]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
7b718769 | 3 | * All Rights Reserved. |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
40ebd81d RD |
18 | #include <linux/log2.h> |
19 | ||
1da177e4 | 20 | #include "xfs.h" |
a844f451 | 21 | #include "xfs_fs.h" |
6ca1c906 | 22 | #include "xfs_format.h" |
1da177e4 | 23 | #include "xfs_log.h" |
a844f451 | 24 | #include "xfs_inum.h" |
1da177e4 | 25 | #include "xfs_trans.h" |
c24b5dfa | 26 | #include "xfs_trans_space.h" |
1da177e4 LT |
27 | #include "xfs_trans_priv.h" |
28 | #include "xfs_sb.h" | |
29 | #include "xfs_ag.h" | |
1da177e4 | 30 | #include "xfs_mount.h" |
c24b5dfa DC |
31 | #include "xfs_da_btree.h" |
32 | #include "xfs_dir2_format.h" | |
33 | #include "xfs_dir2.h" | |
1da177e4 | 34 | #include "xfs_bmap_btree.h" |
a844f451 | 35 | #include "xfs_alloc_btree.h" |
1da177e4 | 36 | #include "xfs_ialloc_btree.h" |
a844f451 | 37 | #include "xfs_attr_sf.h" |
c24b5dfa | 38 | #include "xfs_attr.h" |
1da177e4 | 39 | #include "xfs_dinode.h" |
1da177e4 | 40 | #include "xfs_inode.h" |
1da177e4 | 41 | #include "xfs_buf_item.h" |
a844f451 NS |
42 | #include "xfs_inode_item.h" |
43 | #include "xfs_btree.h" | |
44 | #include "xfs_alloc.h" | |
45 | #include "xfs_ialloc.h" | |
46 | #include "xfs_bmap.h" | |
68988114 | 47 | #include "xfs_bmap_util.h" |
1da177e4 | 48 | #include "xfs_error.h" |
1da177e4 | 49 | #include "xfs_utils.h" |
1da177e4 | 50 | #include "xfs_quota.h" |
2a82b8be | 51 | #include "xfs_filestream.h" |
93848a99 | 52 | #include "xfs_cksum.h" |
0b1b213f | 53 | #include "xfs_trace.h" |
33479e05 | 54 | #include "xfs_icache.h" |
c24b5dfa | 55 | #include "xfs_symlink.h" |
1da177e4 | 56 | |
1da177e4 | 57 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
58 | |
59 | /* | |
8f04c47a | 60 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
61 | * freed from a file in a single transaction. |
62 | */ | |
63 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
64 | ||
65 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
1da177e4 | 66 | |
2a0ec1d9 DC |
67 | /* |
68 | * helper function to extract extent size hint from inode | |
69 | */ | |
70 | xfs_extlen_t | |
71 | xfs_get_extsz_hint( | |
72 | struct xfs_inode *ip) | |
73 | { | |
74 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
75 | return ip->i_d.di_extsize; | |
76 | if (XFS_IS_REALTIME_INODE(ip)) | |
77 | return ip->i_mount->m_sb.sb_rextsize; | |
78 | return 0; | |
79 | } | |
80 | ||
fa96acad DC |
81 | /* |
82 | * This is a wrapper routine around the xfs_ilock() routine used to centralize | |
83 | * some grungy code. It is used in places that wish to lock the inode solely | |
84 | * for reading the extents. The reason these places can't just call | |
85 | * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the | |
86 | * extents from disk for a file in b-tree format. If the inode is in b-tree | |
87 | * format, then we need to lock the inode exclusively until the extents are read | |
88 | * in. Locking it exclusively all the time would limit our parallelism | |
89 | * unnecessarily, though. What we do instead is check to see if the extents | |
90 | * have been read in yet, and only lock the inode exclusively if they have not. | |
91 | * | |
92 | * The function returns a value which should be given to the corresponding | |
93 | * xfs_iunlock_map_shared(). This value is the mode in which the lock was | |
94 | * actually taken. | |
95 | */ | |
96 | uint | |
97 | xfs_ilock_map_shared( | |
98 | xfs_inode_t *ip) | |
99 | { | |
100 | uint lock_mode; | |
101 | ||
102 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
103 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
104 | lock_mode = XFS_ILOCK_EXCL; | |
105 | } else { | |
106 | lock_mode = XFS_ILOCK_SHARED; | |
107 | } | |
108 | ||
109 | xfs_ilock(ip, lock_mode); | |
110 | ||
111 | return lock_mode; | |
112 | } | |
113 | ||
114 | /* | |
115 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
116 | * All it does is call xfs_iunlock() with the given lock_mode. | |
117 | */ | |
118 | void | |
119 | xfs_iunlock_map_shared( | |
120 | xfs_inode_t *ip, | |
121 | unsigned int lock_mode) | |
122 | { | |
123 | xfs_iunlock(ip, lock_mode); | |
124 | } | |
125 | ||
126 | /* | |
127 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
128 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
129 | * allows either or both of the locks to be obtained. | |
130 | * | |
131 | * The 2 locks should always be ordered so that the IO lock is | |
132 | * obtained first in order to prevent deadlock. | |
133 | * | |
134 | * ip -- the inode being locked | |
135 | * lock_flags -- this parameter indicates the inode's locks | |
136 | * to be locked. It can be: | |
137 | * XFS_IOLOCK_SHARED, | |
138 | * XFS_IOLOCK_EXCL, | |
139 | * XFS_ILOCK_SHARED, | |
140 | * XFS_ILOCK_EXCL, | |
141 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
142 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
143 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
144 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
145 | */ | |
146 | void | |
147 | xfs_ilock( | |
148 | xfs_inode_t *ip, | |
149 | uint lock_flags) | |
150 | { | |
151 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
152 | ||
153 | /* | |
154 | * You can't set both SHARED and EXCL for the same lock, | |
155 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
156 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
157 | */ | |
158 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
159 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
160 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
161 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
162 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
163 | ||
164 | if (lock_flags & XFS_IOLOCK_EXCL) | |
165 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
166 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
167 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
168 | ||
169 | if (lock_flags & XFS_ILOCK_EXCL) | |
170 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
171 | else if (lock_flags & XFS_ILOCK_SHARED) | |
172 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
173 | } | |
174 | ||
175 | /* | |
176 | * This is just like xfs_ilock(), except that the caller | |
177 | * is guaranteed not to sleep. It returns 1 if it gets | |
178 | * the requested locks and 0 otherwise. If the IO lock is | |
179 | * obtained but the inode lock cannot be, then the IO lock | |
180 | * is dropped before returning. | |
181 | * | |
182 | * ip -- the inode being locked | |
183 | * lock_flags -- this parameter indicates the inode's locks to be | |
184 | * to be locked. See the comment for xfs_ilock() for a list | |
185 | * of valid values. | |
186 | */ | |
187 | int | |
188 | xfs_ilock_nowait( | |
189 | xfs_inode_t *ip, | |
190 | uint lock_flags) | |
191 | { | |
192 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
193 | ||
194 | /* | |
195 | * You can't set both SHARED and EXCL for the same lock, | |
196 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
197 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
198 | */ | |
199 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
200 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
201 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
202 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
203 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
204 | ||
205 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
206 | if (!mrtryupdate(&ip->i_iolock)) | |
207 | goto out; | |
208 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
209 | if (!mrtryaccess(&ip->i_iolock)) | |
210 | goto out; | |
211 | } | |
212 | if (lock_flags & XFS_ILOCK_EXCL) { | |
213 | if (!mrtryupdate(&ip->i_lock)) | |
214 | goto out_undo_iolock; | |
215 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
216 | if (!mrtryaccess(&ip->i_lock)) | |
217 | goto out_undo_iolock; | |
218 | } | |
219 | return 1; | |
220 | ||
221 | out_undo_iolock: | |
222 | if (lock_flags & XFS_IOLOCK_EXCL) | |
223 | mrunlock_excl(&ip->i_iolock); | |
224 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
225 | mrunlock_shared(&ip->i_iolock); | |
226 | out: | |
227 | return 0; | |
228 | } | |
229 | ||
230 | /* | |
231 | * xfs_iunlock() is used to drop the inode locks acquired with | |
232 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
233 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
234 | * that we know which locks to drop. | |
235 | * | |
236 | * ip -- the inode being unlocked | |
237 | * lock_flags -- this parameter indicates the inode's locks to be | |
238 | * to be unlocked. See the comment for xfs_ilock() for a list | |
239 | * of valid values for this parameter. | |
240 | * | |
241 | */ | |
242 | void | |
243 | xfs_iunlock( | |
244 | xfs_inode_t *ip, | |
245 | uint lock_flags) | |
246 | { | |
247 | /* | |
248 | * You can't set both SHARED and EXCL for the same lock, | |
249 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
250 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
251 | */ | |
252 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
253 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
254 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
255 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
256 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
257 | ASSERT(lock_flags != 0); | |
258 | ||
259 | if (lock_flags & XFS_IOLOCK_EXCL) | |
260 | mrunlock_excl(&ip->i_iolock); | |
261 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
262 | mrunlock_shared(&ip->i_iolock); | |
263 | ||
264 | if (lock_flags & XFS_ILOCK_EXCL) | |
265 | mrunlock_excl(&ip->i_lock); | |
266 | else if (lock_flags & XFS_ILOCK_SHARED) | |
267 | mrunlock_shared(&ip->i_lock); | |
268 | ||
269 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
270 | } | |
271 | ||
272 | /* | |
273 | * give up write locks. the i/o lock cannot be held nested | |
274 | * if it is being demoted. | |
275 | */ | |
276 | void | |
277 | xfs_ilock_demote( | |
278 | xfs_inode_t *ip, | |
279 | uint lock_flags) | |
280 | { | |
281 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
282 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
283 | ||
284 | if (lock_flags & XFS_ILOCK_EXCL) | |
285 | mrdemote(&ip->i_lock); | |
286 | if (lock_flags & XFS_IOLOCK_EXCL) | |
287 | mrdemote(&ip->i_iolock); | |
288 | ||
289 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
290 | } | |
291 | ||
742ae1e3 | 292 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
293 | int |
294 | xfs_isilocked( | |
295 | xfs_inode_t *ip, | |
296 | uint lock_flags) | |
297 | { | |
298 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
299 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
300 | return !!ip->i_lock.mr_writer; | |
301 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
302 | } | |
303 | ||
304 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { | |
305 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
306 | return !!ip->i_iolock.mr_writer; | |
307 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
308 | } | |
309 | ||
310 | ASSERT(0); | |
311 | return 0; | |
312 | } | |
313 | #endif | |
314 | ||
c24b5dfa DC |
315 | #ifdef DEBUG |
316 | int xfs_locked_n; | |
317 | int xfs_small_retries; | |
318 | int xfs_middle_retries; | |
319 | int xfs_lots_retries; | |
320 | int xfs_lock_delays; | |
321 | #endif | |
322 | ||
323 | /* | |
324 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
325 | * a different value | |
326 | */ | |
327 | static inline int | |
328 | xfs_lock_inumorder(int lock_mode, int subclass) | |
329 | { | |
330 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
331 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; | |
332 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) | |
333 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; | |
334 | ||
335 | return lock_mode; | |
336 | } | |
337 | ||
338 | /* | |
339 | * The following routine will lock n inodes in exclusive mode. | |
340 | * We assume the caller calls us with the inodes in i_ino order. | |
341 | * | |
342 | * We need to detect deadlock where an inode that we lock | |
343 | * is in the AIL and we start waiting for another inode that is locked | |
344 | * by a thread in a long running transaction (such as truncate). This can | |
345 | * result in deadlock since the long running trans might need to wait | |
346 | * for the inode we just locked in order to push the tail and free space | |
347 | * in the log. | |
348 | */ | |
349 | void | |
350 | xfs_lock_inodes( | |
351 | xfs_inode_t **ips, | |
352 | int inodes, | |
353 | uint lock_mode) | |
