]>
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" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef DC |
23 | #include "xfs_format.h" |
24 | #include "xfs_log_format.h" | |
25 | #include "xfs_trans_resv.h" | |
a844f451 | 26 | #include "xfs_inum.h" |
1da177e4 LT |
27 | #include "xfs_sb.h" |
28 | #include "xfs_ag.h" | |
1da177e4 | 29 | #include "xfs_mount.h" |
a4fbe6ab | 30 | #include "xfs_inode.h" |
57062787 | 31 | #include "xfs_da_format.h" |
c24b5dfa | 32 | #include "xfs_da_btree.h" |
c24b5dfa | 33 | #include "xfs_dir2.h" |
a844f451 | 34 | #include "xfs_attr_sf.h" |
c24b5dfa | 35 | #include "xfs_attr.h" |
239880ef DC |
36 | #include "xfs_trans_space.h" |
37 | #include "xfs_trans.h" | |
1da177e4 | 38 | #include "xfs_buf_item.h" |
a844f451 | 39 | #include "xfs_inode_item.h" |
a844f451 NS |
40 | #include "xfs_ialloc.h" |
41 | #include "xfs_bmap.h" | |
68988114 | 42 | #include "xfs_bmap_util.h" |
1da177e4 | 43 | #include "xfs_error.h" |
1da177e4 | 44 | #include "xfs_quota.h" |
a4fbe6ab | 45 | #include "xfs_dinode.h" |
2a82b8be | 46 | #include "xfs_filestream.h" |
93848a99 | 47 | #include "xfs_cksum.h" |
0b1b213f | 48 | #include "xfs_trace.h" |
33479e05 | 49 | #include "xfs_icache.h" |
c24b5dfa | 50 | #include "xfs_symlink.h" |
239880ef DC |
51 | #include "xfs_trans_priv.h" |
52 | #include "xfs_log.h" | |
a4fbe6ab | 53 | #include "xfs_bmap_btree.h" |
1da177e4 | 54 | |
1da177e4 | 55 | kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
56 | |
57 | /* | |
8f04c47a | 58 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
1da177e4 LT |
59 | * freed from a file in a single transaction. |
60 | */ | |
61 | #define XFS_ITRUNC_MAX_EXTENTS 2 | |
62 | ||
63 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); | |
1da177e4 | 64 | |
2a0ec1d9 DC |
65 | /* |
66 | * helper function to extract extent size hint from inode | |
67 | */ | |
68 | xfs_extlen_t | |
69 | xfs_get_extsz_hint( | |
70 | struct xfs_inode *ip) | |
71 | { | |
72 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) | |
73 | return ip->i_d.di_extsize; | |
74 | if (XFS_IS_REALTIME_INODE(ip)) | |
75 | return ip->i_mount->m_sb.sb_rextsize; | |
76 | return 0; | |
77 | } | |
78 | ||
fa96acad DC |
79 | /* |
80 | * This is a wrapper routine around the xfs_ilock() routine used to centralize | |
81 | * some grungy code. It is used in places that wish to lock the inode solely | |
82 | * for reading the extents. The reason these places can't just call | |
83 | * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the | |
84 | * extents from disk for a file in b-tree format. If the inode is in b-tree | |
85 | * format, then we need to lock the inode exclusively until the extents are read | |
86 | * in. Locking it exclusively all the time would limit our parallelism | |
87 | * unnecessarily, though. What we do instead is check to see if the extents | |
88 | * have been read in yet, and only lock the inode exclusively if they have not. | |
89 | * | |
90 | * The function returns a value which should be given to the corresponding | |
91 | * xfs_iunlock_map_shared(). This value is the mode in which the lock was | |
92 | * actually taken. | |
93 | */ | |
94 | uint | |
95 | xfs_ilock_map_shared( | |
96 | xfs_inode_t *ip) | |
97 | { | |
98 | uint lock_mode; | |
99 | ||
100 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && | |
101 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { | |
102 | lock_mode = XFS_ILOCK_EXCL; | |
103 | } else { | |
104 | lock_mode = XFS_ILOCK_SHARED; | |
105 | } | |
106 | ||
107 | xfs_ilock(ip, lock_mode); | |
108 | ||
109 | return lock_mode; | |
110 | } | |
111 | ||
112 | /* | |
113 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). | |
114 | * All it does is call xfs_iunlock() with the given lock_mode. | |
115 | */ | |
116 | void | |
117 | xfs_iunlock_map_shared( | |
118 | xfs_inode_t *ip, | |
119 | unsigned int lock_mode) | |
120 | { | |
121 | xfs_iunlock(ip, lock_mode); | |
122 | } | |
123 | ||
124 | /* | |
125 | * The xfs inode contains 2 locks: a multi-reader lock called the | |
126 | * i_iolock and a multi-reader lock called the i_lock. This routine | |
127 | * allows either or both of the locks to be obtained. | |
128 | * | |
129 | * The 2 locks should always be ordered so that the IO lock is | |
130 | * obtained first in order to prevent deadlock. | |
131 | * | |
132 | * ip -- the inode being locked | |
133 | * lock_flags -- this parameter indicates the inode's locks | |
134 | * to be locked. It can be: | |
135 | * XFS_IOLOCK_SHARED, | |
136 | * XFS_IOLOCK_EXCL, | |
137 | * XFS_ILOCK_SHARED, | |
138 | * XFS_ILOCK_EXCL, | |
139 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, | |
140 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, | |
141 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, | |
142 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | |
143 | */ | |
144 | void | |
145 | xfs_ilock( | |
146 | xfs_inode_t *ip, | |
147 | uint lock_flags) | |
148 | { | |
149 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); | |
150 | ||
151 | /* | |
152 | * You can't set both SHARED and EXCL for the same lock, | |
153 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
154 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
155 | */ | |
156 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
157 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
158 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
159 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
160 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
161 | ||
162 | if (lock_flags & XFS_IOLOCK_EXCL) | |
163 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
164 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
165 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); | |
166 | ||
167 | if (lock_flags & XFS_ILOCK_EXCL) | |
168 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
169 | else if (lock_flags & XFS_ILOCK_SHARED) | |
170 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); | |
171 | } | |
172 | ||
173 | /* | |
174 | * This is just like xfs_ilock(), except that the caller | |
175 | * is guaranteed not to sleep. It returns 1 if it gets | |
176 | * the requested locks and 0 otherwise. If the IO lock is | |
177 | * obtained but the inode lock cannot be, then the IO lock | |
178 | * is dropped before returning. | |
179 | * | |
180 | * ip -- the inode being locked | |
181 | * lock_flags -- this parameter indicates the inode's locks to be | |
182 | * to be locked. See the comment for xfs_ilock() for a list | |
183 | * of valid values. | |
184 | */ | |
185 | int | |
186 | xfs_ilock_nowait( | |
187 | xfs_inode_t *ip, | |
188 | uint lock_flags) | |
189 | { | |
190 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); | |
191 | ||
192 | /* | |
193 | * You can't set both SHARED and EXCL for the same lock, | |
194 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
195 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
196 | */ | |
197 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
198 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
199 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
200 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
201 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
202 | ||
203 | if (lock_flags & XFS_IOLOCK_EXCL) { | |
204 | if (!mrtryupdate(&ip->i_iolock)) | |
205 | goto out; | |
206 | } else if (lock_flags & XFS_IOLOCK_SHARED) { | |
207 | if (!mrtryaccess(&ip->i_iolock)) | |
208 | goto out; | |
209 | } | |
210 | if (lock_flags & XFS_ILOCK_EXCL) { | |
211 | if (!mrtryupdate(&ip->i_lock)) | |
212 | goto out_undo_iolock; | |
213 | } else if (lock_flags & XFS_ILOCK_SHARED) { | |
214 | if (!mrtryaccess(&ip->i_lock)) | |
215 | goto out_undo_iolock; | |
216 | } | |
217 | return 1; | |
218 | ||
219 | out_undo_iolock: | |
220 | if (lock_flags & XFS_IOLOCK_EXCL) | |
221 | mrunlock_excl(&ip->i_iolock); | |
222 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
223 | mrunlock_shared(&ip->i_iolock); | |
224 | out: | |
225 | return 0; | |
226 | } | |
227 | ||
228 | /* | |
229 | * xfs_iunlock() is used to drop the inode locks acquired with | |
230 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass | |
231 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so | |
232 | * that we know which locks to drop. | |
233 | * | |
234 | * ip -- the inode being unlocked | |
235 | * lock_flags -- this parameter indicates the inode's locks to be | |
236 | * to be unlocked. See the comment for xfs_ilock() for a list | |
237 | * of valid values for this parameter. | |
238 | * | |
239 | */ | |
240 | void | |
241 | xfs_iunlock( | |
242 | xfs_inode_t *ip, | |
243 | uint lock_flags) | |
244 | { | |
245 | /* | |
246 | * You can't set both SHARED and EXCL for the same lock, | |
247 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, | |
248 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. | |
249 | */ | |
250 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != | |
251 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); | |
252 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != | |
253 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); | |
254 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); | |
255 | ASSERT(lock_flags != 0); | |
256 | ||
257 | if (lock_flags & XFS_IOLOCK_EXCL) | |
258 | mrunlock_excl(&ip->i_iolock); | |
259 | else if (lock_flags & XFS_IOLOCK_SHARED) | |
260 | mrunlock_shared(&ip->i_iolock); | |
261 | ||
262 | if (lock_flags & XFS_ILOCK_EXCL) | |
263 | mrunlock_excl(&ip->i_lock); | |
264 | else if (lock_flags & XFS_ILOCK_SHARED) | |
265 | mrunlock_shared(&ip->i_lock); | |
266 | ||
267 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); | |
268 | } | |
269 | ||
270 | /* | |
271 | * give up write locks. the i/o lock cannot be held nested | |
272 | * if it is being demoted. | |
273 | */ | |
274 | void | |
275 | xfs_ilock_demote( | |
276 | xfs_inode_t *ip, | |
277 | uint lock_flags) | |
278 | { | |
279 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); | |
280 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); | |
281 | ||
282 | if (lock_flags & XFS_ILOCK_EXCL) | |
283 | mrdemote(&ip->i_lock); | |
284 | if (lock_flags & XFS_IOLOCK_EXCL) | |
285 | mrdemote(&ip->i_iolock); | |
286 | ||
287 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); | |
288 | } | |
289 | ||
742ae1e3 | 290 | #if defined(DEBUG) || defined(XFS_WARN) |
fa96acad DC |
291 | int |
292 | xfs_isilocked( | |
293 | xfs_inode_t *ip, | |
294 | uint lock_flags) | |
295 | { | |
296 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { | |
297 | if (!(lock_flags & XFS_ILOCK_SHARED)) | |
298 | return !!ip->i_lock.mr_writer; | |
299 | return rwsem_is_locked(&ip->i_lock.mr_lock); | |
300 | } | |
301 | ||
302 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { | |
303 | if (!(lock_flags & XFS_IOLOCK_SHARED)) | |
304 | return !!ip->i_iolock.mr_writer; | |
305 | return rwsem_is_locked(&ip->i_iolock.mr_lock); | |
306 | } | |
307 | ||
308 | ASSERT(0); | |
309 | return 0; | |
310 | } | |
311 | #endif | |
312 | ||
c24b5dfa DC |
313 | #ifdef DEBUG |
314 | int xfs_locked_n; | |
315 | int xfs_small_retries; | |
316 | int xfs_middle_retries; | |
317 | int xfs_lots_retries; | |
318 | int xfs_lock_delays; | |
319 | #endif | |
320 | ||
321 | /* | |
322 | * Bump the subclass so xfs_lock_inodes() acquires each lock with | |
323 | * a different value | |
324 | */ | |
325 | static inline int | |
326 | xfs_lock_inumorder(int lock_mode, int subclass) | |
327 | { | |
328 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
329 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; | |
330 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) | |
331 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; | |
332 | ||
333 | return lock_mode; | |
334 | } | |
335 | ||
336 | /* | |
337 | * The following routine will lock n inodes in exclusive mode. | |
338 | * We assume the caller calls us with the inodes in i_ino order. | |
339 | * | |
340 | * We need to detect deadlock where an inode that we lock | |
341 | * is in the AIL and we start waiting for another inode that is locked | |
342 | * by a thread in a long running transaction (such as truncate). This can | |
343 | * result in deadlock since the long running trans might need to wait | |
344 | * for the inode we just locked in order to push the tail and free space | |
345 | * in the log. | |
346 | */ | |
347 | void | |
348 | xfs_lock_inodes( | |
349 | xfs_inode_t **ips, | |
350 | int inodes, | |
351 | uint lock_mode) | |
352 | { | |
353 | int attempts = 0, i, j, try_lock; | |
354 | xfs_log_item_t *lp; | |
355 | ||
356 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ | |
357 | ||
358 | try_lock = 0; | |
359 | i = 0; | |
360 | ||
361 | again: | |
362 | for (; i < inodes; i++) { | |
363 | ASSERT(ips[i]); | |
364 | ||
365 | if (i && (ips[i] == ips[i-1])) /* Already locked */ | |
366 | continue; | |
367 | ||
368 | /* | |
369 | * If try_lock is not set yet, make sure all locked inodes | |
370 | * are not in the AIL. | |
371 | * If any are, set try_lock to be used later. | |
372 | */ | |
373 | ||
374 | if (!try_lock) { | |
375 | for (j = (i - 1); j >= 0 && !try_lock; j--) { | |
376 | lp = (xfs_log_item_t *)ips[j]->i_itemp; | |
377 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
378 | try_lock++; | |
379 | } | |
380 | } | |
381 | } | |
382 | ||
383 | /* | |
384 | * If any of the previous locks we have locked is in the AIL, | |
385 | * we must TRY to get the second and subsequent locks. If | |
386 | * we can't get any, we must release all we have | |
387 | * and try again. | |
388 | */ | |
389 | ||
390 | if (try_lock) { | |
391 | /* try_lock must be 0 if i is 0. */ | |
392 | /* | |
393 | * try_lock means we have an inode locked | |
394 | * that is in the AIL. | |
395 | */ | |
396 | ASSERT(i != 0); | |
397 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { | |
398 | attempts++; | |
399 | ||
400 | /* | |
401 | * Unlock all previous guys and try again. | |
402 | * xfs_iunlock will try to push the tail | |
403 | * if the inode is in the AIL. | |
404 | */ | |
405 | ||
406 | for(j = i - 1; j >= 0; j--) { | |
407 | ||
408 | /* | |
409 | * Check to see if we've already | |
410 | * unlocked this one. | |
411 | * Not the first one going back, | |
412 | * and the inode ptr is the same. | |
413 | */ | |
414 | if ((j != (i - 1)) && ips[j] == | |
415 | ips[j+1]) | |
416 | continue; | |
417 | ||
418 | xfs_iunlock(ips[j], lock_mode); | |
419 | } | |
420 | ||
421 | if ((attempts % 5) == 0) { | |
422 | delay(1); /* Don't just spin the CPU */ | |
423 | #ifdef DEBUG | |
424 | xfs_lock_delays++; | |
425 | #endif | |
426 | } | |
427 | i = 0; | |
428 | try_lock = 0; | |
429 | goto again; | |
430 | } | |
431 | } else { | |
432 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); | |
433 | } | |
434 | } | |
435 | ||
436 | #ifdef DEBUG | |
437 | if (attempts) { | |
438 | if (attempts < 5) xfs_small_retries++; | |
439 | else if (attempts < 100) xfs_middle_retries++; | |
440 | else xfs_lots_retries++; | |
441 | } else { | |
442 | xfs_locked_n++; | |
443 | } | |
444 | #endif | |
445 | } | |
446 | ||
447 | /* | |
448 | * xfs_lock_two_inodes() can only be used to lock one type of lock | |
449 | * at a time - the iolock or the ilock, but not both at once. If | |
450 | * we lock both at once, lockdep will report false positives saying | |
451 | * we have violated locking orders. | |
452 | */ | |
453 | void | |
454 | xfs_lock_two_inodes( | |
455 | xfs_inode_t *ip0, | |
456 | xfs_inode_t *ip1, | |
457 | uint lock_mode) | |
458 | { | |
459 | xfs_inode_t *temp; | |
460 | int attempts = 0; | |
461 | xfs_log_item_t *lp; | |
462 | ||
463 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) | |
464 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); | |
