1 // SPDX-License-Identifier: GPL-2.0-or-later 1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* !! 2 /* -*- mode: c; c-basic-offset: 8; -*-
>> 3 * vim: noexpandtab sw=8 ts=8 sts=0:
>> 4 *
3 * dcache.c 5 * dcache.c
4 * 6 *
5 * dentry cache handling code 7 * dentry cache handling code
6 * 8 *
7 * Copyright (C) 2002, 2004 Oracle. All right 9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
8 */ 10 */
9 11
10 #include <linux/fs.h> 12 #include <linux/fs.h>
11 #include <linux/types.h> 13 #include <linux/types.h>
12 #include <linux/slab.h> 14 #include <linux/slab.h>
13 #include <linux/namei.h> 15 #include <linux/namei.h>
14 16
15 #include <cluster/masklog.h> 17 #include <cluster/masklog.h>
16 18
17 #include "ocfs2.h" 19 #include "ocfs2.h"
18 20
19 #include "alloc.h" 21 #include "alloc.h"
20 #include "dcache.h" 22 #include "dcache.h"
21 #include "dlmglue.h" 23 #include "dlmglue.h"
22 #include "file.h" 24 #include "file.h"
23 #include "inode.h" 25 #include "inode.h"
24 #include "ocfs2_trace.h" 26 #include "ocfs2_trace.h"
25 27
26 void ocfs2_dentry_attach_gen(struct dentry *de 28 void ocfs2_dentry_attach_gen(struct dentry *dentry)
27 { 29 {
28 unsigned long gen = 30 unsigned long gen =
29 OCFS2_I(d_inode(dentry->d_pare 31 OCFS2_I(d_inode(dentry->d_parent))->ip_dir_lock_gen;
30 BUG_ON(d_inode(dentry)); 32 BUG_ON(d_inode(dentry));
31 dentry->d_fsdata = (void *)gen; 33 dentry->d_fsdata = (void *)gen;
32 } 34 }
33 35
34 36
35 static int ocfs2_dentry_revalidate(struct dent 37 static int ocfs2_dentry_revalidate(struct dentry *dentry, unsigned int flags)
36 { 38 {
37 struct inode *inode; 39 struct inode *inode;
38 int ret = 0; /* if all else fails, 40 int ret = 0; /* if all else fails, just return false */
39 struct ocfs2_super *osb; 41 struct ocfs2_super *osb;
40 42
41 if (flags & LOOKUP_RCU) 43 if (flags & LOOKUP_RCU)
42 return -ECHILD; 44 return -ECHILD;
43 45
44 inode = d_inode(dentry); 46 inode = d_inode(dentry);
45 osb = OCFS2_SB(dentry->d_sb); 47 osb = OCFS2_SB(dentry->d_sb);
46 48
47 trace_ocfs2_dentry_revalidate(dentry, 49 trace_ocfs2_dentry_revalidate(dentry, dentry->d_name.len,
48 dentry-> 50 dentry->d_name.name);
49 51
50 /* For a negative dentry - 52 /* For a negative dentry -
51 * check the generation number of the 53 * check the generation number of the parent and compare with the
52 * one stored in the inode. 54 * one stored in the inode.