354 | { | |
355 | int attempts = 0, i, j, try_lock; | |
356 | xfs_log_item_t *lp; | |
357 | ||
358 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
359 | ||
360 | try_lock = 0; | |
361 | i = 0; | |
362 | ||
363 | again: | |
364 | for (; i < inodes; i++) { | |
365 | ASSERT(ips[i]); | |
366 | ||
367 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
368 | continue; | |
369 | ||
370 | /* | |
371 | * If try_lock is not set yet, make sure all locked inodes | |
372 | * are not in the AIL. | |
373 | * If any are, set try_lock to be used later. | |
374 | */ | |
375 | ||
376 | if (!try_lock) { | |
377 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
378 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
379 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
380 | try_lock++; | |
381 | } | |
382 | } | |
383 | } | |
384 | ||
385 | /* | |
386 | * If any of the previous locks we have locked is in the AIL, | |
387 | * we must TRY to get the second and subsequent locks. If | |
388 | * we can't get any, we must release all we have | |
389 | * and try again. | |
390 | */ | |
391 | ||
392 | if (try_lock) { | |
393 | /* try_lock must be 0 if i is 0. */ | |
394 | /* | |
395 | * try_lock means we have an inode locked | |
396 | * that is in the AIL. | |
397 | */ | |
398 | ASSERT(i != 0); | |
399 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { | |
400 | attempts++; | |
401 | ||
402 | /* | |
403 | * Unlock all previous guys and try again. | |
404 | * xfs_iunlock will try to push the tail | |
405 | * if the inode is in the AIL. | |
406 | */ | |
407 | ||
408 | for(j = i - 1; j >= 0; j--) { | |
409 | ||
410 | /* | |
411 | * Check to see if we've already | |
412 | * unlocked this one. | |
413 | * Not the first one going back, | |
414 | * and the inode ptr is the same. | |
415 | */ | |
416 | if ((j != (i - 1)) && ips[j] == | |
417 | ips[j+1]) | |
418 | continue; | |
419 | ||
420 | xfs_iunlock(ips[j], lock_mode); | |
421 | } | |
422 | ||
423 | if ((attempts % 5) == 0) { | |
424 | delay(1); /* Don't just spin the CPU */ | |
425 | #ifdef DEBUG | |
426 | xfs_lock_delays++; | |
427 | #endif | |
428 | } | |
429 | i = 0; | |
430 | try_lock = 0; | |
431 | goto again; | |
432 | } | |
433 | } else { | |
434 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
435 | } | |
436 | } | |
437 | ||
438 | #ifdef DEBUG | |
439 | if (attempts) { | |
440 | if (attempts < 5) xfs_small_retries++; | |
441 | else if (attempts < 100) xfs_middle_retries++; | |
442 | else xfs_lots_retries++; | |
443 | } else { | |
444 | xfs_locked_n++; | |
445 | } | |
446 | #endif | |
447 | } | |
448 | ||
449 | /* | |
450 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
451 | * at a time - the iolock or the ilock, but not both at once. If | |
452 | * we lock both at once, lockdep will report false positives saying | |
453 | * we have violated locking orders. | |
454 | */ | |
455 | void | |
456 | xfs_lock_two_inodes( | |
457 | xfs_inode_t *ip0, | |
458 | xfs_inode_t *ip1, | |
459 | uint lock_mode) | |
460 | { | |
461 | xfs_inode_t *temp; | |
462 | int attempts = 0; | |
463 | xfs_log_item_t *lp; | |
464 | ||
465 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
466 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
467 | ASSERT(ip0->i_ino != ip1->i_ino); | |
468 | ||
469 | if (ip0->i_ino > ip1->i_ino) { | |
470 | temp = ip0; | |
471 | ip0 = ip1; | |
472 | ip1 = temp; | |
473 | } | |
474 | ||
475 | again: | |
476 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
477 | ||
478 | /* | |
479 | * If the first lock we have locked is in the AIL, we must TRY to get | |
480 | * the second lock. If we can't get it, we must release the first one | |
481 | * and try again. | |
482 | */ | |
483 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
484 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
485 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
486 | xfs_iunlock(ip0, lock_mode); | |
487 | if ((++attempts % 5) == 0) | |
488 | delay(1); /* Don't just spin the CPU */ | |
489 | goto again; | |
490 | } | |
491 | } else { | |
492 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
493 | } | |
494 | } | |
495 | ||
496 | ||
fa96acad DC |
497 | void |
498 | __xfs_iflock( | |
499 | struct xfs_inode *ip) | |
500 | { | |
501 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
502 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
503 | ||
504 | do { | |
505 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
506 | if (xfs_isiflocked(ip)) | |
507 | io_schedule(); | |
508 | } while (!xfs_iflock_nowait(ip)); | |
509 | ||
510 | finish_wait(wq, &wait.wait); | |
511 | } | |
512 | ||
1da177e4 LT |
513 | STATIC uint |
514 | _xfs_dic2xflags( | |
1da177e4 LT |
515 | __uint16_t di_flags) |
516 | { | |
517 | uint flags = 0; | |
518 | ||
519 | if (di_flags & XFS_DIFLAG_ANY) { | |
520 | if (di_flags & XFS_DIFLAG_REALTIME) | |
521 | flags |= XFS_XFLAG_REALTIME; | |
522 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
523 | flags |= XFS_XFLAG_PREALLOC; | |
524 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
525 | flags |= XFS_XFLAG_IMMUTABLE; | |
526 | if (di_flags & XFS_DIFLAG_APPEND) | |
527 | flags |= XFS_XFLAG_APPEND; | |
528 | if (di_flags & XFS_DIFLAG_SYNC) | |
529 | flags |= XFS_XFLAG_SYNC; | |
530 | if (di_flags & XFS_DIFLAG_NOATIME) | |
531 | flags |= XFS_XFLAG_NOATIME; | |
532 | if (di_flags & XFS_DIFLAG_NODUMP) | |
533 | flags |= XFS_XFLAG_NODUMP; | |
534 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
535 | flags |= XFS_XFLAG_RTINHERIT; | |
536 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
537 | flags |= XFS_XFLAG_PROJINHERIT; | |
538 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
539 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
540 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
541 | flags |= XFS_XFLAG_EXTSIZE; | |
542 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
543 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
544 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
545 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
546 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
547 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
548 | } |
549 | ||
550 | return flags; | |
551 | } | |
552 | ||
553 | uint | |
554 | xfs_ip2xflags( | |
555 | xfs_inode_t *ip) | |
556 | { | |
347d1c01 | 557 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 558 | |
a916e2bd | 559 | return _xfs_dic2xflags(dic->di_flags) | |
45ba598e | 560 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
561 | } |
562 | ||
563 | uint | |
564 | xfs_dic2xflags( | |
45ba598e | 565 | xfs_dinode_t *dip) |
1da177e4 | 566 | { |
81591fe2 | 567 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | |
45ba598e | 568 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
569 | } |
570 | ||
c24b5dfa DC |
571 | /* |
572 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
573 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
574 | * ci_name->name will point to a the actual name (caller must free) or | |
575 | * will be set to NULL if an exact match is found. | |
576 | */ | |
577 | int | |
578 | xfs_lookup( | |
579 | xfs_inode_t *dp, | |
580 | struct xfs_name *name, | |
581 | xfs_inode_t **ipp, | |
582 | struct xfs_name *ci_name) | |
583 | { | |
584 | xfs_ino_t inum; | |
585 | int error; | |
586 | uint lock_mode; | |
587 | ||
588 | trace_xfs_lookup(dp, name); | |
589 | ||
590 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
591 | return XFS_ERROR(EIO); | |
592 | ||
593 | lock_mode = xfs_ilock_map_shared(dp); | |
594 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); | |
595 | xfs_iunlock_map_shared(dp, lock_mode); | |
596 | ||
597 | if (error) | |
598 | goto out; | |
599 | ||
600 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
601 | if (error) | |
602 | goto out_free_name; | |
603 | ||
604 | return 0; | |
605 | ||
606 | out_free_name: | |
607 | if (ci_name) | |
608 | kmem_free(ci_name->name); | |
609 | out: | |
610 | *ipp = NULL; | |
611 | return error; | |
612 | } | |
613 | ||
1da177e4 LT |
614 | /* |
615 | * Allocate an inode on disk and return a copy of its in-core version. | |
616 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
617 | * appropriately within the inode. The uid and gid for the inode are | |
618 | * set according to the contents of the given cred structure. | |
619 | * | |
620 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
621 | * has a free inode available, call xfs_iget() to obtain the in-core |
622 | * version of the allocated inode. Finally, fill in the inode and | |
623 | * log its initial contents. In this case, ialloc_context would be | |
624 | * set to NULL. | |
1da177e4 | 625 | * |
cd856db6 CM |
626 | * If xfs_dialloc() does not have an available inode, it will replenish |
627 | * its supply by doing an allocation. Since we can only do one | |
628 | * allocation within a transaction without deadlocks, we must commit | |
629 | * the current transaction before returning the inode itself. | |
630 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
631 | * The caller should then commit the current transaction, start a new |
632 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
633 | * | |
634 | * To ensure that some other process does not grab the inode that | |
635 | * was allocated during the first call to xfs_ialloc(), this routine | |
636 | * also returns the [locked] bp pointing to the head of the freelist | |
637 | * as ialloc_context. The caller should hold this buffer across | |
638 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
639 | * |
640 | * If we are allocating quota inodes, we do not have a parent inode | |
641 | * to attach to or associate with (i.e. pip == NULL) because they | |
642 | * are not linked into the directory structure - they are attached | |
643 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
644 | */ |
645 | int | |
646 | xfs_ialloc( | |
647 | xfs_trans_t *tp, | |
648 | xfs_inode_t *pip, | |
576b1d67 | 649 | umode_t mode, |
31b084ae | 650 | xfs_nlink_t nlink, |
1da177e4 | 651 | xfs_dev_t rdev, |
6743099c | 652 | prid_t prid, |
1da177e4 LT |
653 | int okalloc, |
654 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
655 | xfs_inode_t **ipp) |
656 | { | |
93848a99 | 657 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
658 | xfs_ino_t ino; |
659 | xfs_inode_t *ip; | |
1da177e4 LT |
660 | uint flags; |
661 | int error; | |
dff35fd4 | 662 | timespec_t tv; |
bf904248 | 663 | int filestreams = 0; |
1da177e4 LT |
664 | |
665 | /* | |
666 | * Call the space management code to pick | |
667 | * the on-disk inode to be allocated. | |
668 | */ | |
b11f94d5 | 669 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 670 | ialloc_context, &ino); |
bf904248 | 671 | if (error) |
1da177e4 | 672 | return error; |
08358906 | 673 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
674 | *ipp = NULL; |
675 | return 0; | |
676 | } | |
677 | ASSERT(*ialloc_context == NULL); | |
678 | ||
679 | /* | |
680 | * Get the in-core inode with the lock held exclusively. | |
681 | * This is because we're setting fields here we need | |
682 | * to prevent others from looking at until we're done. | |
683 | */ | |
93848a99 | 684 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 685 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 686 | if (error) |
1da177e4 | 687 | return error; |
1da177e4 LT |
688 | ASSERT(ip != NULL); |
689 | ||
576b1d67 | 690 | ip->i_d.di_mode = mode; |
1da177e4 LT |
691 | ip->i_d.di_onlink = 0; |
692 | ip->i_d.di_nlink = nlink; | |
693 | ASSERT(ip->i_d.di_nlink == nlink); | |
9e2b2dc4 DH |
694 | ip->i_d.di_uid = current_fsuid(); |
695 | ip->i_d.di_gid = current_fsgid(); | |
6743099c | 696 | xfs_set_projid(ip, prid); |
1da177e4 LT |
697 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
698 | ||
699 | /* | |
700 | * If the superblock version is up to where we support new format | |
701 | * inodes and this is currently an old format inode, then change | |
702 | * the inode version number now. This way we only do the conversion | |
703 | * here rather than here and in the flush/logging code. | |
704 | */ | |
93848a99 | 705 | if (xfs_sb_version_hasnlink(&mp->m_sb) && |
51ce16d5 CH |
706 | ip->i_d.di_version == 1) { |
707 | ip->i_d.di_version = 2; | |
1da177e4 LT |
708 | /* |
709 | * We've already zeroed the old link count, the projid field, | |
710 | * and the pad field. | |
711 | */ | |
712 | } | |
713 | ||
714 | /* | |
715 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
716 | */ | |
51ce16d5 | 717 | if ((prid != 0) && (ip->i_d.di_version == 1)) |
1da177e4 LT |
718 | xfs_bump_ino_vers2(tp, ip); |