465 | ASSERT(ip0->i_ino != ip1->i_ino); | |
466 | ||
467 | if (ip0->i_ino > ip1->i_ino) { | |
468 | temp = ip0; | |
469 | ip0 = ip1; | |
470 | ip1 = temp; | |
471 | } | |
472 | ||
473 | again: | |
474 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); | |
475 | ||
476 | /* | |
477 | * If the first lock we have locked is in the AIL, we must TRY to get | |
478 | * the second lock. If we can't get it, we must release the first one | |
479 | * and try again. | |
480 | */ | |
481 | lp = (xfs_log_item_t *)ip0->i_itemp; | |
482 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { | |
483 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { | |
484 | xfs_iunlock(ip0, lock_mode); | |
485 | if ((++attempts % 5) == 0) | |
486 | delay(1); /* Don't just spin the CPU */ | |
487 | goto again; | |
488 | } | |
489 | } else { | |
490 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); | |
491 | } | |
492 | } | |
493 | ||
494 | ||
fa96acad DC |
495 | void |
496 | __xfs_iflock( | |
497 | struct xfs_inode *ip) | |
498 | { | |
499 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); | |
500 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); | |
501 | ||
502 | do { | |
503 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
504 | if (xfs_isiflocked(ip)) | |
505 | io_schedule(); | |
506 | } while (!xfs_iflock_nowait(ip)); | |
507 | ||
508 | finish_wait(wq, &wait.wait); | |
509 | } | |
510 | ||
1da177e4 LT |
511 | STATIC uint |
512 | _xfs_dic2xflags( | |
1da177e4 LT |
513 | __uint16_t di_flags) |
514 | { | |
515 | uint flags = 0; | |
516 | ||
517 | if (di_flags & XFS_DIFLAG_ANY) { | |
518 | if (di_flags & XFS_DIFLAG_REALTIME) | |
519 | flags |= XFS_XFLAG_REALTIME; | |
520 | if (di_flags & XFS_DIFLAG_PREALLOC) | |
521 | flags |= XFS_XFLAG_PREALLOC; | |
522 | if (di_flags & XFS_DIFLAG_IMMUTABLE) | |
523 | flags |= XFS_XFLAG_IMMUTABLE; | |
524 | if (di_flags & XFS_DIFLAG_APPEND) | |
525 | flags |= XFS_XFLAG_APPEND; | |
526 | if (di_flags & XFS_DIFLAG_SYNC) | |
527 | flags |= XFS_XFLAG_SYNC; | |
528 | if (di_flags & XFS_DIFLAG_NOATIME) | |
529 | flags |= XFS_XFLAG_NOATIME; | |
530 | if (di_flags & XFS_DIFLAG_NODUMP) | |
531 | flags |= XFS_XFLAG_NODUMP; | |
532 | if (di_flags & XFS_DIFLAG_RTINHERIT) | |
533 | flags |= XFS_XFLAG_RTINHERIT; | |
534 | if (di_flags & XFS_DIFLAG_PROJINHERIT) | |
535 | flags |= XFS_XFLAG_PROJINHERIT; | |
536 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) | |
537 | flags |= XFS_XFLAG_NOSYMLINKS; | |
dd9f438e NS |
538 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
539 | flags |= XFS_XFLAG_EXTSIZE; | |
540 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) | |
541 | flags |= XFS_XFLAG_EXTSZINHERIT; | |
d3446eac BN |
542 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
543 | flags |= XFS_XFLAG_NODEFRAG; | |
2a82b8be DC |
544 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
545 | flags |= XFS_XFLAG_FILESTREAM; | |
1da177e4 LT |
546 | } |
547 | ||
548 | return flags; | |
549 | } | |
550 | ||
551 | uint | |
552 | xfs_ip2xflags( | |
553 | xfs_inode_t *ip) | |
554 | { | |
347d1c01 | 555 | xfs_icdinode_t *dic = &ip->i_d; |
1da177e4 | 556 | |
a916e2bd | 557 | return _xfs_dic2xflags(dic->di_flags) | |
45ba598e | 558 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
559 | } |
560 | ||
561 | uint | |
562 | xfs_dic2xflags( | |
45ba598e | 563 | xfs_dinode_t *dip) |
1da177e4 | 564 | { |
81591fe2 | 565 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | |
45ba598e | 566 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
1da177e4 LT |
567 | } |
568 | ||
c24b5dfa DC |
569 | /* |
570 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match | |
571 | * is allowed, otherwise it has to be an exact match. If a CI match is found, | |
572 | * ci_name->name will point to a the actual name (caller must free) or | |
573 | * will be set to NULL if an exact match is found. | |
574 | */ | |
575 | int | |
576 | xfs_lookup( | |
577 | xfs_inode_t *dp, | |
578 | struct xfs_name *name, | |
579 | xfs_inode_t **ipp, | |
580 | struct xfs_name *ci_name) | |
581 | { | |
582 | xfs_ino_t inum; | |
583 | int error; | |
584 | uint lock_mode; | |
585 | ||
586 | trace_xfs_lookup(dp, name); | |
587 | ||
588 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) | |
589 | return XFS_ERROR(EIO); | |
590 | ||
591 | lock_mode = xfs_ilock_map_shared(dp); | |
592 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); | |
593 | xfs_iunlock_map_shared(dp, lock_mode); | |
594 | ||
595 | if (error) | |
596 | goto out; | |
597 | ||
598 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); | |
599 | if (error) | |
600 | goto out_free_name; | |
601 | ||
602 | return 0; | |
603 | ||
604 | out_free_name: | |
605 | if (ci_name) | |
606 | kmem_free(ci_name->name); | |
607 | out: | |
608 | *ipp = NULL; | |
609 | return error; | |
610 | } | |
611 | ||
1da177e4 LT |
612 | /* |
613 | * Allocate an inode on disk and return a copy of its in-core version. | |
614 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev | |
615 | * appropriately within the inode. The uid and gid for the inode are | |
616 | * set according to the contents of the given cred structure. | |
617 | * | |
618 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() | |
cd856db6 CM |
619 | * has a free inode available, call xfs_iget() to obtain the in-core |
620 | * version of the allocated inode. Finally, fill in the inode and | |
621 | * log its initial contents. In this case, ialloc_context would be | |
622 | * set to NULL. | |
1da177e4 | 623 | * |
cd856db6 CM |
624 | * If xfs_dialloc() does not have an available inode, it will replenish |
625 | * its supply by doing an allocation. Since we can only do one | |
626 | * allocation within a transaction without deadlocks, we must commit | |
627 | * the current transaction before returning the inode itself. | |
628 | * In this case, therefore, we will set ialloc_context and return. | |
1da177e4 LT |
629 | * The caller should then commit the current transaction, start a new |
630 | * transaction, and call xfs_ialloc() again to actually get the inode. | |
631 | * | |
632 | * To ensure that some other process does not grab the inode that | |
633 | * was allocated during the first call to xfs_ialloc(), this routine | |
634 | * also returns the [locked] bp pointing to the head of the freelist | |
635 | * as ialloc_context. The caller should hold this buffer across | |
636 | * the commit and pass it back into this routine on the second call. | |
b11f94d5 DC |
637 | * |
638 | * If we are allocating quota inodes, we do not have a parent inode | |
639 | * to attach to or associate with (i.e. pip == NULL) because they | |
640 | * are not linked into the directory structure - they are attached | |
641 | * directly to the superblock - and so have no parent. | |
1da177e4 LT |
642 | */ |
643 | int | |
644 | xfs_ialloc( | |
645 | xfs_trans_t *tp, | |
646 | xfs_inode_t *pip, | |
576b1d67 | 647 | umode_t mode, |
31b084ae | 648 | xfs_nlink_t nlink, |
1da177e4 | 649 | xfs_dev_t rdev, |
6743099c | 650 | prid_t prid, |
1da177e4 LT |
651 | int okalloc, |
652 | xfs_buf_t **ialloc_context, | |
1da177e4 LT |
653 | xfs_inode_t **ipp) |
654 | { | |
93848a99 | 655 | struct xfs_mount *mp = tp->t_mountp; |
1da177e4 LT |
656 | xfs_ino_t ino; |
657 | xfs_inode_t *ip; | |
1da177e4 LT |
658 | uint flags; |
659 | int error; | |
dff35fd4 | 660 | timespec_t tv; |
bf904248 | 661 | int filestreams = 0; |
1da177e4 LT |
662 | |
663 | /* | |
664 | * Call the space management code to pick | |
665 | * the on-disk inode to be allocated. | |
666 | */ | |
b11f94d5 | 667 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
08358906 | 668 | ialloc_context, &ino); |
bf904248 | 669 | if (error) |
1da177e4 | 670 | return error; |
08358906 | 671 | if (*ialloc_context || ino == NULLFSINO) { |
1da177e4 LT |
672 | *ipp = NULL; |
673 | return 0; | |
674 | } | |
675 | ASSERT(*ialloc_context == NULL); | |
676 | ||
677 | /* | |
678 | * Get the in-core inode with the lock held exclusively. | |
679 | * This is because we're setting fields here we need | |
680 | * to prevent others from looking at until we're done. | |
681 | */ | |
93848a99 | 682 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
ec3ba85f | 683 | XFS_ILOCK_EXCL, &ip); |
bf904248 | 684 | if (error) |
1da177e4 | 685 | return error; |
1da177e4 LT |
686 | ASSERT(ip != NULL); |
687 | ||
576b1d67 | 688 | ip->i_d.di_mode = mode; |
1da177e4 LT |
689 | ip->i_d.di_onlink = 0; |
690 | ip->i_d.di_nlink = nlink; | |
691 | ASSERT(ip->i_d.di_nlink == nlink); | |
7aab1b28 DE |
692 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
693 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); | |
6743099c | 694 | xfs_set_projid(ip, prid); |
1da177e4 LT |
695 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
696 | ||
697 | /* | |
698 | * If the superblock version is up to where we support new format | |
699 | * inodes and this is currently an old format inode, then change | |
700 | * the inode version number now. This way we only do the conversion | |
701 | * here rather than here and in the flush/logging code. | |
702 | */ | |
93848a99 | 703 | if (xfs_sb_version_hasnlink(&mp->m_sb) && |
51ce16d5 CH |
704 | ip->i_d.di_version == 1) { |
705 | ip->i_d.di_version = 2; | |
1da177e4 LT |
706 | /* |
707 | * We've already zeroed the old link count, the projid field, | |
708 | * and the pad field. | |
709 | */ | |
710 | } | |
711 | ||
712 | /* | |
713 | * Project ids won't be stored on disk if we are using a version 1 inode. | |
714 | */ | |
51ce16d5 | 715 | if ((prid != 0) && (ip->i_d.di_version == 1)) |
1da177e4 LT |
716 | xfs_bump_ino_vers2(tp, ip); |
717 | ||
bd186aa9 | 718 | if (pip && XFS_INHERIT_GID(pip)) { |
1da177e4 | 719 | ip->i_d.di_gid = pip->i_d.di_gid; |
abbede1b | 720 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
1da177e4 LT |
721 | ip->i_d.di_mode |= S_ISGID; |
722 | } | |
723 | } | |
724 | ||
725 | /* | |
726 | * If the group ID of the new file does not match the effective group | |
727 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared | |
728 | * (and only if the irix_sgid_inherit compatibility variable is set). | |
729 | */ | |
730 | if ((irix_sgid_inherit) && | |
731 | (ip->i_d.di_mode & S_ISGID) && | |
7aab1b28 | 732 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) { |
1da177e4 LT |
733 | ip->i_d.di_mode &= ~S_ISGID; |
734 | } | |
735 | ||
736 | ip->i_d.di_size = 0; | |
737 | ip->i_d.di_nextents = 0; | |
738 | ASSERT(ip->i_d.di_nblocks == 0); | |
dff35fd4 CH |
739 | |
740 | nanotime(&tv); | |
741 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; | |
742 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; | |
743 | ip->i_d.di_atime = ip->i_d.di_mtime; | |
744 | ip->i_d.di_ctime = ip->i_d.di_mtime; | |
745 | ||
1da177e4 LT |
746 | /* |
747 | * di_gen will have been taken care of in xfs_iread. | |
748 | */ | |
749 | ip->i_d.di_extsize = 0; | |
750 | ip->i_d.di_dmevmask = 0; | |
751 | ip->i_d.di_dmstate = 0; | |
752 | ip->i_d.di_flags = 0; | |
93848a99 CH |
753 | |
754 | if (ip->i_d.di_version == 3) { | |
755 | ASSERT(ip->i_d.di_ino == ino); | |
756 | ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid)); | |
757 | ip->i_d.di_crc = 0; | |
758 | ip->i_d.di_changecount = 1; | |
759 | ip->i_d.di_lsn = 0; | |
760 | ip->i_d.di_flags2 = 0; | |
761 | memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2)); | |
762 | ip->i_d.di_crtime = ip->i_d.di_mtime; | |
763 | } | |
764 | ||
765 | ||
1da177e4 LT |
766 | flags = XFS_ILOG_CORE; |
767 | switch (mode & S_IFMT) { | |
768 | case S_IFIFO: | |
769 | case S_IFCHR: | |
770 | case S_IFBLK: | |
771 | case S_IFSOCK: | |
772 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; | |
773 | ip->i_df.if_u2.if_rdev = rdev; | |
774 | ip->i_df.if_flags = 0; | |
775 | flags |= XFS_ILOG_DEV; | |
776 | break; | |
777 | case S_IFREG: | |
bf904248 DC |
778 | /* |
779 | * we can't set up filestreams until after the VFS inode | |
780 | * is set up properly. | |
781 | */ | |
782 | if (pip && xfs_inode_is_filestream(pip)) | |
783 | filestreams = 1; | |
2a82b8be | 784 | /* fall through */ |
1da177e4 | 785 | case S_IFDIR: |
b11f94d5 | 786 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
365ca83d NS |
787 | uint di_flags = 0; |
788 | ||
abbede1b | 789 | if (S_ISDIR(mode)) { |
365ca83d NS |
790 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
791 | di_flags |= XFS_DIFLAG_RTINHERIT; | |
dd9f438e NS |
792 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
793 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; | |
794 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
795 | } | |
abbede1b | 796 | } else if (S_ISREG(mode)) { |
613d7043 | 797 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
365ca83d | 798 | di_flags |= XFS_DIFLAG_REALTIME; |
dd9f438e NS |
799 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
800 | di_flags |= XFS_DIFLAG_EXTSIZE; | |
801 | ip->i_d.di_extsize = pip->i_d.di_extsize; | |
802 | } | |
1da177e4 LT |
803 | } |
804 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && | |
805 | xfs_inherit_noatime) | |
365ca83d | 806 | di_flags |= XFS_DIFLAG_NOATIME; |
1da177e4 LT |
807 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
808 | xfs_inherit_nodump) | |
365ca83d | 809 | di_flags |= XFS_DIFLAG_NODUMP; |
1da177e4 LT |
810 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
811 | xfs_inherit_sync) | |
365ca83d | 812 | di_flags |= XFS_DIFLAG_SYNC; |
1da177e4 LT |
813 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
814 | xfs_inherit_nosymlinks) | |
365ca83d NS |
815 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
816 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
817 | di_flags |= XFS_DIFLAG_PROJINHERIT; | |
d3446eac BN |
818 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
819 | xfs_inherit_nodefrag) | |
820 | di_flags |= XFS_DIFLAG_NODEFRAG; | |
2a82b8be DC |
821 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
822 | di_flags |= XFS_DIFLAG_FILESTREAM; | |
365ca83d | 823 | ip->i_d.di_flags |= di_flags; |
1da177e4 LT |
824 | } |
825 | /* FALLTHROUGH */ | |
826 | case S_IFLNK: | |
827 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; | |
828 | ip->i_df.if_flags = XFS_IFEXTENTS; | |
829 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; | |
830 | ip->i_df.if_u1.if_extents = NULL; | |
831 | break; | |
832 | default: | |
833 | ASSERT(0); | |
834 | } | |
835 | /* | |
836 | * Attribute fork settings for new inode. | |
837 | */ | |
838 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
839 | ip->i_d.di_anextents = 0; | |
840 | ||
841 | /* | |
842 | * Log the new values stuffed into the inode. | |
843 | */ | |
ddc3415a | 844 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
845 | xfs_trans_log_inode(tp, ip, flags); |
846 | ||
b83bd138 | 847 | /* now that we have an i_mode we can setup inode ops and unlock */ |
41be8bed | 848 | xfs_setup_inode(ip); |
1da177e4 | 849 | |
bf904248 DC |
850 | /* now we have set up the vfs inode we can associate the filestream */ |
851 | if (filestreams) { | |
852 | error = xfs_filestream_associate(pip, ip); | |
853 | if (error < 0) | |
854 | return -error; | |
855 | if (!error) | |
856 | xfs_iflags_set(ip, XFS_IFILESTREAM); | |
857 | } | |
858 | ||
1da177e4 LT |
859 | *ipp = ip; |
860 | return 0; | |
861 | } | |
862 | ||
e546cb79 DC |
863 | /* |
864 | * Allocates a new inode from disk and return a pointer to the | |
865 | * incore copy. This routine will internally commit the current | |
866 | * transaction and allocate a new one if the Space Manager needed | |
867 | * to do an allocation to replenish the inode free-list. | |
868 | * | |
869 | * This routine is designed to be called from xfs_create and | |
870 | * xfs_create_dir. | |
871 | * | |
872 | */ | |
873 | int | |
874 | xfs_dir_ialloc( | |
875 | xfs_trans_t **tpp, /* input: current transaction; | |
