blob: e3d75385aa76a6e45b7711a65bb39c268c9f689b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Robert P. J. Day40ebd812007-11-23 16:30:51 +110018#include <linux/log2.h>
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_fs.h"
Dave Chinner6ca1c902013-08-12 20:49:26 +100022#include "xfs_format.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100026#include "xfs_trans_space.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_trans_priv.h"
28#include "xfs_sb.h"
29#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_mount.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100031#include "xfs_da_btree.h"
32#include "xfs_dir2_format.h"
33#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_attr_sf.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100038#include "xfs_attr.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110042#include "xfs_inode_item.h"
43#include "xfs_btree.h"
44#include "xfs_alloc.h"
45#include "xfs_ialloc.h"
46#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100047#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "xfs_quota.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100050#include "xfs_filestream.h"
Christoph Hellwig93848a92013-04-03 16:11:17 +110051#include "xfs_cksum.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000052#include "xfs_trace.h"
Dave Chinner33479e02012-10-08 21:56:11 +110053#include "xfs_icache.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100054#include "xfs_symlink.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +020059 * Used in xfs_itruncate_extents(). This is the maximum number of extents
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * freed from a file in a single transaction.
61 */
62#define XFS_ITRUNC_MAX_EXTENTS 2
63
64STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Dave Chinner2a0ec1d2012-04-23 15:59:02 +100066/*
67 * helper function to extract extent size hint from inode
68 */
69xfs_extlen_t
70xfs_get_extsz_hint(
71 struct xfs_inode *ip)
72{
73 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
74 return ip->i_d.di_extsize;
75 if (XFS_IS_REALTIME_INODE(ip))
76 return ip->i_mount->m_sb.sb_rextsize;
77 return 0;
78}
79
Dave Chinnerfa96aca2012-10-08 21:56:10 +110080/*
81 * This is a wrapper routine around the xfs_ilock() routine used to centralize
82 * some grungy code. It is used in places that wish to lock the inode solely
83 * for reading the extents. The reason these places can't just call
84 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
85 * extents from disk for a file in b-tree format. If the inode is in b-tree
86 * format, then we need to lock the inode exclusively until the extents are read
87 * in. Locking it exclusively all the time would limit our parallelism
88 * unnecessarily, though. What we do instead is check to see if the extents
89 * have been read in yet, and only lock the inode exclusively if they have not.
90 *
91 * The function returns a value which should be given to the corresponding
92 * xfs_iunlock_map_shared(). This value is the mode in which the lock was
93 * actually taken.
94 */
95uint
96xfs_ilock_map_shared(
97 xfs_inode_t *ip)
98{
99 uint lock_mode;
100
101 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
102 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
103 lock_mode = XFS_ILOCK_EXCL;
104 } else {
105 lock_mode = XFS_ILOCK_SHARED;
106 }
107
108 xfs_ilock(ip, lock_mode);
109
110 return lock_mode;
111}
112
113/*
114 * This is simply the unlock routine to go with xfs_ilock_map_shared().
115 * All it does is call xfs_iunlock() with the given lock_mode.
116 */
117void
118xfs_iunlock_map_shared(
119 xfs_inode_t *ip,
120 unsigned int lock_mode)
121{
122 xfs_iunlock(ip, lock_mode);
123}
124
125/*
126 * The xfs inode contains 2 locks: a multi-reader lock called the
127 * i_iolock and a multi-reader lock called the i_lock. This routine
128 * allows either or both of the locks to be obtained.
129 *
130 * The 2 locks should always be ordered so that the IO lock is
131 * obtained first in order to prevent deadlock.
132 *
133 * ip -- the inode being locked
134 * lock_flags -- this parameter indicates the inode's locks
135 * to be locked. It can be:
136 * XFS_IOLOCK_SHARED,
137 * XFS_IOLOCK_EXCL,
138 * XFS_ILOCK_SHARED,
139 * XFS_ILOCK_EXCL,
140 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
141 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
142 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
143 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
144 */
145void
146xfs_ilock(
147 xfs_inode_t *ip,
148 uint lock_flags)
149{
150 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
151
152 /*
153 * You can't set both SHARED and EXCL for the same lock,
154 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
155 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
156 */
157 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
158 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
159 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
160 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
161 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
162
163 if (lock_flags & XFS_IOLOCK_EXCL)
164 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
165 else if (lock_flags & XFS_IOLOCK_SHARED)
166 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
167
168 if (lock_flags & XFS_ILOCK_EXCL)
169 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
170 else if (lock_flags & XFS_ILOCK_SHARED)
171 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
172}
173
174/*
175 * This is just like xfs_ilock(), except that the caller
176 * is guaranteed not to sleep. It returns 1 if it gets
177 * the requested locks and 0 otherwise. If the IO lock is
178 * obtained but the inode lock cannot be, then the IO lock
179 * is dropped before returning.
180 *
181 * ip -- the inode being locked
182 * lock_flags -- this parameter indicates the inode's locks to be
183 * to be locked. See the comment for xfs_ilock() for a list
184 * of valid values.
185 */
186int
187xfs_ilock_nowait(
188 xfs_inode_t *ip,
189 uint lock_flags)
190{
191 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
192
193 /*
194 * You can't set both SHARED and EXCL for the same lock,
195 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
196 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
197 */
198 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
199 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
200 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
201 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
202 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
203
204 if (lock_flags & XFS_IOLOCK_EXCL) {
205 if (!mrtryupdate(&ip->i_iolock))
206 goto out;
207 } else if (lock_flags & XFS_IOLOCK_SHARED) {
208 if (!mrtryaccess(&ip->i_iolock))
209 goto out;
210 }
211 if (lock_flags & XFS_ILOCK_EXCL) {
212 if (!mrtryupdate(&ip->i_lock))
213 goto out_undo_iolock;
214 } else if (lock_flags & XFS_ILOCK_SHARED) {
215 if (!mrtryaccess(&ip->i_lock))
216 goto out_undo_iolock;
217 }
218 return 1;
219
220 out_undo_iolock:
221 if (lock_flags & XFS_IOLOCK_EXCL)
222 mrunlock_excl(&ip->i_iolock);
223 else if (lock_flags & XFS_IOLOCK_SHARED)
224 mrunlock_shared(&ip->i_iolock);
225 out:
226 return 0;
227}
228
229/*
230 * xfs_iunlock() is used to drop the inode locks acquired with
231 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
232 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
233 * that we know which locks to drop.
234 *
235 * ip -- the inode being unlocked
236 * lock_flags -- this parameter indicates the inode's locks to be
237 * to be unlocked. See the comment for xfs_ilock() for a list
238 * of valid values for this parameter.
239 *
240 */
241void
242xfs_iunlock(
243 xfs_inode_t *ip,
244 uint lock_flags)
245{
246 /*
247 * You can't set both SHARED and EXCL for the same lock,
248 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
249 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
250 */
251 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
252 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
253 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
254 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
255 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
256 ASSERT(lock_flags != 0);
257
258 if (lock_flags & XFS_IOLOCK_EXCL)
259 mrunlock_excl(&ip->i_iolock);
260 else if (lock_flags & XFS_IOLOCK_SHARED)
261 mrunlock_shared(&ip->i_iolock);
262
263 if (lock_flags & XFS_ILOCK_EXCL)
264 mrunlock_excl(&ip->i_lock);
265 else if (lock_flags & XFS_ILOCK_SHARED)
266 mrunlock_shared(&ip->i_lock);
267
268 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
269}
270
271/*
272 * give up write locks. the i/o lock cannot be held nested
273 * if it is being demoted.
274 */
275void
276xfs_ilock_demote(
277 xfs_inode_t *ip,
278 uint lock_flags)
279{
280 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
281 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
282
283 if (lock_flags & XFS_ILOCK_EXCL)
284 mrdemote(&ip->i_lock);
285 if (lock_flags & XFS_IOLOCK_EXCL)
286 mrdemote(&ip->i_iolock);
287
288 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
289}
290
Dave Chinner742ae1e2013-04-30 21:39:34 +1000291#if defined(DEBUG) || defined(XFS_WARN)
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100292int
293xfs_isilocked(
294 xfs_inode_t *ip,
295 uint lock_flags)
296{
297 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
298 if (!(lock_flags & XFS_ILOCK_SHARED))
299 return !!ip->i_lock.mr_writer;
300 return rwsem_is_locked(&ip->i_lock.mr_lock);
301 }
302
303 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
304 if (!(lock_flags & XFS_IOLOCK_SHARED))
305 return !!ip->i_iolock.mr_writer;
306 return rwsem_is_locked(&ip->i_iolock.mr_lock);
307 }
308
309 ASSERT(0);
310 return 0;
311}
312#endif
313
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000314#ifdef DEBUG
315int xfs_locked_n;
316int xfs_small_retries;
317int xfs_middle_retries;
318int xfs_lots_retries;
319int xfs_lock_delays;
320#endif
321
322/*
323 * Bump the subclass so xfs_lock_inodes() acquires each lock with
324 * a different value
325 */
326static inline int
327xfs_lock_inumorder(int lock_mode, int subclass)
328{
329 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
330 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
331 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
332 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
333
334 return lock_mode;
335}
336
337/*
338 * The following routine will lock n inodes in exclusive mode.
339 * We assume the caller calls us with the inodes in i_ino order.
340 *
341 * We need to detect deadlock where an inode that we lock
342 * is in the AIL and we start waiting for another inode that is locked
343 * by a thread in a long running transaction (such as truncate). This can
344 * result in deadlock since the long running trans might need to wait
345 * for the inode we just locked in order to push the tail and free space
346 * in the log.