53 */ 55 */
54 if (inode == NULL) { 56 if (inode == NULL) {
55 unsigned long gen = (unsigned 57 unsigned long gen = (unsigned long) dentry->d_fsdata;
56 unsigned long pgen; 58 unsigned long pgen;
57 spin_lock(&dentry->d_lock); 59 spin_lock(&dentry->d_lock);
58 pgen = OCFS2_I(d_inode(dentry- 60 pgen = OCFS2_I(d_inode(dentry->d_parent))->ip_dir_lock_gen;
59 spin_unlock(&dentry->d_lock); 61 spin_unlock(&dentry->d_lock);
60 trace_ocfs2_dentry_revalidate_ 62 trace_ocfs2_dentry_revalidate_negative(dentry->d_name.len,
61 63 dentry->d_name.name,
62 64 pgen, gen);
63 if (gen != pgen) 65 if (gen != pgen)
64 goto bail; 66 goto bail;
65 goto valid; 67 goto valid;
66 } 68 }
67 69
68 BUG_ON(!osb); 70 BUG_ON(!osb);
69 71
70 if (inode == osb->root_inode || is_bad 72 if (inode == osb->root_inode || is_bad_inode(inode))
71 goto bail; 73 goto bail;
72 74
73 spin_lock(&OCFS2_I(inode)->ip_lock); 75 spin_lock(&OCFS2_I(inode)->ip_lock);
74 /* did we or someone else delete this 76 /* did we or someone else delete this inode? */
75 if (OCFS2_I(inode)->ip_flags & OCFS2_I 77 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
76 spin_unlock(&OCFS2_I(inode)->i 78 spin_unlock(&OCFS2_I(inode)->ip_lock);
77 trace_ocfs2_dentry_revalidate_ 79 trace_ocfs2_dentry_revalidate_delete(
78 (unsigned long 80 (unsigned long long)OCFS2_I(inode)->ip_blkno);
79 goto bail; 81 goto bail;
80 } 82 }
81 spin_unlock(&OCFS2_I(inode)->ip_lock); 83 spin_unlock(&OCFS2_I(inode)->ip_lock);
82 84
83 /* 85 /*
84 * We don't need a cluster lock to tes 86 * We don't need a cluster lock to test this because once an
85 * inode nlink hits zero, it never goe 87 * inode nlink hits zero, it never goes back.
86 */ 88 */
87 if (inode->i_nlink == 0) { 89 if (inode->i_nlink == 0) {
88 trace_ocfs2_dentry_revalidate_ 90 trace_ocfs2_dentry_revalidate_orphaned(
89 (unsigned long long)OC 91 (unsigned long long)OCFS2_I(inode)->ip_blkno,
90 S_ISDIR(inode->i_mode) 92 S_ISDIR(inode->i_mode));
91 goto bail; 93 goto bail;
92 } 94 }
93 95
94 /* 96 /*
95 * If the last lookup failed to create 97 * If the last lookup failed to create dentry lock, let us
96 * redo it. 98 * redo it.
97 */ 99 */
98 if (!dentry->d_fsdata) { 100 if (!dentry->d_fsdata) {
99 trace_ocfs2_dentry_revalidate_ 101 trace_ocfs2_dentry_revalidate_nofsdata(
100 (unsigned long 102 (unsigned long long)OCFS2_I(inode)->ip_blkno);
101 goto bail; 103 goto bail;
102 } 104 }
103 105
104 valid: 106 valid:
105 ret = 1; 107 ret = 1;
106 108
107 bail: 109 bail:
108 trace_ocfs2_dentry_revalidate_ret(ret) 110 trace_ocfs2_dentry_revalidate_ret(ret);
109 return ret; 111 return ret;
110 } 112 }
111 113
112 static int ocfs2_match_dentry(struct dentry *d 114 static int ocfs2_match_dentry(struct dentry *dentry,
113 u64 parent_blkno 115 u64 parent_blkno,
114 int skip_unhashe 116 int skip_unhashed)
115 { 117 {
116 struct inode *parent; 118 struct inode *parent;
117 119
118 /* 120 /*
119 * ocfs2_lookup() does a d_splice_alia 121 * ocfs2_lookup() does a d_splice_alias() _before_ attaching
120 * to the lock data, so we skip those 122 * to the lock data, so we skip those here, otherwise
121 * ocfs2_dentry_attach_lock() will get 123 * ocfs2_dentry_attach_lock() will get its original dentry
122 * back. 124 * back.
123 */ 125 */
124 if (!dentry->d_fsdata) 126 if (!dentry->d_fsdata)
125 return 0; 127 return 0;
126 128
>> 129 if (!dentry->d_parent)
>> 130 return 0;
>> 131
127 if (skip_unhashed && d_unhashed(dentry 132 if (skip_unhashed && d_unhashed(dentry))
128 return 0; 133 return 0;
129 134
130 parent = d_inode(dentry->d_parent); 135 parent = d_inode(dentry->d_parent);
>> 136 /* Negative parent dentry? */
>> 137 if (!parent)
>> 138 return 0;
>> 139
131 /* Name is in a different directory. * 140 /* Name is in a different directory. */
132 if (OCFS2_I(parent)->ip_blkno != paren 141 if (OCFS2_I(parent)->ip_blkno != parent_blkno)
133 return 0; 142 return 0;
134 143
135 return 1; 144 return 1;
136 } 145 }
137 146
138 /* 147 /*
139 * Walk the inode alias list, and find a dentr 148 * Walk the inode alias list, and find a dentry which has a given
140 * parent. ocfs2_dentry_attach_lock() wants to 149 * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
141 * is looking for a dentry_lock reference. The 150 * is looking for a dentry_lock reference. The downconvert thread is
142 * looking to unhash aliases, so we allow it t 151 * looking to unhash aliases, so we allow it to skip any that already
143 * have that property. 152 * have that property.