876 | output: may be a new transaction. */ | |
877 | xfs_inode_t *dp, /* directory within whose allocate | |
878 | the inode. */ | |
879 | umode_t mode, | |
880 | xfs_nlink_t nlink, | |
881 | xfs_dev_t rdev, | |
882 | prid_t prid, /* project id */ | |
883 | int okalloc, /* ok to allocate new space */ | |
884 | xfs_inode_t **ipp, /* pointer to inode; it will be | |
885 | locked. */ | |
886 | int *committed) | |
887 | ||
888 | { | |
889 | xfs_trans_t *tp; | |
890 | xfs_trans_t *ntp; | |
891 | xfs_inode_t *ip; | |
892 | xfs_buf_t *ialloc_context = NULL; | |
893 | int code; | |
e546cb79 DC |
894 | void *dqinfo; |
895 | uint tflags; | |
896 | ||
897 | tp = *tpp; | |
898 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); | |
899 | ||
900 | /* | |
901 | * xfs_ialloc will return a pointer to an incore inode if | |
902 | * the Space Manager has an available inode on the free | |
903 | * list. Otherwise, it will do an allocation and replenish | |
904 | * the freelist. Since we can only do one allocation per | |
905 | * transaction without deadlocks, we will need to commit the | |
906 | * current transaction and start a new one. We will then | |
907 | * need to call xfs_ialloc again to get the inode. | |
908 | * | |
909 | * If xfs_ialloc did an allocation to replenish the freelist, | |
910 | * it returns the bp containing the head of the freelist as | |
911 | * ialloc_context. We will hold a lock on it across the | |
912 | * transaction commit so that no other process can steal | |
913 | * the inode(s) that we've just allocated. | |
914 | */ | |
915 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, | |
916 | &ialloc_context, &ip); | |
917 | ||
918 | /* | |
919 | * Return an error if we were unable to allocate a new inode. | |
920 | * This should only happen if we run out of space on disk or | |
921 | * encounter a disk error. | |
922 | */ | |
923 | if (code) { | |
924 | *ipp = NULL; | |
925 | return code; | |
926 | } | |
927 | if (!ialloc_context && !ip) { | |
928 | *ipp = NULL; | |
929 | return XFS_ERROR(ENOSPC); | |
930 | } | |
931 | ||
932 | /* | |
933 | * If the AGI buffer is non-NULL, then we were unable to get an | |
934 | * inode in one operation. We need to commit the current | |
935 | * transaction and call xfs_ialloc() again. It is guaranteed | |
936 | * to succeed the second time. | |
937 | */ | |
938 | if (ialloc_context) { | |
3d3c8b52 JL |
939 | struct xfs_trans_res tres; |
940 | ||
e546cb79 DC |
941 | /* |
942 | * Normally, xfs_trans_commit releases all the locks. | |
943 | * We call bhold to hang on to the ialloc_context across | |
944 | * the commit. Holding this buffer prevents any other | |
945 | * processes from doing any allocations in this | |
946 | * allocation group. | |
947 | */ | |
948 | xfs_trans_bhold(tp, ialloc_context); | |
949 | /* | |
950 | * Save the log reservation so we can use | |
951 | * them in the next transaction. | |
952 | */ | |
3d3c8b52 JL |
953 | tres.tr_logres = xfs_trans_get_log_res(tp); |
954 | tres.tr_logcount = xfs_trans_get_log_count(tp); | |
e546cb79 DC |
955 | |
956 | /* | |
957 | * We want the quota changes to be associated with the next | |
958 | * transaction, NOT this one. So, detach the dqinfo from this | |
959 | * and attach it to the next transaction. | |
960 | */ | |
961 | dqinfo = NULL; | |
962 | tflags = 0; | |
963 | if (tp->t_dqinfo) { | |
964 | dqinfo = (void *)tp->t_dqinfo; | |
965 | tp->t_dqinfo = NULL; | |
966 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; | |
967 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); | |
968 | } | |
969 | ||
970 | ntp = xfs_trans_dup(tp); | |
971 | code = xfs_trans_commit(tp, 0); | |
972 | tp = ntp; | |
973 | if (committed != NULL) { | |
974 | *committed = 1; | |
975 | } | |
976 | /* | |
977 | * If we get an error during the commit processing, | |
978 | * release the buffer that is still held and return | |
979 | * to the caller. | |
980 | */ | |
981 | if (code) { | |
982 | xfs_buf_relse(ialloc_context); | |
983 | if (dqinfo) { | |
984 | tp->t_dqinfo = dqinfo; | |
985 | xfs_trans_free_dqinfo(tp); | |
986 | } | |
987 | *tpp = ntp; | |
988 | *ipp = NULL; | |
989 | return code; | |
990 | } | |
991 | ||
992 | /* | |
993 | * transaction commit worked ok so we can drop the extra ticket | |
994 | * reference that we gained in xfs_trans_dup() | |
995 | */ | |
996 | xfs_log_ticket_put(tp->t_ticket); | |
3d3c8b52 JL |
997 | tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; |
998 | code = xfs_trans_reserve(tp, &tres, 0, 0); | |
999 | ||
e546cb79 DC |
1000 | /* |
1001 | * Re-attach the quota info that we detached from prev trx. | |
1002 | */ | |
1003 | if (dqinfo) { | |
1004 | tp->t_dqinfo = dqinfo; | |
1005 | tp->t_flags |= tflags; | |
1006 | } | |
1007 | ||
1008 | if (code) { | |
1009 | xfs_buf_relse(ialloc_context); | |
1010 | *tpp = ntp; | |
1011 | *ipp = NULL; | |
1012 | return code; | |
1013 | } | |
1014 | xfs_trans_bjoin(tp, ialloc_context); | |
1015 | ||
1016 | /* | |
1017 | * Call ialloc again. Since we've locked out all | |
1018 | * other allocations in this allocation group, | |
1019 | * this call should always succeed. | |
1020 | */ | |
1021 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, | |
1022 | okalloc, &ialloc_context, &ip); | |
1023 | ||
1024 | /* | |
1025 | * If we get an error at this point, return to the caller | |
1026 | * so that the current transaction can be aborted. | |
1027 | */ | |
1028 | if (code) { | |
1029 | *tpp = tp; | |
1030 | *ipp = NULL; | |
1031 | return code; | |
1032 | } | |
1033 | ASSERT(!ialloc_context && ip); | |
1034 | ||
1035 | } else { | |
1036 | if (committed != NULL) | |
1037 | *committed = 0; | |
1038 | } | |
1039 | ||
1040 | *ipp = ip; | |
1041 | *tpp = tp; | |
1042 | ||
1043 | return 0; | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Decrement the link count on an inode & log the change. | |
1048 | * If this causes the link count to go to zero, initiate the | |
1049 | * logging activity required to truncate a file. | |
1050 | */ | |
1051 | int /* error */ | |
1052 | xfs_droplink( | |
1053 | xfs_trans_t *tp, | |
1054 | xfs_inode_t *ip) | |
1055 | { | |
1056 | int error; | |
1057 | ||
1058 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1059 | ||
1060 | ASSERT (ip->i_d.di_nlink > 0); | |
1061 | ip->i_d.di_nlink--; | |
1062 | drop_nlink(VFS_I(ip)); | |
1063 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1064 | ||
1065 | error = 0; | |
1066 | if (ip->i_d.di_nlink == 0) { | |
1067 | /* | |
1068 | * We're dropping the last link to this file. | |
1069 | * Move the on-disk inode to the AGI unlinked list. | |
1070 | * From xfs_inactive() we will pull the inode from | |
1071 | * the list and free it. | |
1072 | */ | |
1073 | error = xfs_iunlink(tp, ip); | |
1074 | } | |
1075 | return error; | |
1076 | } | |
1077 | ||
1078 | /* | |
1079 | * This gets called when the inode's version needs to be changed from 1 to 2. | |
1080 | * Currently this happens when the nlink field overflows the old 16-bit value | |
1081 | * or when chproj is called to change the project for the first time. | |
1082 | * As a side effect the superblock version will also get rev'd | |
1083 | * to contain the NLINK bit. | |
1084 | */ | |
1085 | void | |
1086 | xfs_bump_ino_vers2( | |
1087 | xfs_trans_t *tp, | |
1088 | xfs_inode_t *ip) | |
1089 | { | |
1090 | xfs_mount_t *mp; | |
1091 | ||
1092 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); | |
1093 | ASSERT(ip->i_d.di_version == 1); | |
1094 | ||
1095 | ip->i_d.di_version = 2; | |
1096 | ip->i_d.di_onlink = 0; | |
1097 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); | |
1098 | mp = tp->t_mountp; | |
1099 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { | |
1100 | spin_lock(&mp->m_sb_lock); | |
1101 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { | |
1102 | xfs_sb_version_addnlink(&mp->m_sb); | |
1103 | spin_unlock(&mp->m_sb_lock); | |
1104 | xfs_mod_sb(tp, XFS_SB_VERSIONNUM); | |
1105 | } else { | |
1106 | spin_unlock(&mp->m_sb_lock); | |
1107 | } | |
1108 | } | |
1109 | /* Caller must log the inode */ | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * Increment the link count on an inode & log the change. | |
1114 | */ | |
1115 | int | |
1116 | xfs_bumplink( | |
1117 | xfs_trans_t *tp, | |
1118 | xfs_inode_t *ip) | |
1119 | { | |
1120 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); | |
1121 | ||
1122 | ASSERT(ip->i_d.di_nlink > 0); | |
1123 | ip->i_d.di_nlink++; | |
1124 | inc_nlink(VFS_I(ip)); | |
1125 | if ((ip->i_d.di_version == 1) && | |
1126 | (ip->i_d.di_nlink > XFS_MAXLINK_1)) { | |
1127 | /* | |
1128 | * The inode has increased its number of links beyond | |
1129 | * what can fit in an old format inode. It now needs | |
1130 | * to be converted to a version 2 inode with a 32 bit | |
1131 | * link count. If this is the first inode in the file | |
1132 | * system to do this, then we need to bump the superblock | |
1133 | * version number as well. | |
1134 | */ | |
1135 | xfs_bump_ino_vers2(tp, ip); | |
1136 | } | |
1137 | ||
1138 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1139 | return 0; | |
1140 | } | |
1141 | ||
c24b5dfa DC |
1142 | int |
1143 | xfs_create( | |
1144 | xfs_inode_t *dp, | |
1145 | struct xfs_name *name, | |
1146 | umode_t mode, | |
1147 | xfs_dev_t rdev, | |
1148 | xfs_inode_t **ipp) | |
1149 | { | |
1150 | int is_dir = S_ISDIR(mode); | |
1151 | struct xfs_mount *mp = dp->i_mount; | |
1152 | struct xfs_inode *ip = NULL; | |
1153 | struct xfs_trans *tp = NULL; | |
1154 | int error; | |
1155 | xfs_bmap_free_t free_list; | |
1156 | xfs_fsblock_t first_block; | |
1157 | bool unlock_dp_on_error = false; | |
1158 | uint cancel_flags; | |
1159 | int committed; | |
1160 | prid_t prid; | |
1161 | struct xfs_dquot *udqp = NULL; | |
1162 | struct xfs_dquot *gdqp = NULL; | |
1163 | struct xfs_dquot *pdqp = NULL; | |
3d3c8b52 | 1164 | struct xfs_trans_res tres; |
c24b5dfa | 1165 | uint resblks; |
c24b5dfa DC |
1166 | |
1167 | trace_xfs_create(dp, name); | |
1168 | ||
1169 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1170 | return XFS_ERROR(EIO); | |
1171 | ||
1172 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) | |
1173 | prid = xfs_get_projid(dp); | |
1174 | else | |
1175 | prid = XFS_PROJID_DEFAULT; | |
1176 | ||
1177 | /* | |
1178 | * Make sure that we have allocated dquot(s) on disk. | |
1179 | */ | |
7aab1b28 DE |
1180 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
1181 | xfs_kgid_to_gid(current_fsgid()), prid, | |
c24b5dfa DC |
1182 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
1183 | &udqp, &gdqp, &pdqp); | |
1184 | if (error) | |
1185 | return error; | |
1186 | ||
1187 | if (is_dir) { | |
1188 | rdev = 0; | |
1189 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); | |
3d3c8b52 JL |
1190 | tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres; |
1191 | tres.tr_logcount = XFS_MKDIR_LOG_COUNT; | |
c24b5dfa DC |
1192 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); |
1193 | } else { | |
1194 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); | |
3d3c8b52 JL |
1195 | tres.tr_logres = M_RES(mp)->tr_create.tr_logres; |
1196 | tres.tr_logcount = XFS_CREATE_LOG_COUNT; | |
c24b5dfa DC |
1197 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); |
1198 | } | |
1199 | ||
1200 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1201 | ||
1202 | /* | |
1203 | * Initially assume that the file does not exist and | |
1204 | * reserve the resources for that case. If that is not | |
1205 | * the case we'll drop the one we have and get a more | |
1206 | * appropriate transaction later. | |
1207 | */ | |
3d3c8b52 JL |
1208 | tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; |
1209 | error = xfs_trans_reserve(tp, &tres, resblks, 0); | |
c24b5dfa DC |
1210 | if (error == ENOSPC) { |
1211 | /* flush outstanding delalloc blocks and retry */ | |
1212 | xfs_flush_inodes(mp); | |
3d3c8b52 | 1213 | error = xfs_trans_reserve(tp, &tres, resblks, 0); |
c24b5dfa DC |
1214 | } |
1215 | if (error == ENOSPC) { | |
1216 | /* No space at all so try a "no-allocation" reservation */ | |
1217 | resblks = 0; | |
3d3c8b52 | 1218 | error = xfs_trans_reserve(tp, &tres, 0, 0); |
c24b5dfa DC |
1219 | } |
1220 | if (error) { | |
1221 | cancel_flags = 0; | |
1222 | goto out_trans_cancel; | |
1223 | } | |
1224 | ||
1225 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); | |
1226 | unlock_dp_on_error = true; | |
1227 | ||
1228 | xfs_bmap_init(&free_list, &first_block); | |
1229 | ||
1230 | /* | |
1231 | * Reserve disk quota and the inode. | |
1232 | */ | |
1233 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, | |
1234 | pdqp, resblks, 1, 0); | |
1235 | if (error) | |
1236 | goto out_trans_cancel; | |
1237 | ||
1238 | error = xfs_dir_canenter(tp, dp, name, resblks); | |
1239 | if (error) | |
1240 | goto out_trans_cancel; | |
1241 | ||
1242 | /* | |
1243 | * A newly created regular or special file just has one directory | |
1244 | * entry pointing to them, but a directory also the "." entry | |
1245 | * pointing to itself. | |
1246 | */ | |
1247 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, | |
1248 | prid, resblks > 0, &ip, &committed); | |
1249 | if (error) { | |
1250 | if (error == ENOSPC) | |
1251 | goto out_trans_cancel; | |
1252 | goto out_trans_abort; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * Now we join the directory inode to the transaction. We do not do it | |
1257 | * earlier because xfs_dir_ialloc might commit the previous transaction | |
1258 | * (and release all the locks). An error from here on will result in | |
1259 | * the transaction cancel unlocking dp so don't do it explicitly in the | |
1260 | * error path. | |
1261 | */ | |
1262 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
1263 | unlock_dp_on_error = false; | |
1264 | ||
1265 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, | |
1266 | &first_block, &free_list, resblks ? | |
1267 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); | |
1268 | if (error) { | |
1269 | ASSERT(error != ENOSPC); | |
1270 | goto out_trans_abort; | |
1271 | } | |
1272 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1273 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
1274 | ||
1275 | if (is_dir) { | |
1276 | error = xfs_dir_init(tp, ip, dp); | |
1277 | if (error) | |
1278 | goto out_bmap_cancel; | |
1279 | ||
1280 | error = xfs_bumplink(tp, dp); | |
1281 | if (error) | |
1282 | goto out_bmap_cancel; | |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | * If this is a synchronous mount, make sure that the | |
1287 | * create transaction goes to disk before returning to | |
1288 | * the user. | |
1289 | */ | |
1290 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
1291 | xfs_trans_set_sync(tp); | |
1292 | ||
1293 | /* | |
1294 | * Attach the dquot(s) to the inodes and modify them incore. | |
1295 | * These ids of the inode couldn't have changed since the new | |
1296 | * inode has been locked ever since it was created. | |
1297 | */ | |
1298 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); | |
1299 | ||
1300 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1301 | if (error) | |
1302 | goto out_bmap_cancel; | |
1303 | ||
1304 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1305 | if (error) | |
1306 | goto out_release_inode; | |
1307 | ||
1308 | xfs_qm_dqrele(udqp); | |
1309 | xfs_qm_dqrele(gdqp); | |
1310 | xfs_qm_dqrele(pdqp); | |
1311 | ||
1312 | *ipp = ip; | |
1313 | return 0; | |
1314 | ||
1315 | out_bmap_cancel: | |
1316 | xfs_bmap_cancel(&free_list); | |
1317 | out_trans_abort: | |
1318 | cancel_flags |= XFS_TRANS_ABORT; | |
1319 | out_trans_cancel: | |
1320 | xfs_trans_cancel(tp, cancel_flags); | |
1321 | out_release_inode: | |
1322 | /* | |
1323 | * Wait until after the current transaction is aborted to | |
