seq = 0; | 3055 | int len, seq = 0; |
3046 | int error = 0; | 3056 | int error = 0; |
3047 | 3057 | ||
3048 | if (buflen < 2) | 3058 | if (buflen < 2) |
3049 | goto Elong; | 3059 | goto Elong; |
3050 | 3060 | ||
3051 | rcu_read_lock(); | 3061 | rcu_read_lock(); |
3052 | restart: | 3062 | restart: |
3053 | dentry = d; | 3063 | dentry = d; |
3054 | end = buf + buflen; | 3064 | end = buf + buflen; |
3055 | len = buflen; | 3065 | len = buflen; |
3056 | prepend(&end, &len, "\0", 1); | 3066 | prepend(&end, &len, "\0", 1); |
3057 | /* Get '/' right */ | 3067 | /* Get '/' right */ |
3058 | retval = end-1; | 3068 | retval = end-1; |
3059 | *retval = '/'; | 3069 | *retval = '/'; |
3060 | read_seqbegin_or_lock(&rename_lock, &seq); | 3070 | read_seqbegin_or_lock(&rename_lock, &seq); |
3061 | while (!IS_ROOT(dentry)) { | 3071 | while (!IS_ROOT(dentry)) { |
3062 | struct dentry *parent = dentry->d_parent; | 3072 | struct dentry *parent = dentry->d_parent; |
3063 | 3073 | ||
3064 | prefetch(parent); | 3074 | prefetch(parent); |
3065 | error = prepend_name(&end, &len, &dentry->d_name); | 3075 | error = prepend_name(&end, &len, &dentry->d_name); |
3066 | if (error) | 3076 | if (error) |
3067 | break; | 3077 | break; |
3068 | 3078 | ||
3069 | retval = end; | 3079 | retval = end; |
3070 | dentry = parent; | 3080 | dentry = parent; |
3071 | } | 3081 | } |
3072 | if (!(seq & 1)) | 3082 | if (!(seq & 1)) |
3073 | rcu_read_unlock(); | 3083 | rcu_read_unlock(); |
3074 | if (need_seqretry(&rename_lock, seq)) { | 3084 | if (need_seqretry(&rename_lock, seq)) { |
3075 | seq = 1; | 3085 | seq = 1; |
3076 | goto restart; | 3086 | goto restart; |
3077 | } | 3087 | } |
3078 | done_seqretry(&rename_lock, seq); | 3088 | done_seqretry(&rename_lock, seq); |
3079 | if (error) | 3089 | if (error) |
3080 | goto Elong; | 3090 | goto Elong; |
3081 | return retval; | 3091 | return retval; |
3082 | Elong: | 3092 | Elong: |
3083 | return ERR_PTR(-ENAMETOOLONG); | 3093 | return ERR_PTR(-ENAMETOOLONG); |
3084 | } | 3094 | } |
3085 | 3095 | ||
3086 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | 3096 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) |
3087 | { | 3097 | { |
3088 | return __dentry_path(dentry, buf, buflen); | 3098 | return __dentry_path(dentry, buf, buflen); |
3089 | } | 3099 | } |
3090 | EXPORT_SYMBOL(dentry_path_raw); | 3100 | EXPORT_SYMBOL(dentry_path_raw); |
3091 | 3101 | ||
3092 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | 3102 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) |
3093 | { | 3103 | { |
3094 | char *p = NULL; | 3104 | char *p = NULL; |
3095 | char *retval; | 3105 | char *retval; |
3096 | 3106 | ||
3097 | if (d_unlinked(dentry)) { | 3107 | if (d_unlinked(dentry)) { |
3098 | p = buf + buflen; | 3108 | p = buf + buflen; |
3099 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | 3109 | if (prepend(&p, &buflen, "//deleted", 10) != 0) |
3100 | goto Elong; | 3110 | goto Elong; |
3101 | buflen++; | 3111 | buflen++; |
3102 | } | 3112 | } |
3103 | retval = __dentry_path(dentry, buf, buflen); | 3113 | retval = __dentry_path(dentry, buf, buflen); |
3104 | if (!IS_ERR(retval) && p) | 3114 | if (!IS_ERR(retval) && p) |
3105 | *p = '/'; /* restore '/' overriden with '\0' */ | 3115 | *p = '/'; /* restore '/' overriden with '\0' */ |
3106 | return retval; | 3116 | return retval; |
3107 | Elong: | 3117 | Elong: |