144 */ 153 */
145 struct dentry *ocfs2_find_local_alias(struct i 154 struct dentry *ocfs2_find_local_alias(struct inode *inode,
146 u64 pare 155 u64 parent_blkno,
147 int skip 156 int skip_unhashed)
148 { 157 {
149 struct dentry *dentry; 158 struct dentry *dentry;
150 159
151 spin_lock(&inode->i_lock); 160 spin_lock(&inode->i_lock);
152 hlist_for_each_entry(dentry, &inode->i 161 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
153 spin_lock(&dentry->d_lock); 162 spin_lock(&dentry->d_lock);
154 if (ocfs2_match_dentry(dentry, 163 if (ocfs2_match_dentry(dentry, parent_blkno, skip_unhashed)) {
155 trace_ocfs2_find_local 164 trace_ocfs2_find_local_alias(dentry->d_name.len,
156 165 dentry->d_name.name);
157 166
158 dget_dlock(dentry); 167 dget_dlock(dentry);
159 spin_unlock(&dentry->d 168 spin_unlock(&dentry->d_lock);
160 spin_unlock(&inode->i_ 169 spin_unlock(&inode->i_lock);
161 return dentry; 170 return dentry;
162 } 171 }
163 spin_unlock(&dentry->d_lock); 172 spin_unlock(&dentry->d_lock);
164 } 173 }
165 spin_unlock(&inode->i_lock); 174 spin_unlock(&inode->i_lock);
166 return NULL; 175 return NULL;
167 } 176 }
168 177
169 DEFINE_SPINLOCK(dentry_attach_lock); 178 DEFINE_SPINLOCK(dentry_attach_lock);
170 179
171 /* 180 /*
172 * Attach this dentry to a cluster lock. 181 * Attach this dentry to a cluster lock.
173 * 182 *
174 * Dentry locks cover all links in a given dir 183 * Dentry locks cover all links in a given directory to a particular
175 * inode. We do this so that ocfs2 can build a 184 * inode. We do this so that ocfs2 can build a lock name which all
176 * nodes in the cluster can agree on at all ti 185 * nodes in the cluster can agree on at all times. Shoving full names
177 * in the cluster lock won't work due to size 186 * in the cluster lock won't work due to size restrictions. Covering
178 * links inside of a directory is a good compr 187 * links inside of a directory is a good compromise because it still
179 * allows us to use the parent directory lock 188 * allows us to use the parent directory lock to synchronize
180 * operations. 189 * operations.
181 * 190 *
182 * Call this function with the parent dir sema 191 * Call this function with the parent dir semaphore and the parent dir
183 * cluster lock held. 192 * cluster lock held.
184 * 193 *
185 * The dir semaphore will protect us from havi 194 * The dir semaphore will protect us from having to worry about
186 * concurrent processes on our node trying to 195 * concurrent processes on our node trying to attach a lock at the
187 * same time. 196 * same time.
188 * 197 *
189 * The dir cluster lock (held at either PR or 198 * The dir cluster lock (held at either PR or EX mode) protects us
190 * from unlink and rename on other nodes. 199 * from unlink and rename on other nodes.
191 * 200 *
192 * A dput() can happen asynchronously due to p 201 * A dput() can happen asynchronously due to pruning, so we cover
193 * attaching and detaching the dentry lock wit 202 * attaching and detaching the dentry lock with a
194 * dentry_attach_lock. 203 * dentry_attach_lock.