1324 | * release the inode. This prevents recursive transactions | |
1325 | * and deadlocks from xfs_inactive. | |
1326 | */ | |
1327 | if (ip) | |
1328 | IRELE(ip); | |
1329 | ||
1330 | xfs_qm_dqrele(udqp); | |
1331 | xfs_qm_dqrele(gdqp); | |
1332 | xfs_qm_dqrele(pdqp); | |
1333 | ||
1334 | if (unlock_dp_on_error) | |
1335 | xfs_iunlock(dp, XFS_ILOCK_EXCL); | |
1336 | return error; | |
1337 | } | |
1338 | ||
1339 | int | |
1340 | xfs_link( | |
1341 | xfs_inode_t *tdp, | |
1342 | xfs_inode_t *sip, | |
1343 | struct xfs_name *target_name) | |
1344 | { | |
1345 | xfs_mount_t *mp = tdp->i_mount; | |
1346 | xfs_trans_t *tp; | |
1347 | int error; | |
1348 | xfs_bmap_free_t free_list; | |
1349 | xfs_fsblock_t first_block; | |
1350 | int cancel_flags; | |
1351 | int committed; | |
1352 | int resblks; | |
1353 | ||
1354 | trace_xfs_link(tdp, target_name); | |
1355 | ||
1356 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); | |
1357 | ||
1358 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1359 | return XFS_ERROR(EIO); | |
1360 | ||
1361 | error = xfs_qm_dqattach(sip, 0); | |
1362 | if (error) | |
1363 | goto std_return; | |
1364 | ||
1365 | error = xfs_qm_dqattach(tdp, 0); | |
1366 | if (error) | |
1367 | goto std_return; | |
1368 | ||
1369 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); | |
1370 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
1371 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); | |
3d3c8b52 | 1372 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0); |
c24b5dfa DC |
1373 | if (error == ENOSPC) { |
1374 | resblks = 0; | |
3d3c8b52 | 1375 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0); |
c24b5dfa DC |
1376 | } |
1377 | if (error) { | |
1378 | cancel_flags = 0; | |
1379 | goto error_return; | |
1380 | } | |
1381 | ||
1382 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); | |
1383 | ||
1384 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); | |
1385 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); | |
1386 | ||
1387 | /* | |
1388 | * If we are using project inheritance, we only allow hard link | |
1389 | * creation in our tree when the project IDs are the same; else | |
1390 | * the tree quota mechanism could be circumvented. | |
1391 | */ | |
1392 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
1393 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { | |
1394 | error = XFS_ERROR(EXDEV); | |
1395 | goto error_return; | |
1396 | } | |
1397 | ||
1398 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); | |
1399 | if (error) | |
1400 | goto error_return; | |
1401 | ||
1402 | xfs_bmap_init(&free_list, &first_block); | |
1403 | ||
1404 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, | |
1405 | &first_block, &free_list, resblks); | |
1406 | if (error) | |
1407 | goto abort_return; | |
1408 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
1409 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); | |
1410 | ||
1411 | error = xfs_bumplink(tp, sip); | |
1412 | if (error) | |
1413 | goto abort_return; | |
1414 | ||
1415 | /* | |
1416 | * If this is a synchronous mount, make sure that the | |
1417 | * link transaction goes to disk before returning to | |
1418 | * the user. | |
1419 | */ | |
1420 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
1421 | xfs_trans_set_sync(tp); | |
1422 | } | |
1423 | ||
1424 | error = xfs_bmap_finish (&tp, &free_list, &committed); | |
1425 | if (error) { | |
1426 | xfs_bmap_cancel(&free_list); | |
1427 | goto abort_return; | |
1428 | } | |
1429 | ||
1430 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1431 | ||
1432 | abort_return: | |
1433 | cancel_flags |= XFS_TRANS_ABORT; | |
1434 | error_return: | |
1435 | xfs_trans_cancel(tp, cancel_flags); | |
1436 | std_return: | |
1437 | return error; | |
1438 | } | |
1439 | ||
1da177e4 | 1440 | /* |
8f04c47a CH |
1441 | * Free up the underlying blocks past new_size. The new size must be smaller |
1442 | * than the current size. This routine can be used both for the attribute and | |
1443 | * data fork, and does not modify the inode size, which is left to the caller. | |
1da177e4 | 1444 | * |
f6485057 DC |
1445 | * The transaction passed to this routine must have made a permanent log |
1446 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the | |
1447 | * given transaction and start new ones, so make sure everything involved in | |
1448 | * the transaction is tidy before calling here. Some transaction will be | |
1449 | * returned to the caller to be committed. The incoming transaction must | |
1450 | * already include the inode, and both inode locks must be held exclusively. | |
1451 | * The inode must also be "held" within the transaction. On return the inode | |
1452 | * will be "held" within the returned transaction. This routine does NOT | |
1453 | * require any disk space to be reserved for it within the transaction. | |
1da177e4 | 1454 | * |
f6485057 DC |
1455 | * If we get an error, we must return with the inode locked and linked into the |
1456 | * current transaction. This keeps things simple for the higher level code, | |
1457 | * because it always knows that the inode is locked and held in the transaction | |
1458 | * that returns to it whether errors occur or not. We don't mark the inode | |
1459 | * dirty on error so that transactions can be easily aborted if possible. | |
1da177e4 LT |
1460 | */ |
1461 | int | |
8f04c47a CH |
1462 | xfs_itruncate_extents( |
1463 | struct xfs_trans **tpp, | |
1464 | struct xfs_inode *ip, | |
1465 | int whichfork, | |
1466 | xfs_fsize_t new_size) | |
1da177e4 | 1467 | { |
8f04c47a CH |
1468 | struct xfs_mount *mp = ip->i_mount; |
1469 | struct xfs_trans *tp = *tpp; | |
1470 | struct xfs_trans *ntp; | |
1471 | xfs_bmap_free_t free_list; | |
1472 | xfs_fsblock_t first_block; | |
1473 | xfs_fileoff_t first_unmap_block; | |
1474 | xfs_fileoff_t last_block; | |
1475 | xfs_filblks_t unmap_len; | |
1476 | int committed; | |
1477 | int error = 0; | |
1478 | int done = 0; | |
1da177e4 | 1479 | |
0b56185b CH |
1480 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1481 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || | |
1482 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); | |
ce7ae151 | 1483 | ASSERT(new_size <= XFS_ISIZE(ip)); |
8f04c47a | 1484 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
1da177e4 | 1485 | ASSERT(ip->i_itemp != NULL); |
898621d5 | 1486 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
8f04c47a | 1487 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
1da177e4 | 1488 | |
673e8e59 CH |
1489 | trace_xfs_itruncate_extents_start(ip, new_size); |
1490 | ||
1da177e4 LT |
1491 | /* |
1492 | * Since it is possible for space to become allocated beyond | |
1493 | * the end of the file (in a crash where the space is allocated | |
1494 | * but the inode size is not yet updated), simply remove any | |
1495 | * blocks which show up between the new EOF and the maximum | |
1496 | * possible file size. If the first block to be removed is | |
1497 | * beyond the maximum file size (ie it is the same as last_block), | |
1498 | * then there is nothing to do. | |
1499 | */ | |
8f04c47a | 1500 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
32972383 | 1501 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
8f04c47a CH |
1502 | if (first_unmap_block == last_block) |
1503 | return 0; | |
1504 | ||
1505 | ASSERT(first_unmap_block < last_block); | |
1506 | unmap_len = last_block - first_unmap_block + 1; | |
1da177e4 | 1507 | while (!done) { |
9d87c319 | 1508 | xfs_bmap_init(&free_list, &first_block); |
8f04c47a | 1509 | error = xfs_bunmapi(tp, ip, |
3e57ecf6 | 1510 | first_unmap_block, unmap_len, |
8f04c47a | 1511 | xfs_bmapi_aflag(whichfork), |
1da177e4 | 1512 | XFS_ITRUNC_MAX_EXTENTS, |
3e57ecf6 | 1513 | &first_block, &free_list, |
b4e9181e | 1514 | &done); |
8f04c47a CH |
1515 | if (error) |
1516 | goto out_bmap_cancel; | |
1da177e4 LT |
1517 | |
1518 | /* | |
1519 | * Duplicate the transaction that has the permanent | |
1520 | * reservation and commit the old transaction. | |
1521 | */ | |
8f04c47a | 1522 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
898621d5 | 1523 | if (committed) |
ddc3415a | 1524 | xfs_trans_ijoin(tp, ip, 0); |
8f04c47a CH |
1525 | if (error) |
1526 | goto out_bmap_cancel; | |
1da177e4 LT |
1527 | |
1528 | if (committed) { | |
1529 | /* | |
f6485057 | 1530 | * Mark the inode dirty so it will be logged and |
e5720eec | 1531 | * moved forward in the log as part of every commit. |
1da177e4 | 1532 | */ |
8f04c47a | 1533 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
1da177e4 | 1534 | } |
f6485057 | 1535 | |
8f04c47a CH |
1536 | ntp = xfs_trans_dup(tp); |
1537 | error = xfs_trans_commit(tp, 0); | |
1538 | tp = ntp; | |
e5720eec | 1539 | |
ddc3415a | 1540 | xfs_trans_ijoin(tp, ip, 0); |
f6485057 | 1541 | |
cc09c0dc | 1542 | if (error) |
8f04c47a CH |
1543 | goto out; |
1544 | ||
cc09c0dc | 1545 | /* |
8f04c47a | 1546 | * Transaction commit worked ok so we can drop the extra ticket |
cc09c0dc DC |
1547 | * reference that we gained in xfs_trans_dup() |
1548 | */ | |
8f04c47a | 1549 | xfs_log_ticket_put(tp->t_ticket); |
3d3c8b52 | 1550 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); |
f6485057 | 1551 | if (error) |
8f04c47a | 1552 | goto out; |
1da177e4 | 1553 | } |
8f04c47a | 1554 | |
673e8e59 CH |
1555 | /* |
1556 | * Always re-log the inode so that our permanent transaction can keep | |
1557 | * on rolling it forward in the log. | |
1558 | */ | |
1559 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1560 | ||
1561 | trace_xfs_itruncate_extents_end(ip, new_size); | |
1562 | ||
8f04c47a CH |
1563 | out: |
1564 | *tpp = tp; | |
1565 | return error; | |
1566 | out_bmap_cancel: | |
1da177e4 | 1567 | /* |
8f04c47a CH |
1568 | * If the bunmapi call encounters an error, return to the caller where |
1569 | * the transaction can be properly aborted. We just need to make sure | |
1570 | * we're not holding any resources that we were not when we came in. | |
1da177e4 | 1571 | */ |
8f04c47a CH |
1572 | xfs_bmap_cancel(&free_list); |
1573 | goto out; | |
1574 | } | |
1575 | ||
c24b5dfa DC |
1576 | int |
1577 | xfs_release( | |
1578 | xfs_inode_t *ip) | |
1579 | { | |
1580 | xfs_mount_t *mp = ip->i_mount; | |
1581 | int error; | |
1582 | ||
1583 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) | |
1584 | return 0; | |
1585 | ||
1586 | /* If this is a read-only mount, don't do this (would generate I/O) */ | |
1587 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1588 | return 0; | |
1589 | ||
1590 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1591 | int truncated; | |
1592 | ||
1593 | /* | |
1594 | * If we are using filestreams, and we have an unlinked | |
1595 | * file that we are processing the last close on, then nothing | |
1596 | * will be able to reopen and write to this file. Purge this | |
1597 | * inode from the filestreams cache so that it doesn't delay | |
1598 | * teardown of the inode. | |
1599 | */ | |
1600 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) | |
1601 | xfs_filestream_deassociate(ip); | |
1602 | ||
1603 | /* | |
1604 | * If we previously truncated this file and removed old data | |
1605 | * in the process, we want to initiate "early" writeout on | |
1606 | * the last close. This is an attempt to combat the notorious | |
1607 | * NULL files problem which is particularly noticeable from a | |
1608 | * truncate down, buffered (re-)write (delalloc), followed by | |
1609 | * a crash. What we are effectively doing here is | |
1610 | * significantly reducing the time window where we'd otherwise | |
1611 | * be exposed to that problem. | |
1612 | */ | |
1613 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); | |
1614 | if (truncated) { | |
1615 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); | |
1616 | if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) { | |
1617 | error = -filemap_flush(VFS_I(ip)->i_mapping); | |
1618 | if (error) | |
1619 | return error; | |
1620 | } | |
1621 | } | |
1622 | } | |
1623 | ||
1624 | if (ip->i_d.di_nlink == 0) | |
1625 | return 0; | |
1626 | ||
1627 | if (xfs_can_free_eofblocks(ip, false)) { | |
1628 | ||
1629 | /* | |
1630 | * If we can't get the iolock just skip truncating the blocks | |
1631 | * past EOF because we could deadlock with the mmap_sem | |
1632 | * otherwise. We'll get another chance to drop them once the | |
1633 | * last reference to the inode is dropped, so we'll never leak | |
1634 | * blocks permanently. | |
1635 | * | |
1636 | * Further, check if the inode is being opened, written and | |
1637 | * closed frequently and we have delayed allocation blocks | |
1638 | * outstanding (e.g. streaming writes from the NFS server), | |
1639 | * truncating the blocks past EOF will cause fragmentation to | |
1640 | * occur. | |
1641 | * | |
1642 | * In this case don't do the truncation, either, but we have to | |
1643 | * be careful how we detect this case. Blocks beyond EOF show | |
1644 | * up as i_delayed_blks even when the inode is clean, so we | |
1645 | * need to truncate them away first before checking for a dirty | |
1646 | * release. Hence on the first dirty close we will still remove | |
1647 | * the speculative allocation, but after that we will leave it | |
1648 | * in place. | |
1649 | */ | |
1650 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) | |
1651 | return 0; | |
1652 | ||
1653 | error = xfs_free_eofblocks(mp, ip, true); | |
1654 | if (error && error != EAGAIN) | |
1655 | return error; | |
1656 | ||
1657 | /* delalloc blocks after truncation means it really is dirty */ | |
1658 | if (ip->i_delayed_blks) | |
1659 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); | |
1660 | } | |
1661 | return 0; | |
1662 | } | |
1663 | ||
f7be2d7f BF |
1664 | /* |
1665 | * xfs_inactive_truncate | |
1666 | * | |
1667 | * Called to perform a truncate when an inode becomes unlinked. | |
1668 | */ | |
1669 | STATIC int | |
1670 | xfs_inactive_truncate( | |
1671 | struct xfs_inode *ip) | |
1672 | { | |
1673 | struct xfs_mount *mp = ip->i_mount; | |
1674 | struct xfs_trans *tp; | |
1675 | int error; | |
1676 | ||
1677 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1678 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); | |
1679 | if (error) { | |
1680 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1681 | xfs_trans_cancel(tp, 0); | |
1682 | return error; | |
1683 | } | |
1684 | ||
1685 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1686 | xfs_trans_ijoin(tp, ip, 0); | |
1687 | ||
1688 | /* | |
1689 | * Log the inode size first to prevent stale data exposure in the event | |
1690 | * of a system crash before the truncate completes. See the related | |
1691 | * comment in xfs_setattr_size() for details. | |
1692 | */ | |
1693 | ip->i_d.di_size = 0; | |
1694 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1695 | ||
1696 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); | |
1697 | if (error) | |
1698 | goto error_trans_cancel; | |
1699 | ||
1700 | ASSERT(ip->i_d.di_nextents == 0); | |
1701 | ||
1702 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1703 | if (error) | |
1704 | goto error_unlock; | |
1705 | ||
1706 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1707 | return 0; | |
1708 | ||
1709 | error_trans_cancel: | |
1710 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
1711 | error_unlock: | |
1712 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1713 | return error; | |
1714 | } | |
1715 | ||
88877d2b BF |
1716 | /* |
1717 | * xfs_inactive_ifree() | |
1718 | * | |
1719 | * Perform the inode free when an inode is unlinked. | |
1720 | */ | |
1721 | STATIC int | |
1722 | xfs_inactive_ifree( | |
1723 | struct xfs_inode *ip) | |
1724 | { | |