719 | ||
bd186aa9 | 720 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 721 | ip->i_d.di_gid = pip->i_d.di_gid; |
abbede1b | 722 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
1da177e4 LT |
723 | ip->i_d.di_mode |= S_ISGID; |
724 | } | |
725 | } | |
726 | ||
727 | /* | |
728 | * If the group ID of the new file does not match the effective group | |
729 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
730 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
731 | */ | |
732 | if ((irix_sgid_inherit) && | |
733 | (ip->i_d.di_mode & S_ISGID) && | |
734 | (!in_group_p((gid_t)ip->i_d.di_gid))) { | |
735 | ip->i_d.di_mode &= ~S_ISGID; | |
736 | } | |
737 | ||
738 | ip->i_d.di_size = 0; | |
739 | ip->i_d.di_nextents = 0; | |
740 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 CH |
741 | |
742 | nanotime(&tv); | |
743 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; | |
744 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; | |
745 | ip->i_d.di_atime = ip->i_d.di_mtime; | |
746 | ip->i_d.di_ctime = ip->i_d.di_mtime; | |
747 | ||
1da177e4 LT |
748 | /* |
749 | * di_gen will have been taken care of in xfs_iread. | |
750 | */ | |
751 | ip->i_d.di_extsize = 0; | |
752 | ip->i_d.di_dmevmask = 0; | |
753 | ip->i_d.di_dmstate = 0; | |
754 | ip->i_d.di_flags = 0; | |
93848a99 CH |
755 | |
756 | if (ip->i_d.di_version == 3) { | |
757 | ASSERT(ip->i_d.di_ino == ino); | |
758 | ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid)); | |
759 | ip->i_d.di_crc = 0; | |
760 | ip->i_d.di_changecount = 1; | |
761 | ip->i_d.di_lsn = 0; | |
762 | ip->i_d.di_flags2 = 0; | |
763 | memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2)); | |
764 | ip->i_d.di_crtime = ip->i_d.di_mtime; | |
765 | } | |
766 | ||
767 | ||
1da177e4 LT |
768 | flags = XFS_ILOG_CORE; |
769 | switch (mode & S_IFMT) { | |
770 | case S_IFIFO: | |
771 | case S_IFCHR: | |
772 | case S_IFBLK: | |
773 | case S_IFSOCK: | |
774 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
775 | ip->i_df.if_u2.if_rdev = rdev; | |
776 | ip->i_df.if_flags = 0; | |
777 | flags |= XFS_ILOG_DEV; | |
778 | break; | |
779 | case S_IFREG: | |
bf904248 DC |
780 | /* |
781 | * we can't set up filestreams until after the VFS inode | |
782 | * is set up properly. | |
783 | */ | |
784 | if (pip && xfs_inode_is_filestream(pip)) | |
785 | filestreams = 1; | |
2a82b8be | 786 | /* fall through */ |
1da177e4 | 787 | case S_IFDIR: |
b11f94d5 | 788 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
789 | uint di_flags = 0; |
790 | ||
abbede1b | 791 | if (S_ISDIR(mode)) { |
365ca83d NS |
792 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
793 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
794 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
795 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
796 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
797 | } | |
abbede1b | 798 | } else if (S_ISREG(mode)) { |
613d7043 | 799 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 800 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
801 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
802 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
803 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
804 | } | |
1da177e4 LT |
805 | } |
806 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
807 | xfs_inherit_noatime) | |
365ca83d | 808 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
809 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
810 | xfs_inherit_nodump) | |
365ca83d | 811 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
812 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
813 | xfs_inherit_sync) | |
365ca83d | 814 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
815 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
816 | xfs_inherit_nosymlinks) | |
365ca83d NS |
817 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
818 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
819 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
820 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
821 | xfs_inherit_nodefrag) | |
822 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
823 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
824 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 825 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
826 | } |
827 | /* FALLTHROUGH */ | |
828 | case S_IFLNK: | |
829 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
830 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
831 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
832 | ip->i_df.if_u1.if_extents = NULL; | |
833 | break; | |
834 | default: | |
835 | ASSERT(0); | |
836 | } | |
837 | /* | |
838 | * Attribute fork settings for new inode. | |
839 | */ | |
840 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
841 | ip->i_d.di_anextents = 0; | |
842 | ||
843 | /* | |
844 | * Log the new values stuffed into the inode. | |
845 | */ | |
ddc3415a | 846 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
847 | xfs_trans_log_inode(tp, ip, flags); |
848 | ||
b83bd138 | 849 | /* now that we have an i_mode we can setup inode ops and unlock */ |
41be8bed | 850 | xfs_setup_inode(ip); |
1da177e4 | 851 | |
bf904248 DC |
852 | /* now we have set up the vfs inode we can associate the filestream */ |
853 | if (filestreams) { | |
854 | error = xfs_filestream_associate(pip, ip); | |
855 | if (error < 0) | |
856 | return -error; | |
857 | if (!error) | |
858 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
859 | } | |
860 | ||
1da177e4 LT |
861 | *ipp = ip; |
862 | return 0; | |
863 | } | |
864 | ||
c24b5dfa DC |
865 | int |
866 | xfs_create( | |
867 | xfs_inode_t *dp, | |
868 | struct xfs_name *name, | |
869 | umode_t mode, | |
870 | xfs_dev_t rdev, | |
871 | xfs_inode_t **ipp) | |
872 | { | |
873 | int is_dir = S_ISDIR(mode); | |
874 | struct xfs_mount *mp = dp->i_mount; | |
875 | struct xfs_inode *ip = NULL; | |
876 | struct xfs_trans *tp = NULL; | |
877 | int error; | |
878 | xfs_bmap_free_t free_list; | |
879 | xfs_fsblock_t first_block; | |
880 | bool unlock_dp_on_error = false; | |
881 | uint cancel_flags; | |
882 | int committed; | |
883 | prid_t prid; | |
884 | struct xfs_dquot *udqp = NULL; | |
885 | struct xfs_dquot *gdqp = NULL; | |
886 | struct xfs_dquot *pdqp = NULL; | |
887 | uint resblks; | |
888 | uint log_res; | |
889 | uint log_count; | |
890 | ||
891 | trace_xfs_create(dp, name); | |
892 | ||
893 | if (XFS_FORCED_SHUTDOWN(mp)) | |
894 | return XFS_ERROR(EIO); | |
895 | ||
896 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
897 | prid = xfs_get_projid(dp); | |
898 | else | |
899 | prid = XFS_PROJID_DEFAULT; | |
900 | ||
901 | /* | |
902 | * Make sure that we have allocated dquot(s) on disk. | |
903 | */ | |
904 | error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid, | |
905 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, | |
906 | &udqp, &gdqp, &pdqp); | |
907 | if (error) | |
908 | return error; | |
909 | ||
910 | if (is_dir) { | |
911 | rdev = 0; | |
912 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
913 | log_res = XFS_MKDIR_LOG_RES(mp); | |
914 | log_count = XFS_MKDIR_LOG_COUNT; | |
915 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); | |
916 | } else { | |
917 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
918 | log_res = XFS_CREATE_LOG_RES(mp); | |
919 | log_count = XFS_CREATE_LOG_COUNT; | |
920 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); | |
921 | } | |
922 | ||
923 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
924 | ||
925 | /* | |
926 | * Initially assume that the file does not exist and | |
927 | * reserve the resources for that case. If that is not | |
928 | * the case we'll drop the one we have and get a more | |
929 | * appropriate transaction later. | |
930 | */ | |
931 | error = xfs_trans_reserve(tp, resblks, log_res, 0, | |
932 | XFS_TRANS_PERM_LOG_RES, log_count); | |
933 | if (error == ENOSPC) { | |
934 | /* flush outstanding delalloc blocks and retry */ | |
935 | xfs_flush_inodes(mp); | |
936 | error = xfs_trans_reserve(tp, resblks, log_res, 0, | |
937 | XFS_TRANS_PERM_LOG_RES, log_count); | |
938 | } | |
939 | if (error == ENOSPC) { | |
940 | /* No space at all so try a "no-allocation" reservation */ | |
941 | resblks = 0; | |
942 | error = xfs_trans_reserve(tp, 0, log_res, 0, | |
943 | XFS_TRANS_PERM_LOG_RES, log_count); | |
944 | } | |
945 | if (error) { | |
946 | cancel_flags = 0; | |
947 | goto out_trans_cancel; | |
948 | } | |
949 | ||
950 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); | |
951 | unlock_dp_on_error = true; | |
952 | ||
953 | xfs_bmap_init(&free_list, &first_block); | |
954 | ||
955 | /* | |
956 | * Reserve disk quota and the inode. | |
957 | */ | |
958 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
959 | pdqp, resblks, 1, 0); | |
960 | if (error) | |
961 | goto out_trans_cancel; | |
962 | ||
963 | error = xfs_dir_canenter(tp, dp, name, resblks); | |
964 | if (error) | |
965 | goto out_trans_cancel; | |
966 | ||
967 | /* | |
968 | * A newly created regular or special file just has one directory | |
969 | * entry pointing to them, but a directory also the "." entry | |
970 | * pointing to itself. | |
971 | */ | |
972 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, | |
973 | prid, resblks > 0, &ip, &committed); | |
974 | if (error) { | |
975 | if (error == ENOSPC) | |
976 | goto out_trans_cancel; | |
977 | goto out_trans_abort; | |
978 | } | |
979 | ||
980 | /* | |
981 | * Now we join the directory inode to the transaction. We do not do it | |
982 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
983 | * (and release all the locks). An error from here on will result in | |
984 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
985 | * error path. | |
986 | */ | |
987 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
988 | unlock_dp_on_error = false; | |
989 | ||
990 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, | |
991 | &first_block, &free_list, resblks ? | |
992 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
993 | if (error) { | |
994 | ASSERT(error != ENOSPC); | |
995 | goto out_trans_abort; | |
996 | } | |
997 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
998 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
999 | ||
1000 | if (is_dir) { | |
1001 | error = xfs_dir_init(tp, ip, dp); | |
1002 | if (error) | |
1003 | goto out_bmap_cancel; | |
1004 | ||
1005 | error = xfs_bumplink(tp, dp); | |
1006 | if (error) | |
1007 | goto out_bmap_cancel; | |
1008 | } | |
1009 | ||
1010 | /* | |
1011 | * If this is a synchronous mount, make sure that the | |
1012 | * create transaction goes to disk before returning to | |
1013 | * the user. | |
1014 | */ | |
1015 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1016 | xfs_trans_set_sync(tp); | |
1017 | ||
1018 | /* | |
1019 | * Attach the dquot(s) to the inodes and modify them incore. | |
1020 | * These ids of the inode couldn't have changed since the new | |
1021 | * inode has been locked ever since it was created. | |
1022 | */ | |
1023 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1024 | ||
1025 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1026 | if (error) | |
1027 | goto out_bmap_cancel; | |
1028 | ||
1029 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1030 | if (error) | |
1031 | goto out_release_inode; | |
1032 | ||
1033 | xfs_qm_dqrele(udqp); | |
1034 | xfs_qm_dqrele(gdqp); | |
1035 | xfs_qm_dqrele(pdqp); | |
1036 | ||
1037 | *ipp = ip; | |
1038 | return 0; | |
1039 | ||
1040 | out_bmap_cancel: | |
1041 | xfs_bmap_cancel(&free_list); | |
1042 | out_trans_abort: | |
1043 | cancel_flags |= XFS_TRANS_ABORT; | |
1044 | out_trans_cancel: | |
1045 | xfs_trans_cancel(tp, cancel_flags); | |
1046 | out_release_inode: | |
1047 | /* | |
1048 | * Wait until after the current transaction is aborted to | |
1049 | * release the inode. This prevents recursive transactions | |
1050 | * and deadlocks from xfs_inactive. | |
1051 | */ | |
1052 | if (ip) | |
1053 | IRELE(ip); | |
1054 | ||
1055 | xfs_qm_dqrele(udqp); | |
1056 | xfs_qm_dqrele(gdqp); | |
1057 | xfs_qm_dqrele(pdqp); | |
1058 | ||
1059 | if (unlock_dp_on_error) | |
1060 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1061 | return error; | |
1062 | } | |
1063 | ||
1064 | int | |
1065 | xfs_link( | |
1066 | xfs_inode_t *tdp, | |
1067 | xfs_inode_t *sip, | |
1068 | struct xfs_name *target_name) | |
1069 | { | |
1070 | xfs_mount_t *mp = tdp->i_mount; | |
1071 | xfs_trans_t *tp; | |
1072 | int error; | |
1073 | xfs_bmap_free_t free_list; | |
1074 | xfs_fsblock_t first_block; | |