347 */
348void
349xfs_lock_inodes(
350 xfs_inode_t **ips,
351 int inodes,
352 uint lock_mode)
353{
354 int attempts = 0, i, j, try_lock;
355 xfs_log_item_t *lp;
356
357 ASSERT(ips && (inodes >= 2)); /* we need at least two */
358
359 try_lock = 0;
360 i = 0;
361
362again:
363 for (; i < inodes; i++) {
364 ASSERT(ips[i]);
365
366 if (i && (ips[i] == ips[i-1])) /* Already locked */
367 continue;
368
369 /*
370 * If try_lock is not set yet, make sure all locked inodes
371 * are not in the AIL.
372 * If any are, set try_lock to be used later.
373 */
374
375 if (!try_lock) {
376 for (j = (i - 1); j >= 0 && !try_lock; j--) {
377 lp = (xfs_log_item_t *)ips[j]->i_itemp;
378 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
379 try_lock++;
380 }
381 }
382 }
383
384 /*
385 * If any of the previous locks we have locked is in the AIL,
386 * we must TRY to get the second and subsequent locks. If
387 * we can't get any, we must release all we have
388 * and try again.
389 */
390
391 if (try_lock) {
392 /* try_lock must be 0 if i is 0. */
393 /*
394 * try_lock means we have an inode locked
395 * that is in the AIL.
396 */
397 ASSERT(i != 0);
398 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
399 attempts++;
400
401 /*
402 * Unlock all previous guys and try again.
403 * xfs_iunlock will try to push the tail
404 * if the inode is in the AIL.
405 */
406
407 for(j = i - 1; j >= 0; j--) {
408
409 /*
410 * Check to see if we've already
411 * unlocked this one.
412 * Not the first one going back,
413 * and the inode ptr is the same.
414 */
415 if ((j != (i - 1)) && ips[j] ==
416 ips[j+1])
417 continue;
418
419 xfs_iunlock(ips[j], lock_mode);
420 }
421
422 if ((attempts % 5) == 0) {
423 delay(1); /* Don't just spin the CPU */
424#ifdef DEBUG
425 xfs_lock_delays++;
426#endif
427 }
428 i = 0;
429 try_lock = 0;
430 goto again;
431 }
432 } else {
433 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
434 }
435 }
436
437#ifdef DEBUG
438 if (attempts) {
439 if (attempts < 5) xfs_small_retries++;
440 else if (attempts < 100) xfs_middle_retries++;
441 else xfs_lots_retries++;
442 } else {
443 xfs_locked_n++;
444 }
445#endif
446}
447
448/*
449 * xfs_lock_two_inodes() can only be used to lock one type of lock
450 * at a time - the iolock or the ilock, but not both at once. If
451 * we lock both at once, lockdep will report false positives saying
452 * we have violated locking orders.
453 */
454void
455xfs_lock_two_inodes(
456 xfs_inode_t *ip0,
457 xfs_inode_t *ip1,
458 uint lock_mode)
459{
460 xfs_inode_t *temp;
461 int attempts = 0;
462 xfs_log_item_t *lp;
463
464 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
465 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
466 ASSERT(ip0->i_ino != ip1->i_ino);
467
468 if (ip0->i_ino > ip1->i_ino) {
469 temp = ip0;
470 ip0 = ip1;
471 ip1 = temp;
472 }
473
474 again:
475 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
476
477 /*
478 * If the first lock we have locked is in the AIL, we must TRY to get
479 * the second lock. If we can't get it, we must release the first one
480 * and try again.
481 */
482 lp = (xfs_log_item_t *)ip0->i_itemp;
483 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
484 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
485 xfs_iunlock(ip0, lock_mode);
486 if ((++attempts % 5) == 0)
487 delay(1); /* Don't just spin the CPU */
488 goto again;
489 }
490 } else {
491 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
492 }
493}
494
495
Dave Chinnerfa96aca2012-10-08 21:56:10 +1100496void
497__xfs_iflock(
498 struct xfs_inode *ip)
499{
500 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
501 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
502
503 do {
504 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
505 if (xfs_isiflocked(ip))
506 io_schedule();
507 } while (!xfs_iflock_nowait(ip));
508
509 finish_wait(wq, &wait.wait);
510}
511
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512STATIC uint
513_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 __uint16_t di_flags)
515{
516 uint flags = 0;
517
518 if (di_flags & XFS_DIFLAG_ANY) {
519 if (di_flags & XFS_DIFLAG_REALTIME)
520 flags |= XFS_XFLAG_REALTIME;
521 if (di_flags & XFS_DIFLAG_PREALLOC)
522 flags |= XFS_XFLAG_PREALLOC;
523 if (di_flags & XFS_DIFLAG_IMMUTABLE)
524 flags |= XFS_XFLAG_IMMUTABLE;
525 if (di_flags & XFS_DIFLAG_APPEND)
526 flags |= XFS_XFLAG_APPEND;
527 if (di_flags & XFS_DIFLAG_SYNC)
528 flags |= XFS_XFLAG_SYNC;
529 if (di_flags & XFS_DIFLAG_NOATIME)
530 flags |= XFS_XFLAG_NOATIME;
531 if (di_flags & XFS_DIFLAG_NODUMP)
532 flags |= XFS_XFLAG_NODUMP;
533 if (di_flags & XFS_DIFLAG_RTINHERIT)
534 flags |= XFS_XFLAG_RTINHERIT;
535 if (di_flags & XFS_DIFLAG_PROJINHERIT)
536 flags |= XFS_XFLAG_PROJINHERIT;
537 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
538 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100539 if (di_flags & XFS_DIFLAG_EXTSIZE)
540 flags |= XFS_XFLAG_EXTSIZE;
541 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
542 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000543 if (di_flags & XFS_DIFLAG_NODEFRAG)
544 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000545 if (di_flags & XFS_DIFLAG_FILESTREAM)
546 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 }
548
549 return flags;
550}
551
552uint
553xfs_ip2xflags(
554 xfs_inode_t *ip)
555{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000556 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Nathan Scotta916e2b2006-06-09 17:12:17 +1000558 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100559 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
561
562uint
563xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100564 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100566 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100567 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570/*
Dave Chinnerc24b5df2013-08-12 20:49:45 +1000571 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
572 * is allowed, otherwise it has to be an exact match. If a CI match is found,
573 * ci_name->name will point to a the actual name (caller must free) or
574 * will be set to NULL if an exact match is found.
575 */
576int
577xfs_lookup(
578 xfs_inode_t *dp,
579 struct xfs_name *name,
580 xfs_inode_t **ipp,
581 struct xfs_name *ci_name)
582{
583 xfs_ino_t inum;
584 int error;
585 uint lock_mode;
586
587 trace_xfs_lookup(dp, name);
588
589 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
590 return XFS_ERROR(EIO);
591
592 lock_mode = xfs_ilock_map_shared(dp);
593 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
594 xfs_iunlock_map_shared(dp, lock_mode);
595
596 if (error)
597 goto out;
598
599 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
600 if (error)
601 goto out_free_name;
602
603 return 0;
604
605out_free_name:
606 if (ci_name)
607 kmem_free(ci_name->name);
608out:
609 *ipp = NULL;
610 return error;
611}
612
613/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 * Allocate an inode on disk and return a copy of its in-core version.
615 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
616 * appropriately within the inode. The uid and gid for the inode are
617 * set according to the contents of the given cred structure.
618 *
619 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300620 * has a free inode available, call xfs_iget() to obtain the in-core
621 * version of the allocated inode. Finally, fill in the inode and
622 * log its initial contents. In this case, ialloc_context would be
623 * set to NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 *
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300625 * If xfs_dialloc() does not have an available inode, it will replenish
626 * its supply by doing an allocation. Since we can only do one
627 * allocation within a transaction without deadlocks, we must commit
628 * the current transaction before returning the inode itself.
629 * In this case, therefore, we will set ialloc_context and return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 * The caller should then commit the current transaction, start a new
631 * transaction, and call xfs_ialloc() again to actually get the inode.
632 *
633 * To ensure that some other process does not grab the inode that
634 * was allocated during the first call to xfs_ialloc(), this routine
635 * also returns the [locked] bp pointing to the head of the freelist
636 * as ialloc_context. The caller should hold this buffer across
637 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000638 *
639 * If we are allocating quota inodes, we do not have a parent inode
640 * to attach to or associate with (i.e. pip == NULL) because they
641 * are not linked into the directory structure - they are attached
642 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 */
644int
645xfs_ialloc(
646 xfs_trans_t *tp,
647 xfs_inode_t *pip,
Al Viro576b1d62011-07-26 02:50:15 -0400648 umode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700649 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 xfs_dev_t rdev,
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000651 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 int okalloc,
653 xfs_buf_t **ialloc_context,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 xfs_inode_t **ipp)
655{
Christoph Hellwig93848a92013-04-03 16:11:17 +1100656 struct xfs_mount *mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 xfs_ino_t ino;
658 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 uint flags;
660 int error;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000661 timespec_t tv;
David Chinnerbf904242008-10-30 17:36:14 +1100662 int filestreams = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 /*
665 * Call the space management code to pick
666 * the on-disk inode to be allocated.
667 */
David Chinnerb11f94d2007-07-11 11:09:33 +1000668 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Christoph Hellwig08358902012-07-04 10:54:47 -0400669 ialloc_context, &ino);
David Chinnerbf904242008-10-30 17:36:14 +1100670 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 return error;
Christoph Hellwig08358902012-07-04 10:54:47 -0400672 if (*ialloc_context || ino == NULLFSINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *ipp = NULL;
674 return 0;
675 }
676 ASSERT(*ialloc_context == NULL);
677
678 /*
679 * Get the in-core inode with the lock held exclusively.
680 * This is because we're setting fields here we need
681 * to prevent others from looking at until we're done.