195 * 204 *
196 * A node which has done lookup on a name reta 205 * A node which has done lookup on a name retains a protected read
197 * lock until final dput. If the user requests 206 * lock until final dput. If the user requests and unlink or rename,
198 * the protected read is upgraded to an exclus 207 * the protected read is upgraded to an exclusive lock. Other nodes
199 * who have seen the dentry will then be infor 208 * who have seen the dentry will then be informed that they need to
200 * downgrade their lock, which will involve d_ 209 * downgrade their lock, which will involve d_delete on the
201 * dentry. This happens in ocfs2_dentry_conver 210 * dentry. This happens in ocfs2_dentry_convert_worker().
202 */ 211 */
203 int ocfs2_dentry_attach_lock(struct dentry *de 212 int ocfs2_dentry_attach_lock(struct dentry *dentry,
204 struct inode *ino 213 struct inode *inode,
205 u64 parent_blkno) 214 u64 parent_blkno)
206 { 215 {
207 int ret; 216 int ret;
208 struct dentry *alias; 217 struct dentry *alias;
209 struct ocfs2_dentry_lock *dl = dentry- 218 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
210 219
211 trace_ocfs2_dentry_attach_lock(dentry- 220 trace_ocfs2_dentry_attach_lock(dentry->d_name.len, dentry->d_name.name,
212 (unsign 221 (unsigned long long)parent_blkno, dl);
213 222
214 /* 223 /*
215 * Negative dentry. We ignore these fo 224 * Negative dentry. We ignore these for now.
216 * 225 *
217 * XXX: Could we can improve ocfs2_den 226 * XXX: Could we can improve ocfs2_dentry_revalidate() by
218 * tracking these? 227 * tracking these?
219 */ 228 */
220 if (!inode) 229 if (!inode)
221 return 0; 230 return 0;
222 231
223 if (d_really_is_negative(dentry) && de 232 if (d_really_is_negative(dentry) && dentry->d_fsdata) {
224 /* Converting a negative dentr 233 /* Converting a negative dentry to positive
225 Clear dentry->d_fsdata */ 234 Clear dentry->d_fsdata */
226 dentry->d_fsdata = dl = NULL; 235 dentry->d_fsdata = dl = NULL;
227 } 236 }
228 237
229 if (dl) { 238 if (dl) {
230 mlog_bug_on_msg(dl->dl_parent_ 239 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
231 " \"%pd\": old 240 " \"%pd\": old parent: %llu, new: %llu\n",
232 dentry, 241 dentry,
233 (unsigned long 242 (unsigned long long)parent_blkno,
234 (unsigned long 243 (unsigned long long)dl->dl_parent_blkno);
235 return 0; 244 return 0;
236 } 245 }
237 246
238 alias = ocfs2_find_local_alias(inode, 247 alias = ocfs2_find_local_alias(inode, parent_blkno, 0);
239 if (alias) { 248 if (alias) {
240 /* 249 /*
241 * Great, an alias exists, whi 250 * Great, an alias exists, which means we must have a
242 * dentry lock already. We can 251 * dentry lock already. We can just grab the lock off
243 * the alias and add it to the 252 * the alias and add it to the list.
244 * 253 *
245 * We're depending here on the 254 * We're depending here on the fact that this dentry
246 * was found and exists in the 255 * was found and exists in the dcache and so must have
247 * a reference to the dentry_l 256 * a reference to the dentry_lock because we can't
248 * race creates. Final dput() 257 * race creates. Final dput() cannot happen on it
249 * since we have it pinned, so 258 * since we have it pinned, so our reference is safe.