1075 | int cancel_flags; | |
1076 | int committed; | |
1077 | int resblks; | |
1078 | ||
1079 | trace_xfs_link(tdp, target_name); | |
1080 | ||
1081 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); | |
1082 | ||
1083 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1084 | return XFS_ERROR(EIO); | |
1085 | ||
1086 | error = xfs_qm_dqattach(sip, 0); | |
1087 | if (error) | |
1088 | goto std_return; | |
1089 | ||
1090 | error = xfs_qm_dqattach(tdp, 0); | |
1091 | if (error) | |
1092 | goto std_return; | |
1093 | ||
1094 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
1095 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1096 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); | |
1097 | error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0, | |
1098 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1099 | if (error == ENOSPC) { | |
1100 | resblks = 0; | |
1101 | error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0, | |
1102 | XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT); | |
1103 | } | |
1104 | if (error) { | |
1105 | cancel_flags = 0; | |
1106 | goto error_return; | |
1107 | } | |
1108 | ||
1109 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); | |
1110 | ||
1111 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
1112 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
1113 | ||
1114 | /* | |
1115 | * If we are using project inheritance, we only allow hard link | |
1116 | * creation in our tree when the project IDs are the same; else | |
1117 | * the tree quota mechanism could be circumvented. | |
1118 | */ | |
1119 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1120 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
1121 | error = XFS_ERROR(EXDEV); | |
1122 | goto error_return; | |
1123 | } | |
1124 | ||
1125 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); | |
1126 | if (error) | |
1127 | goto error_return; | |
1128 | ||
1129 | xfs_bmap_init(&free_list, &first_block); | |
1130 | ||
1131 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, | |
1132 | &first_block, &free_list, resblks); | |
1133 | if (error) | |
1134 | goto abort_return; | |
1135 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1136 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1137 | ||
1138 | error = xfs_bumplink(tp, sip); | |
1139 | if (error) | |
1140 | goto abort_return; | |
1141 | ||
1142 | /* | |
1143 | * If this is a synchronous mount, make sure that the | |
1144 | * link transaction goes to disk before returning to | |
1145 | * the user. | |
1146 | */ | |
1147 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1148 | xfs_trans_set_sync(tp); | |
1149 | } | |
1150 | ||
1151 | error = xfs_bmap_finish (&tp, &free_list, &committed); | |
1152 | if (error) { | |
1153 | xfs_bmap_cancel(&free_list); | |
1154 | goto abort_return; | |
1155 | } | |
1156 | ||
1157 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1158 | ||
1159 | abort_return: | |
1160 | cancel_flags |= XFS_TRANS_ABORT; | |
1161 | error_return: | |
1162 | xfs_trans_cancel(tp, cancel_flags); | |
1163 | std_return: | |
1164 | return error; | |
1165 | } | |
1166 | ||
1da177e4 | 1167 | /* |
8f04c47a CH |
1168 | * Free up the underlying blocks past new_size. The new size must be smaller |
1169 | * than the current size. This routine can be used both for the attribute and | |
1170 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1171 | * |
f6485057 DC |
1172 | * The transaction passed to this routine must have made a permanent log |
1173 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1174 | * given transaction and start new ones, so make sure everything involved in | |
1175 | * the transaction is tidy before calling here. Some transaction will be | |
1176 | * returned to the caller to be committed. The incoming transaction must | |
1177 | * already include the inode, and both inode locks must be held exclusively. | |
1178 | * The inode must also be "held" within the transaction. On return the inode | |
1179 | * will be "held" within the returned transaction. This routine does NOT | |
1180 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1181 | * |
f6485057 DC |
1182 | * If we get an error, we must return with the inode locked and linked into the |
1183 | * current transaction. This keeps things simple for the higher level code, | |
1184 | * because it always knows that the inode is locked and held in the transaction | |
1185 | * that returns to it whether errors occur or not. We don't mark the inode | |
1186 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1187 | */ |
1188 | int | |
8f04c47a CH |
1189 | xfs_itruncate_extents( |
1190 | struct xfs_trans **tpp, | |
1191 | struct xfs_inode *ip, | |
1192 | int whichfork, | |
1193 | xfs_fsize_t new_size) | |
1da177e4 | 1194 | { |
8f04c47a CH |
1195 | struct xfs_mount *mp = ip->i_mount; |
1196 | struct xfs_trans *tp = *tpp; | |
1197 | struct xfs_trans *ntp; | |
1198 | xfs_bmap_free_t free_list; | |
1199 | xfs_fsblock_t first_block; | |
1200 | xfs_fileoff_t first_unmap_block; | |
1201 | xfs_fileoff_t last_block; | |
1202 | xfs_filblks_t unmap_len; | |
1203 | int committed; | |
1204 | int error = 0; | |
1205 | int done = 0; | |
1da177e4 | 1206 | |
0b56185b CH |
1207 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1208 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1209 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1210 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1211 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1212 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1213 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1214 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1215 | |
673e8e59 CH |
1216 | trace_xfs_itruncate_extents_start(ip, new_size); |
1217 | ||
1da177e4 LT |
1218 | /* |
1219 | * Since it is possible for space to become allocated beyond | |
1220 | * the end of the file (in a crash where the space is allocated | |
1221 | * but the inode size is not yet updated), simply remove any | |
1222 | * blocks which show up between the new EOF and the maximum | |
1223 | * possible file size. If the first block to be removed is | |
1224 | * beyond the maximum file size (ie it is the same as last_block), | |
1225 | * then there is nothing to do. | |
1226 | */ | |
8f04c47a | 1227 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1228 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1229 | if (first_unmap_block == last_block) |
1230 | return 0; | |
1231 | ||
1232 | ASSERT(first_unmap_block < last_block); | |
1233 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1234 | while (!done) { |
9d87c319 | 1235 | xfs_bmap_init(&free_list, &first_block); |
8f04c47a | 1236 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1237 | first_unmap_block, unmap_len, |
8f04c47a | 1238 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1239 | XFS_ITRUNC_MAX_EXTENTS, |
3e57ecf6 | 1240 | &first_block, &free_list, |
b4e9181e | 1241 | &done); |
8f04c47a CH |
1242 | if (error) |
1243 | goto out_bmap_cancel; | |
1da177e4 LT |
1244 | |
1245 | /* | |
1246 | * Duplicate the transaction that has the permanent | |
1247 | * reservation and commit the old transaction. | |
1248 | */ | |
8f04c47a | 1249 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
898621d5 | 1250 | if (committed) |
ddc3415a | 1251 | xfs_trans_ijoin(tp, ip, 0); |
8f04c47a CH |
1252 | if (error) |
1253 | goto out_bmap_cancel; | |
1da177e4 LT |
1254 | |
1255 | if (committed) { | |
1256 | /* | |
f6485057 | 1257 | * Mark the inode dirty so it will be logged and |
e5720eec | 1258 | * moved forward in the log as part of every commit. |
1da177e4 | 1259 | */ |
8f04c47a | 1260 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1da177e4 | 1261 | } |
f6485057 | 1262 | |
8f04c47a CH |
1263 | ntp = xfs_trans_dup(tp); |
1264 | error = xfs_trans_commit(tp, 0); | |
1265 | tp = ntp; | |
e5720eec | 1266 | |
ddc3415a | 1267 | xfs_trans_ijoin(tp, ip, 0); |
f6485057 | 1268 | |
cc09c0dc | 1269 | if (error) |
8f04c47a CH |
1270 | goto out; |
1271 | ||
cc09c0dc | 1272 | /* |
8f04c47a | 1273 | * Transaction commit worked ok so we can drop the extra ticket |
cc09c0dc DC |
1274 | * reference that we gained in xfs_trans_dup() |
1275 | */ | |
8f04c47a CH |
1276 | xfs_log_ticket_put(tp->t_ticket); |
1277 | error = xfs_trans_reserve(tp, 0, | |
f6485057 DC |
1278 | XFS_ITRUNCATE_LOG_RES(mp), 0, |
1279 | XFS_TRANS_PERM_LOG_RES, | |
1280 | XFS_ITRUNCATE_LOG_COUNT); | |
1281 | if (error) | |
8f04c47a | 1282 | goto out; |
1da177e4 | 1283 | } |
8f04c47a | 1284 | |
673e8e59 CH |
1285 | /* |
1286 | * Always re-log the inode so that our permanent transaction can keep | |
1287 | * on rolling it forward in the log. | |
1288 | */ | |
1289 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1290 | ||
1291 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1292 | ||
8f04c47a CH |
1293 | out: |
1294 | *tpp = tp; | |
1295 | return error; | |
1296 | out_bmap_cancel: | |
1da177e4 | 1297 | /* |
8f04c47a CH |
1298 | * If the bunmapi call encounters an error, return to the caller where |
1299 | * the transaction can be properly aborted. We just need to make sure | |
1300 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1301 | */ |
8f04c47a CH |
1302 | xfs_bmap_cancel(&free_list); |
1303 | goto out; | |
1304 | } | |
1305 | ||
c24b5dfa DC |
1306 | int |
1307 | xfs_release( | |
1308 | xfs_inode_t *ip) | |
1309 | { | |
1310 | xfs_mount_t *mp = ip->i_mount; | |
1311 | int error; | |
1312 | ||
1313 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) | |
1314 | return 0; | |
1315 | ||
1316 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1317 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1318 | return 0; | |
1319 | ||
1320 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1321 | int truncated; | |
1322 | ||
1323 | /* | |
1324 | * If we are using filestreams, and we have an unlinked | |
1325 | * file that we are processing the last close on, then nothing | |
1326 | * will be able to reopen and write to this file. Purge this | |
1327 | * inode from the filestreams cache so that it doesn't delay | |
1328 | * teardown of the inode. | |
1329 | */ | |
1330 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1331 | xfs_filestream_deassociate(ip); | |
1332 | ||
1333 | /* | |
1334 | * If we previously truncated this file and removed old data | |
1335 | * in the process, we want to initiate "early" writeout on | |
1336 | * the last close. This is an attempt to combat the notorious | |
1337 | * NULL files problem which is particularly noticeable from a | |
1338 | * truncate down, buffered (re-)write (delalloc), followed by | |
1339 | * a crash. What we are effectively doing here is | |
1340 | * significantly reducing the time window where we'd otherwise | |
1341 | * be exposed to that problem. | |
1342 | */ | |
1343 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1344 | if (truncated) { | |
1345 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
1346 | if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) { | |
1347 | error = -filemap_flush(VFS_I(ip)->i_mapping); | |
1348 | if (error) | |
1349 | return error; | |
1350 | } | |
1351 | } | |
1352 | } | |
1353 | ||
1354 | if (ip->i_d.di_nlink == 0) | |
1355 | return 0; | |
1356 | ||
1357 | if (xfs_can_free_eofblocks(ip, false)) { | |
1358 | ||
1359 | /* | |
1360 | * If we can't get the iolock just skip truncating the blocks | |
1361 | * past EOF because we could deadlock with the mmap_sem | |
1362 | * otherwise. We'll get another chance to drop them once the | |
1363 | * last reference to the inode is dropped, so we'll never leak | |
1364 | * blocks permanently. | |
1365 | * | |
1366 | * Further, check if the inode is being opened, written and | |
1367 | * closed frequently and we have delayed allocation blocks | |
1368 | * outstanding (e.g. streaming writes from the NFS server), | |
1369 | * truncating the blocks past EOF will cause fragmentation to | |
1370 | * occur. | |
1371 | * | |
1372 | * In this case don't do the truncation, either, but we have to | |
1373 | * be careful how we detect this case. Blocks beyond EOF show | |
1374 | * up as i_delayed_blks even when the inode is clean, so we | |
1375 | * need to truncate them away first before checking for a dirty | |
1376 | * release. Hence on the first dirty close we will still remove | |
1377 | * the speculative allocation, but after that we will leave it | |
1378 | * in place. | |
1379 | */ | |
1380 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1381 | return 0; | |
1382 | ||
1383 | error = xfs_free_eofblocks(mp, ip, true); | |
1384 | if (error && error != EAGAIN) | |