3108 | return ERR_PTR(-ENAMETOOLONG); | 3118 | return ERR_PTR(-ENAMETOOLONG); |
3109 | } | 3119 | } |
3110 | 3120 | ||
3111 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, | 3121 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3112 | struct path *pwd) | 3122 | struct path *pwd) |
3113 | { | 3123 | { |
3114 | unsigned seq; | 3124 | unsigned seq; |
3115 | 3125 | ||
3116 | do { | 3126 | do { |
3117 | seq = read_seqcount_begin(&fs->seq); | 3127 | seq = read_seqcount_begin(&fs->seq); |
3118 | *root = fs->root; | 3128 | *root = fs->root; |
3119 | *pwd = fs->pwd; | 3129 | *pwd = fs->pwd; |
3120 | } while (read_seqcount_retry(&fs->seq, seq)); | 3130 | } while (read_seqcount_retry(&fs->seq, seq)); |
3121 | } | 3131 | } |
3122 | 3132 | ||
3123 | /* | 3133 | /* |
3124 | * NOTE! The user-level library version returns a | 3134 | * NOTE! The user-level library version returns a |
3125 | * character pointer. The kernel system call just | 3135 | * character pointer. The kernel system call just |
3126 | * returns the length of the buffer filled (which | 3136 | * returns the length of the buffer filled (which |
3127 | * includes the ending '\0' character), or a negative | 3137 | * includes the ending '\0' character), or a negative |
3128 | * error value. So libc would do something like | 3138 | * error value. So libc would do something like |
3129 | * | 3139 | * |
3130 | * char *getcwd(char * buf, size_t size) | 3140 | * char *getcwd(char * buf, size_t size) |
3131 | * { | 3141 | * { |
3132 | * int retval; | 3142 | * int retval; |
3133 | * | 3143 | * |
3134 | * retval = sys_getcwd(buf, size); | 3144 | * retval = sys_getcwd(buf, size); |
3135 | * if (retval >= 0) | 3145 | * if (retval >= 0) |
3136 | * return buf; | 3146 | * return buf; |
3137 | * errno = -retval; | 3147 | * errno = -retval; |
3138 | * return NULL; | 3148 | * return NULL; |
3139 | * } | 3149 | * } |
3140 | */ | 3150 | */ |
3141 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) | 3151 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
3142 | { | 3152 | { |
3143 | int error; | 3153 | int error; |
3144 | struct path pwd, root; | 3154 | struct path pwd, root; |
3145 | char *page = __getname(); | 3155 | char *page = __getname(); |
3146 | 3156 | ||
3147 | if (!page) | 3157 | if (!page) |
3148 | return -ENOMEM; | 3158 | return -ENOMEM; |
3149 | 3159 | ||
3150 | rcu_read_lock(); | 3160 | rcu_read_lock(); |
3151 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | 3161 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); |
3152 | 3162 | ||
3153 | error = -ENOENT; | 3163 | error = -ENOENT; |
3154 | if (!d_unlinked(pwd.dentry)) { | 3164 | if (!d_unlinked(pwd.dentry)) { |
3155 | unsigned long len; | 3165 | unsigned long len; |
3156 | char *cwd = page + PATH_MAX; | 3166 | char *cwd = page + PATH_MAX; |
3157 | int buflen = PATH_MAX; | 3167 | int buflen = PATH_MAX; |
3158 | 3168 | ||
3159 | prepend(&cwd, &buflen, "\0", 1); | 3169 | prepend(&cwd, &buflen, "\0", 1); |
3160 | error = prepend_path(&pwd, &root, &cwd, &buflen); | 3170 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
3161 | rcu_read_unlock(); | 3171 | rcu_read_unlock(); |
3162 | 3172 | ||
3163 | if (error < 0) | 3173 | if (error < 0) |
3164 | goto out; | 3174 | goto out; |
3165 | 3175 | ||
3166 | /* Unreachable from current root */ | 3176 | /* Unreachable from current root */ |