682 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100683 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
Christoph Hellwigec3ba852011-02-13 13:26:42 +0000684 XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +1100685 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 ASSERT(ip != NULL);
688
Al Viro576b1d62011-07-26 02:50:15 -0400689 ip->i_d.di_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 ip->i_d.di_onlink = 0;
691 ip->i_d.di_nlink = nlink;
692 ASSERT(ip->i_d.di_nlink == nlink);
Dwight Engen7aab1b22013-08-15 14:08:01 -0400693 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
694 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000695 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
697
698 /*
699 * If the superblock version is up to where we support new format
700 * inodes and this is currently an old format inode, then change
701 * the inode version number now. This way we only do the conversion
702 * here rather than here and in the flush/logging code.
703 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100704 if (xfs_sb_version_hasnlink(&mp->m_sb) &&
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100705 ip->i_d.di_version == 1) {
706 ip->i_d.di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 /*
708 * We've already zeroed the old link count, the projid field,
709 * and the pad field.
710 */
711 }
712
713 /*
714 * Project ids won't be stored on disk if we are using a version 1 inode.
715 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100716 if ((prid != 0) && (ip->i_d.di_version == 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 xfs_bump_ino_vers2(tp, ip);
718
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000719 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 ip->i_d.di_gid = pip->i_d.di_gid;
Al Viroabbede12011-07-26 02:31:30 -0400721 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 ip->i_d.di_mode |= S_ISGID;
723 }
724 }
725
726 /*
727 * If the group ID of the new file does not match the effective group
728 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
729 * (and only if the irix_sgid_inherit compatibility variable is set).
730 */
731 if ((irix_sgid_inherit) &&
732 (ip->i_d.di_mode & S_ISGID) &&
Dwight Engen7aab1b22013-08-15 14:08:01 -0400733 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 ip->i_d.di_mode &= ~S_ISGID;
735 }
736
737 ip->i_d.di_size = 0;
738 ip->i_d.di_nextents = 0;
739 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000740
741 nanotime(&tv);
742 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
743 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
744 ip->i_d.di_atime = ip->i_d.di_mtime;
745 ip->i_d.di_ctime = ip->i_d.di_mtime;
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 /*
748 * di_gen will have been taken care of in xfs_iread.
749 */
750 ip->i_d.di_extsize = 0;
751 ip->i_d.di_dmevmask = 0;
752 ip->i_d.di_dmstate = 0;
753 ip->i_d.di_flags = 0;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100754
755 if (ip->i_d.di_version == 3) {
756 ASSERT(ip->i_d.di_ino == ino);
757 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
758 ip->i_d.di_crc = 0;
759 ip->i_d.di_changecount = 1;
760 ip->i_d.di_lsn = 0;
761 ip->i_d.di_flags2 = 0;
762 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
763 ip->i_d.di_crtime = ip->i_d.di_mtime;
764 }
765
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 flags = XFS_ILOG_CORE;
768 switch (mode & S_IFMT) {
769 case S_IFIFO:
770 case S_IFCHR:
771 case S_IFBLK:
772 case S_IFSOCK:
773 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
774 ip->i_df.if_u2.if_rdev = rdev;
775 ip->i_df.if_flags = 0;
776 flags |= XFS_ILOG_DEV;
777 break;
778 case S_IFREG:
David Chinnerbf904242008-10-30 17:36:14 +1100779 /*
780 * we can't set up filestreams until after the VFS inode
781 * is set up properly.
782 */
783 if (pip && xfs_inode_is_filestream(pip))
784 filestreams = 1;
David Chinner2a82b8b2007-07-11 11:09:12 +1000785 /* fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +1000787 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000788 uint di_flags = 0;
789
Al Viroabbede12011-07-26 02:31:30 -0400790 if (S_ISDIR(mode)) {
Nathan Scott365ca832005-06-21 15:39:12 +1000791 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
792 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100793 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
794 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
795 ip->i_d.di_extsize = pip->i_d.di_extsize;
796 }
Al Viroabbede12011-07-26 02:31:30 -0400797 } else if (S_ISREG(mode)) {
Christoph Hellwig613d7042007-10-11 17:44:08 +1000798 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +1000799 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100800 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
801 di_flags |= XFS_DIFLAG_EXTSIZE;
802 ip->i_d.di_extsize = pip->i_d.di_extsize;
803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
805 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
806 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +1000807 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
809 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +1000810 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
812 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +1000813 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
815 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +1000816 di_flags |= XFS_DIFLAG_NOSYMLINKS;
817 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
818 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000819 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
820 xfs_inherit_nodefrag)
821 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000822 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
823 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +1000824 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 }
826 /* FALLTHROUGH */
827 case S_IFLNK:
828 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
829 ip->i_df.if_flags = XFS_IFEXTENTS;
830 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
831 ip->i_df.if_u1.if_extents = NULL;
832 break;
833 default:
834 ASSERT(0);
835 }
836 /*
837 * Attribute fork settings for new inode.
838 */
839 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
840 ip->i_d.di_anextents = 0;
841
842 /*
843 * Log the new values stuffed into the inode.
844 */
Christoph Hellwigddc34152011-09-19 15:00:54 +0000845 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 xfs_trans_log_inode(tp, ip, flags);
847
Nathan Scottb83bd132006-06-09 16:48:30 +1000848 /* now that we have an i_mode we can setup inode ops and unlock */
Christoph Hellwig41be8be2008-08-13 16:23:13 +1000849 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
David Chinnerbf904242008-10-30 17:36:14 +1100851 /* now we have set up the vfs inode we can associate the filestream */
852 if (filestreams) {
853 error = xfs_filestream_associate(pip, ip);
854 if (error < 0)
855 return -error;
856 if (!error)
857 xfs_iflags_set(ip, XFS_IFILESTREAM);
858 }
859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 *ipp = ip;
861 return 0;
862}
863
Dave Chinnere546cb72013-08-12 20:49:47 +1000864/*
865 * Allocates a new inode from disk and return a pointer to the
866 * incore copy. This routine will internally commit the current
867 * transaction and allocate a new one if the Space Manager needed
868 * to do an allocation to replenish the inode free-list.
869 *
870 * This routine is designed to be called from xfs_create and
871 * xfs_create_dir.
872 *
873 */
874int
875xfs_dir_ialloc(
876 xfs_trans_t **tpp, /* input: current transaction;
877 output: may be a new transaction. */
878 xfs_inode_t *dp, /* directory within whose allocate
879 the inode. */
880 umode_t mode,
881 xfs_nlink_t nlink,
882 xfs_dev_t rdev,
883 prid_t prid, /* project id */
884 int okalloc, /* ok to allocate new space */
885 xfs_inode_t **ipp, /* pointer to inode; it will be
886 locked. */
887 int *committed)
888
889{
890 xfs_trans_t *tp;
891 xfs_trans_t *ntp;
892 xfs_inode_t *ip;
893 xfs_buf_t *ialloc_context = NULL;
894 int code;
Dave Chinnere546cb72013-08-12 20:49:47 +1000895 void *dqinfo;
896 uint tflags;
897
898 tp = *tpp;
899 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
900
901 /*
902 * xfs_ialloc will return a pointer to an incore inode if
903 * the Space Manager has an available inode on the free
904 * list. Otherwise, it will do an allocation and replenish
905 * the freelist. Since we can only do one allocation per
906 * transaction without deadlocks, we will need to commit the
907 * current transaction and start a new one. We will then
908 * need to call xfs_ialloc again to get the inode.
909 *
910 * If xfs_ialloc did an allocation to replenish the freelist,
911 * it returns the bp containing the head of the freelist as
912 * ialloc_context. We will hold a lock on it across the
913 * transaction commit so that no other process can steal
914 * the inode(s) that we've just allocated.
915 */
916 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
917 &ialloc_context, &ip);
918
919 /*
920 * Return an error if we were unable to allocate a new inode.
921 * This should only happen if we run out of space on disk or
922 * encounter a disk error.
923 */
924 if (code) {
925 *ipp = NULL;
926 return code;
927 }
928 if (!ialloc_context && !ip) {
929 *ipp = NULL;
930 return XFS_ERROR(ENOSPC);
931 }
932
933 /*
934 * If the AGI buffer is non-NULL, then we were unable to get an
935 * inode in one operation. We need to commit the current
936 * transaction and call xfs_ialloc() again. It is guaranteed
937 * to succeed the second time.
938 */
939 if (ialloc_context) {
Jie Liu3d3c8b52013-08-12 20:49:59 +1000940 struct xfs_trans_res tres;
941
Dave Chinnere546cb72013-08-12 20:49:47 +1000942 /*
943 * Normally, xfs_trans_commit releases all the locks.
944 * We call bhold to hang on to the ialloc_context across
945 * the commit. Holding this buffer prevents any other
946 * processes from doing any allocations in this
947 * allocation group.
948 */
949 xfs_trans_bhold(tp, ialloc_context);
950 /*
951 * Save the log reservation so we can use
952 * them in the next transaction.
953 */
Jie Liu3d3c8b52013-08-12 20:49:59 +1000954 tres.tr_logres = xfs_trans_get_log_res(tp);
955 tres.tr_logcount = xfs_trans_get_log_count(tp);
Dave Chinnere546cb72013-08-12 20:49:47 +1000956
957 /*
958 * We want the quota changes to be associated with the next
959 * transaction, NOT this one. So, detach the dqinfo from this
960 * and attach it to the next transaction.
961 */
962 dqinfo = NULL;
963 tflags = 0;
964 if (tp->t_dqinfo) {
965 dqinfo = (void *)tp->t_dqinfo;
966 tp->t_dqinfo = NULL;
967 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
968 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
969 }
970
971 ntp = xfs_trans_dup(tp);
972 code = xfs_trans_commit(tp, 0);
973 tp = ntp;
974 if (committed != NULL) {
975 *committed = 1;
976 }
977 /*
978 * If we get an error during the commit processing,
979 * release the buffer that is still held and return
980 * to the caller.