1385 | return error; | |
1386 | ||
1387 | /* delalloc blocks after truncation means it really is dirty */ | |
1388 | if (ip->i_delayed_blks) | |
1389 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1390 | } | |
1391 | return 0; | |
1392 | } | |
1393 | ||
1394 | /* | |
1395 | * xfs_inactive | |
1396 | * | |
1397 | * This is called when the vnode reference count for the vnode | |
1398 | * goes to zero. If the file has been unlinked, then it must | |
1399 | * now be truncated. Also, we clear all of the read-ahead state | |
1400 | * kept for the inode here since the file is now closed. | |
1401 | */ | |
1402 | int | |
1403 | xfs_inactive( | |
1404 | xfs_inode_t *ip) | |
1405 | { | |
1406 | xfs_bmap_free_t free_list; | |
1407 | xfs_fsblock_t first_block; | |
1408 | int committed; | |
1409 | xfs_trans_t *tp; | |
1410 | xfs_mount_t *mp; | |
1411 | int error; | |
1412 | int truncate = 0; | |
1413 | ||
1414 | /* | |
1415 | * If the inode is already free, then there can be nothing | |
1416 | * to clean up here. | |
1417 | */ | |
1418 | if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) { | |
1419 | ASSERT(ip->i_df.if_real_bytes == 0); | |
1420 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
1421 | return VN_INACTIVE_CACHE; | |
1422 | } | |
1423 | ||
1424 | mp = ip->i_mount; | |
1425 | ||
1426 | error = 0; | |
1427 | ||
1428 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1429 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1430 | goto out; | |
1431 | ||
1432 | if (ip->i_d.di_nlink != 0) { | |
1433 | /* | |
1434 | * force is true because we are evicting an inode from the | |
1435 | * cache. Post-eof blocks must be freed, lest we end up with | |
1436 | * broken free space accounting. | |
1437 | */ | |
1438 | if (xfs_can_free_eofblocks(ip, true)) { | |
1439 | error = xfs_free_eofblocks(mp, ip, false); | |
1440 | if (error) | |
1441 | return VN_INACTIVE_CACHE; | |
1442 | } | |
1443 | goto out; | |
1444 | } | |
1445 | ||
1446 | if (S_ISREG(ip->i_d.di_mode) && | |
1447 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || | |
1448 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
1449 | truncate = 1; | |
1450 | ||
1451 | error = xfs_qm_dqattach(ip, 0); | |
1452 | if (error) | |
1453 | return VN_INACTIVE_CACHE; | |
1454 | ||
1455 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1456 | error = xfs_trans_reserve(tp, 0, | |
1457 | (truncate || S_ISLNK(ip->i_d.di_mode)) ? | |
1458 | XFS_ITRUNCATE_LOG_RES(mp) : | |
1459 | XFS_IFREE_LOG_RES(mp), | |
1460 | 0, | |
1461 | XFS_TRANS_PERM_LOG_RES, | |
1462 | XFS_ITRUNCATE_LOG_COUNT); | |
1463 | if (error) { | |
1464 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1465 | xfs_trans_cancel(tp, 0); | |
1466 | return VN_INACTIVE_CACHE; | |
1467 | } | |
1468 | ||
1469 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1470 | xfs_trans_ijoin(tp, ip, 0); | |
1471 | ||
1472 | if (S_ISLNK(ip->i_d.di_mode)) { | |
1473 | error = xfs_inactive_symlink(ip, &tp); | |
1474 | if (error) | |
1475 | goto out_cancel; | |
1476 | } else if (truncate) { | |
1477 | ip->i_d.di_size = 0; | |
1478 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1479 | ||
1480 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1481 | if (error) | |
1482 | goto out_cancel; | |
1483 | ||
1484 | ASSERT(ip->i_d.di_nextents == 0); | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * If there are attributes associated with the file then blow them away | |
1489 | * now. The code calls a routine that recursively deconstructs the | |
1490 | * attribute fork. We need to just commit the current transaction | |
1491 | * because we can't use it for xfs_attr_inactive(). | |
1492 | */ | |
1493 | if (ip->i_d.di_anextents > 0) { | |
1494 | ASSERT(ip->i_d.di_forkoff != 0); | |
1495 | ||
1496 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1497 | if (error) | |
1498 | goto out_unlock; | |
1499 | ||
1500 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1501 | ||
1502 | error = xfs_attr_inactive(ip); | |
1503 | if (error) | |
1504 | goto out; | |
1505 | ||
1506 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1507 | error = xfs_trans_reserve(tp, 0, | |
1508 | XFS_IFREE_LOG_RES(mp), | |
1509 | 0, XFS_TRANS_PERM_LOG_RES, | |
1510 | XFS_INACTIVE_LOG_COUNT); | |
1511 | if (error) { | |
1512 | xfs_trans_cancel(tp, 0); | |
1513 | goto out; | |
1514 | } | |
1515 | ||
1516 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1517 | xfs_trans_ijoin(tp, ip, 0); | |
1518 | } | |
1519 | ||
1520 | if (ip->i_afp) | |
1521 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1522 | ||
1523 | ASSERT(ip->i_d.di_anextents == 0); | |
1524 | ||
1525 | /* | |
1526 | * Free the inode. | |
1527 | */ | |
1528 | xfs_bmap_init(&free_list, &first_block); | |
1529 | error = xfs_ifree(tp, ip, &free_list); | |
1530 | if (error) { | |
1531 | /* | |
1532 | * If we fail to free the inode, shut down. The cancel | |
1533 | * might do that, we need to make sure. Otherwise the | |
1534 | * inode might be lost for a long time or forever. | |
1535 | */ | |
1536 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1537 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1538 | __func__, error); | |
1539 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1540 | } | |
1541 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1542 | } else { | |
1543 | /* | |
1544 | * Credit the quota account(s). The inode is gone. | |
1545 | */ | |
1546 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1547 | ||
1548 | /* | |
1549 | * Just ignore errors at this point. There is nothing we can | |
1550 | * do except to try to keep going. Make sure it's not a silent | |
1551 | * error. | |
1552 | */ | |
1553 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1554 | if (error) | |
1555 | xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", | |
1556 | __func__, error); | |
1557 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1558 | if (error) | |
1559 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1560 | __func__, error); | |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * Release the dquots held by inode, if any. | |
1565 | */ | |
1566 | xfs_qm_dqdetach(ip); | |
1567 | out_unlock: | |
1568 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1569 | out: | |
1570 | return VN_INACTIVE_CACHE; | |
1571 | out_cancel: | |
1572 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1573 | goto out_unlock; | |
1574 | } | |
1575 | ||
1da177e4 LT |
1576 | /* |
1577 | * This is called when the inode's link count goes to 0. | |
1578 | * We place the on-disk inode on a list in the AGI. It | |
1579 | * will be pulled from this list when the inode is freed. | |
1580 | */ | |
1581 | int | |
1582 | xfs_iunlink( | |
1583 | xfs_trans_t *tp, | |
1584 | xfs_inode_t *ip) | |
1585 | { | |
1586 | xfs_mount_t *mp; | |
1587 | xfs_agi_t *agi; | |
1588 | xfs_dinode_t *dip; | |
1589 | xfs_buf_t *agibp; | |
1590 | xfs_buf_t *ibp; | |
1da177e4 LT |
1591 | xfs_agino_t agino; |
1592 | short bucket_index; | |
1593 | int offset; | |
1594 | int error; | |
1da177e4 LT |
1595 | |
1596 | ASSERT(ip->i_d.di_nlink == 0); | |
1597 | ASSERT(ip->i_d.di_mode != 0); | |
1da177e4 LT |
1598 | |
1599 | mp = tp->t_mountp; | |
1600 | ||
1da177e4 LT |
1601 | /* |
1602 | * Get the agi buffer first. It ensures lock ordering | |
1603 | * on the list. | |
1604 | */ | |
5e1be0fb | 1605 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1606 | if (error) |
1da177e4 | 1607 | return error; |
1da177e4 | 1608 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1609 | |
1da177e4 LT |
1610 | /* |
1611 | * Get the index into the agi hash table for the | |
1612 | * list this inode will go on. | |
1613 | */ | |
1614 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1615 | ASSERT(agino != 0); | |
1616 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1617 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1618 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1619 | |
69ef921b | 1620 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1621 | /* |
1622 | * There is already another inode in the bucket we need | |
1623 | * to add ourselves to. Add us at the front of the list. | |
1624 | * Here we put the head pointer into our next pointer, | |
1625 | * and then we fall through to point the head at us. | |
1626 | */ | |
475ee413 CH |
1627 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1628 | 0, 0); | |
c319b58b VA |
1629 | if (error) |
1630 | return error; | |
1631 | ||
69ef921b | 1632 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 1633 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 1634 | offset = ip->i_imap.im_boffset + |
1da177e4 | 1635 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
1636 | |
1637 | /* need to recalc the inode CRC if appropriate */ | |
1638 | xfs_dinode_calc_crc(mp, dip); | |
1639 | ||
1da177e4 LT |
1640 | xfs_trans_inode_buf(tp, ibp); |
1641 | xfs_trans_log_buf(tp, ibp, offset, | |
1642 | (offset + sizeof(xfs_agino_t) - 1)); | |
1643 | xfs_inobp_check(mp, ibp); | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * Point the bucket head pointer at the inode being inserted. | |
1648 | */ | |
1649 | ASSERT(agino != 0); | |
16259e7d | 1650 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1651 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1652 | (sizeof(xfs_agino_t) * bucket_index); | |
1653 | xfs_trans_log_buf(tp, agibp, offset, | |
1654 | (offset + sizeof(xfs_agino_t) - 1)); | |
1655 | return 0; | |
1656 | } | |
1657 | ||
1658 | /* | |
1659 | * Pull the on-disk inode from the AGI unlinked list. | |
1660 | */ | |
1661 | STATIC int | |
1662 | xfs_iunlink_remove( | |
1663 | xfs_trans_t *tp, | |
1664 | xfs_inode_t *ip) | |
1665 | { | |
1666 | xfs_ino_t next_ino; | |
1667 | xfs_mount_t *mp; | |
1668 | xfs_agi_t *agi; | |
1669 | xfs_dinode_t *dip; | |
1670 | xfs_buf_t *agibp; | |
1671 | xfs_buf_t *ibp; | |
1672 | xfs_agnumber_t agno; | |
1da177e4 LT |
1673 | xfs_agino_t agino; |
1674 | xfs_agino_t next_agino; | |
1675 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 1676 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 1677 | short bucket_index; |
6fdf8ccc | 1678 | int offset, last_offset = 0; |
1da177e4 | 1679 | int error; |
1da177e4 | 1680 | |
1da177e4 | 1681 | mp = tp->t_mountp; |
1da177e4 | 1682 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
1683 | |
1684 | /* | |
1685 | * Get the agi buffer first. It ensures lock ordering | |
1686 | * on the list. | |
1687 | */ | |
5e1be0fb CH |
1688 | error = xfs_read_agi(mp, tp, agno, &agibp); |
1689 | if (error) | |
1da177e4 | 1690 | return error; |
5e1be0fb | 1691 | |
1da177e4 | 1692 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1693 | |
1da177e4 LT |
1694 | /* |
1695 | * Get the index into the agi hash table for the | |
1696 | * list this inode will go on. | |
1697 | */ | |
1698 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1699 | ASSERT(agino != 0); | |
1700 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 1701 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
1702 | ASSERT(agi->agi_unlinked[bucket_index]); |
1703 | ||
16259e7d | 1704 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 1705 | /* |
475ee413 CH |
1706 | * We're at the head of the list. Get the inode's on-disk |
1707 | * buffer to see if there is anyone after us on the list. | |
1708 | * Only modify our next pointer if it is not already NULLAGINO. | |
1709 | * This saves us the overhead of dealing with the buffer when | |
1710 | * there is no need to change it. | |
1da177e4 | 1711 | */ |
475ee413 CH |
1712 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1713 | 0, 0); | |
1da177e4 | 1714 | if (error) { |
475ee413 | 1715 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 1716 | __func__, error); |
1da177e4 LT |
1717 | return error; |
1718 | } | |
347d1c01 | 1719 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
1720 | ASSERT(next_agino != 0); |
1721 | if (next_agino != NULLAGINO) { | |
347d1c01 | 1722 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 1723 | offset = ip->i_imap.im_boffset + |
1da177e4 | 1724 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
1725 | |
1726 | /* need to recalc the inode CRC if appropriate */ | |
1727 | xfs_dinode_calc_crc(mp, dip); | |
1728 | ||
1da177e4 LT |
1729 | xfs_trans_inode_buf(tp, ibp); |
1730 | xfs_trans_log_buf(tp, ibp, offset, | |
1731 | (offset + sizeof(xfs_agino_t) - 1)); | |
1732 | xfs_inobp_check(mp, ibp); | |
1733 | } else { | |
1734 | xfs_trans_brelse(tp, ibp); | |
1735 | } | |
1736 | /* | |
1737 | * Point the bucket head pointer at the next inode. | |
1738 | */ | |
1739 | ASSERT(next_agino != 0); | |