1725 | xfs_bmap_free_t free_list; | |
1726 | xfs_fsblock_t first_block; | |
1727 | int committed; | |
1728 | struct xfs_mount *mp = ip->i_mount; | |
1729 | struct xfs_trans *tp; | |
1730 | int error; | |
1731 | ||
1732 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); | |
1733 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0); | |
1734 | if (error) { | |
1735 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); | |
1736 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1737 | return error; | |
1738 | } | |
1739 | ||
1740 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1741 | xfs_trans_ijoin(tp, ip, 0); | |
1742 | ||
1743 | xfs_bmap_init(&free_list, &first_block); | |
1744 | error = xfs_ifree(tp, ip, &free_list); | |
1745 | if (error) { | |
1746 | /* | |
1747 | * If we fail to free the inode, shut down. The cancel | |
1748 | * might do that, we need to make sure. Otherwise the | |
1749 | * inode might be lost for a long time or forever. | |
1750 | */ | |
1751 | if (!XFS_FORCED_SHUTDOWN(mp)) { | |
1752 | xfs_notice(mp, "%s: xfs_ifree returned error %d", | |
1753 | __func__, error); | |
1754 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1755 | } | |
1756 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); | |
1757 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1758 | return error; | |
1759 | } | |
1760 | ||
1761 | /* | |
1762 | * Credit the quota account(s). The inode is gone. | |
1763 | */ | |
1764 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); | |
1765 | ||
1766 | /* | |
1767 | * Just ignore errors at this point. There is nothing we can | |
1768 | * do except to try to keep going. Make sure it's not a silent | |
1769 | * error. | |
1770 | */ | |
1771 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1772 | if (error) | |
1773 | xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", | |
1774 | __func__, error); | |
1775 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
1776 | if (error) | |
1777 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", | |
1778 | __func__, error); | |
1779 | ||
1780 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1781 | return 0; | |
1782 | } | |
1783 | ||
c24b5dfa DC |
1784 | /* |
1785 | * xfs_inactive | |
1786 | * | |
1787 | * This is called when the vnode reference count for the vnode | |
1788 | * goes to zero. If the file has been unlinked, then it must | |
1789 | * now be truncated. Also, we clear all of the read-ahead state | |
1790 | * kept for the inode here since the file is now closed. | |
1791 | */ | |
74564fb4 | 1792 | void |
c24b5dfa DC |
1793 | xfs_inactive( |
1794 | xfs_inode_t *ip) | |
1795 | { | |
3d3c8b52 | 1796 | struct xfs_mount *mp; |
3d3c8b52 JL |
1797 | int error; |
1798 | int truncate = 0; | |
c24b5dfa DC |
1799 | |
1800 | /* | |
1801 | * If the inode is already free, then there can be nothing | |
1802 | * to clean up here. | |
1803 | */ | |
d948709b | 1804 | if (ip->i_d.di_mode == 0) { |
c24b5dfa DC |
1805 | ASSERT(ip->i_df.if_real_bytes == 0); |
1806 | ASSERT(ip->i_df.if_broot_bytes == 0); | |
74564fb4 | 1807 | return; |
c24b5dfa DC |
1808 | } |
1809 | ||
1810 | mp = ip->i_mount; | |
1811 | ||
c24b5dfa DC |
1812 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
1813 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
74564fb4 | 1814 | return; |
c24b5dfa DC |
1815 | |
1816 | if (ip->i_d.di_nlink != 0) { | |
1817 | /* | |
1818 | * force is true because we are evicting an inode from the | |
1819 | * cache. Post-eof blocks must be freed, lest we end up with | |
1820 | * broken free space accounting. | |
1821 | */ | |
74564fb4 BF |
1822 | if (xfs_can_free_eofblocks(ip, true)) |
1823 | xfs_free_eofblocks(mp, ip, false); | |
1824 | ||
1825 | return; | |
c24b5dfa DC |
1826 | } |
1827 | ||
1828 | if (S_ISREG(ip->i_d.di_mode) && | |
1829 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || | |
1830 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) | |
1831 | truncate = 1; | |
1832 | ||
1833 | error = xfs_qm_dqattach(ip, 0); | |
1834 | if (error) | |
74564fb4 | 1835 | return; |
c24b5dfa | 1836 | |
f7be2d7f | 1837 | if (S_ISLNK(ip->i_d.di_mode)) |
36b21dde | 1838 | error = xfs_inactive_symlink(ip); |
f7be2d7f BF |
1839 | else if (truncate) |
1840 | error = xfs_inactive_truncate(ip); | |
1841 | if (error) | |
74564fb4 | 1842 | return; |
c24b5dfa DC |
1843 | |
1844 | /* | |
1845 | * If there are attributes associated with the file then blow them away | |
1846 | * now. The code calls a routine that recursively deconstructs the | |
1847 | * attribute fork. We need to just commit the current transaction | |
1848 | * because we can't use it for xfs_attr_inactive(). | |
1849 | */ | |
1850 | if (ip->i_d.di_anextents > 0) { | |
1851 | ASSERT(ip->i_d.di_forkoff != 0); | |
1852 | ||
c24b5dfa DC |
1853 | error = xfs_attr_inactive(ip); |
1854 | if (error) | |
74564fb4 | 1855 | return; |
c24b5dfa DC |
1856 | } |
1857 | ||
1858 | if (ip->i_afp) | |
1859 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); | |
1860 | ||
1861 | ASSERT(ip->i_d.di_anextents == 0); | |
1862 | ||
1863 | /* | |
1864 | * Free the inode. | |
1865 | */ | |
88877d2b BF |
1866 | error = xfs_inactive_ifree(ip); |
1867 | if (error) | |
74564fb4 | 1868 | return; |
c24b5dfa DC |
1869 | |
1870 | /* | |
1871 | * Release the dquots held by inode, if any. | |
1872 | */ | |
1873 | xfs_qm_dqdetach(ip); | |
c24b5dfa DC |
1874 | } |
1875 | ||
1da177e4 LT |
1876 | /* |
1877 | * This is called when the inode's link count goes to 0. | |
1878 | * We place the on-disk inode on a list in the AGI. It | |
1879 | * will be pulled from this list when the inode is freed. | |
1880 | */ | |
1881 | int | |
1882 | xfs_iunlink( | |
1883 | xfs_trans_t *tp, | |
1884 | xfs_inode_t *ip) | |
1885 | { | |
1886 | xfs_mount_t *mp; | |
1887 | xfs_agi_t *agi; | |
1888 | xfs_dinode_t *dip; | |
1889 | xfs_buf_t *agibp; | |
1890 | xfs_buf_t *ibp; | |
1da177e4 LT |
1891 | xfs_agino_t agino; |
1892 | short bucket_index; | |
1893 | int offset; | |
1894 | int error; | |
1da177e4 LT |
1895 | |
1896 | ASSERT(ip->i_d.di_nlink == 0); | |
1897 | ASSERT(ip->i_d.di_mode != 0); | |
1da177e4 LT |
1898 | |
1899 | mp = tp->t_mountp; | |
1900 | ||
1da177e4 LT |
1901 | /* |
1902 | * Get the agi buffer first. It ensures lock ordering | |
1903 | * on the list. | |
1904 | */ | |
5e1be0fb | 1905 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
859d7182 | 1906 | if (error) |
1da177e4 | 1907 | return error; |
1da177e4 | 1908 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1909 | |
1da177e4 LT |
1910 | /* |
1911 | * Get the index into the agi hash table for the | |
1912 | * list this inode will go on. | |
1913 | */ | |
1914 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1915 | ASSERT(agino != 0); | |
1916 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
1917 | ASSERT(agi->agi_unlinked[bucket_index]); | |
16259e7d | 1918 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
1da177e4 | 1919 | |
69ef921b | 1920 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
1da177e4 LT |
1921 | /* |
1922 | * There is already another inode in the bucket we need | |
1923 | * to add ourselves to. Add us at the front of the list. | |
1924 | * Here we put the head pointer into our next pointer, | |
1925 | * and then we fall through to point the head at us. | |
1926 | */ | |
475ee413 CH |
1927 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
1928 | 0, 0); | |
c319b58b VA |
1929 | if (error) |
1930 | return error; | |
1931 | ||
69ef921b | 1932 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
1da177e4 | 1933 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
92bfc6e7 | 1934 | offset = ip->i_imap.im_boffset + |
1da177e4 | 1935 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
1936 | |
1937 | /* need to recalc the inode CRC if appropriate */ | |
1938 | xfs_dinode_calc_crc(mp, dip); | |
1939 | ||
1da177e4 LT |
1940 | xfs_trans_inode_buf(tp, ibp); |
1941 | xfs_trans_log_buf(tp, ibp, offset, | |
1942 | (offset + sizeof(xfs_agino_t) - 1)); | |
1943 | xfs_inobp_check(mp, ibp); | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * Point the bucket head pointer at the inode being inserted. | |
1948 | */ | |
1949 | ASSERT(agino != 0); | |
16259e7d | 1950 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
1da177e4 LT |
1951 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
1952 | (sizeof(xfs_agino_t) * bucket_index); | |
1953 | xfs_trans_log_buf(tp, agibp, offset, | |
1954 | (offset + sizeof(xfs_agino_t) - 1)); | |
1955 | return 0; | |
1956 | } | |
1957 | ||
1958 | /* | |
1959 | * Pull the on-disk inode from the AGI unlinked list. | |
1960 | */ | |
1961 | STATIC int | |
1962 | xfs_iunlink_remove( | |
1963 | xfs_trans_t *tp, | |
1964 | xfs_inode_t *ip) | |
1965 | { | |
1966 | xfs_ino_t next_ino; | |
1967 | xfs_mount_t *mp; | |
1968 | xfs_agi_t *agi; | |
1969 | xfs_dinode_t *dip; | |
1970 | xfs_buf_t *agibp; | |
1971 | xfs_buf_t *ibp; | |
1972 | xfs_agnumber_t agno; | |
1da177e4 LT |
1973 | xfs_agino_t agino; |
1974 | xfs_agino_t next_agino; | |
1975 | xfs_buf_t *last_ibp; | |
6fdf8ccc | 1976 | xfs_dinode_t *last_dip = NULL; |
1da177e4 | 1977 | short bucket_index; |
6fdf8ccc | 1978 | int offset, last_offset = 0; |
1da177e4 | 1979 | int error; |
1da177e4 | 1980 | |
1da177e4 | 1981 | mp = tp->t_mountp; |
1da177e4 | 1982 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
1da177e4 LT |
1983 | |
1984 | /* | |
1985 | * Get the agi buffer first. It ensures lock ordering | |
1986 | * on the list. | |
1987 | */ | |
5e1be0fb CH |
1988 | error = xfs_read_agi(mp, tp, agno, &agibp); |
1989 | if (error) | |
1da177e4 | 1990 | return error; |
5e1be0fb | 1991 | |
1da177e4 | 1992 | agi = XFS_BUF_TO_AGI(agibp); |
5e1be0fb | 1993 | |
1da177e4 LT |
1994 | /* |
1995 | * Get the index into the agi hash table for the | |
1996 | * list this inode will go on. | |
1997 | */ | |
1998 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); | |
1999 | ASSERT(agino != 0); | |
2000 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; | |
69ef921b | 2001 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
1da177e4 LT |
2002 | ASSERT(agi->agi_unlinked[bucket_index]); |
2003 | ||
16259e7d | 2004 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
1da177e4 | 2005 | /* |
475ee413 CH |
2006 | * We're at the head of the list. Get the inode's on-disk |
2007 | * buffer to see if there is anyone after us on the list. | |
2008 | * Only modify our next pointer if it is not already NULLAGINO. | |
2009 | * This saves us the overhead of dealing with the buffer when | |
2010 | * there is no need to change it. | |
1da177e4 | 2011 | */ |
475ee413 CH |
2012 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2013 | 0, 0); | |
1da177e4 | 2014 | if (error) { |
475ee413 | 2015 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2016 | __func__, error); |
1da177e4 LT |
2017 | return error; |
2018 | } | |
347d1c01 | 2019 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2020 | ASSERT(next_agino != 0); |
2021 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2022 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2023 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2024 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2025 | |
2026 | /* need to recalc the inode CRC if appropriate */ | |
2027 | xfs_dinode_calc_crc(mp, dip); | |
2028 | ||
1da177e4 LT |
2029 | xfs_trans_inode_buf(tp, ibp); |
2030 | xfs_trans_log_buf(tp, ibp, offset, | |
2031 | (offset + sizeof(xfs_agino_t) - 1)); | |
2032 | xfs_inobp_check(mp, ibp); | |
2033 | } else { | |
2034 | xfs_trans_brelse(tp, ibp); | |
2035 | } | |
2036 | /* | |
2037 | * Point the bucket head pointer at the next inode. | |
2038 | */ | |
2039 | ASSERT(next_agino != 0); | |
2040 | ASSERT(next_agino != agino); | |
16259e7d | 2041 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
1da177e4 LT |
2042 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
2043 | (sizeof(xfs_agino_t) * bucket_index); | |
2044 | xfs_trans_log_buf(tp, agibp, offset, | |
2045 | (offset + sizeof(xfs_agino_t) - 1)); | |
2046 | } else { | |
2047 | /* | |
2048 | * We need to search the list for the inode being freed. | |
2049 | */ | |
16259e7d | 2050 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
1da177e4 LT |
2051 | last_ibp = NULL; |
2052 | while (next_agino != agino) { | |
129dbc9a CH |
2053 | struct xfs_imap imap; |
2054 | ||
2055 | if (last_ibp) | |
1da177e4 | 2056 | xfs_trans_brelse(tp, last_ibp); |
129dbc9a CH |
2057 | |
2058 | imap.im_blkno = 0; | |
1da177e4 | 2059 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
129dbc9a CH |
2060 | |
2061 | error = xfs_imap(mp, tp, next_ino, &imap, 0); | |
2062 | if (error) { | |
2063 | xfs_warn(mp, | |
2064 | "%s: xfs_imap returned error %d.", | |
2065 | __func__, error); | |
2066 | return error; | |
2067 | } | |
2068 | ||
2069 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, | |
2070 | &last_ibp, 0, 0); | |
1da177e4 | 2071 | if (error) { |
0b932ccc | 2072 | xfs_warn(mp, |
129dbc9a | 2073 | "%s: xfs_imap_to_bp returned error %d.", |
0b932ccc | 2074 | __func__, error); |
1da177e4 LT |
2075 | return error; |
2076 | } | |
129dbc9a CH |
2077 | |
2078 | last_offset = imap.im_boffset; | |
347d1c01 | 2079 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
1da177e4 LT |
2080 | ASSERT(next_agino != NULLAGINO); |
2081 | ASSERT(next_agino != 0); | |
2082 | } | |
475ee413 | 2083 | |
1da177e4 | 2084 | /* |
475ee413 CH |
2085 | * Now last_ibp points to the buffer previous to us on the |
2086 | * unlinked list. Pull us from the list. | |
1da177e4 | 2087 | */ |
475ee413 CH |
2088 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
2089 | 0, 0); | |
1da177e4 | 2090 | if (error) { |
475ee413 | 2091 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
0b932ccc | 2092 | __func__, error); |
1da177e4 LT |
2093 | return error; |
2094 | } | |
347d1c01 | 2095 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
1da177e4 LT |
2096 | ASSERT(next_agino != 0); |
2097 | ASSERT(next_agino != agino); | |
2098 | if (next_agino != NULLAGINO) { | |
347d1c01 | 2099 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
92bfc6e7 | 2100 | offset = ip->i_imap.im_boffset + |
1da177e4 | 2101 | offsetof(xfs_dinode_t, di_next_unlinked); |
0a32c26e DC |
2102 | |
2103 | /* need to recalc the inode CRC if appropriate */ | |
2104 | xfs_dinode_calc_crc(mp, dip); | |
2105 | ||
1da177e4 LT |
2106 | xfs_trans_inode_buf(tp, ibp); |
2107 | xfs_trans_log_buf(tp, ibp, offset, | |
2108 | (offset + sizeof(xfs_agino_t) - 1)); | |
2109 | xfs_inobp_check(mp, ibp); | |
2110 | } else { | |
2111 | xfs_trans_brelse(tp, ibp); | |
2112 | } | |
2113 | /* | |
2114 | * Point the previous inode on the list to the next inode. | |
2115 | */ | |
347d1c01 | 2116 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
1da177e4 LT |
2117 | ASSERT(next_agino != 0); |
2118 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); | |
0a32c26e DC |
2119 | |
2120 | /* need to recalc the inode CRC if appropriate */ | |
2121 | xfs_dinode_calc_crc(mp, last_dip); | |
2122 | ||
1da177e4 LT |
2123 | xfs_trans_inode_buf(tp, last_ibp); |
2124 | xfs_trans_log_buf(tp, last_ibp, offset, | |
2125 | (offset + sizeof(xfs_agino_t) - 1)); | |
2126 | xfs_inobp_check(mp, last_ibp); | |
2127 | } | |
2128 | return 0; | |
2129 | } | |
2130 | ||
5b3eed75 | 2131 | /* |
0b8182db | 2132 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
5b3eed75 DC |
2133 | * inodes that are in memory - they all must be marked stale and attached to |
2134 | * the cluster buffer. | |
2135 | */ | |
2a30f36d | 2136 | STATIC int |
1da177e4 LT |
2137 | xfs_ifree_cluster( |
2138 | xfs_inode_t *free_ip, | |
2139 | xfs_trans_t *tp, | |
2140 | xfs_ino_t inum) | |
2141 | { | |
2142 | xfs_mount_t *mp = free_ip->i_mount; | |
2143 | int blks_per_cluster; | |
2144 | int nbufs; | |