3167 | if (error > 0) { | 3177 | if (error > 0) { |
3168 | error = prepend_unreachable(&cwd, &buflen); | 3178 | error = prepend_unreachable(&cwd, &buflen); |
3169 | if (error) | 3179 | if (error) |
3170 | goto out; | 3180 | goto out; |
3171 | } | 3181 | } |
3172 | 3182 | ||
3173 | error = -ERANGE; | 3183 | error = -ERANGE; |
3174 | len = PATH_MAX + page - cwd; | 3184 | len = PATH_MAX + page - cwd; |
3175 | if (len <= size) { | 3185 | if (len <= size) { |
3176 | error = len; | 3186 | error = len; |
3177 | if (copy_to_user(buf, cwd, len)) | 3187 | if (copy_to_user(buf, cwd, len)) |
3178 | error = -EFAULT; | 3188 | error = -EFAULT; |
3179 | } | 3189 | } |
3180 | } else { | 3190 | } else { |
3181 | rcu_read_unlock(); | 3191 | rcu_read_unlock(); |
3182 | } | 3192 | } |
3183 | 3193 | ||
3184 | out: | 3194 | out: |
3185 | __putname(page); | 3195 | __putname(page); |
3186 | return error; | 3196 | return error; |
3187 | } | 3197 | } |
3188 | 3198 | ||
3189 | /* | 3199 | /* |
3190 | * Test whether new_dentry is a subdirectory of old_dentry. | 3200 | * Test whether new_dentry is a subdirectory of old_dentry. |
3191 | * | 3201 | * |
3192 | * Trivially implemented using the dcache structure | 3202 | * Trivially implemented using the dcache structure |
3193 | */ | 3203 | */ |
3194 | 3204 | ||
3195 | /** | 3205 | /** |
3196 | * is_subdir - is new dentry a subdirectory of old_dentry | 3206 | * is_subdir - is new dentry a subdirectory of old_dentry |
3197 | * @new_dentry: new dentry | 3207 | * @new_dentry: new dentry |
3198 | * @old_dentry: old dentry | 3208 | * @old_dentry: old dentry |
3199 | * | 3209 | * |
3200 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | 3210 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). |
3201 | * Returns 0 otherwise. | 3211 | * Returns 0 otherwise. |
3202 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | 3212 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3203 | */ | 3213 | */ |
3204 | 3214 | ||
3205 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) | 3215 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
3206 | { | 3216 | { |
3207 | int result; | 3217 | int result; |
3208 | unsigned seq; | 3218 | unsigned seq; |
3209 | 3219 | ||
3210 | if (new_dentry == old_dentry) | 3220 | if (new_dentry == old_dentry) |
3211 | return 1; | 3221 | return 1; |
3212 | 3222 | ||
3213 | do { | 3223 | do { |
3214 | /* for restarting inner loop in case of seq retry */ | 3224 | /* for restarting inner loop in case of seq retry */ |
3215 | seq = read_seqbegin(&rename_lock); | 3225 | seq = read_seqbegin(&rename_lock); |
3216 | /* | 3226 | /* |
3217 | * Need rcu_readlock to protect against the d_parent trashing | 3227 | * Need rcu_readlock to protect against the d_parent trashing |
3218 | * due to d_move | 3228 | * due to d_move |
3219 | */ | 3229 | */ |
3220 | rcu_read_lock(); | 3230 | rcu_read_lock(); |
3221 | if (d_ancestor(old_dentry, new_dentry)) | 3231 | if (d_ancestor(old_dentry, new_dentry)) |
3222 | result = 1; | 3232 | result = 1; |
3223 | else | 3233 | else |
3224 | result = 0; | 3234 | result = 0; |
3225 | rcu_read_unlock(); | 3235 | rcu_read_unlock(); |
3226 | } while (read_seqretry(&rename_lock, seq)); | 3236 | } while (read_seqretry(&rename_lock, seq)); |
3227 | 3237 | ||
3228 | return result; | 3238 | return result; |
3229 | } | 3239 | } |
3230 | 3240 | ||