1740 | ASSERT(next_agino != agino); | |
16259e7d | 1741 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
1742 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1743 | (sizeof(xfs_agino_t) * bucket_index); | |
1744 | xfs_trans_log_buf(tp, agibp, offset, | |
1745 | (offset + sizeof(xfs_agino_t) - 1)); | |
1746 | } else { | |
1747 | /* | |
1748 | * We need to search the list for the inode being freed. | |
1749 | */ | |
16259e7d | 1750 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
1751 | last_ibp = NULL; |
1752 | while (next_agino != agino) { | |
129dbc9a CH |
1753 | struct xfs_imap imap; |
1754 | ||
1755 | if (last_ibp) | |
1da177e4 | 1756 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
1757 | |
1758 | imap.im_blkno = 0; | |
1da177e4 | 1759 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
1760 | |
1761 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
1762 | if (error) { | |
1763 | xfs_warn(mp, | |
1764 | "%s: xfs_imap returned error %d.", | |
1765 | __func__, error); | |
1766 | return error; | |
1767 | } | |
1768 | ||
1769 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
1770 | &last_ibp, 0, 0); | |
1da177e4 | 1771 | if (error) { |
0b932ccc | 1772 | xfs_warn(mp, |
129dbc9a | 1773 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 1774 | __func__, error); |
1da177e4 LT |
1775 | return error; |
1776 | } | |
129dbc9a CH |
1777 | |
1778 | last_offset = imap.im_boffset; | |
347d1c01 | 1779 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
1780 | ASSERT(next_agino != NULLAGINO); |
1781 | ASSERT(next_agino != 0); | |
1782 | } | |
475ee413 | 1783 | |
1da177e4 | 1784 | /* |
475ee413 CH |
1785 | * Now last_ibp points to the buffer previous to us on the |
1786 | * unlinked list. Pull us from the list. | |
1da177e4 | 1787 | */ |
475ee413 CH |
1788 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1789 | 0, 0); | |
1da177e4 | 1790 | if (error) { |
475ee413 | 1791 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 1792 | __func__, error); |
1da177e4 LT |
1793 | return error; |
1794 | } | |
347d1c01 | 1795 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
1796 | ASSERT(next_agino != 0); |
1797 | ASSERT(next_agino != agino); | |
1798 | if (next_agino != NULLAGINO) { | |
347d1c01 | 1799 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 1800 | offset = ip->i_imap.im_boffset + |
1da177e4 | 1801 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
1802 | |
1803 | /* need to recalc the inode CRC if appropriate */ | |
1804 | xfs_dinode_calc_crc(mp, dip); | |
1805 | ||
1da177e4 LT |
1806 | xfs_trans_inode_buf(tp, ibp); |
1807 | xfs_trans_log_buf(tp, ibp, offset, | |
1808 | (offset + sizeof(xfs_agino_t) - 1)); | |
1809 | xfs_inobp_check(mp, ibp); | |
1810 | } else { | |
1811 | xfs_trans_brelse(tp, ibp); | |
1812 | } | |
1813 | /* | |
1814 | * Point the previous inode on the list to the next inode. | |
1815 | */ | |
347d1c01 | 1816 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
1817 | ASSERT(next_agino != 0); |
1818 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
1819 | |
1820 | /* need to recalc the inode CRC if appropriate */ | |
1821 | xfs_dinode_calc_crc(mp, last_dip); | |
1822 | ||
1da177e4 LT |
1823 | xfs_trans_inode_buf(tp, last_ibp); |
1824 | xfs_trans_log_buf(tp, last_ibp, offset, | |
1825 | (offset + sizeof(xfs_agino_t) - 1)); | |
1826 | xfs_inobp_check(mp, last_ibp); | |
1827 | } | |
1828 | return 0; | |
1829 | } | |
1830 | ||
5b3eed75 DC |
1831 | /* |
1832 | * A big issue when freeing the inode cluster is is that we _cannot_ skip any | |
1833 | * inodes that are in memory - they all must be marked stale and attached to | |
1834 | * the cluster buffer. | |
1835 | */ | |
2a30f36d | 1836 | STATIC int |
1da177e4 LT |
1837 | xfs_ifree_cluster( |
1838 | xfs_inode_t *free_ip, | |
1839 | xfs_trans_t *tp, | |
1840 | xfs_ino_t inum) | |
1841 | { | |
1842 | xfs_mount_t *mp = free_ip->i_mount; | |
1843 | int blks_per_cluster; | |
1844 | int nbufs; | |
1845 | int ninodes; | |
5b257b4a | 1846 | int i, j; |
1da177e4 LT |
1847 | xfs_daddr_t blkno; |
1848 | xfs_buf_t *bp; | |
5b257b4a | 1849 | xfs_inode_t *ip; |
1da177e4 LT |
1850 | xfs_inode_log_item_t *iip; |
1851 | xfs_log_item_t *lip; | |
5017e97d | 1852 | struct xfs_perag *pag; |
1da177e4 | 1853 | |
5017e97d | 1854 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
1da177e4 LT |
1855 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { |
1856 | blks_per_cluster = 1; | |
1857 | ninodes = mp->m_sb.sb_inopblock; | |
1858 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
1859 | } else { | |
1860 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
1861 | mp->m_sb.sb_blocksize; | |
1862 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
1863 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
1864 | } | |
1865 | ||
1da177e4 LT |
1866 | for (j = 0; j < nbufs; j++, inum += ninodes) { |
1867 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
1868 | XFS_INO_TO_AGBNO(mp, inum)); | |
1869 | ||
5b257b4a DC |
1870 | /* |
1871 | * We obtain and lock the backing buffer first in the process | |
1872 | * here, as we have to ensure that any dirty inode that we | |
1873 | * can't get the flush lock on is attached to the buffer. | |
1874 | * If we scan the in-memory inodes first, then buffer IO can | |
1875 | * complete before we get a lock on it, and hence we may fail | |
1876 | * to mark all the active inodes on the buffer stale. | |
1877 | */ | |
1878 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
1879 | mp->m_bsize * blks_per_cluster, |
1880 | XBF_UNMAPPED); | |
5b257b4a | 1881 | |
2a30f36d CS |
1882 | if (!bp) |
1883 | return ENOMEM; | |
b0f539de DC |
1884 | |
1885 | /* | |
1886 | * This buffer may not have been correctly initialised as we | |
1887 | * didn't read it from disk. That's not important because we are | |
1888 | * only using to mark the buffer as stale in the log, and to | |
1889 | * attach stale cached inodes on it. That means it will never be | |
1890 | * dispatched for IO. If it is, we want to know about it, and we | |
1891 | * want it to fail. We can acheive this by adding a write | |
1892 | * verifier to the buffer. | |
1893 | */ | |
1813dd64 | 1894 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 1895 | |
5b257b4a DC |
1896 | /* |
1897 | * Walk the inodes already attached to the buffer and mark them | |
1898 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
1899 | * in-memory inode walk can't lock them. By marking them all |
1900 | * stale first, we will not attempt to lock them in the loop | |
1901 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 1902 | */ |
adadbeef | 1903 | lip = bp->b_fspriv; |
5b257b4a DC |
1904 | while (lip) { |
1905 | if (lip->li_type == XFS_LI_INODE) { | |
1906 | iip = (xfs_inode_log_item_t *)lip; | |
1907 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 1908 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
1909 | xfs_trans_ail_copy_lsn(mp->m_ail, |
1910 | &iip->ili_flush_lsn, | |
1911 | &iip->ili_item.li_lsn); | |
1912 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
1913 | } |
1914 | lip = lip->li_bio_list; | |
1915 | } | |
1da177e4 | 1916 | |
5b3eed75 | 1917 | |
1da177e4 | 1918 | /* |
5b257b4a DC |
1919 | * For each inode in memory attempt to add it to the inode |
1920 | * buffer and set it up for being staled on buffer IO | |
1921 | * completion. This is safe as we've locked out tail pushing | |
1922 | * and flushing by locking the buffer. | |
1da177e4 | 1923 | * |
5b257b4a DC |
1924 | * We have already marked every inode that was part of a |
1925 | * transaction stale above, which means there is no point in | |
1926 | * even trying to lock them. | |
1da177e4 | 1927 | */ |
1da177e4 | 1928 | for (i = 0; i < ninodes; i++) { |
5b3eed75 | 1929 | retry: |
1a3e8f3d | 1930 | rcu_read_lock(); |
da353b0d DC |
1931 | ip = radix_tree_lookup(&pag->pag_ici_root, |
1932 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 1933 | |
1a3e8f3d DC |
1934 | /* Inode not in memory, nothing to do */ |
1935 | if (!ip) { | |
1936 | rcu_read_unlock(); | |
1da177e4 LT |
1937 | continue; |
1938 | } | |
1939 | ||
1a3e8f3d DC |
1940 | /* |
1941 | * because this is an RCU protected lookup, we could | |
1942 | * find a recently freed or even reallocated inode | |
1943 | * during the lookup. We need to check under the | |
1944 | * i_flags_lock for a valid inode here. Skip it if it | |
1945 | * is not valid, the wrong inode or stale. | |
1946 | */ | |
1947 | spin_lock(&ip->i_flags_lock); | |
1948 | if (ip->i_ino != inum + i || | |
1949 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
1950 | spin_unlock(&ip->i_flags_lock); | |
1951 | rcu_read_unlock(); | |
1952 | continue; | |
1953 | } | |
1954 | spin_unlock(&ip->i_flags_lock); | |
1955 | ||
5b3eed75 DC |
1956 | /* |
1957 | * Don't try to lock/unlock the current inode, but we | |
1958 | * _cannot_ skip the other inodes that we did not find | |
1959 | * in the list attached to the buffer and are not | |
1960 | * already marked stale. If we can't lock it, back off | |
1961 | * and retry. | |
1962 | */ | |
5b257b4a DC |
1963 | if (ip != free_ip && |
1964 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1a3e8f3d | 1965 | rcu_read_unlock(); |
5b3eed75 DC |
1966 | delay(1); |
1967 | goto retry; | |
1da177e4 | 1968 | } |
1a3e8f3d | 1969 | rcu_read_unlock(); |
1da177e4 | 1970 | |
5b3eed75 | 1971 | xfs_iflock(ip); |
5b257b4a | 1972 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 1973 | |
5b3eed75 DC |
1974 | /* |
1975 | * we don't need to attach clean inodes or those only | |
1976 | * with unlogged changes (which we throw away, anyway). | |
1977 | */ | |
1da177e4 | 1978 | iip = ip->i_itemp; |
5b3eed75 | 1979 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 1980 | ASSERT(ip != free_ip); |
1da177e4 LT |
1981 | xfs_ifunlock(ip); |
1982 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1983 | continue; | |
1984 | } | |
1985 | ||
f5d8d5c4 CH |
1986 | iip->ili_last_fields = iip->ili_fields; |
1987 | iip->ili_fields = 0; | |
1da177e4 | 1988 | iip->ili_logged = 1; |
7b2e2a31 DC |
1989 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
1990 | &iip->ili_item.li_lsn); | |
1da177e4 | 1991 | |
ca30b2a7 CH |
1992 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
1993 | &iip->ili_item); | |
5b257b4a DC |
1994 | |
1995 | if (ip != free_ip) | |
1da177e4 | 1996 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
1997 | } |
1998 | ||
5b3eed75 | 1999 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2000 | xfs_trans_binval(tp, bp); |
2001 | } | |
2002 | ||
5017e97d | 2003 | xfs_perag_put(pag); |
2a30f36d | 2004 | return 0; |
1da177e4 LT |
2005 | } |
2006 | ||
2007 | /* | |
2008 | * This is called to return an inode to the inode free list. | |
2009 | * The inode should already be truncated to 0 length and have | |
2010 | * no pages associated with it. This routine also assumes that | |
2011 | * the inode is already a part of the transaction. | |
2012 | * | |
2013 | * The on-disk copy of the inode will have been added to the list | |
2014 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2015 | * that list atomically with respect to freeing it here. | |
2016 | */ | |
2017 | int | |
2018 | xfs_ifree( | |
2019 | xfs_trans_t *tp, | |
2020 | xfs_inode_t *ip, | |
2021 | xfs_bmap_free_t *flist) | |
2022 | { | |
2023 | int error; | |
2024 | int delete; | |
2025 | xfs_ino_t first_ino; | |
2026 | ||
579aa9ca | 2027 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
2028 | ASSERT(ip->i_d.di_nlink == 0); |
2029 | ASSERT(ip->i_d.di_nextents == 0); | |
2030 | ASSERT(ip->i_d.di_anextents == 0); | |
ce7ae151 | 2031 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
1da177e4 LT |
2032 | ASSERT(ip->i_d.di_nblocks == 0); |
2033 | ||
2034 | /* | |
2035 | * Pull the on-disk inode from the AGI unlinked list. | |
2036 | */ | |
2037 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2038 | if (error) |
1da177e4 | 2039 | return error; |
1da177e4 LT |
2040 | |
2041 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
1baaed8f | 2042 | if (error) |
1da177e4 | 2043 | return error; |
1baaed8f | 2044 | |
1da177e4 LT |
2045 | ip->i_d.di_mode = 0; /* mark incore inode as free */ |
2046 | ip->i_d.di_flags = 0; | |
2047 | ip->i_d.di_dmevmask = 0; | |
2048 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2049 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2050 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2051 | /* | |
2052 | * Bump the generation count so no one will be confused | |
2053 | * by reincarnations of this inode. | |
2054 | */ | |
2055 | ip->i_d.di_gen++; | |
2056 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2057 | ||