2145 | int ninodes; | |
5b257b4a | 2146 | int i, j; |
1da177e4 LT |
2147 | xfs_daddr_t blkno; |
2148 | xfs_buf_t *bp; | |
5b257b4a | 2149 | xfs_inode_t *ip; |
1da177e4 LT |
2150 | xfs_inode_log_item_t *iip; |
2151 | xfs_log_item_t *lip; | |
5017e97d | 2152 | struct xfs_perag *pag; |
1da177e4 | 2153 | |
5017e97d | 2154 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
1da177e4 LT |
2155 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { |
2156 | blks_per_cluster = 1; | |
2157 | ninodes = mp->m_sb.sb_inopblock; | |
2158 | nbufs = XFS_IALLOC_BLOCKS(mp); | |
2159 | } else { | |
2160 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / | |
2161 | mp->m_sb.sb_blocksize; | |
2162 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; | |
2163 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; | |
2164 | } | |
2165 | ||
1da177e4 LT |
2166 | for (j = 0; j < nbufs; j++, inum += ninodes) { |
2167 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), | |
2168 | XFS_INO_TO_AGBNO(mp, inum)); | |
2169 | ||
5b257b4a DC |
2170 | /* |
2171 | * We obtain and lock the backing buffer first in the process | |
2172 | * here, as we have to ensure that any dirty inode that we | |
2173 | * can't get the flush lock on is attached to the buffer. | |
2174 | * If we scan the in-memory inodes first, then buffer IO can | |
2175 | * complete before we get a lock on it, and hence we may fail | |
2176 | * to mark all the active inodes on the buffer stale. | |
2177 | */ | |
2178 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, | |
b6aff29f DC |
2179 | mp->m_bsize * blks_per_cluster, |
2180 | XBF_UNMAPPED); | |
5b257b4a | 2181 | |
2a30f36d CS |
2182 | if (!bp) |
2183 | return ENOMEM; | |
b0f539de DC |
2184 | |
2185 | /* | |
2186 | * This buffer may not have been correctly initialised as we | |
2187 | * didn't read it from disk. That's not important because we are | |
2188 | * only using to mark the buffer as stale in the log, and to | |
2189 | * attach stale cached inodes on it. That means it will never be | |
2190 | * dispatched for IO. If it is, we want to know about it, and we | |
2191 | * want it to fail. We can acheive this by adding a write | |
2192 | * verifier to the buffer. | |
2193 | */ | |
1813dd64 | 2194 | bp->b_ops = &xfs_inode_buf_ops; |
b0f539de | 2195 | |
5b257b4a DC |
2196 | /* |
2197 | * Walk the inodes already attached to the buffer and mark them | |
2198 | * stale. These will all have the flush locks held, so an | |
5b3eed75 DC |
2199 | * in-memory inode walk can't lock them. By marking them all |
2200 | * stale first, we will not attempt to lock them in the loop | |
2201 | * below as the XFS_ISTALE flag will be set. | |
5b257b4a | 2202 | */ |
adadbeef | 2203 | lip = bp->b_fspriv; |
5b257b4a DC |
2204 | while (lip) { |
2205 | if (lip->li_type == XFS_LI_INODE) { | |
2206 | iip = (xfs_inode_log_item_t *)lip; | |
2207 | ASSERT(iip->ili_logged == 1); | |
ca30b2a7 | 2208 | lip->li_cb = xfs_istale_done; |
5b257b4a DC |
2209 | xfs_trans_ail_copy_lsn(mp->m_ail, |
2210 | &iip->ili_flush_lsn, | |
2211 | &iip->ili_item.li_lsn); | |
2212 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); | |
5b257b4a DC |
2213 | } |
2214 | lip = lip->li_bio_list; | |
2215 | } | |
1da177e4 | 2216 | |
5b3eed75 | 2217 | |
1da177e4 | 2218 | /* |
5b257b4a DC |
2219 | * For each inode in memory attempt to add it to the inode |
2220 | * buffer and set it up for being staled on buffer IO | |
2221 | * completion. This is safe as we've locked out tail pushing | |
2222 | * and flushing by locking the buffer. | |
1da177e4 | 2223 | * |
5b257b4a DC |
2224 | * We have already marked every inode that was part of a |
2225 | * transaction stale above, which means there is no point in | |
2226 | * even trying to lock them. | |
1da177e4 | 2227 | */ |
1da177e4 | 2228 | for (i = 0; i < ninodes; i++) { |
5b3eed75 | 2229 | retry: |
1a3e8f3d | 2230 | rcu_read_lock(); |
da353b0d DC |
2231 | ip = radix_tree_lookup(&pag->pag_ici_root, |
2232 | XFS_INO_TO_AGINO(mp, (inum + i))); | |
1da177e4 | 2233 | |
1a3e8f3d DC |
2234 | /* Inode not in memory, nothing to do */ |
2235 | if (!ip) { | |
2236 | rcu_read_unlock(); | |
1da177e4 LT |
2237 | continue; |
2238 | } | |
2239 | ||
1a3e8f3d DC |
2240 | /* |
2241 | * because this is an RCU protected lookup, we could | |
2242 | * find a recently freed or even reallocated inode | |
2243 | * during the lookup. We need to check under the | |
2244 | * i_flags_lock for a valid inode here. Skip it if it | |
2245 | * is not valid, the wrong inode or stale. | |
2246 | */ | |
2247 | spin_lock(&ip->i_flags_lock); | |
2248 | if (ip->i_ino != inum + i || | |
2249 | __xfs_iflags_test(ip, XFS_ISTALE)) { | |
2250 | spin_unlock(&ip->i_flags_lock); | |
2251 | rcu_read_unlock(); | |
2252 | continue; | |
2253 | } | |
2254 | spin_unlock(&ip->i_flags_lock); | |
2255 | ||
5b3eed75 DC |
2256 | /* |
2257 | * Don't try to lock/unlock the current inode, but we | |
2258 | * _cannot_ skip the other inodes that we did not find | |
2259 | * in the list attached to the buffer and are not | |
2260 | * already marked stale. If we can't lock it, back off | |
2261 | * and retry. | |
2262 | */ | |
5b257b4a DC |
2263 | if (ip != free_ip && |
2264 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { | |
1a3e8f3d | 2265 | rcu_read_unlock(); |
5b3eed75 DC |
2266 | delay(1); |
2267 | goto retry; | |
1da177e4 | 2268 | } |
1a3e8f3d | 2269 | rcu_read_unlock(); |
1da177e4 | 2270 | |
5b3eed75 | 2271 | xfs_iflock(ip); |
5b257b4a | 2272 | xfs_iflags_set(ip, XFS_ISTALE); |
1da177e4 | 2273 | |
5b3eed75 DC |
2274 | /* |
2275 | * we don't need to attach clean inodes or those only | |
2276 | * with unlogged changes (which we throw away, anyway). | |
2277 | */ | |
1da177e4 | 2278 | iip = ip->i_itemp; |
5b3eed75 | 2279 | if (!iip || xfs_inode_clean(ip)) { |
5b257b4a | 2280 | ASSERT(ip != free_ip); |
1da177e4 LT |
2281 | xfs_ifunlock(ip); |
2282 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
2283 | continue; | |
2284 | } | |
2285 | ||
f5d8d5c4 CH |
2286 | iip->ili_last_fields = iip->ili_fields; |
2287 | iip->ili_fields = 0; | |
1da177e4 | 2288 | iip->ili_logged = 1; |
7b2e2a31 DC |
2289 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
2290 | &iip->ili_item.li_lsn); | |
1da177e4 | 2291 | |
ca30b2a7 CH |
2292 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
2293 | &iip->ili_item); | |
5b257b4a DC |
2294 | |
2295 | if (ip != free_ip) | |
1da177e4 | 2296 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1da177e4 LT |
2297 | } |
2298 | ||
5b3eed75 | 2299 | xfs_trans_stale_inode_buf(tp, bp); |
1da177e4 LT |
2300 | xfs_trans_binval(tp, bp); |
2301 | } | |
2302 | ||
5017e97d | 2303 | xfs_perag_put(pag); |
2a30f36d | 2304 | return 0; |
1da177e4 LT |
2305 | } |
2306 | ||
2307 | /* | |
2308 | * This is called to return an inode to the inode free list. | |
2309 | * The inode should already be truncated to 0 length and have | |
2310 | * no pages associated with it. This routine also assumes that | |
2311 | * the inode is already a part of the transaction. | |
2312 | * | |
2313 | * The on-disk copy of the inode will have been added to the list | |
2314 | * of unlinked inodes in the AGI. We need to remove the inode from | |
2315 | * that list atomically with respect to freeing it here. | |
2316 | */ | |
2317 | int | |
2318 | xfs_ifree( | |
2319 | xfs_trans_t *tp, | |
2320 | xfs_inode_t *ip, | |
2321 | xfs_bmap_free_t *flist) | |
2322 | { | |
2323 | int error; | |
2324 | int delete; | |
2325 | xfs_ino_t first_ino; | |
2326 | ||
579aa9ca | 2327 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
2328 | ASSERT(ip->i_d.di_nlink == 0); |
2329 | ASSERT(ip->i_d.di_nextents == 0); | |
2330 | ASSERT(ip->i_d.di_anextents == 0); | |
ce7ae151 | 2331 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
1da177e4 LT |
2332 | ASSERT(ip->i_d.di_nblocks == 0); |
2333 | ||
2334 | /* | |
2335 | * Pull the on-disk inode from the AGI unlinked list. | |
2336 | */ | |
2337 | error = xfs_iunlink_remove(tp, ip); | |
1baaed8f | 2338 | if (error) |
1da177e4 | 2339 | return error; |
1da177e4 LT |
2340 | |
2341 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); | |
1baaed8f | 2342 | if (error) |
1da177e4 | 2343 | return error; |
1baaed8f | 2344 | |
1da177e4 LT |
2345 | ip->i_d.di_mode = 0; /* mark incore inode as free */ |
2346 | ip->i_d.di_flags = 0; | |
2347 | ip->i_d.di_dmevmask = 0; | |
2348 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ | |
1da177e4 LT |
2349 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
2350 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; | |
2351 | /* | |
2352 | * Bump the generation count so no one will be confused | |
2353 | * by reincarnations of this inode. | |
2354 | */ | |
2355 | ip->i_d.di_gen++; | |
2356 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
2357 | ||
1baaed8f | 2358 | if (delete) |
2a30f36d | 2359 | error = xfs_ifree_cluster(ip, tp, first_ino); |
1da177e4 | 2360 | |
2a30f36d | 2361 | return error; |
1da177e4 LT |
2362 | } |
2363 | ||
1da177e4 | 2364 | /* |
60ec6783 CH |
2365 | * This is called to unpin an inode. The caller must have the inode locked |
2366 | * in at least shared mode so that the buffer cannot be subsequently pinned | |
2367 | * once someone is waiting for it to be unpinned. | |
1da177e4 | 2368 | */ |
60ec6783 | 2369 | static void |
f392e631 | 2370 | xfs_iunpin( |
60ec6783 | 2371 | struct xfs_inode *ip) |
1da177e4 | 2372 | { |
579aa9ca | 2373 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
1da177e4 | 2374 | |
4aaf15d1 DC |
2375 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
2376 | ||
a3f74ffb | 2377 | /* Give the log a push to start the unpinning I/O */ |
60ec6783 | 2378 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
a14a348b | 2379 | |
a3f74ffb | 2380 | } |
1da177e4 | 2381 | |
f392e631 CH |
2382 | static void |
2383 | __xfs_iunpin_wait( | |
2384 | struct xfs_inode *ip) | |
2385 | { | |
2386 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); | |
2387 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); | |
2388 | ||
2389 | xfs_iunpin(ip); | |
2390 | ||
2391 | do { | |
2392 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2393 | if (xfs_ipincount(ip)) | |
2394 | io_schedule(); | |
2395 | } while (xfs_ipincount(ip)); | |
2396 | finish_wait(wq, &wait.wait); | |
2397 | } | |
2398 | ||
777df5af | 2399 | void |
a3f74ffb | 2400 | xfs_iunpin_wait( |
60ec6783 | 2401 | struct xfs_inode *ip) |
a3f74ffb | 2402 | { |
f392e631 CH |
2403 | if (xfs_ipincount(ip)) |
2404 | __xfs_iunpin_wait(ip); | |
1da177e4 LT |
2405 | } |
2406 | ||
27320369 DC |
2407 | /* |
2408 | * Removing an inode from the namespace involves removing the directory entry | |
2409 | * and dropping the link count on the inode. Removing the directory entry can | |
2410 | * result in locking an AGF (directory blocks were freed) and removing a link | |
2411 | * count can result in placing the inode on an unlinked list which results in | |
2412 | * locking an AGI. | |
2413 | * | |
2414 | * The big problem here is that we have an ordering constraint on AGF and AGI | |
2415 | * locking - inode allocation locks the AGI, then can allocate a new extent for | |
2416 | * new inodes, locking the AGF after the AGI. Similarly, freeing the inode | |
2417 | * removes the inode from the unlinked list, requiring that we lock the AGI | |
2418 | * first, and then freeing the inode can result in an inode chunk being freed | |
2419 | * and hence freeing disk space requiring that we lock an AGF. | |
2420 | * | |
2421 | * Hence the ordering that is imposed by other parts of the code is AGI before | |
2422 | * AGF. This means we cannot remove the directory entry before we drop the inode | |
2423 | * reference count and put it on the unlinked list as this results in a lock | |
2424 | * order of AGF then AGI, and this can deadlock against inode allocation and | |
2425 | * freeing. Therefore we must drop the link counts before we remove the | |
2426 | * directory entry. | |
2427 | * | |
2428 | * This is still safe from a transactional point of view - it is not until we | |
2429 | * get to xfs_bmap_finish() that we have the possibility of multiple | |
2430 | * transactions in this operation. Hence as long as we remove the directory | |
2431 | * entry and drop the link count in the first transaction of the remove | |
2432 | * operation, there are no transactional constraints on the ordering here. | |
2433 | */ | |
c24b5dfa DC |
2434 | int |
2435 | xfs_remove( | |
2436 | xfs_inode_t *dp, | |
2437 | struct xfs_name *name, | |
2438 | xfs_inode_t *ip) | |
2439 | { | |
2440 | xfs_mount_t *mp = dp->i_mount; | |
2441 | xfs_trans_t *tp = NULL; | |
2442 | int is_dir = S_ISDIR(ip->i_d.di_mode); | |
2443 | int error = 0; | |
2444 | xfs_bmap_free_t free_list; | |
2445 | xfs_fsblock_t first_block; | |
2446 | int cancel_flags; | |
2447 | int committed; | |
2448 | int link_zero; | |
2449 | uint resblks; | |
2450 | uint log_count; | |
2451 | ||
2452 | trace_xfs_remove(dp, name); | |
2453 | ||
2454 | if (XFS_FORCED_SHUTDOWN(mp)) | |
2455 | return XFS_ERROR(EIO); | |
2456 | ||
2457 | error = xfs_qm_dqattach(dp, 0); | |
2458 | if (error) | |
2459 | goto std_return; | |
2460 | ||
2461 | error = xfs_qm_dqattach(ip, 0); | |
2462 | if (error) | |
2463 | goto std_return; | |
2464 | ||
2465 | if (is_dir) { | |
2466 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); | |
2467 | log_count = XFS_DEFAULT_LOG_COUNT; | |
2468 | } else { | |
2469 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); | |
2470 | log_count = XFS_REMOVE_LOG_COUNT; | |
2471 | } | |
2472 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2473 | ||
2474 | /* | |
2475 | * We try to get the real space reservation first, | |
2476 | * allowing for directory btree deletion(s) implying | |
2477 | * possible bmap insert(s). If we can't get the space | |
2478 | * reservation then we use 0 instead, and avoid the bmap | |
2479 | * btree insert(s) in the directory code by, if the bmap | |
2480 | * insert tries to happen, instead trimming the LAST | |
2481 | * block from the directory. | |
2482 | */ | |
2483 | resblks = XFS_REMOVE_SPACE_RES(mp); | |
3d3c8b52 | 2484 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0); |
c24b5dfa DC |
2485 | if (error == ENOSPC) { |
2486 | resblks = 0; | |
3d3c8b52 | 2487 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0); |
c24b5dfa DC |
2488 | } |
2489 | if (error) { | |
2490 | ASSERT(error != ENOSPC); | |
2491 | cancel_flags = 0; | |
2492 | goto out_trans_cancel; | |
2493 | } | |
2494 | ||
2495 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); | |
2496 | ||
2497 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); | |
2498 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
2499 | ||
2500 | /* | |
2501 | * If we're removing a directory perform some additional validation. | |
2502 | */ | |
27320369 | 2503 | cancel_flags |= XFS_TRANS_ABORT; |
c24b5dfa DC |
2504 | if (is_dir) { |
2505 | ASSERT(ip->i_d.di_nlink >= 2); | |
2506 | if (ip->i_d.di_nlink != 2) { | |
2507 | error = XFS_ERROR(ENOTEMPTY); | |
2508 | goto out_trans_cancel; | |
2509 | } | |
2510 | if (!xfs_dir_isempty(ip)) { | |
2511 | error = XFS_ERROR(ENOTEMPTY); | |
2512 | goto out_trans_cancel; | |
2513 | } | |
c24b5dfa | 2514 | |
27320369 | 2515 | /* Drop the link from ip's "..". */ |
c24b5dfa DC |
2516 | error = xfs_droplink(tp, dp); |
2517 | if (error) | |
27320369 | 2518 | goto out_trans_cancel; |
c24b5dfa | 2519 | |
27320369 | 2520 | /* Drop the "." link from ip to self. */ |
c24b5dfa DC |
2521 | error = xfs_droplink(tp, ip); |
2522 | if (error) | |
27320369 | 2523 | goto out_trans_cancel; |
c24b5dfa DC |
2524 | } else { |
2525 | /* | |
2526 | * When removing a non-directory we need to log the parent | |
2527 | * inode here. For a directory this is done implicitly | |