250 */ 259 */
251 dl = alias->d_fsdata; 260 dl = alias->d_fsdata;
252 mlog_bug_on_msg(!dl, "parent % 261 mlog_bug_on_msg(!dl, "parent %llu, ino %llu\n",
253 (unsigned long 262 (unsigned long long)parent_blkno,
254 (unsigned long 263 (unsigned long long)OCFS2_I(inode)->ip_blkno);
255 264
256 mlog_bug_on_msg(dl->dl_parent_ 265 mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
257 " \"%pd\": old 266 " \"%pd\": old parent: %llu, new: %llu\n",
258 dentry, 267 dentry,
259 (unsigned long 268 (unsigned long long)parent_blkno,
260 (unsigned long 269 (unsigned long long)dl->dl_parent_blkno);
261 270
262 trace_ocfs2_dentry_attach_lock 271 trace_ocfs2_dentry_attach_lock_found(dl->dl_lockres.l_name,
263 (unsigned long 272 (unsigned long long)parent_blkno,
264 (unsigned long 273 (unsigned long long)OCFS2_I(inode)->ip_blkno);
265 274
266 goto out_attach; 275 goto out_attach;
267 } 276 }
268 277
269 /* 278 /*
270 * There are no other aliases 279 * There are no other aliases
271 */ 280 */
272 dl = kmalloc(sizeof(*dl), GFP_NOFS); 281 dl = kmalloc(sizeof(*dl), GFP_NOFS);
273 if (!dl) { 282 if (!dl) {
274 ret = -ENOMEM; 283 ret = -ENOMEM;
275 mlog_errno(ret); 284 mlog_errno(ret);
276 return ret; 285 return ret;
277 } 286 }
278 287
279 dl->dl_count = 0; 288 dl->dl_count = 0;
280 /* 289 /*
281 * Does this have to happen below, for 290 * Does this have to happen below, for all attaches, in case
282 * the struct inode gets blown away by 291 * the struct inode gets blown away by the downconvert thread?
283 */ 292 */
284 dl->dl_inode = igrab(inode); 293 dl->dl_inode = igrab(inode);
285 dl->dl_parent_blkno = parent_blkno; 294 dl->dl_parent_blkno = parent_blkno;
286 ocfs2_dentry_lock_res_init(dl, parent_ 295 ocfs2_dentry_lock_res_init(dl, parent_blkno, inode);
287 296
288 out_attach: 297 out_attach:
289 spin_lock(&dentry_attach_lock); 298 spin_lock(&dentry_attach_lock);
290 if (unlikely(dentry->d_fsdata && !alia 299 if (unlikely(dentry->d_fsdata && !alias)) {
291 /* d_fsdata is set by a racing 300 /* d_fsdata is set by a racing thread which is doing
292 * the same thing as this thre 301 * the same thing as this thread is doing. Leave the racing
293 * thread going ahead and we r 302 * thread going ahead and we return here.
294 */ 303 */
295 spin_unlock(&dentry_attach_loc 304 spin_unlock(&dentry_attach_lock);
296 iput(dl->dl_inode); 305 iput(dl->dl_inode);
297 ocfs2_lock_res_free(&dl->dl_lo 306 ocfs2_lock_res_free(&dl->dl_lockres);
298 kfree(dl); 307 kfree(dl);
299 return 0; 308 return 0;
300 } 309 }
301 310
302 dentry->d_fsdata = dl; 311 dentry->d_fsdata = dl;
303 dl->dl_count++; 312 dl->dl_count++;
304 spin_unlock(&dentry_attach_lock); 313 spin_unlock(&dentry_attach_lock);
305 314
306 /* 315 /*
307 * This actually gets us our PRMODE le 316 * This actually gets us our PRMODE level lock. From now on,
308 * we'll have a notification if one of 317 * we'll have a notification if one of these names is
309 * destroyed on another node. 318 * destroyed on another node.
310 */ 319 */
311 ret = ocfs2_dentry_lock(dentry, 0); 320 ret = ocfs2_dentry_lock(dentry, 0);
312 if (!ret) 321 if (!ret)
313 ocfs2_dentry_unlock(dentry, 0) 322 ocfs2_dentry_unlock(dentry, 0);
314 else 323 else
315 mlog_errno(ret); 324 mlog_errno(ret);
316 325
317 /* 326 /*
318 * In case of error, manually free the 327 * In case of error, manually free the allocation and do the iput().
319 * We need to do this because error he 328 * We need to do this because error here means no d_instantiate(),
320 * which means iput() will not be call 329 * which means iput() will not be called during dput(dentry).