3231 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) | 3241 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
3232 | { | 3242 | { |
3233 | struct dentry *root = data; | 3243 | struct dentry *root = data; |
3234 | if (dentry != root) { | 3244 | if (dentry != root) { |
3235 | if (d_unhashed(dentry) || !dentry->d_inode) | 3245 | if (d_unhashed(dentry) || !dentry->d_inode) |
3236 | return D_WALK_SKIP; | 3246 | return D_WALK_SKIP; |
3237 | 3247 | ||
3238 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { | 3248 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3239 | dentry->d_flags |= DCACHE_GENOCIDE; | 3249 | dentry->d_flags |= DCACHE_GENOCIDE; |
3240 | dentry->d_lockref.count--; | 3250 | dentry->d_lockref.count--; |
3241 | } | 3251 | } |
3242 | } | 3252 | } |
3243 | return D_WALK_CONTINUE; | 3253 | return D_WALK_CONTINUE; |
3244 | } | 3254 | } |
3245 | 3255 | ||
3246 | void d_genocide(struct dentry *parent) | 3256 | void d_genocide(struct dentry *parent) |
3247 | { | 3257 | { |
3248 | d_walk(parent, parent, d_genocide_kill, NULL); | 3258 | d_walk(parent, parent, d_genocide_kill, NULL); |
3249 | } | 3259 | } |
3250 | 3260 | ||
3251 | void d_tmpfile(struct dentry *dentry, struct inode *inode) | 3261 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
3252 | { | 3262 | { |
3253 | inode_dec_link_count(inode); | 3263 | inode_dec_link_count(inode); |
3254 | BUG_ON(dentry->d_name.name != dentry->d_iname || | 3264 | BUG_ON(dentry->d_name.name != dentry->d_iname || |
3255 | !hlist_unhashed(&dentry->d_alias) || | 3265 | !hlist_unhashed(&dentry->d_alias) || |
3256 | !d_unlinked(dentry)); | 3266 | !d_unlinked(dentry)); |
3257 | spin_lock(&dentry->d_parent->d_lock); | 3267 | spin_lock(&dentry->d_parent->d_lock); |
3258 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | 3268 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
3259 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | 3269 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", |
3260 | (unsigned long long)inode->i_ino); | 3270 | (unsigned long long)inode->i_ino); |
3261 | spin_unlock(&dentry->d_lock); | 3271 | spin_unlock(&dentry->d_lock); |
3262 | spin_unlock(&dentry->d_parent->d_lock); | 3272 | spin_unlock(&dentry->d_parent->d_lock); |
3263 | d_instantiate(dentry, inode); | 3273 | d_instantiate(dentry, inode); |
3264 | } | 3274 | } |
3265 | EXPORT_SYMBOL(d_tmpfile); | 3275 | EXPORT_SYMBOL(d_tmpfile); |
3266 | 3276 | ||
3267 | static __initdata unsigned long dhash_entries; | 3277 | static __initdata unsigned long dhash_entries; |
3268 | static int __init set_dhash_entries(char *str) | 3278 | static int __init set_dhash_entries(char *str) |
3269 | { | 3279 | { |
3270 | if (!str) | 3280 | if (!str) |
3271 | return 0; | 3281 | return 0; |
3272 | dhash_entries = simple_strtoul(str, &str, 0); | 3282 | dhash_entries = simple_strtoul(str, &str, 0); |
3273 | return 1; | 3283 | return 1; |
3274 | } | 3284 | } |
3275 | __setup("dhash_entries=", set_dhash_entries); | 3285 | __setup("dhash_entries=", set_dhash_entries); |
3276 | 3286 | ||
3277 | static void __init dcache_init_early(void) | 3287 | static void __init dcache_init_early(void) |
3278 | { | 3288 | { |
3279 | unsigned int loop; | 3289 | unsigned int loop; |
3280 | 3290 | ||
3281 | /* If hashes are distributed across NUMA nodes, defer | 3291 | /* If hashes are distributed across NUMA nodes, defer |