981 */
982 if (code) {
983 xfs_buf_relse(ialloc_context);
984 if (dqinfo) {
985 tp->t_dqinfo = dqinfo;
986 xfs_trans_free_dqinfo(tp);
987 }
988 *tpp = ntp;
989 *ipp = NULL;
990 return code;
991 }
992
993 /*
994 * transaction commit worked ok so we can drop the extra ticket
995 * reference that we gained in xfs_trans_dup()
996 */
997 xfs_log_ticket_put(tp->t_ticket);
Jie Liu3d3c8b52013-08-12 20:49:59 +1000998 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
999 code = xfs_trans_reserve(tp, &tres, 0, 0);
1000
Dave Chinnere546cb72013-08-12 20:49:47 +10001001 /*
1002 * Re-attach the quota info that we detached from prev trx.
1003 */
1004 if (dqinfo) {
1005 tp->t_dqinfo = dqinfo;
1006 tp->t_flags |= tflags;
1007 }
1008
1009 if (code) {
1010 xfs_buf_relse(ialloc_context);
1011 *tpp = ntp;
1012 *ipp = NULL;
1013 return code;
1014 }
1015 xfs_trans_bjoin(tp, ialloc_context);
1016
1017 /*
1018 * Call ialloc again. Since we've locked out all
1019 * other allocations in this allocation group,
1020 * this call should always succeed.
1021 */
1022 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1023 okalloc, &ialloc_context, &ip);
1024
1025 /*
1026 * If we get an error at this point, return to the caller
1027 * so that the current transaction can be aborted.
1028 */
1029 if (code) {
1030 *tpp = tp;
1031 *ipp = NULL;
1032 return code;
1033 }
1034 ASSERT(!ialloc_context && ip);
1035
1036 } else {
1037 if (committed != NULL)
1038 *committed = 0;
1039 }
1040
1041 *ipp = ip;
1042 *tpp = tp;
1043
1044 return 0;
1045}
1046
1047/*
1048 * Decrement the link count on an inode & log the change.
1049 * If this causes the link count to go to zero, initiate the
1050 * logging activity required to truncate a file.
1051 */
1052int /* error */
1053xfs_droplink(
1054 xfs_trans_t *tp,
1055 xfs_inode_t *ip)
1056{
1057 int error;
1058
1059 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1060
1061 ASSERT (ip->i_d.di_nlink > 0);
1062 ip->i_d.di_nlink--;
1063 drop_nlink(VFS_I(ip));
1064 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1065
1066 error = 0;
1067 if (ip->i_d.di_nlink == 0) {
1068 /*
1069 * We're dropping the last link to this file.
1070 * Move the on-disk inode to the AGI unlinked list.
1071 * From xfs_inactive() we will pull the inode from
1072 * the list and free it.
1073 */
1074 error = xfs_iunlink(tp, ip);
1075 }
1076 return error;
1077}
1078
1079/*
1080 * This gets called when the inode's version needs to be changed from 1 to 2.
1081 * Currently this happens when the nlink field overflows the old 16-bit value
1082 * or when chproj is called to change the project for the first time.
1083 * As a side effect the superblock version will also get rev'd
1084 * to contain the NLINK bit.
1085 */
1086void
1087xfs_bump_ino_vers2(
1088 xfs_trans_t *tp,
1089 xfs_inode_t *ip)
1090{
1091 xfs_mount_t *mp;
1092
1093 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1094 ASSERT(ip->i_d.di_version == 1);
1095
1096 ip->i_d.di_version = 2;
1097 ip->i_d.di_onlink = 0;
1098 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1099 mp = tp->t_mountp;
1100 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1101 spin_lock(&mp->m_sb_lock);
1102 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1103 xfs_sb_version_addnlink(&mp->m_sb);
1104 spin_unlock(&mp->m_sb_lock);
1105 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
1106 } else {
1107 spin_unlock(&mp->m_sb_lock);
1108 }
1109 }
1110 /* Caller must log the inode */
1111}
1112
1113/*
1114 * Increment the link count on an inode & log the change.
1115 */
1116int
1117xfs_bumplink(
1118 xfs_trans_t *tp,
1119 xfs_inode_t *ip)
1120{
1121 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1122
1123 ASSERT(ip->i_d.di_nlink > 0);
1124 ip->i_d.di_nlink++;
1125 inc_nlink(VFS_I(ip));
1126 if ((ip->i_d.di_version == 1) &&
1127 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
1128 /*
1129 * The inode has increased its number of links beyond
1130 * what can fit in an old format inode. It now needs
1131 * to be converted to a version 2 inode with a 32 bit
1132 * link count. If this is the first inode in the file
1133 * system to do this, then we need to bump the superblock
1134 * version number as well.
1135 */
1136 xfs_bump_ino_vers2(tp, ip);
1137 }
1138
1139 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1140 return 0;
1141}
1142
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001143int
1144xfs_create(
1145 xfs_inode_t *dp,
1146 struct xfs_name *name,
1147 umode_t mode,
1148 xfs_dev_t rdev,
1149 xfs_inode_t **ipp)
1150{
1151 int is_dir = S_ISDIR(mode);
1152 struct xfs_mount *mp = dp->i_mount;
1153 struct xfs_inode *ip = NULL;
1154 struct xfs_trans *tp = NULL;
1155 int error;
1156 xfs_bmap_free_t free_list;
1157 xfs_fsblock_t first_block;
1158 bool unlock_dp_on_error = false;
1159 uint cancel_flags;
1160 int committed;
1161 prid_t prid;
1162 struct xfs_dquot *udqp = NULL;
1163 struct xfs_dquot *gdqp = NULL;
1164 struct xfs_dquot *pdqp = NULL;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001165 struct xfs_trans_res tres;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001166 uint resblks;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001167
1168 trace_xfs_create(dp, name);
1169
1170 if (XFS_FORCED_SHUTDOWN(mp))
1171 return XFS_ERROR(EIO);
1172
1173 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1174 prid = xfs_get_projid(dp);
1175 else
1176 prid = XFS_PROJID_DEFAULT;
1177
1178 /*
1179 * Make sure that we have allocated dquot(s) on disk.
1180 */
Dwight Engen7aab1b22013-08-15 14:08:01 -04001181 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1182 xfs_kgid_to_gid(current_fsgid()), prid,
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001183 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1184 &udqp, &gdqp, &pdqp);
1185 if (error)
1186 return error;
1187
1188 if (is_dir) {
1189 rdev = 0;
1190 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001191 tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1192 tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001193 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1194 } else {
1195 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001196 tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1197 tres.tr_logcount = XFS_CREATE_LOG_COUNT;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001198 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1199 }
1200
1201 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1202
1203 /*
1204 * Initially assume that the file does not exist and
1205 * reserve the resources for that case. If that is not
1206 * the case we'll drop the one we have and get a more
1207 * appropriate transaction later.
1208 */
Jie Liu3d3c8b52013-08-12 20:49:59 +10001209 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1210 error = xfs_trans_reserve(tp, &tres, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001211 if (error == ENOSPC) {
1212 /* flush outstanding delalloc blocks and retry */
1213 xfs_flush_inodes(mp);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001214 error = xfs_trans_reserve(tp, &tres, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001215 }
1216 if (error == ENOSPC) {
1217 /* No space at all so try a "no-allocation" reservation */
1218 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001219 error = xfs_trans_reserve(tp, &tres, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001220 }
1221 if (error) {
1222 cancel_flags = 0;
1223 goto out_trans_cancel;
1224 }
1225
1226 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1227 unlock_dp_on_error = true;
1228
1229 xfs_bmap_init(&free_list, &first_block);
1230
1231 /*
1232 * Reserve disk quota and the inode.
1233 */
1234 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1235 pdqp, resblks, 1, 0);
1236 if (error)
1237 goto out_trans_cancel;
1238
1239 error = xfs_dir_canenter(tp, dp, name, resblks);
1240 if (error)
1241 goto out_trans_cancel;
1242
1243 /*
1244 * A newly created regular or special file just has one directory
1245 * entry pointing to them, but a directory also the "." entry
1246 * pointing to itself.
1247 */
1248 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1249 prid, resblks > 0, &ip, &committed);
1250 if (error) {
1251 if (error == ENOSPC)
1252 goto out_trans_cancel;
1253 goto out_trans_abort;
1254 }
1255
1256 /*
1257 * Now we join the directory inode to the transaction. We do not do it
1258 * earlier because xfs_dir_ialloc might commit the previous transaction
1259 * (and release all the locks). An error from here on will result in
1260 * the transaction cancel unlocking dp so don't do it explicitly in the
1261 * error path.
1262 */
1263 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1264 unlock_dp_on_error = false;
1265
1266 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1267 &first_block, &free_list, resblks ?
1268 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1269 if (error) {
1270 ASSERT(error != ENOSPC);
1271 goto out_trans_abort;
1272 }
1273 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1274 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1275
1276 if (is_dir) {
1277 error = xfs_dir_init(tp, ip, dp);
1278 if (error)
1279 goto out_bmap_cancel;
1280
1281 error = xfs_bumplink(tp, dp);
1282 if (error)
1283 goto out_bmap_cancel;
1284 }
1285
1286 /*
1287 * If this is a synchronous mount, make sure that the
1288 * create transaction goes to disk before returning to
1289 * the user.
1290 */
1291 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1292 xfs_trans_set_sync(tp);
1293
1294 /*
1295 * Attach the dquot(s) to the inodes and modify them incore.
1296 * These ids of the inode couldn't have changed since the new
1297 * inode has been locked ever since it was created.