2528 | * by the xfs_droplink call for the ".." entry. | |
2529 | */ | |
2530 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); | |
2531 | } | |
27320369 | 2532 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
c24b5dfa | 2533 | |
27320369 | 2534 | /* Drop the link from dp to ip. */ |
c24b5dfa DC |
2535 | error = xfs_droplink(tp, ip); |
2536 | if (error) | |
27320369 | 2537 | goto out_trans_cancel; |
c24b5dfa | 2538 | |
27320369 | 2539 | /* Determine if this is the last link while the inode is locked */ |
c24b5dfa DC |
2540 | link_zero = (ip->i_d.di_nlink == 0); |
2541 | ||
27320369 DC |
2542 | xfs_bmap_init(&free_list, &first_block); |
2543 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, | |
2544 | &first_block, &free_list, resblks); | |
2545 | if (error) { | |
2546 | ASSERT(error != ENOENT); | |
2547 | goto out_bmap_cancel; | |
2548 | } | |
2549 | ||
c24b5dfa DC |
2550 | /* |
2551 | * If this is a synchronous mount, make sure that the | |
2552 | * remove transaction goes to disk before returning to | |
2553 | * the user. | |
2554 | */ | |
2555 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) | |
2556 | xfs_trans_set_sync(tp); | |
2557 | ||
2558 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
2559 | if (error) | |
2560 | goto out_bmap_cancel; | |
2561 | ||
2562 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
2563 | if (error) | |
2564 | goto std_return; | |
2565 | ||
2566 | /* | |
2567 | * If we are using filestreams, kill the stream association. | |
2568 | * If the file is still open it may get a new one but that | |
2569 | * will get killed on last close in xfs_close() so we don't | |
2570 | * have to worry about that. | |
2571 | */ | |
2572 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) | |
2573 | xfs_filestream_deassociate(ip); | |
2574 | ||
2575 | return 0; | |
2576 | ||
2577 | out_bmap_cancel: | |
2578 | xfs_bmap_cancel(&free_list); | |
c24b5dfa DC |
2579 | out_trans_cancel: |
2580 | xfs_trans_cancel(tp, cancel_flags); | |
2581 | std_return: | |
2582 | return error; | |
2583 | } | |
2584 | ||
f6bba201 DC |
2585 | /* |
2586 | * Enter all inodes for a rename transaction into a sorted array. | |
2587 | */ | |
2588 | STATIC void | |
2589 | xfs_sort_for_rename( | |
2590 | xfs_inode_t *dp1, /* in: old (source) directory inode */ | |
2591 | xfs_inode_t *dp2, /* in: new (target) directory inode */ | |
2592 | xfs_inode_t *ip1, /* in: inode of old entry */ | |
2593 | xfs_inode_t *ip2, /* in: inode of new entry, if it | |
2594 | already exists, NULL otherwise. */ | |
2595 | xfs_inode_t **i_tab,/* out: array of inode returned, sorted */ | |
2596 | int *num_inodes) /* out: number of inodes in array */ | |
2597 | { | |
2598 | xfs_inode_t *temp; | |
2599 | int i, j; | |
2600 | ||
2601 | /* | |
2602 | * i_tab contains a list of pointers to inodes. We initialize | |
2603 | * the table here & we'll sort it. We will then use it to | |
2604 | * order the acquisition of the inode locks. | |
2605 | * | |
2606 | * Note that the table may contain duplicates. e.g., dp1 == dp2. | |
2607 | */ | |
2608 | i_tab[0] = dp1; | |
2609 | i_tab[1] = dp2; | |
2610 | i_tab[2] = ip1; | |
2611 | if (ip2) { | |
2612 | *num_inodes = 4; | |
2613 | i_tab[3] = ip2; | |
2614 | } else { | |
2615 | *num_inodes = 3; | |
2616 | i_tab[3] = NULL; | |
2617 | } | |
2618 | ||
2619 | /* | |
2620 | * Sort the elements via bubble sort. (Remember, there are at | |
2621 | * most 4 elements to sort, so this is adequate.) | |
2622 | */ | |
2623 | for (i = 0; i < *num_inodes; i++) { | |
2624 | for (j = 1; j < *num_inodes; j++) { | |
2625 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { | |
2626 | temp = i_tab[j]; | |
2627 | i_tab[j] = i_tab[j-1]; | |
2628 | i_tab[j-1] = temp; | |
2629 | } | |
2630 | } | |
2631 | } | |
2632 | } | |
2633 | ||
2634 | /* | |
2635 | * xfs_rename | |
2636 | */ | |
2637 | int | |
2638 | xfs_rename( | |
2639 | xfs_inode_t *src_dp, | |
2640 | struct xfs_name *src_name, | |
2641 | xfs_inode_t *src_ip, | |
2642 | xfs_inode_t *target_dp, | |
2643 | struct xfs_name *target_name, | |
2644 | xfs_inode_t *target_ip) | |
2645 | { | |
2646 | xfs_trans_t *tp = NULL; | |
2647 | xfs_mount_t *mp = src_dp->i_mount; | |
2648 | int new_parent; /* moving to a new dir */ | |
2649 | int src_is_directory; /* src_name is a directory */ | |
2650 | int error; | |
2651 | xfs_bmap_free_t free_list; | |
2652 | xfs_fsblock_t first_block; | |
2653 | int cancel_flags; | |
2654 | int committed; | |
2655 | xfs_inode_t *inodes[4]; | |
2656 | int spaceres; | |
2657 | int num_inodes; | |
2658 | ||
2659 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); | |
2660 | ||
2661 | new_parent = (src_dp != target_dp); | |
2662 | src_is_directory = S_ISDIR(src_ip->i_d.di_mode); | |
2663 | ||
2664 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, | |
2665 | inodes, &num_inodes); | |
2666 | ||
2667 | xfs_bmap_init(&free_list, &first_block); | |
2668 | tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME); | |
2669 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; | |
2670 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); | |
3d3c8b52 | 2671 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0); |
f6bba201 DC |
2672 | if (error == ENOSPC) { |
2673 | spaceres = 0; | |
3d3c8b52 | 2674 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0); |
f6bba201 DC |
2675 | } |
2676 | if (error) { | |
2677 | xfs_trans_cancel(tp, 0); | |
2678 | goto std_return; | |
2679 | } | |
2680 | ||
2681 | /* | |
2682 | * Attach the dquots to the inodes | |
2683 | */ | |
2684 | error = xfs_qm_vop_rename_dqattach(inodes); | |
2685 | if (error) { | |
2686 | xfs_trans_cancel(tp, cancel_flags); | |
2687 | goto std_return; | |
2688 | } | |
2689 | ||
2690 | /* | |
2691 | * Lock all the participating inodes. Depending upon whether | |
2692 | * the target_name exists in the target directory, and | |
2693 | * whether the target directory is the same as the source | |
2694 | * directory, we can lock from 2 to 4 inodes. | |
2695 | */ | |
2696 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); | |
2697 | ||
2698 | /* | |
2699 | * Join all the inodes to the transaction. From this point on, | |
2700 | * we can rely on either trans_commit or trans_cancel to unlock | |
2701 | * them. | |
2702 | */ | |
2703 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); | |
2704 | if (new_parent) | |
2705 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); | |
2706 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); | |
2707 | if (target_ip) | |
2708 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); | |
2709 | ||
2710 | /* | |
2711 | * If we are using project inheritance, we only allow renames | |
2712 | * into our tree when the project IDs are the same; else the | |
2713 | * tree quota mechanism would be circumvented. | |
2714 | */ | |
2715 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && | |
2716 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { | |
2717 | error = XFS_ERROR(EXDEV); | |
2718 | goto error_return; | |
2719 | } | |
2720 | ||
2721 | /* | |
2722 | * Set up the target. | |
2723 | */ | |
2724 | if (target_ip == NULL) { | |
2725 | /* | |
2726 | * If there's no space reservation, check the entry will | |
2727 | * fit before actually inserting it. | |
2728 | */ | |
2729 | error = xfs_dir_canenter(tp, target_dp, target_name, spaceres); | |
2730 | if (error) | |
2731 | goto error_return; | |
2732 | /* | |
2733 | * If target does not exist and the rename crosses | |
2734 | * directories, adjust the target directory link count | |
2735 | * to account for the ".." reference from the new entry. | |
2736 | */ | |
2737 | error = xfs_dir_createname(tp, target_dp, target_name, | |
2738 | src_ip->i_ino, &first_block, | |
2739 | &free_list, spaceres); | |
2740 | if (error == ENOSPC) | |
2741 | goto error_return; | |
2742 | if (error) | |
2743 | goto abort_return; | |
2744 | ||
2745 | xfs_trans_ichgtime(tp, target_dp, | |
2746 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2747 | ||
2748 | if (new_parent && src_is_directory) { | |
2749 | error = xfs_bumplink(tp, target_dp); | |
2750 | if (error) | |
2751 | goto abort_return; | |
2752 | } | |
2753 | } else { /* target_ip != NULL */ | |
2754 | /* | |
2755 | * If target exists and it's a directory, check that both | |
2756 | * target and source are directories and that target can be | |
2757 | * destroyed, or that neither is a directory. | |
2758 | */ | |
2759 | if (S_ISDIR(target_ip->i_d.di_mode)) { | |
2760 | /* | |
2761 | * Make sure target dir is empty. | |
2762 | */ | |
2763 | if (!(xfs_dir_isempty(target_ip)) || | |
2764 | (target_ip->i_d.di_nlink > 2)) { | |
2765 | error = XFS_ERROR(EEXIST); | |
2766 | goto error_return; | |
2767 | } | |
2768 | } | |
2769 | ||
2770 | /* | |
2771 | * Link the source inode under the target name. | |
2772 | * If the source inode is a directory and we are moving | |
2773 | * it across directories, its ".." entry will be | |
2774 | * inconsistent until we replace that down below. | |
2775 | * | |
2776 | * In case there is already an entry with the same | |
2777 | * name at the destination directory, remove it first. | |
2778 | */ | |
2779 | error = xfs_dir_replace(tp, target_dp, target_name, | |
2780 | src_ip->i_ino, | |
2781 | &first_block, &free_list, spaceres); | |
2782 | if (error) | |
2783 | goto abort_return; | |
2784 | ||
2785 | xfs_trans_ichgtime(tp, target_dp, | |
2786 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2787 | ||
2788 | /* | |
2789 | * Decrement the link count on the target since the target | |
2790 | * dir no longer points to it. | |
2791 | */ | |
2792 | error = xfs_droplink(tp, target_ip); | |
2793 | if (error) | |
2794 | goto abort_return; | |
2795 | ||
2796 | if (src_is_directory) { | |
2797 | /* | |
2798 | * Drop the link from the old "." entry. | |
2799 | */ | |
2800 | error = xfs_droplink(tp, target_ip); | |
2801 | if (error) | |
2802 | goto abort_return; | |
2803 | } | |
2804 | } /* target_ip != NULL */ | |
2805 | ||
2806 | /* | |
2807 | * Remove the source. | |
2808 | */ | |
2809 | if (new_parent && src_is_directory) { | |
2810 | /* | |
2811 | * Rewrite the ".." entry to point to the new | |
2812 | * directory. | |
2813 | */ | |
2814 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, | |
2815 | target_dp->i_ino, | |
2816 | &first_block, &free_list, spaceres); | |
2817 | ASSERT(error != EEXIST); | |
2818 | if (error) | |
2819 | goto abort_return; | |
2820 | } | |
2821 | ||
2822 | /* | |
2823 | * We always want to hit the ctime on the source inode. | |
2824 | * | |
2825 | * This isn't strictly required by the standards since the source | |
2826 | * inode isn't really being changed, but old unix file systems did | |
2827 | * it and some incremental backup programs won't work without it. | |
2828 | */ | |
2829 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); | |
2830 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); | |
2831 | ||
2832 | /* | |
2833 | * Adjust the link count on src_dp. This is necessary when | |
2834 | * renaming a directory, either within one parent when | |
2835 | * the target existed, or across two parent directories. | |
2836 | */ | |
2837 | if (src_is_directory && (new_parent || target_ip != NULL)) { | |
2838 | ||
2839 | /* | |
2840 | * Decrement link count on src_directory since the | |
2841 | * entry that's moved no longer points to it. | |
2842 | */ | |
2843 | error = xfs_droplink(tp, src_dp); | |
2844 | if (error) | |
2845 | goto abort_return; | |
2846 | } | |
2847 | ||
2848 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, | |
2849 | &first_block, &free_list, spaceres); | |
2850 | if (error) | |
2851 | goto abort_return; | |
2852 | ||
2853 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); | |
2854 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); | |
2855 | if (new_parent) | |
2856 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); | |
2857 | ||
2858 | /* | |
2859 | * If this is a synchronous mount, make sure that the | |
2860 | * rename transaction goes to disk before returning to | |
2861 | * the user. | |
2862 | */ | |
2863 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { | |
2864 | xfs_trans_set_sync(tp); | |
2865 | } | |
2866 | ||
2867 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
2868 | if (error) { | |
2869 | xfs_bmap_cancel(&free_list); | |
2870 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | | |
2871 | XFS_TRANS_ABORT)); | |
2872 | goto std_return; | |
2873 | } | |
2874 | ||
2875 | /* | |
2876 | * trans_commit will unlock src_ip, target_ip & decrement | |
2877 | * the vnode references. | |
2878 | */ | |
2879 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); | |
2880 | ||
2881 | abort_return: | |
2882 | cancel_flags |= XFS_TRANS_ABORT; | |
2883 | error_return: | |
2884 | xfs_bmap_cancel(&free_list); | |
2885 | xfs_trans_cancel(tp, cancel_flags); | |
2886 | std_return: | |
2887 | return error; | |
2888 | } | |
2889 | ||
5c4d97d0 DC |
2890 | STATIC int |
2891 | xfs_iflush_cluster( | |
2892 | xfs_inode_t *ip, | |
2893 | xfs_buf_t *bp) | |
1da177e4 | 2894 | { |
5c4d97d0 DC |
2895 | xfs_mount_t *mp = ip->i_mount; |
2896 | struct xfs_perag *pag; | |
2897 | unsigned long first_index, mask; | |
2898 | unsigned long inodes_per_cluster; | |
2899 | int ilist_size; | |
2900 | xfs_inode_t **ilist; | |
2901 | xfs_inode_t *iq; | |
2902 | int nr_found; | |
2903 | int clcount = 0; | |
2904 | int bufwasdelwri; | |
1da177e4 | 2905 | int i; |
1da177e4 | 2906 | |
5c4d97d0 | 2907 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
1da177e4 | 2908 | |
5c4d97d0 DC |
2909 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
2910 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); | |
2911 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); | |
2912 | if (!ilist) | |
2913 | goto out_put; | |
1da177e4 | 2914 | |
5c4d97d0 DC |
2915 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
2916 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; | |
2917 | rcu_read_lock(); | |
2918 | /* really need a gang lookup range call here */ | |
2919 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, | |
2920 | first_index, inodes_per_cluster); | |
2921 | if (nr_found == 0) | |
2922 | goto out_free; | |
2923 | ||
2924 | for (i = 0; i < nr_found; i++) { | |
2925 | iq = ilist[i]; | |
2926 | if (iq == ip) | |
bad55843 | 2927 | continue; |
1a3e8f3d DC |
2928 | |
2929 | /* | |
2930 | * because this is an RCU protected lookup, we could find a | |
2931 | * recently freed or even reallocated inode during the lookup. | |
2932 | * We need to check under the i_flags_lock for a valid inode | |
2933 | * here. Skip it if it is not valid or the wrong inode. | |
2934 | */ | |
2935 | spin_lock(&ip->i_flags_lock); | |
2936 | if (!ip->i_ino || | |
2937 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { | |
2938 | spin_unlock(&ip->i_flags_lock); | |
2939 | continue; | |
2940 | } | |
2941 | spin_unlock(&ip->i_flags_lock); | |
2942 | ||
bad55843 DC |
2943 | /* |
2944 | * Do an un-protected check to see if the inode is dirty and | |
2945 | * is a candidate for flushing. These checks will be repeated | |
2946 | * later after the appropriate locks are acquired. | |
2947 | */ | |
33540408 | 2948 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
bad55843 | 2949 | continue; |
bad55843 DC |
2950 | |
2951 | /* | |
2952 | * Try to get locks. If any are unavailable or it is pinned, | |
2953 | * then this inode cannot be flushed and is skipped. | |
2954 | */ | |
2955 | ||
2956 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) | |
2957 | continue; | |
2958 | if (!xfs_iflock_nowait(iq)) { | |
2959 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2960 | continue; | |
2961 | } | |
2962 | if (xfs_ipincount(iq)) { | |
2963 | xfs_ifunlock(iq); | |