3282 | * hash allocation until vmalloc space is available. | 3292 | * hash allocation until vmalloc space is available. |
3283 | */ | 3293 | */ |
3284 | if (hashdist) | 3294 | if (hashdist) |
3285 | return; | 3295 | return; |
3286 | 3296 | ||
3287 | dentry_hashtable = | 3297 | dentry_hashtable = |
3288 | alloc_large_system_hash("Dentry cache", | 3298 | alloc_large_system_hash("Dentry cache", |
3289 | sizeof(struct hlist_bl_head), | 3299 | sizeof(struct hlist_bl_head), |
3290 | dhash_entries, | 3300 | dhash_entries, |
3291 | 13, | 3301 | 13, |
3292 | HASH_EARLY, | 3302 | HASH_EARLY, |
3293 | &d_hash_shift, | 3303 | &d_hash_shift, |
3294 | &d_hash_mask, | 3304 | &d_hash_mask, |
3295 | 0, | 3305 | 0, |
3296 | 0); | 3306 | 0); |
3297 | 3307 | ||
3298 | for (loop = 0; loop < (1U << d_hash_shift); loop++) | 3308 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
3299 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); | 3309 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
3300 | } | 3310 | } |
3301 | 3311 | ||
3302 | static void __init dcache_init(void) | 3312 | static void __init dcache_init(void) |
3303 | { | 3313 | { |
3304 | unsigned int loop; | 3314 | unsigned int loop; |
3305 | 3315 | ||
3306 | /* | 3316 | /* |
3307 | * A constructor could be added for stable state like the lists, | 3317 | * A constructor could be added for stable state like the lists, |
3308 | * but it is probably not worth it because of the cache nature | 3318 | * but it is probably not worth it because of the cache nature |
3309 | * of the dcache. | 3319 | * of the dcache. |
3310 | */ | 3320 | */ |
3311 | dentry_cache = KMEM_CACHE(dentry, | 3321 | dentry_cache = KMEM_CACHE(dentry, |
3312 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | 3322 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); |
3313 | 3323 | ||
3314 | /* Hash may have been set up in dcache_init_early */ | 3324 | /* Hash may have been set up in dcache_init_early */ |
3315 | if (!hashdist) | 3325 | if (!hashdist) |
3316 | return; | 3326 | return; |
3317 | 3327 | ||
3318 | dentry_hashtable = | 3328 | dentry_hashtable = |
3319 | alloc_large_system_hash("Dentry cache", | 3329 | alloc_large_system_hash("Dentry cache", |
3320 | sizeof(struct hlist_bl_head), | 3330 | sizeof(struct hlist_bl_head), |
3321 | dhash_entries, | 3331 | dhash_entries, |
3322 | 13, | 3332 | 13, |
3323 | 0, | 3333 | 0, |
3324 | &d_hash_shift, | 3334 | &d_hash_shift, |
3325 | &d_hash_mask, | 3335 | &d_hash_mask, |
3326 | 0, | 3336 | 0, |
3327 | 0); | 3337 | 0); |
3328 | 3338 | ||
3329 | for (loop = 0; loop < (1U << d_hash_shift); loop++) | 3339 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
3330 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); | 3340 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
3331 | } | 3341 | } |
3332 | 3342 | ||
3333 | /* SLAB cache for __getname() consumers */ | 3343 | /* SLAB cache for __getname() consumers */ |
3334 | struct kmem_cache *names_cachep __read_mostly; | 3344 | struct kmem_cache *names_cachep __read_mostly; |
3335 | EXPORT_SYMBOL(names_cachep); | 3345 | EXPORT_SYMBOL(names_cachep); |
3336 | 3346 | ||
3337 | EXPORT_SYMBOL(d_genocide); | 3347 | EXPORT_SYMBOL(d_genocide); |
3338 | 3348 | ||
3339 | void __init vfs_caches_init_early(void) | 3349 | void __init vfs_caches_init_early(void) |
3340 | { | 3350 | { |