1298 */
1299 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1300
1301 error = xfs_bmap_finish(&tp, &free_list, &committed);
1302 if (error)
1303 goto out_bmap_cancel;
1304
1305 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1306 if (error)
1307 goto out_release_inode;
1308
1309 xfs_qm_dqrele(udqp);
1310 xfs_qm_dqrele(gdqp);
1311 xfs_qm_dqrele(pdqp);
1312
1313 *ipp = ip;
1314 return 0;
1315
1316 out_bmap_cancel:
1317 xfs_bmap_cancel(&free_list);
1318 out_trans_abort:
1319 cancel_flags |= XFS_TRANS_ABORT;
1320 out_trans_cancel:
1321 xfs_trans_cancel(tp, cancel_flags);
1322 out_release_inode:
1323 /*
1324 * Wait until after the current transaction is aborted to
1325 * release the inode. This prevents recursive transactions
1326 * and deadlocks from xfs_inactive.
1327 */
1328 if (ip)
1329 IRELE(ip);
1330
1331 xfs_qm_dqrele(udqp);
1332 xfs_qm_dqrele(gdqp);
1333 xfs_qm_dqrele(pdqp);
1334
1335 if (unlock_dp_on_error)
1336 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1337 return error;
1338}
1339
1340int
1341xfs_link(
1342 xfs_inode_t *tdp,
1343 xfs_inode_t *sip,
1344 struct xfs_name *target_name)
1345{
1346 xfs_mount_t *mp = tdp->i_mount;
1347 xfs_trans_t *tp;
1348 int error;
1349 xfs_bmap_free_t free_list;
1350 xfs_fsblock_t first_block;
1351 int cancel_flags;
1352 int committed;
1353 int resblks;
1354
1355 trace_xfs_link(tdp, target_name);
1356
1357 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1358
1359 if (XFS_FORCED_SHUTDOWN(mp))
1360 return XFS_ERROR(EIO);
1361
1362 error = xfs_qm_dqattach(sip, 0);
1363 if (error)
1364 goto std_return;
1365
1366 error = xfs_qm_dqattach(tdp, 0);
1367 if (error)
1368 goto std_return;
1369
1370 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1371 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1372 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001373 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001374 if (error == ENOSPC) {
1375 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10001376 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001377 }
1378 if (error) {
1379 cancel_flags = 0;
1380 goto error_return;
1381 }
1382
1383 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1384
1385 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1386 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1387
1388 /*
1389 * If we are using project inheritance, we only allow hard link
1390 * creation in our tree when the project IDs are the same; else
1391 * the tree quota mechanism could be circumvented.
1392 */
1393 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1394 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1395 error = XFS_ERROR(EXDEV);
1396 goto error_return;
1397 }
1398
1399 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1400 if (error)
1401 goto error_return;
1402
1403 xfs_bmap_init(&free_list, &first_block);
1404
1405 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1406 &first_block, &free_list, resblks);
1407 if (error)
1408 goto abort_return;
1409 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1410 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1411
1412 error = xfs_bumplink(tp, sip);
1413 if (error)
1414 goto abort_return;
1415
1416 /*
1417 * If this is a synchronous mount, make sure that the
1418 * link transaction goes to disk before returning to
1419 * the user.
1420 */
1421 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1422 xfs_trans_set_sync(tp);
1423 }
1424
1425 error = xfs_bmap_finish (&tp, &free_list, &committed);
1426 if (error) {
1427 xfs_bmap_cancel(&free_list);
1428 goto abort_return;
1429 }
1430
1431 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1432
1433 abort_return:
1434 cancel_flags |= XFS_TRANS_ABORT;
1435 error_return:
1436 xfs_trans_cancel(tp, cancel_flags);
1437 std_return:
1438 return error;
1439}
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441/*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001442 * Free up the underlying blocks past new_size. The new size must be smaller
1443 * than the current size. This routine can be used both for the attribute and
1444 * data fork, and does not modify the inode size, which is left to the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 *
David Chinnerf6485052008-04-17 16:50:04 +10001446 * The transaction passed to this routine must have made a permanent log
1447 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1448 * given transaction and start new ones, so make sure everything involved in
1449 * the transaction is tidy before calling here. Some transaction will be
1450 * returned to the caller to be committed. The incoming transaction must
1451 * already include the inode, and both inode locks must be held exclusively.
1452 * The inode must also be "held" within the transaction. On return the inode
1453 * will be "held" within the returned transaction. This routine does NOT
1454 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 *
David Chinnerf6485052008-04-17 16:50:04 +10001456 * If we get an error, we must return with the inode locked and linked into the
1457 * current transaction. This keeps things simple for the higher level code,
1458 * because it always knows that the inode is locked and held in the transaction
1459 * that returns to it whether errors occur or not. We don't mark the inode
1460 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 */
1462int
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001463xfs_itruncate_extents(
1464 struct xfs_trans **tpp,
1465 struct xfs_inode *ip,
1466 int whichfork,
1467 xfs_fsize_t new_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468{
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001469 struct xfs_mount *mp = ip->i_mount;
1470 struct xfs_trans *tp = *tpp;
1471 struct xfs_trans *ntp;
1472 xfs_bmap_free_t free_list;
1473 xfs_fsblock_t first_block;
1474 xfs_fileoff_t first_unmap_block;
1475 xfs_fileoff_t last_block;
1476 xfs_filblks_t unmap_len;
1477 int committed;
1478 int error = 0;
1479 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Christoph Hellwig0b561852012-07-04 11:13:31 -04001481 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1482 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1483 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00001484 ASSERT(new_size <= XFS_ISIZE(ip));
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001485 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001487 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001488 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001490 trace_xfs_itruncate_extents_start(ip, new_size);
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 /*
1493 * Since it is possible for space to become allocated beyond
1494 * the end of the file (in a crash where the space is allocated
1495 * but the inode size is not yet updated), simply remove any
1496 * blocks which show up between the new EOF and the maximum
1497 * possible file size. If the first block to be removed is
1498 * beyond the maximum file size (ie it is the same as last_block),
1499 * then there is nothing to do.
1500 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001501 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Dave Chinner32972382012-06-08 15:44:54 +10001502 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001503 if (first_unmap_block == last_block)
1504 return 0;
1505
1506 ASSERT(first_unmap_block < last_block);
1507 unmap_len = last_block - first_unmap_block + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 while (!done) {
Eric Sandeen9d87c312009-01-14 23:22:07 -06001509 xfs_bmap_init(&free_list, &first_block);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001510 error = xfs_bunmapi(tp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001511 first_unmap_block, unmap_len,
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001512 xfs_bmapi_aflag(whichfork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001514 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001515 &done);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001516 if (error)
1517 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519 /*
1520 * Duplicate the transaction that has the permanent
1521 * reservation and commit the old transaction.
1522 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001523 error = xfs_bmap_finish(&tp, &free_list, &committed);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001524 if (committed)
Christoph Hellwigddc34152011-09-19 15:00:54 +00001525 xfs_trans_ijoin(tp, ip, 0);
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001526 if (error)
1527 goto out_bmap_cancel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 if (committed) {
1530 /*
David Chinnerf6485052008-04-17 16:50:04 +10001531 * Mark the inode dirty so it will be logged and
David Chinnere5720ee2008-04-10 12:21:18 +10001532 * moved forward in the log as part of every commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001534 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
David Chinnerf6485052008-04-17 16:50:04 +10001536
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001537 ntp = xfs_trans_dup(tp);
1538 error = xfs_trans_commit(tp, 0);
1539 tp = ntp;
David Chinnere5720ee2008-04-10 12:21:18 +10001540
Christoph Hellwigddc34152011-09-19 15:00:54 +00001541 xfs_trans_ijoin(tp, ip, 0);
David Chinnerf6485052008-04-17 16:50:04 +10001542
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001543 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001544 goto out;
1545
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001546 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001547 * Transaction commit worked ok so we can drop the extra ticket
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001548 * reference that we gained in xfs_trans_dup()
1549 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001550 xfs_log_ticket_put(tp->t_ticket);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001551 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
David Chinnerf6485052008-04-17 16:50:04 +10001552 if (error)
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001553 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 }
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001555
Christoph Hellwig673e8e52011-12-18 20:00:04 +00001556 /*
1557 * Always re-log the inode so that our permanent transaction can keep
1558 * on rolling it forward in the log.
1559 */
1560 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1561
1562 trace_xfs_itruncate_extents_end(ip, new_size);
1563
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001564out:
1565 *tpp = tp;
1566 return error;
1567out_bmap_cancel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 /*
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001569 * If the bunmapi call encounters an error, return to the caller where
1570 * the transaction can be properly aborted. We just need to make sure
1571 * we're not holding any resources that we were not when we came in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 */
Christoph Hellwig8f04c472011-07-08 14:34:34 +02001573 xfs_bmap_cancel(&free_list);
1574 goto out;
1575}
1576
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001577int
1578xfs_release(
1579 xfs_inode_t *ip)
1580{
1581 xfs_mount_t *mp = ip->i_mount;
1582 int error;
1583
1584 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1585 return 0;
1586
1587 /* If this is a read-only mount, don't do this (would generate I/O) */
1588 if (mp->m_flags & XFS_MOUNT_RDONLY)
1589 return 0;
1590
1591 if (!XFS_FORCED_SHUTDOWN(mp)) {
1592 int truncated;
1593
1594 /*
1595 * If we are using filestreams, and we have an unlinked
1596 * file that we are processing the last close on, then nothing
1597 * will be able to reopen and write to this file. Purge this
1598 * inode from the filestreams cache so that it doesn't delay
1599 * teardown of the inode.
1600 */
1601 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1602 xfs_filestream_deassociate(ip);
1603
1604 /*
1605 * If we previously truncated this file and removed old data
1606 * in the process, we want to initiate "early" writeout on
1607 * the last close. This is an attempt to combat the notorious
1608 * NULL files problem which is particularly noticeable from a
1609 * truncate down, buffered (re-)write (delalloc), followed by
1610 * a crash. What we are effectively doing here is
1611 * significantly reducing the time window where we'd otherwise
1612 * be exposed to that problem.