2964 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2965 | continue; | |
2966 | } | |
2967 | ||
2968 | /* | |
2969 | * arriving here means that this inode can be flushed. First | |
2970 | * re-check that it's dirty before flushing. | |
2971 | */ | |
33540408 DC |
2972 | if (!xfs_inode_clean(iq)) { |
2973 | int error; | |
bad55843 DC |
2974 | error = xfs_iflush_int(iq, bp); |
2975 | if (error) { | |
2976 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2977 | goto cluster_corrupt_out; | |
2978 | } | |
2979 | clcount++; | |
2980 | } else { | |
2981 | xfs_ifunlock(iq); | |
2982 | } | |
2983 | xfs_iunlock(iq, XFS_ILOCK_SHARED); | |
2984 | } | |
2985 | ||
2986 | if (clcount) { | |
2987 | XFS_STATS_INC(xs_icluster_flushcnt); | |
2988 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); | |
2989 | } | |
2990 | ||
2991 | out_free: | |
1a3e8f3d | 2992 | rcu_read_unlock(); |
f0e2d93c | 2993 | kmem_free(ilist); |
44b56e0a DC |
2994 | out_put: |
2995 | xfs_perag_put(pag); | |
bad55843 DC |
2996 | return 0; |
2997 | ||
2998 | ||
2999 | cluster_corrupt_out: | |
3000 | /* | |
3001 | * Corruption detected in the clustering loop. Invalidate the | |
3002 | * inode buffer and shut down the filesystem. | |
3003 | */ | |
1a3e8f3d | 3004 | rcu_read_unlock(); |
bad55843 | 3005 | /* |
43ff2122 | 3006 | * Clean up the buffer. If it was delwri, just release it -- |
bad55843 DC |
3007 | * brelse can handle it with no problems. If not, shut down the |
3008 | * filesystem before releasing the buffer. | |
3009 | */ | |
43ff2122 | 3010 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
bad55843 DC |
3011 | if (bufwasdelwri) |
3012 | xfs_buf_relse(bp); | |
3013 | ||
3014 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
3015 | ||
3016 | if (!bufwasdelwri) { | |
3017 | /* | |
3018 | * Just like incore_relse: if we have b_iodone functions, | |
3019 | * mark the buffer as an error and call them. Otherwise | |
3020 | * mark it as stale and brelse. | |
3021 | */ | |
cb669ca5 | 3022 | if (bp->b_iodone) { |
bad55843 | 3023 | XFS_BUF_UNDONE(bp); |
c867cb61 | 3024 | xfs_buf_stale(bp); |
5a52c2a5 | 3025 | xfs_buf_ioerror(bp, EIO); |
1a1a3e97 | 3026 | xfs_buf_ioend(bp, 0); |
bad55843 | 3027 | } else { |
c867cb61 | 3028 | xfs_buf_stale(bp); |
bad55843 DC |
3029 | xfs_buf_relse(bp); |
3030 | } | |
3031 | } | |
3032 | ||
3033 | /* | |
3034 | * Unlocks the flush lock | |
3035 | */ | |
04913fdd | 3036 | xfs_iflush_abort(iq, false); |
f0e2d93c | 3037 | kmem_free(ilist); |
44b56e0a | 3038 | xfs_perag_put(pag); |
bad55843 DC |
3039 | return XFS_ERROR(EFSCORRUPTED); |
3040 | } | |
3041 | ||
1da177e4 | 3042 | /* |
4c46819a CH |
3043 | * Flush dirty inode metadata into the backing buffer. |
3044 | * | |
3045 | * The caller must have the inode lock and the inode flush lock held. The | |
3046 | * inode lock will still be held upon return to the caller, and the inode | |
3047 | * flush lock will be released after the inode has reached the disk. | |
3048 | * | |
3049 | * The caller must write out the buffer returned in *bpp and release it. | |
1da177e4 LT |
3050 | */ |
3051 | int | |
3052 | xfs_iflush( | |
4c46819a CH |
3053 | struct xfs_inode *ip, |
3054 | struct xfs_buf **bpp) | |
1da177e4 | 3055 | { |
4c46819a CH |
3056 | struct xfs_mount *mp = ip->i_mount; |
3057 | struct xfs_buf *bp; | |
3058 | struct xfs_dinode *dip; | |
1da177e4 | 3059 | int error; |
1da177e4 LT |
3060 | |
3061 | XFS_STATS_INC(xs_iflush_count); | |
3062 | ||
579aa9ca | 3063 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3064 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3065 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3066 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
1da177e4 | 3067 | |
4c46819a | 3068 | *bpp = NULL; |
1da177e4 | 3069 | |
1da177e4 LT |
3070 | xfs_iunpin_wait(ip); |
3071 | ||
4b6a4688 DC |
3072 | /* |
3073 | * For stale inodes we cannot rely on the backing buffer remaining | |
3074 | * stale in cache for the remaining life of the stale inode and so | |
475ee413 | 3075 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
4b6a4688 DC |
3076 | * inodes below. We have to check this after ensuring the inode is |
3077 | * unpinned so that it is safe to reclaim the stale inode after the | |
3078 | * flush call. | |
3079 | */ | |
3080 | if (xfs_iflags_test(ip, XFS_ISTALE)) { | |
3081 | xfs_ifunlock(ip); | |
3082 | return 0; | |
3083 | } | |
3084 | ||
1da177e4 LT |
3085 | /* |
3086 | * This may have been unpinned because the filesystem is shutting | |
3087 | * down forcibly. If that's the case we must not write this inode | |
32ce90a4 CH |
3088 | * to disk, because the log record didn't make it to disk. |
3089 | * | |
3090 | * We also have to remove the log item from the AIL in this case, | |
3091 | * as we wait for an empty AIL as part of the unmount process. | |
1da177e4 LT |
3092 | */ |
3093 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
32ce90a4 CH |
3094 | error = XFS_ERROR(EIO); |
3095 | goto abort_out; | |
1da177e4 LT |
3096 | } |
3097 | ||
a3f74ffb DC |
3098 | /* |
3099 | * Get the buffer containing the on-disk inode. | |
3100 | */ | |
475ee413 CH |
3101 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
3102 | 0); | |
a3f74ffb DC |
3103 | if (error || !bp) { |
3104 | xfs_ifunlock(ip); | |
3105 | return error; | |
3106 | } | |
3107 | ||
1da177e4 LT |
3108 | /* |
3109 | * First flush out the inode that xfs_iflush was called with. | |
3110 | */ | |
3111 | error = xfs_iflush_int(ip, bp); | |
bad55843 | 3112 | if (error) |
1da177e4 | 3113 | goto corrupt_out; |
1da177e4 | 3114 | |
a3f74ffb DC |
3115 | /* |
3116 | * If the buffer is pinned then push on the log now so we won't | |
3117 | * get stuck waiting in the write for too long. | |
3118 | */ | |
811e64c7 | 3119 | if (xfs_buf_ispinned(bp)) |
a14a348b | 3120 | xfs_log_force(mp, 0); |
a3f74ffb | 3121 | |
1da177e4 LT |
3122 | /* |
3123 | * inode clustering: | |
3124 | * see if other inodes can be gathered into this write | |
3125 | */ | |
bad55843 DC |
3126 | error = xfs_iflush_cluster(ip, bp); |
3127 | if (error) | |
3128 | goto cluster_corrupt_out; | |
1da177e4 | 3129 | |
4c46819a CH |
3130 | *bpp = bp; |
3131 | return 0; | |
1da177e4 LT |
3132 | |
3133 | corrupt_out: | |
3134 | xfs_buf_relse(bp); | |
7d04a335 | 3135 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
1da177e4 | 3136 | cluster_corrupt_out: |
32ce90a4 CH |
3137 | error = XFS_ERROR(EFSCORRUPTED); |
3138 | abort_out: | |
1da177e4 LT |
3139 | /* |
3140 | * Unlocks the flush lock | |
3141 | */ | |
04913fdd | 3142 | xfs_iflush_abort(ip, false); |
32ce90a4 | 3143 | return error; |
1da177e4 LT |
3144 | } |
3145 | ||
1da177e4 LT |
3146 | STATIC int |
3147 | xfs_iflush_int( | |
93848a99 CH |
3148 | struct xfs_inode *ip, |
3149 | struct xfs_buf *bp) | |
1da177e4 | 3150 | { |
93848a99 CH |
3151 | struct xfs_inode_log_item *iip = ip->i_itemp; |
3152 | struct xfs_dinode *dip; | |
3153 | struct xfs_mount *mp = ip->i_mount; | |
1da177e4 | 3154 | |
579aa9ca | 3155 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
474fce06 | 3156 | ASSERT(xfs_isiflocked(ip)); |
1da177e4 | 3157 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
8096b1eb | 3158 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
93848a99 | 3159 | ASSERT(iip != NULL && iip->ili_fields != 0); |
1da177e4 | 3160 | |
1da177e4 | 3161 | /* set *dip = inode's place in the buffer */ |
92bfc6e7 | 3162 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); |
1da177e4 | 3163 | |
69ef921b | 3164 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
1da177e4 | 3165 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
6a19d939 DC |
3166 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3167 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", | |
3168 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); | |
1da177e4 LT |
3169 | goto corrupt_out; |
3170 | } | |
3171 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, | |
3172 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { | |
6a19d939 DC |
3173 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3174 | "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", | |
3175 | __func__, ip->i_ino, ip, ip->i_d.di_magic); | |
1da177e4 LT |
3176 | goto corrupt_out; |
3177 | } | |
abbede1b | 3178 | if (S_ISREG(ip->i_d.di_mode)) { |
1da177e4 LT |
3179 | if (XFS_TEST_ERROR( |
3180 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3181 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), | |
3182 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { | |
6a19d939 DC |
3183 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3184 | "%s: Bad regular inode %Lu, ptr 0x%p", | |
3185 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3186 | goto corrupt_out; |
3187 | } | |
abbede1b | 3188 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
1da177e4 LT |
3189 | if (XFS_TEST_ERROR( |
3190 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && | |
3191 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && | |
3192 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), | |
3193 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { | |
6a19d939 DC |
3194 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3195 | "%s: Bad directory inode %Lu, ptr 0x%p", | |
3196 | __func__, ip->i_ino, ip); | |
1da177e4 LT |
3197 | goto corrupt_out; |
3198 | } | |
3199 | } | |
3200 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > | |
3201 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, | |
3202 | XFS_RANDOM_IFLUSH_5)) { | |
6a19d939 DC |
3203 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3204 | "%s: detected corrupt incore inode %Lu, " | |
3205 | "total extents = %d, nblocks = %Ld, ptr 0x%p", | |
3206 | __func__, ip->i_ino, | |
1da177e4 | 3207 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
6a19d939 | 3208 | ip->i_d.di_nblocks, ip); |
1da177e4 LT |
3209 | goto corrupt_out; |
3210 | } | |
3211 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, | |
3212 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { | |
6a19d939 DC |
3213 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
3214 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", | |
3215 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); | |
1da177e4 LT |
3216 | goto corrupt_out; |
3217 | } | |
e60896d8 | 3218 | |
1da177e4 | 3219 | /* |
e60896d8 DC |
3220 | * Inode item log recovery for v1/v2 inodes are dependent on the |
3221 | * di_flushiter count for correct sequencing. We bump the flush | |
3222 | * iteration count so we can detect flushes which postdate a log record | |
3223 | * during recovery. This is redundant as we now log every change and | |
3224 | * hence this can't happen but we need to still do it to ensure | |
3225 | * backwards compatibility with old kernels that predate logging all | |
3226 | * inode changes. | |
1da177e4 | 3227 | */ |
e60896d8 DC |
3228 | if (ip->i_d.di_version < 3) |
3229 | ip->i_d.di_flushiter++; | |
1da177e4 LT |
3230 | |
3231 | /* | |
3232 | * Copy the dirty parts of the inode into the on-disk | |
3233 | * inode. We always copy out the core of the inode, | |
3234 | * because if the inode is dirty at all the core must | |
3235 | * be. | |
3236 | */ | |
81591fe2 | 3237 | xfs_dinode_to_disk(dip, &ip->i_d); |
1da177e4 LT |
3238 | |
3239 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ | |
3240 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) | |
3241 | ip->i_d.di_flushiter = 0; | |
3242 | ||
3243 | /* | |
3244 | * If this is really an old format inode and the superblock version | |
3245 | * has not been updated to support only new format inodes, then | |
3246 | * convert back to the old inode format. If the superblock version | |
3247 | * has been updated, then make the conversion permanent. | |
3248 | */ | |
51ce16d5 CH |
3249 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
3250 | if (ip->i_d.di_version == 1) { | |
62118709 | 3251 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
1da177e4 LT |
3252 | /* |
3253 | * Convert it back. | |
3254 | */ | |
3255 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); | |
81591fe2 | 3256 | dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
1da177e4 LT |
3257 | } else { |
3258 | /* | |
3259 | * The superblock version has already been bumped, | |
3260 | * so just make the conversion to the new inode | |
3261 | * format permanent. | |
3262 | */ | |
51ce16d5 CH |
3263 | ip->i_d.di_version = 2; |
3264 | dip->di_version = 2; | |
1da177e4 | 3265 | ip->i_d.di_onlink = 0; |
81591fe2 | 3266 | dip->di_onlink = 0; |
1da177e4 | 3267 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
81591fe2 CH |
3268 | memset(&(dip->di_pad[0]), 0, |
3269 | sizeof(dip->di_pad)); | |
6743099c | 3270 | ASSERT(xfs_get_projid(ip) == 0); |
1da177e4 LT |
3271 | } |
3272 | } | |
3273 | ||
e4ac967b DC |
3274 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
3275 | if (XFS_IFORK_Q(ip)) | |
3276 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); | |
1da177e4 LT |
3277 | xfs_inobp_check(mp, bp); |
3278 | ||
3279 | /* | |
f5d8d5c4 CH |
3280 | * We've recorded everything logged in the inode, so we'd like to clear |
3281 | * the ili_fields bits so we don't log and flush things unnecessarily. | |
3282 | * However, we can't stop logging all this information until the data | |
3283 | * we've copied into the disk buffer is written to disk. If we did we | |
3284 | * might overwrite the copy of the inode in the log with all the data | |
3285 | * after re-logging only part of it, and in the face of a crash we | |
3286 | * wouldn't have all the data we need to recover. | |
1da177e4 | 3287 | * |
f5d8d5c4 CH |
3288 | * What we do is move the bits to the ili_last_fields field. When |
3289 | * logging the inode, these bits are moved back to the ili_fields field. | |
3290 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we | |
3291 | * know that the information those bits represent is permanently on | |
3292 | * disk. As long as the flush completes before the inode is logged | |
3293 | * again, then both ili_fields and ili_last_fields will be cleared. | |
1da177e4 | 3294 | * |
f5d8d5c4 CH |
3295 | * We can play with the ili_fields bits here, because the inode lock |
3296 | * must be held exclusively in order to set bits there and the flush | |
3297 | * lock protects the ili_last_fields bits. Set ili_logged so the flush | |
3298 | * done routine can tell whether or not to look in the AIL. Also, store | |
3299 | * the current LSN of the inode so that we can tell whether the item has | |
3300 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we | |
3301 | * need the AIL lock, because it is a 64 bit value that cannot be read | |
3302 | * atomically. | |
1da177e4 | 3303 | */ |
93848a99 CH |
3304 | iip->ili_last_fields = iip->ili_fields; |
3305 | iip->ili_fields = 0; | |
3306 | iip->ili_logged = 1; | |
1da177e4 | 3307 | |
93848a99 CH |
3308 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
3309 | &iip->ili_item.li_lsn); | |
1da177e4 | 3310 | |
93848a99 CH |
3311 | /* |
3312 | * Attach the function xfs_iflush_done to the inode's | |
3313 | * buffer. This will remove the inode from the AIL | |
3314 | * and unlock the inode's flush lock when the inode is | |
3315 | * completely written to disk. | |
3316 | */ | |
3317 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); | |
1da177e4 | 3318 | |
93848a99 CH |
3319 | /* update the lsn in the on disk inode if required */ |
3320 | if (ip->i_d.di_version == 3) | |
3321 | dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); | |
3322 | ||
3323 | /* generate the checksum. */ | |
3324 | xfs_dinode_calc_crc(mp, dip); | |
1da177e4 | 3325 | |
93848a99 CH |
3326 | ASSERT(bp->b_fspriv != NULL); |
3327 | ASSERT(bp->b_iodone != NULL); | |
1da177e4 LT |
3328 | return 0; |
3329 | ||
3330 | corrupt_out: | |
3331 | return XFS_ERROR(EFSCORRUPTED); | |
3332 | } |