321 */ 330 */
322 if (ret < 0 && !alias) { 331 if (ret < 0 && !alias) {
323 ocfs2_lock_res_free(&dl->dl_lo 332 ocfs2_lock_res_free(&dl->dl_lockres);
324 BUG_ON(dl->dl_count != 1); 333 BUG_ON(dl->dl_count != 1);
325 spin_lock(&dentry_attach_lock) 334 spin_lock(&dentry_attach_lock);
326 dentry->d_fsdata = NULL; 335 dentry->d_fsdata = NULL;
327 spin_unlock(&dentry_attach_loc 336 spin_unlock(&dentry_attach_lock);
328 kfree(dl); 337 kfree(dl);
329 iput(inode); 338 iput(inode);
330 } 339 }
331 340
332 dput(alias); 341 dput(alias);
333 342
334 return ret; 343 return ret;
335 } 344 }
336 345
337 /* 346 /*
338 * ocfs2_dentry_iput() and friends. 347 * ocfs2_dentry_iput() and friends.
339 * 348 *
340 * At this point, our particular dentry is det 349 * At this point, our particular dentry is detached from the inodes
341 * alias list, so there's no way that the lock 350 * alias list, so there's no way that the locking code can find it.
342 * 351 *
343 * The interesting stuff happens when we deter 352 * The interesting stuff happens when we determine that our lock needs
344 * to go away because this is the last subdir 353 * to go away because this is the last subdir alias in the
345 * system. This function needs to handle a cou 354 * system. This function needs to handle a couple things:
346 * 355 *
347 * 1) Synchronizing lock shutdown with the dow 356 * 1) Synchronizing lock shutdown with the downconvert threads. This
348 * is already handled for us via the lockre 357 * is already handled for us via the lockres release drop function
349 * called in ocfs2_release_dentry_lock() 358 * called in ocfs2_release_dentry_lock()
350 * 359 *
351 * 2) A race may occur when we're doing our lo 360 * 2) A race may occur when we're doing our lock shutdown and
352 * another process wants to create a new de 361 * another process wants to create a new dentry lock. Right now we
353 * let them race, which means that for a ve 362 * let them race, which means that for a very short while, this
354 * node might have two locks on a lock reso 363 * node might have two locks on a lock resource. This should be a
355 * problem though because one of them is in 364 * problem though because one of them is in the process of being
356 * thrown out. 365 * thrown out.
357 */ 366 */
358 static void ocfs2_drop_dentry_lock(struct ocfs 367 static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
359 struct ocfs 368 struct ocfs2_dentry_lock *dl)
360 { 369 {
361 iput(dl->dl_inode); 370 iput(dl->dl_inode);
362 ocfs2_simple_drop_lockres(osb, &dl->dl 371 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
363 ocfs2_lock_res_free(&dl->dl_lockres); 372 ocfs2_lock_res_free(&dl->dl_lockres);
364 kfree(dl); 373 kfree(dl);
365 } 374 }
366 375
367 void ocfs2_dentry_lock_put(struct ocfs2_super 376 void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
368 struct ocfs2_dentry 377 struct ocfs2_dentry_lock *dl)
369 { 378 {
370 int unlock = 0; 379 int unlock = 0;
371 380
372 BUG_ON(dl->dl_count == 0); 381 BUG_ON(dl->dl_count == 0);
373 382
374 spin_lock(&dentry_attach_lock); 383 spin_lock(&dentry_attach_lock);
375 dl->dl_count--; 384 dl->dl_count--;
376 unlock = !dl->dl_count; 385 unlock = !dl->dl_count;
377 spin_unlock(&dentry_attach_lock); 386 spin_unlock(&dentry_attach_lock);
378 387
379 if (unlock) 388 if (unlock)
380 ocfs2_drop_dentry_lock(osb, dl 389 ocfs2_drop_dentry_lock(osb, dl);
381 } 390 }
382 391
383 static void ocfs2_dentry_iput(struct dentry *d 392 static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
384 { 393 {
385 struct ocfs2_dentry_lock *dl = dentry- 394 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
386 395
387 if (!dl) { 396 if (!dl) {
388 /* 397 /*
389 * No dentry lock is ok if we' 398 * No dentry lock is ok if we're disconnected or
390 * unhashed. 399 * unhashed.