1613 */
1614 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1615 if (truncated) {
1616 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1617 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1618 error = -filemap_flush(VFS_I(ip)->i_mapping);
1619 if (error)
1620 return error;
1621 }
1622 }
1623 }
1624
1625 if (ip->i_d.di_nlink == 0)
1626 return 0;
1627
1628 if (xfs_can_free_eofblocks(ip, false)) {
1629
1630 /*
1631 * If we can't get the iolock just skip truncating the blocks
1632 * past EOF because we could deadlock with the mmap_sem
1633 * otherwise. We'll get another chance to drop them once the
1634 * last reference to the inode is dropped, so we'll never leak
1635 * blocks permanently.
1636 *
1637 * Further, check if the inode is being opened, written and
1638 * closed frequently and we have delayed allocation blocks
1639 * outstanding (e.g. streaming writes from the NFS server),
1640 * truncating the blocks past EOF will cause fragmentation to
1641 * occur.
1642 *
1643 * In this case don't do the truncation, either, but we have to
1644 * be careful how we detect this case. Blocks beyond EOF show
1645 * up as i_delayed_blks even when the inode is clean, so we
1646 * need to truncate them away first before checking for a dirty
1647 * release. Hence on the first dirty close we will still remove
1648 * the speculative allocation, but after that we will leave it
1649 * in place.
1650 */
1651 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1652 return 0;
1653
1654 error = xfs_free_eofblocks(mp, ip, true);
1655 if (error && error != EAGAIN)
1656 return error;
1657
1658 /* delalloc blocks after truncation means it really is dirty */
1659 if (ip->i_delayed_blks)
1660 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1661 }
1662 return 0;
1663}
1664
1665/*
1666 * xfs_inactive
1667 *
1668 * This is called when the vnode reference count for the vnode
1669 * goes to zero. If the file has been unlinked, then it must
1670 * now be truncated. Also, we clear all of the read-ahead state
1671 * kept for the inode here since the file is now closed.
1672 */
1673int
1674xfs_inactive(
1675 xfs_inode_t *ip)
1676{
Jie Liu3d3c8b52013-08-12 20:49:59 +10001677 xfs_bmap_free_t free_list;
1678 xfs_fsblock_t first_block;
1679 int committed;
1680 struct xfs_trans *tp;
1681 struct xfs_mount *mp;
1682 struct xfs_trans_res *resp;
1683 int error;
1684 int truncate = 0;
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001685
1686 /*
1687 * If the inode is already free, then there can be nothing
1688 * to clean up here.
1689 */
1690 if (ip->i_d.di_mode == 0 || is_bad_inode(VFS_I(ip))) {
1691 ASSERT(ip->i_df.if_real_bytes == 0);
1692 ASSERT(ip->i_df.if_broot_bytes == 0);
1693 return VN_INACTIVE_CACHE;
1694 }
1695
1696 mp = ip->i_mount;
1697
1698 error = 0;
1699
1700 /* If this is a read-only mount, don't do this (would generate I/O) */
1701 if (mp->m_flags & XFS_MOUNT_RDONLY)
1702 goto out;
1703
1704 if (ip->i_d.di_nlink != 0) {
1705 /*
1706 * force is true because we are evicting an inode from the
1707 * cache. Post-eof blocks must be freed, lest we end up with
1708 * broken free space accounting.
1709 */
1710 if (xfs_can_free_eofblocks(ip, true)) {
1711 error = xfs_free_eofblocks(mp, ip, false);
1712 if (error)
1713 return VN_INACTIVE_CACHE;
1714 }
1715 goto out;
1716 }
1717
1718 if (S_ISREG(ip->i_d.di_mode) &&
1719 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1720 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1721 truncate = 1;
1722
1723 error = xfs_qm_dqattach(ip, 0);
1724 if (error)
1725 return VN_INACTIVE_CACHE;
1726
1727 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001728 resp = (truncate || S_ISLNK(ip->i_d.di_mode)) ?
1729 &M_RES(mp)->tr_itruncate : &M_RES(mp)->tr_ifree;
1730
1731 error = xfs_trans_reserve(tp, resp, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001732 if (error) {
1733 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1734 xfs_trans_cancel(tp, 0);
1735 return VN_INACTIVE_CACHE;
1736 }
1737
1738 xfs_ilock(ip, XFS_ILOCK_EXCL);
1739 xfs_trans_ijoin(tp, ip, 0);
1740
1741 if (S_ISLNK(ip->i_d.di_mode)) {
1742 error = xfs_inactive_symlink(ip, &tp);
1743 if (error)
1744 goto out_cancel;
1745 } else if (truncate) {
1746 ip->i_d.di_size = 0;
1747 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1748
1749 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1750 if (error)
1751 goto out_cancel;
1752
1753 ASSERT(ip->i_d.di_nextents == 0);
1754 }
1755
1756 /*
1757 * If there are attributes associated with the file then blow them away
1758 * now. The code calls a routine that recursively deconstructs the
1759 * attribute fork. We need to just commit the current transaction
1760 * because we can't use it for xfs_attr_inactive().
1761 */
1762 if (ip->i_d.di_anextents > 0) {
1763 ASSERT(ip->i_d.di_forkoff != 0);
1764
1765 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1766 if (error)
1767 goto out_unlock;
1768
1769 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1770
1771 error = xfs_attr_inactive(ip);
1772 if (error)
1773 goto out;
1774
1775 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
Jie Liu3d3c8b52013-08-12 20:49:59 +10001776 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10001777 if (error) {
1778 xfs_trans_cancel(tp, 0);
1779 goto out;
1780 }
1781
1782 xfs_ilock(ip, XFS_ILOCK_EXCL);
1783 xfs_trans_ijoin(tp, ip, 0);
1784 }
1785
1786 if (ip->i_afp)
1787 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1788
1789 ASSERT(ip->i_d.di_anextents == 0);
1790
1791 /*
1792 * Free the inode.
1793 */
1794 xfs_bmap_init(&free_list, &first_block);
1795 error = xfs_ifree(tp, ip, &free_list);
1796 if (error) {
1797 /*
1798 * If we fail to free the inode, shut down. The cancel
1799 * might do that, we need to make sure. Otherwise the
1800 * inode might be lost for a long time or forever.
1801 */
1802 if (!XFS_FORCED_SHUTDOWN(mp)) {
1803 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1804 __func__, error);
1805 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1806 }
1807 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1808 } else {
1809 /*
1810 * Credit the quota account(s). The inode is gone.
1811 */
1812 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1813
1814 /*
1815 * Just ignore errors at this point. There is nothing we can
1816 * do except to try to keep going. Make sure it's not a silent
1817 * error.
1818 */
1819 error = xfs_bmap_finish(&tp, &free_list, &committed);
1820 if (error)
1821 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1822 __func__, error);
1823 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1824 if (error)
1825 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1826 __func__, error);
1827 }
1828
1829 /*
1830 * Release the dquots held by inode, if any.
1831 */
1832 xfs_qm_dqdetach(ip);
1833out_unlock:
1834 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1835out:
1836 return VN_INACTIVE_CACHE;
1837out_cancel:
1838 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1839 goto out_unlock;
1840}
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842/*
1843 * This is called when the inode's link count goes to 0.
1844 * We place the on-disk inode on a list in the AGI. It
1845 * will be pulled from this list when the inode is freed.
1846 */
1847int
1848xfs_iunlink(
1849 xfs_trans_t *tp,
1850 xfs_inode_t *ip)
1851{
1852 xfs_mount_t *mp;
1853 xfs_agi_t *agi;
1854 xfs_dinode_t *dip;
1855 xfs_buf_t *agibp;
1856 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 xfs_agino_t agino;
1858 short bucket_index;
1859 int offset;
1860 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
1862 ASSERT(ip->i_d.di_nlink == 0);
1863 ASSERT(ip->i_d.di_mode != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 mp = tp->t_mountp;
1866
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 /*
1868 * Get the agi buffer first. It ensures lock ordering
1869 * on the list.
1870 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001871 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001872 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 /*
1877 * Get the index into the agi hash table for the
1878 * list this inode will go on.
1879 */
1880 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1881 ASSERT(agino != 0);
1882 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1883 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001884 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001886 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 /*
1888 * There is already another inode in the bucket we need
1889 * to add ourselves to. Add us at the front of the list.
1890 * Here we put the head pointer into our next pointer,
1891 * and then we fall through to point the head at us.
1892 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001893 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1894 0, 0);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001895 if (error)
1896 return error;
1897
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001898 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001900 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10001902
1903 /* need to recalc the inode CRC if appropriate */
1904 xfs_dinode_calc_crc(mp, dip);
1905
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 xfs_trans_inode_buf(tp, ibp);
1907 xfs_trans_log_buf(tp, ibp, offset,
1908 (offset + sizeof(xfs_agino_t) - 1));
1909 xfs_inobp_check(mp, ibp);
1910 }
1911
1912 /*
1913 * Point the bucket head pointer at the inode being inserted.
1914 */
1915 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001916 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 offset = offsetof(xfs_agi_t, agi_unlinked) +
1918 (sizeof(xfs_agino_t) * bucket_index);
1919 xfs_trans_log_buf(tp, agibp, offset,
1920 (offset + sizeof(xfs_agino_t) - 1));
1921 return 0;
1922}
1923
1924/*
1925 * Pull the on-disk inode from the AGI unlinked list.