1baaed8f | 2058 | if (delete) |
2a30f36d | 2059 | error = xfs_ifree_cluster(ip, tp, first_ino); |
1da177e4 | 2060 | |
2a30f36d | 2061 | return error; |
1da177e4 LT |
2062 | } |
2063 | ||
1da177e4 | 2064 | /* |
60ec6783 CH |
2065 | * This is called to unpin an inode. The caller must have the inode locked |
2066 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2067 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2068 | */ |
60ec6783 | 2069 | static void |
f392e631 | 2070 | xfs_iunpin( |
60ec6783 | 2071 | struct xfs_inode *ip) |
1da177e4 | 2072 | { |
579aa9ca | 2073 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2074 | |
4aaf15d1 DC |
2075 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2076 | ||
a3f74ffb | 2077 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2078 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2079 | |
a3f74ffb | 2080 | } |
1da177e4 | 2081 | |
f392e631 CH |
2082 | static void |
2083 | __xfs_iunpin_wait( | |
2084 | struct xfs_inode *ip) | |
2085 | { | |
2086 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2087 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2088 | ||
2089 | xfs_iunpin(ip); | |
2090 | ||
2091 | do { | |
2092 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2093 | if (xfs_ipincount(ip)) | |
2094 | io_schedule(); | |
2095 | } while (xfs_ipincount(ip)); | |
2096 | finish_wait(wq, &wait.wait); | |
2097 | } | |
2098 | ||
777df5af | 2099 | void |
a3f74ffb | 2100 | xfs_iunpin_wait( |
60ec6783 | 2101 | struct xfs_inode *ip) |
a3f74ffb | 2102 | { |
f392e631 CH |
2103 | if (xfs_ipincount(ip)) |
2104 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2105 | } |
2106 | ||
c24b5dfa DC |
2107 | int |
2108 | xfs_remove( | |
2109 | xfs_inode_t *dp, | |
2110 | struct xfs_name *name, | |
2111 | xfs_inode_t *ip) | |
2112 | { | |
2113 | xfs_mount_t *mp = dp->i_mount; | |
2114 | xfs_trans_t *tp = NULL; | |
2115 | int is_dir = S_ISDIR(ip->i_d.di_mode); | |
2116 | int error = 0; | |
2117 | xfs_bmap_free_t free_list; | |
2118 | xfs_fsblock_t first_block; | |
2119 | int cancel_flags; | |
2120 | int committed; | |
2121 | int link_zero; | |
2122 | uint resblks; | |
2123 | uint log_count; | |
2124 | ||
2125 | trace_xfs_remove(dp, name); | |
2126 | ||
2127 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2128 | return XFS_ERROR(EIO); | |
2129 | ||
2130 | error = xfs_qm_dqattach(dp, 0); | |
2131 | if (error) | |
2132 | goto std_return; | |
2133 | ||
2134 | error = xfs_qm_dqattach(ip, 0); | |
2135 | if (error) | |
2136 | goto std_return; | |
2137 | ||
2138 | if (is_dir) { | |
2139 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
2140 | log_count = XFS_DEFAULT_LOG_COUNT; | |
2141 | } else { | |
2142 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2143 | log_count = XFS_REMOVE_LOG_COUNT; | |
2144 | } | |
2145 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2146 | ||
2147 | /* | |
2148 | * We try to get the real space reservation first, | |
2149 | * allowing for directory btree deletion(s) implying | |
2150 | * possible bmap insert(s). If we can't get the space | |
2151 | * reservation then we use 0 instead, and avoid the bmap | |
2152 | * btree insert(s) in the directory code by, if the bmap | |
2153 | * insert tries to happen, instead trimming the LAST | |
2154 | * block from the directory. | |
2155 | */ | |
2156 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
2157 | error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0, | |
2158 | XFS_TRANS_PERM_LOG_RES, log_count); | |
2159 | if (error == ENOSPC) { | |
2160 | resblks = 0; | |
2161 | error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0, | |
2162 | XFS_TRANS_PERM_LOG_RES, log_count); | |
2163 | } | |
2164 | if (error) { | |
2165 | ASSERT(error != ENOSPC); | |
2166 | cancel_flags = 0; | |
2167 | goto out_trans_cancel; | |
2168 | } | |
2169 | ||
2170 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); | |
2171 | ||
2172 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2173 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2174 | ||
2175 | /* | |
2176 | * If we're removing a directory perform some additional validation. | |
2177 | */ | |
2178 | if (is_dir) { | |
2179 | ASSERT(ip->i_d.di_nlink >= 2); | |
2180 | if (ip->i_d.di_nlink != 2) { | |
2181 | error = XFS_ERROR(ENOTEMPTY); | |
2182 | goto out_trans_cancel; | |
2183 | } | |
2184 | if (!xfs_dir_isempty(ip)) { | |
2185 | error = XFS_ERROR(ENOTEMPTY); | |
2186 | goto out_trans_cancel; | |
2187 | } | |
2188 | } | |
2189 | ||
2190 | xfs_bmap_init(&free_list, &first_block); | |
2191 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, | |
2192 | &first_block, &free_list, resblks); | |
2193 | if (error) { | |
2194 | ASSERT(error != ENOENT); | |
2195 | goto out_bmap_cancel; | |
2196 | } | |
2197 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2198 | ||
2199 | if (is_dir) { | |
2200 | /* | |
2201 | * Drop the link from ip's "..". | |
2202 | */ | |
2203 | error = xfs_droplink(tp, dp); | |
2204 | if (error) | |
2205 | goto out_bmap_cancel; | |
2206 | ||
2207 | /* | |
2208 | * Drop the "." link from ip to self. | |
2209 | */ | |
2210 | error = xfs_droplink(tp, ip); | |
2211 | if (error) | |
2212 | goto out_bmap_cancel; | |
2213 | } else { | |
2214 | /* | |
2215 | * When removing a non-directory we need to log the parent | |
2216 | * inode here. For a directory this is done implicitly | |
2217 | * by the xfs_droplink call for the ".." entry. | |
2218 | */ | |
2219 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2220 | } | |
2221 | ||
2222 | /* | |
2223 | * Drop the link from dp to ip. | |
2224 | */ | |
2225 | error = xfs_droplink(tp, ip); | |
2226 | if (error) | |
2227 | goto out_bmap_cancel; | |
2228 | ||
2229 | /* | |
2230 | * Determine if this is the last link while | |
2231 | * we are in the transaction. | |
2232 | */ | |
2233 | link_zero = (ip->i_d.di_nlink == 0); | |
2234 | ||
2235 | /* | |
2236 | * If this is a synchronous mount, make sure that the | |
2237 | * remove transaction goes to disk before returning to | |
2238 | * the user. | |
2239 | */ | |
2240 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2241 | xfs_trans_set_sync(tp); | |
2242 | ||
2243 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
2244 | if (error) | |
2245 | goto out_bmap_cancel; | |
2246 | ||
2247 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
2248 | if (error) | |
2249 | goto std_return; | |
2250 | ||
2251 | /* | |
2252 | * If we are using filestreams, kill the stream association. | |
2253 | * If the file is still open it may get a new one but that | |
2254 | * will get killed on last close in xfs_close() so we don't | |
2255 | * have to worry about that. | |
2256 | */ | |
2257 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) | |
2258 | xfs_filestream_deassociate(ip); | |
2259 | ||
2260 | return 0; | |
2261 | ||
2262 | out_bmap_cancel: | |
2263 | xfs_bmap_cancel(&free_list); | |
2264 | cancel_flags |= XFS_TRANS_ABORT; | |
2265 | out_trans_cancel: | |
2266 | xfs_trans_cancel(tp, cancel_flags); | |
2267 | std_return: | |
2268 | return error; | |
2269 | } | |
2270 | ||
5c4d97d0 DC |
2271 | STATIC int |
2272 | xfs_iflush_cluster( | |
2273 | xfs_inode_t *ip, | |
2274 | xfs_buf_t *bp) | |
1da177e4 | 2275 | { |
5c4d97d0 DC |
2276 | xfs_mount_t *mp = ip->i_mount; |
2277 | struct xfs_perag *pag; | |
2278 | unsigned long first_index, mask; | |
2279 | unsigned long inodes_per_cluster; | |
2280 | int ilist_size; | |
2281 | xfs_inode_t **ilist; | |
2282 | xfs_inode_t *iq; | |
2283 | int nr_found; | |
2284 | int clcount = 0; | |
2285 | int bufwasdelwri; | |
1da177e4 | 2286 | int i; |
1da177e4 | 2287 | |
5c4d97d0 | 2288 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 2289 | |
5c4d97d0 DC |
2290 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
2291 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); | |
2292 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); | |
2293 | if (!ilist) | |
2294 | goto out_put; | |
1da177e4 | 2295 | |
5c4d97d0 DC |
2296 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
2297 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; | |
2298 | rcu_read_lock(); | |
2299 | /* really need a gang lookup range call here */ | |
2300 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
2301 | first_index, inodes_per_cluster); | |
2302 | if (nr_found == 0) | |
2303 | goto out_free; | |
2304 | ||
2305 | for (i = 0; i < nr_found; i++) { | |
2306 | iq = ilist[i]; | |
2307 | if (iq == ip) | |
bad55843 | 2308 | continue; |
1a3e8f3d DC |
2309 | |
2310 | /* | |
2311 | * because this is an RCU protected lookup, we could find a | |
2312 | * recently freed or even reallocated inode during the lookup. | |
2313 | * We need to check under the i_flags_lock for a valid inode | |
2314 | * here. Skip it if it is not valid or the wrong inode. | |
2315 | */ | |
2316 | spin_lock(&ip->i_flags_lock); | |
2317 | if (!ip->i_ino || | |
2318 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { | |
2319 | spin_unlock(&ip->i_flags_lock); | |
2320 | continue; | |
2321 | } | |
2322 | spin_unlock(&ip->i_flags_lock); | |
2323 | ||
bad55843 DC |
2324 | /* |
2325 | * Do an un-protected check to see if the inode is dirty and | |
2326 | * is a candidate for flushing. These checks will be repeated | |
2327 | * later after the appropriate locks are acquired. | |
2328 | */ | |
33540408 | 2329 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 2330 | continue; |
bad55843 DC |
2331 | |
2332 | /* | |
2333 | * Try to get locks. If any are unavailable or it is pinned, | |
2334 | * then this inode cannot be flushed and is skipped. | |
2335 | */ | |
2336 | ||
2337 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
2338 | continue; | |
2339 | if (!xfs_iflock_nowait(iq)) { | |
2340 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2341 | continue; | |
2342 | } | |
2343 | if (xfs_ipincount(iq)) { | |
2344 | xfs_ifunlock(iq); | |
2345 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2346 | continue; | |
2347 | } | |
2348 | ||
2349 | /* | |
2350 | * arriving here means that this inode can be flushed. First | |
2351 | * re-check that it's dirty before flushing. | |
2352 | */ | |
33540408 DC |
2353 | if (!xfs_inode_clean(iq)) { |
2354 | int error; | |
bad55843 DC |
2355 | error = xfs_iflush_int(iq, bp); |
2356 | if (error) { | |
2357 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2358 | goto cluster_corrupt_out; | |
2359 | } | |
2360 | clcount++; | |
2361 | } else { | |
2362 | xfs_ifunlock(iq); | |
2363 | } | |
2364 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2365 | } | |
2366 | ||
2367 | if (clcount) { | |
2368 | XFS_STATS_INC(xs_icluster_flushcnt); | |
2369 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
2370 | } | |
2371 | ||
2372 | out_free: | |
1a3e8f3d | 2373 | rcu_read_unlock(); |
f0e2d93c | 2374 | kmem_free(ilist); |
44b56e0a DC |
2375 | out_put: |
2376 | xfs_perag_put(pag); | |
bad55843 DC |
2377 | return 0; |
2378 | ||
2379 | ||
2380 | cluster_corrupt_out: | |
2381 | /* | |
2382 | * Corruption detected in the clustering loop. Invalidate the | |
2383 | * inode buffer and shut down the filesystem. | |
2384 | */ | |
1a3e8f3d | 2385 | rcu_read_unlock(); |
bad55843 | 2386 | /* |
43ff2122 | 2387 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
2388 | * brelse can handle it with no problems. If not, shut down the |
2389 | * filesystem before releasing the buffer. | |
2390 | */ | |
43ff2122 | 2391 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
2392 | if (bufwasdelwri) |
2393 | xfs_buf_relse(bp); | |
2394 | ||
2395 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
2396 | ||
2397 | if (!bufwasdelwri) { | |
2398 | /* | |
2399 | * Just like incore_relse: if we have b_iodone functions, | |
2400 | * mark the buffer as an error and call them. Otherwise | |
2401 | * mark it as stale and brelse. | |
2402 | */ | |
cb669ca5 | 2403 | if (bp->b_iodone) { |
bad55843 | 2404 | XFS_BUF_UNDONE(bp); |
c867cb61 | 2405 | xfs_buf_stale(bp); |
5a52c2a5 | 2406 | xfs_buf_ioerror(bp, EIO); |
1a1a3e97 | 2407 | xfs_buf_ioend(bp, 0); |
bad55843 | 2408 | } else { |
c867cb61 | 2409 | xfs_buf_stale(bp); |
bad55843 DC |
2410 | xfs_buf_relse(bp); |
2411 | } | |
2412 | } | |
2413 | ||
2414 | /* | |
2415 | * Unlocks the flush lock | |
2416 | */ | |
04913fdd | 2417 | xfs_iflush_abort(iq, false); |
f0e2d93c | 2418 | kmem_free(ilist); |
44b56e0a | 2419 | xfs_perag_put(pag); |
bad55843 DC |
2420 | return XFS_ERROR(EFSCORRUPTED); |
2421 | } | |
2422 | ||
1da177e4 | 2423 | /* |
4c46819a CH |
2424 | * Flush dirty inode metadata into the backing buffer. |
2425 | * | |