391 */ 400 */
392 if (!(dentry->d_flags & DCACHE 401 if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
393 !d_unhashed(dentry)) { 402 !d_unhashed(dentry)) {
394 unsigned long long ino 403 unsigned long long ino = 0ULL;
395 if (inode) 404 if (inode)
396 ino = (unsigne 405 ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
397 mlog(ML_ERROR, "Dentry 406 mlog(ML_ERROR, "Dentry is missing cluster lock. "
398 "inode: %llu, d_f 407 "inode: %llu, d_flags: 0x%x, d_name: %pd\n",
399 ino, dentry->d_fl 408 ino, dentry->d_flags, dentry);
400 } 409 }
401 410
402 goto out; 411 goto out;
403 } 412 }
404 413
405 mlog_bug_on_msg(dl->dl_count == 0, "de 414 mlog_bug_on_msg(dl->dl_count == 0, "dentry: %pd, count: %u\n",
406 dentry, dl->dl_count); 415 dentry, dl->dl_count);
407 416
408 ocfs2_dentry_lock_put(OCFS2_SB(dentry- 417 ocfs2_dentry_lock_put(OCFS2_SB(dentry->d_sb), dl);
409 418
410 out: 419 out:
411 iput(inode); 420 iput(inode);
412 } 421 }
413 422
414 /* 423 /*
415 * d_move(), but keep the locks in sync. 424 * d_move(), but keep the locks in sync.
416 * 425 *
417 * When we are done, "dentry" will have the pa 426 * When we are done, "dentry" will have the parent dir and name of
418 * "target", which will be thrown away. 427 * "target", which will be thrown away.
419 * 428 *
420 * We manually update the lock of "dentry" if 429 * We manually update the lock of "dentry" if need be.
421 * 430 *
422 * "target" doesn't have it's dentry lock touc 431 * "target" doesn't have it's dentry lock touched - we allow the later
423 * dput() to handle this for us. 432 * dput() to handle this for us.
424 * 433 *
425 * This is called during ocfs2_rename(), while 434 * This is called during ocfs2_rename(), while holding parent
426 * directory locks. The dentries have already 435 * directory locks. The dentries have already been deleted on other
427 * nodes via ocfs2_remote_dentry_delete(). 436 * nodes via ocfs2_remote_dentry_delete().
428 * 437 *
429 * Normally, the VFS handles the d_move() for 438 * Normally, the VFS handles the d_move() for the file system, after
430 * the ->rename() callback. OCFS2 wants to han 439 * the ->rename() callback. OCFS2 wants to handle this internally, so
431 * the new lock can be created atomically with 440 * the new lock can be created atomically with respect to the cluster.
432 */ 441 */
433 void ocfs2_dentry_move(struct dentry *dentry, 442 void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
434 struct inode *old_dir, 443 struct inode *old_dir, struct inode *new_dir)
435 { 444 {
436 int ret; 445 int ret;
437 struct ocfs2_super *osb = OCFS2_SB(old 446 struct ocfs2_super *osb = OCFS2_SB(old_dir->i_sb);
438 struct inode *inode = d_inode(dentry); 447 struct inode *inode = d_inode(dentry);
439 448
440 /* 449 /*
441 * Move within the same directory, so 450 * Move within the same directory, so the actual lock info won't
442 * change. 451 * change.
443 * 452 *
444 * XXX: Is there any advantage to drop 453 * XXX: Is there any advantage to dropping the lock here?
445 */ 454 */
446 if (old_dir == new_dir) 455 if (old_dir == new_dir)
447 goto out_move; 456 goto out_move;
448 457
449 ocfs2_dentry_lock_put(osb, dentry->d_f 458 ocfs2_dentry_lock_put(osb, dentry->d_fsdata);
450 459
451 dentry->d_fsdata = NULL; 460 dentry->d_fsdata = NULL;
452 ret = ocfs2_dentry_attach_lock(dentry, 461 ret = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(new_dir)->ip_blkno);
453 if (ret) 462 if (ret)
454 mlog_errno(ret); 463 mlog_errno(ret);
455 464
456 out_move: 465 out_move:
457 d_move(dentry, target); 466 d_move(dentry, target);
458 } 467 }
459 468
460 const struct dentry_operations ocfs2_dentry_op 469 const struct dentry_operations ocfs2_dentry_ops = {
461 .d_revalidate = ocfs2_dentry 470 .d_revalidate = ocfs2_dentry_revalidate,
462 .d_iput = ocfs2_dentry 471 .d_iput = ocfs2_dentry_iput,
463 }; 472 };
464 473
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.