1926 */
1927STATIC int
1928xfs_iunlink_remove(
1929 xfs_trans_t *tp,
1930 xfs_inode_t *ip)
1931{
1932 xfs_ino_t next_ino;
1933 xfs_mount_t *mp;
1934 xfs_agi_t *agi;
1935 xfs_dinode_t *dip;
1936 xfs_buf_t *agibp;
1937 xfs_buf_t *ibp;
1938 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 xfs_agino_t agino;
1940 xfs_agino_t next_agino;
1941 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001942 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001944 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
1950 /*
1951 * Get the agi buffer first. It ensures lock ordering
1952 * on the list.
1953 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001954 error = xfs_read_agi(mp, tp, agno, &agibp);
1955 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001957
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 /*
1961 * Get the index into the agi hash table for the
1962 * list this inode will go on.
1963 */
1964 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1965 ASSERT(agino != 0);
1966 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001967 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 ASSERT(agi->agi_unlinked[bucket_index]);
1969
Christoph Hellwig16259e72005-11-02 15:11:25 +11001970 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04001972 * We're at the head of the list. Get the inode's on-disk
1973 * buffer to see if there is anyone after us on the list.
1974 * Only modify our next pointer if it is not already NULLAGINO.
1975 * This saves us the overhead of dealing with the buffer when
1976 * there is no need to change it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04001978 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1979 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04001981 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11001982 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 return error;
1984 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001985 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 ASSERT(next_agino != 0);
1987 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001988 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001989 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10001991
1992 /* need to recalc the inode CRC if appropriate */
1993 xfs_dinode_calc_crc(mp, dip);
1994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 xfs_trans_inode_buf(tp, ibp);
1996 xfs_trans_log_buf(tp, ibp, offset,
1997 (offset + sizeof(xfs_agino_t) - 1));
1998 xfs_inobp_check(mp, ibp);
1999 } else {
2000 xfs_trans_brelse(tp, ibp);
2001 }
2002 /*
2003 * Point the bucket head pointer at the next inode.
2004 */
2005 ASSERT(next_agino != 0);
2006 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002007 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 offset = offsetof(xfs_agi_t, agi_unlinked) +
2009 (sizeof(xfs_agino_t) * bucket_index);
2010 xfs_trans_log_buf(tp, agibp, offset,
2011 (offset + sizeof(xfs_agino_t) - 1));
2012 } else {
2013 /*
2014 * We need to search the list for the inode being freed.
2015 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002016 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 last_ibp = NULL;
2018 while (next_agino != agino) {
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002019 struct xfs_imap imap;
2020
2021 if (last_ibp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 xfs_trans_brelse(tp, last_ibp);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002023
2024 imap.im_blkno = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002026
2027 error = xfs_imap(mp, tp, next_ino, &imap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11002029 xfs_warn(mp,
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002030 "%s: xfs_imap returned error %d.",
2031 __func__, error);
2032 return error;
2033 }
2034
2035 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2036 &last_ibp, 0, 0);
2037 if (error) {
2038 xfs_warn(mp,
2039 "%s: xfs_imap_to_bp returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002040 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 return error;
2042 }
Christoph Hellwig129dbc92012-07-03 12:21:51 -04002043
2044 last_offset = imap.im_boffset;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002045 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 ASSERT(next_agino != NULLAGINO);
2047 ASSERT(next_agino != 0);
2048 }
Christoph Hellwig475ee412012-07-03 12:21:22 -04002049
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 /*
Christoph Hellwig475ee412012-07-03 12:21:22 -04002051 * Now last_ibp points to the buffer previous to us on the
2052 * unlinked list. Pull us from the list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04002054 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2055 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (error) {
Christoph Hellwig475ee412012-07-03 12:21:22 -04002057 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
Dave Chinner0b932cc2011-03-07 10:08:35 +11002058 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 return error;
2060 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002061 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 ASSERT(next_agino != 0);
2063 ASSERT(next_agino != agino);
2064 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002065 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002066 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002068
2069 /* need to recalc the inode CRC if appropriate */
2070 xfs_dinode_calc_crc(mp, dip);
2071
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 xfs_trans_inode_buf(tp, ibp);
2073 xfs_trans_log_buf(tp, ibp, offset,
2074 (offset + sizeof(xfs_agino_t) - 1));
2075 xfs_inobp_check(mp, ibp);
2076 } else {
2077 xfs_trans_brelse(tp, ibp);
2078 }
2079 /*
2080 * Point the previous inode on the list to the next inode.
2081 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002082 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 ASSERT(next_agino != 0);
2084 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
Dave Chinner0a32c262013-06-05 12:09:08 +10002085
2086 /* need to recalc the inode CRC if appropriate */
2087 xfs_dinode_calc_crc(mp, last_dip);
2088
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 xfs_trans_inode_buf(tp, last_ibp);
2090 xfs_trans_log_buf(tp, last_ibp, offset,
2091 (offset + sizeof(xfs_agino_t) - 1));
2092 xfs_inobp_check(mp, last_ibp);
2093 }
2094 return 0;
2095}
2096
Dave Chinner5b3eed72010-08-24 11:42:41 +10002097/*
Zhi Yong Wu0b8182d2013-08-12 03:14:59 +00002098 * A big issue when freeing the inode cluster is that we _cannot_ skip any
Dave Chinner5b3eed72010-08-24 11:42:41 +10002099 * inodes that are in memory - they all must be marked stale and attached to
2100 * the cluster buffer.
2101 */
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002102STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103xfs_ifree_cluster(
2104 xfs_inode_t *free_ip,
2105 xfs_trans_t *tp,
2106 xfs_ino_t inum)
2107{
2108 xfs_mount_t *mp = free_ip->i_mount;
2109 int blks_per_cluster;
2110 int nbufs;
2111 int ninodes;
Dave Chinner5b257b42010-06-03 16:22:29 +10002112 int i, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 xfs_daddr_t blkno;
2114 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10002115 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 xfs_inode_log_item_t *iip;
2117 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00002118 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Dave Chinner5017e972010-01-11 11:47:40 +00002120 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2122 blks_per_cluster = 1;
2123 ninodes = mp->m_sb.sb_inopblock;
2124 nbufs = XFS_IALLOC_BLOCKS(mp);
2125 } else {
2126 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2127 mp->m_sb.sb_blocksize;
2128 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2129 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2130 }
2131
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 for (j = 0; j < nbufs; j++, inum += ninodes) {
2133 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2134 XFS_INO_TO_AGBNO(mp, inum));
2135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10002137 * We obtain and lock the backing buffer first in the process
2138 * here, as we have to ensure that any dirty inode that we
2139 * can't get the flush lock on is attached to the buffer.
2140 * If we scan the in-memory inodes first, then buffer IO can
2141 * complete before we get a lock on it, and hence we may fail
2142 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002144 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Dave Chinnerb6aff292012-11-02 11:38:42 +11002145 mp->m_bsize * blks_per_cluster,
2146 XBF_UNMAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002148 if (!bp)
2149 return ENOMEM;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002150
2151 /*
2152 * This buffer may not have been correctly initialised as we
2153 * didn't read it from disk. That's not important because we are
2154 * only using to mark the buffer as stale in the log, and to
2155 * attach stale cached inodes on it. That means it will never be
2156 * dispatched for IO. If it is, we want to know about it, and we
2157 * want it to fail. We can acheive this by adding a write
2158 * verifier to the buffer.
2159 */
Dave Chinner1813dd62012-11-14 17:54:40 +11002160 bp->b_ops = &xfs_inode_buf_ops;
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002161
Dave Chinner5b257b42010-06-03 16:22:29 +10002162 /*
2163 * Walk the inodes already attached to the buffer and mark them
2164 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10002165 * in-memory inode walk can't lock them. By marking them all
2166 * stale first, we will not attempt to lock them in the loop
2167 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10002168 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +02002169 lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 while (lip) {
2171 if (lip->li_type == XFS_LI_INODE) {
2172 iip = (xfs_inode_log_item_t *)lip;
2173 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002174 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002175 xfs_trans_ail_copy_lsn(mp->m_ail,
2176 &iip->ili_flush_lsn,
2177 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002178 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 }
2180 lip = lip->li_bio_list;
2181 }
2182
Dave Chinner5b3eed72010-08-24 11:42:41 +10002183
Dave Chinner5b257b42010-06-03 16:22:29 +10002184 /*
2185 * For each inode in memory attempt to add it to the inode
2186 * buffer and set it up for being staled on buffer IO
2187 * completion. This is safe as we've locked out tail pushing
2188 * and flushing by locking the buffer.
2189 *
2190 * We have already marked every inode that was part of a
2191 * transaction stale above, which means there is no point in
2192 * even trying to lock them.
2193 */
2194 for (i = 0; i < ninodes; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10002195retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002196 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002197 ip = radix_tree_lookup(&pag->pag_ici_root,
2198 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002200 /* Inode not in memory, nothing to do */
2201 if (!ip) {
2202 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002203 continue;
2204 }
2205
Dave Chinner5b3eed72010-08-24 11:42:41 +10002206 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002207 * because this is an RCU protected lookup, we could
2208 * find a recently freed or even reallocated inode
2209 * during the lookup. We need to check under the
2210 * i_flags_lock for a valid inode here. Skip it if it
2211 * is not valid, the wrong inode or stale.
2212 */
2213 spin_lock(&ip->i_flags_lock);
2214 if (ip->i_ino != inum + i ||
2215 __xfs_iflags_test(ip, XFS_ISTALE)) {
2216 spin_unlock(&ip->i_flags_lock);
2217 rcu_read_unlock();
2218 continue;
2219 }
2220 spin_unlock(&ip->i_flags_lock);
2221
2222 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10002223 * Don't try to lock/unlock the current inode, but we
2224 * _cannot_ skip the other inodes that we did not find
2225 * in the list attached to the buffer and are not
2226 * already marked stale. If we can't lock it, back off
2227 * and retry.