2426 | * The caller must have the inode lock and the inode flush lock held. The | |
2427 | * inode lock will still be held upon return to the caller, and the inode | |
2428 | * flush lock will be released after the inode has reached the disk. | |
2429 | * | |
2430 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
2431 | */ |
2432 | int | |
2433 | xfs_iflush( | |
4c46819a CH |
2434 | struct xfs_inode *ip, |
2435 | struct xfs_buf **bpp) | |
1da177e4 | 2436 | { |
4c46819a CH |
2437 | struct xfs_mount *mp = ip->i_mount; |
2438 | struct xfs_buf *bp; | |
2439 | struct xfs_dinode *dip; | |
1da177e4 | 2440 | int error; |
1da177e4 LT |
2441 | |
2442 | XFS_STATS_INC(xs_iflush_count); | |
2443 | ||
579aa9ca | 2444 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 2445 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 2446 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 2447 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 2448 | |
4c46819a | 2449 | *bpp = NULL; |
1da177e4 | 2450 | |
1da177e4 LT |
2451 | xfs_iunpin_wait(ip); |
2452 | ||
4b6a4688 DC |
2453 | /* |
2454 | * For stale inodes we cannot rely on the backing buffer remaining | |
2455 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 2456 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
2457 | * inodes below. We have to check this after ensuring the inode is |
2458 | * unpinned so that it is safe to reclaim the stale inode after the | |
2459 | * flush call. | |
2460 | */ | |
2461 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
2462 | xfs_ifunlock(ip); | |
2463 | return 0; | |
2464 | } | |
2465 | ||
1da177e4 LT |
2466 | /* |
2467 | * This may have been unpinned because the filesystem is shutting | |
2468 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
2469 | * to disk, because the log record didn't make it to disk. |
2470 | * | |
2471 | * We also have to remove the log item from the AIL in this case, | |
2472 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
2473 | */ |
2474 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
32ce90a4 CH |
2475 | error = XFS_ERROR(EIO); |
2476 | goto abort_out; | |
1da177e4 LT |
2477 | } |
2478 | ||
a3f74ffb DC |
2479 | /* |
2480 | * Get the buffer containing the on-disk inode. | |
2481 | */ | |
475ee413 CH |
2482 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
2483 | 0); | |
a3f74ffb DC |
2484 | if (error || !bp) { |
2485 | xfs_ifunlock(ip); | |
2486 | return error; | |
2487 | } | |
2488 | ||
1da177e4 LT |
2489 | /* |
2490 | * First flush out the inode that xfs_iflush was called with. | |
2491 | */ | |
2492 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 2493 | if (error) |
1da177e4 | 2494 | goto corrupt_out; |
1da177e4 | 2495 | |
a3f74ffb DC |
2496 | /* |
2497 | * If the buffer is pinned then push on the log now so we won't | |
2498 | * get stuck waiting in the write for too long. | |
2499 | */ | |
811e64c7 | 2500 | if (xfs_buf_ispinned(bp)) |
a14a348b | 2501 | xfs_log_force(mp, 0); |
a3f74ffb | 2502 | |
1da177e4 LT |
2503 | /* |
2504 | * inode clustering: | |
2505 | * see if other inodes can be gathered into this write | |
2506 | */ | |
bad55843 DC |
2507 | error = xfs_iflush_cluster(ip, bp); |
2508 | if (error) | |
2509 | goto cluster_corrupt_out; | |
1da177e4 | 2510 | |
4c46819a CH |
2511 | *bpp = bp; |
2512 | return 0; | |
1da177e4 LT |
2513 | |
2514 | corrupt_out: | |
2515 | xfs_buf_relse(bp); | |
7d04a335 | 2516 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 2517 | cluster_corrupt_out: |
32ce90a4 CH |
2518 | error = XFS_ERROR(EFSCORRUPTED); |
2519 | abort_out: | |
1da177e4 LT |
2520 | /* |
2521 | * Unlocks the flush lock | |
2522 | */ | |
04913fdd | 2523 | xfs_iflush_abort(ip, false); |
32ce90a4 | 2524 | return error; |
1da177e4 LT |
2525 | } |
2526 | ||
2527 | ||
2528 | STATIC int | |
2529 | xfs_iflush_int( | |
93848a99 CH |
2530 | struct xfs_inode *ip, |
2531 | struct xfs_buf *bp) | |
1da177e4 | 2532 | { |
93848a99 CH |
2533 | struct xfs_inode_log_item *iip = ip->i_itemp; |
2534 | struct xfs_dinode *dip; | |
2535 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 2536 | |
579aa9ca | 2537 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 2538 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 2539 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 2540 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 2541 | ASSERT(iip != NULL && iip->ili_fields != 0); |
1da177e4 | 2542 | |
1da177e4 | 2543 | /* set *dip = inode's place in the buffer */ |
92bfc6e7 | 2544 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 2545 | |
69ef921b | 2546 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
1da177e4 | 2547 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
6a19d939 DC |
2548 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2549 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
2550 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
2551 | goto corrupt_out; |
2552 | } | |
2553 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
2554 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
6a19d939 DC |
2555 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2556 | "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
2557 | __func__, ip->i_ino, ip, ip->i_d.di_magic); | |
1da177e4 LT |
2558 | goto corrupt_out; |
2559 | } | |
abbede1b | 2560 | if (S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
2561 | if (XFS_TEST_ERROR( |
2562 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
2563 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
2564 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
6a19d939 DC |
2565 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2566 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
2567 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
2568 | goto corrupt_out; |
2569 | } | |
abbede1b | 2570 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
2571 | if (XFS_TEST_ERROR( |
2572 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
2573 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
2574 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
2575 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
6a19d939 DC |
2576 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2577 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
2578 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
2579 | goto corrupt_out; |
2580 | } | |
2581 | } | |
2582 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
2583 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
2584 | XFS_RANDOM_IFLUSH_5)) { | |
6a19d939 DC |
2585 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2586 | "%s: detected corrupt incore inode %Lu, " | |
2587 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
2588 | __func__, ip->i_ino, | |
1da177e4 | 2589 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 2590 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
2591 | goto corrupt_out; |
2592 | } | |
2593 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
2594 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
6a19d939 DC |
2595 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
2596 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
2597 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
2598 | goto corrupt_out; |
2599 | } | |
e60896d8 | 2600 | |
1da177e4 | 2601 | /* |
e60896d8 DC |
2602 | * Inode item log recovery for v1/v2 inodes are dependent on the |
2603 | * di_flushiter count for correct sequencing. We bump the flush | |
2604 | * iteration count so we can detect flushes which postdate a log record | |
2605 | * during recovery. This is redundant as we now log every change and | |
2606 | * hence this can't happen but we need to still do it to ensure | |
2607 | * backwards compatibility with old kernels that predate logging all | |
2608 | * inode changes. | |
1da177e4 | 2609 | */ |
e60896d8 DC |
2610 | if (ip->i_d.di_version < 3) |
2611 | ip->i_d.di_flushiter++; | |
1da177e4 LT |
2612 | |
2613 | /* | |
2614 | * Copy the dirty parts of the inode into the on-disk | |
2615 | * inode. We always copy out the core of the inode, | |
2616 | * because if the inode is dirty at all the core must | |
2617 | * be. | |
2618 | */ | |
81591fe2 | 2619 | xfs_dinode_to_disk(dip, &ip->i_d); |
1da177e4 LT |
2620 | |
2621 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
2622 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
2623 | ip->i_d.di_flushiter = 0; | |
2624 | ||
2625 | /* | |
2626 | * If this is really an old format inode and the superblock version | |
2627 | * has not been updated to support only new format inodes, then | |
2628 | * convert back to the old inode format. If the superblock version | |
2629 | * has been updated, then make the conversion permanent. | |
2630 | */ | |
51ce16d5 CH |
2631 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
2632 | if (ip->i_d.di_version == 1) { | |
62118709 | 2633 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
2634 | /* |
2635 | * Convert it back. | |
2636 | */ | |
2637 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
81591fe2 | 2638 | dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
2639 | } else { |
2640 | /* | |
2641 | * The superblock version has already been bumped, | |
2642 | * so just make the conversion to the new inode | |
2643 | * format permanent. | |
2644 | */ | |
51ce16d5 CH |
2645 | ip->i_d.di_version = 2; |
2646 | dip->di_version = 2; | |
1da177e4 | 2647 | ip->i_d.di_onlink = 0; |
81591fe2 | 2648 | dip->di_onlink = 0; |
1da177e4 | 2649 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
81591fe2 CH |
2650 | memset(&(dip->di_pad[0]), 0, |
2651 | sizeof(dip->di_pad)); | |
6743099c | 2652 | ASSERT(xfs_get_projid(ip) == 0); |
1da177e4 LT |
2653 | } |
2654 | } | |
2655 | ||
e4ac967b DC |
2656 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
2657 | if (XFS_IFORK_Q(ip)) | |
2658 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
1da177e4 LT |
2659 | xfs_inobp_check(mp, bp); |
2660 | ||
2661 | /* | |
f5d8d5c4 CH |
2662 | * We've recorded everything logged in the inode, so we'd like to clear |
2663 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
2664 | * However, we can't stop logging all this information until the data | |
2665 | * we've copied into the disk buffer is written to disk. If we did we | |
2666 | * might overwrite the copy of the inode in the log with all the data | |
2667 | * after re-logging only part of it, and in the face of a crash we | |
2668 | * wouldn't have all the data we need to recover. | |
1da177e4 | 2669 | * |
f5d8d5c4 CH |
2670 | * What we do is move the bits to the ili_last_fields field. When |
2671 | * logging the inode, these bits are moved back to the ili_fields field. | |
2672 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
2673 | * know that the information those bits represent is permanently on | |
2674 | * disk. As long as the flush completes before the inode is logged | |
2675 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 2676 | * |
f5d8d5c4 CH |
2677 | * We can play with the ili_fields bits here, because the inode lock |
2678 | * must be held exclusively in order to set bits there and the flush | |
2679 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
2680 | * done routine can tell whether or not to look in the AIL. Also, store | |
2681 | * the current LSN of the inode so that we can tell whether the item has | |
2682 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
2683 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
2684 | * atomically. | |
1da177e4 | 2685 | */ |
93848a99 CH |
2686 | iip->ili_last_fields = iip->ili_fields; |
2687 | iip->ili_fields = 0; | |
2688 | iip->ili_logged = 1; | |
1da177e4 | 2689 | |
93848a99 CH |
2690 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2691 | &iip->ili_item.li_lsn); | |
1da177e4 | 2692 | |
93848a99 CH |
2693 | /* |
2694 | * Attach the function xfs_iflush_done to the inode's | |
2695 | * buffer. This will remove the inode from the AIL | |
2696 | * and unlock the inode's flush lock when the inode is | |
2697 | * completely written to disk. | |
2698 | */ | |
2699 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 2700 | |
93848a99 CH |
2701 | /* update the lsn in the on disk inode if required */ |
2702 | if (ip->i_d.di_version == 3) | |
2703 | dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); | |
2704 | ||
2705 | /* generate the checksum. */ | |
2706 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 2707 | |
93848a99 CH |
2708 | ASSERT(bp->b_fspriv != NULL); |
2709 | ASSERT(bp->b_iodone != NULL); | |
1da177e4 LT |
2710 | return 0; |
2711 | ||
2712 | corrupt_out: | |
2713 | return XFS_ERROR(EFSCORRUPTED); | |
2714 | } |