2228 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002229 if (ip != free_ip &&
2230 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002231 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10002232 delay(1);
2233 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10002234 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002235 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002236
Dave Chinner5b3eed72010-08-24 11:42:41 +10002237 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10002238 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10002239
Dave Chinner5b3eed72010-08-24 11:42:41 +10002240 /*
2241 * we don't need to attach clean inodes or those only
2242 * with unlogged changes (which we throw away, anyway).
2243 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002244 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10002245 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10002246 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 xfs_ifunlock(ip);
2248 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2249 continue;
2250 }
2251
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00002252 iip->ili_last_fields = iip->ili_fields;
2253 iip->ili_fields = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002255 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2256 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002258 xfs_buf_attach_iodone(bp, xfs_istale_done,
2259 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10002260
2261 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 }
2264
Dave Chinner5b3eed72010-08-24 11:42:41 +10002265 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 xfs_trans_binval(tp, bp);
2267 }
2268
Dave Chinner5017e972010-01-11 11:47:40 +00002269 xfs_perag_put(pag);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002270 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271}
2272
2273/*
2274 * This is called to return an inode to the inode free list.
2275 * The inode should already be truncated to 0 length and have
2276 * no pages associated with it. This routine also assumes that
2277 * the inode is already a part of the transaction.
2278 *
2279 * The on-disk copy of the inode will have been added to the list
2280 * of unlinked inodes in the AGI. We need to remove the inode from
2281 * that list atomically with respect to freeing it here.
2282 */
2283int
2284xfs_ifree(
2285 xfs_trans_t *tp,
2286 xfs_inode_t *ip,
2287 xfs_bmap_free_t *flist)
2288{
2289 int error;
2290 int delete;
2291 xfs_ino_t first_ino;
2292
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002293 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 ASSERT(ip->i_d.di_nlink == 0);
2295 ASSERT(ip->i_d.di_nextents == 0);
2296 ASSERT(ip->i_d.di_anextents == 0);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +00002297 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 ASSERT(ip->i_d.di_nblocks == 0);
2299
2300 /*
2301 * Pull the on-disk inode from the AGI unlinked list.
2302 */
2303 error = xfs_iunlink_remove(tp, ip);
Dave Chinner1baaed82013-06-27 16:04:50 +10002304 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
2307 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
Dave Chinner1baaed82013-06-27 16:04:50 +10002308 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 return error;
Dave Chinner1baaed82013-06-27 16:04:50 +10002310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 ip->i_d.di_mode = 0; /* mark incore inode as free */
2312 ip->i_d.di_flags = 0;
2313 ip->i_d.di_dmevmask = 0;
2314 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2316 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2317 /*
2318 * Bump the generation count so no one will be confused
2319 * by reincarnations of this inode.
2320 */
2321 ip->i_d.di_gen++;
2322 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2323
Dave Chinner1baaed82013-06-27 16:04:50 +10002324 if (delete)
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002325 error = xfs_ifree_cluster(ip, tp, first_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +00002327 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328}
2329
2330/*
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002331 * This is called to unpin an inode. The caller must have the inode locked
2332 * in at least shared mode so that the buffer cannot be subsequently pinned
2333 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002335static void
Christoph Hellwigf392e632011-12-18 20:00:10 +00002336xfs_iunpin(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002337 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002338{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002339 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002340
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002341 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2342
David Chinnera3f74ff2008-03-06 13:43:42 +11002343 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002344 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002345
David Chinnera3f74ff2008-03-06 13:43:42 +11002346}
2347
Christoph Hellwigf392e632011-12-18 20:00:10 +00002348static void
2349__xfs_iunpin_wait(
2350 struct xfs_inode *ip)
2351{
2352 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2353 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2354
2355 xfs_iunpin(ip);
2356
2357 do {
2358 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2359 if (xfs_ipincount(ip))
2360 io_schedule();
2361 } while (xfs_ipincount(ip));
2362 finish_wait(wq, &wait.wait);
2363}
2364
Dave Chinner777df5a2010-02-06 12:37:26 +11002365void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366xfs_iunpin_wait(
Christoph Hellwig60ec67832010-02-17 19:43:56 +00002367 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368{
Christoph Hellwigf392e632011-12-18 20:00:10 +00002369 if (xfs_ipincount(ip))
2370 __xfs_iunpin_wait(ip);
David Chinnera3f74ff2008-03-06 13:43:42 +11002371}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002373int
2374xfs_remove(
2375 xfs_inode_t *dp,
2376 struct xfs_name *name,
2377 xfs_inode_t *ip)
2378{
2379 xfs_mount_t *mp = dp->i_mount;
2380 xfs_trans_t *tp = NULL;
2381 int is_dir = S_ISDIR(ip->i_d.di_mode);
2382 int error = 0;
2383 xfs_bmap_free_t free_list;
2384 xfs_fsblock_t first_block;
2385 int cancel_flags;
2386 int committed;
2387 int link_zero;
2388 uint resblks;
2389 uint log_count;
2390
2391 trace_xfs_remove(dp, name);
2392
2393 if (XFS_FORCED_SHUTDOWN(mp))
2394 return XFS_ERROR(EIO);
2395
2396 error = xfs_qm_dqattach(dp, 0);
2397 if (error)
2398 goto std_return;
2399
2400 error = xfs_qm_dqattach(ip, 0);
2401 if (error)
2402 goto std_return;
2403
2404 if (is_dir) {
2405 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2406 log_count = XFS_DEFAULT_LOG_COUNT;
2407 } else {
2408 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2409 log_count = XFS_REMOVE_LOG_COUNT;
2410 }
2411 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2412
2413 /*
2414 * We try to get the real space reservation first,
2415 * allowing for directory btree deletion(s) implying
2416 * possible bmap insert(s). If we can't get the space
2417 * reservation then we use 0 instead, and avoid the bmap
2418 * btree insert(s) in the directory code by, if the bmap
2419 * insert tries to happen, instead trimming the LAST
2420 * block from the directory.
2421 */
2422 resblks = XFS_REMOVE_SPACE_RES(mp);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002423 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002424 if (error == ENOSPC) {
2425 resblks = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002426 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
Dave Chinnerc24b5df2013-08-12 20:49:45 +10002427 }
2428 if (error) {
2429 ASSERT(error != ENOSPC);
2430 cancel_flags = 0;
2431 goto out_trans_cancel;
2432 }
2433
2434 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2435
2436 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2437 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2438
2439 /*
2440 * If we're removing a directory perform some additional validation.
2441 */
2442 if (is_dir) {
2443 ASSERT(ip->i_d.di_nlink >= 2);
2444 if (ip->i_d.di_nlink != 2) {
2445 error = XFS_ERROR(ENOTEMPTY);
2446 goto out_trans_cancel;
2447 }
2448 if (!xfs_dir_isempty(ip)) {
2449 error = XFS_ERROR(ENOTEMPTY);
2450 goto out_trans_cancel;
2451 }
2452 }
2453
2454 xfs_bmap_init(&free_list, &first_block);
2455 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2456 &first_block, &free_list, resblks);
2457 if (error) {
2458 ASSERT(error != ENOENT);
2459 goto out_bmap_cancel;
2460 }
2461 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2462
2463 if (is_dir) {
2464 /*
2465 * Drop the link from ip's "..".
2466 */
2467 error = xfs_droplink(tp, dp);
2468 if (error)
2469 goto out_bmap_cancel;
2470
2471 /*
2472 * Drop the "." link from ip to self.
2473 */
2474 error = xfs_droplink(tp, ip);
2475 if (error)
2476 goto out_bmap_cancel;
2477 } else {
2478 /*
2479 * When removing a non-directory we need to log the parent
2480 * inode here. For a directory this is done implicitly
2481 * by the xfs_droplink call for the ".." entry.
2482 */
2483 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2484 }
2485
2486 /*
2487 * Drop the link from dp to ip.
2488 */
2489 error = xfs_droplink(tp, ip);
2490 if (error)
2491 goto out_bmap_cancel;
2492
2493 /*
2494 * Determine if this is the last link while
2495 * we are in the transaction.
2496 */
2497 link_zero = (ip->i_d.di_nlink == 0);
2498
2499 /*
2500 * If this is a synchronous mount, make sure that the
2501 * remove transaction goes to disk before returning to
2502 * the user.
2503 */
2504 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2505 xfs_trans_set_sync(tp);
2506
2507 error = xfs_bmap_finish(&tp, &free_list, &committed);
2508 if (error)
2509 goto out_bmap_cancel;
2510
2511 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2512 if (error)
2513 goto std_return;
2514
2515 /*
2516 * If we are using filestreams, kill the stream association.
2517 * If the file is still open it may get a new one but that
2518 * will get killed on last close in xfs_close() so we don't
2519 * have to worry about that.
2520 */
2521 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2522 xfs_filestream_deassociate(ip);
2523
2524 return 0;
2525
2526 out_bmap_cancel:
2527 xfs_bmap_cancel(&free_list);
2528 cancel_flags |= XFS_TRANS_ABORT;
2529 out_trans_cancel:
2530 xfs_trans_cancel(tp, cancel_flags);
2531 std_return:
2532 return error;
2533}
2534
Dave Chinnerf6bba202013-08-12 20:49:46 +10002535/*
2536 * Enter all inodes for a rename transaction into a sorted array.
2537 */
2538STATIC void
2539xfs_sort_for_rename(
2540 xfs_inode_t *dp1, /* in: old (source) directory inode */
2541 xfs_inode_t *dp2, /* in: new (target) directory inode */
2542 xfs_inode_t *ip1, /* in: inode of old entry */
2543 xfs_inode_t *ip2, /* in: inode of new entry, if it
2544 already exists, NULL otherwise. */
2545 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2546 int *num_inodes) /* out: number of inodes in array */
2547{
2548 xfs_inode_t *temp;
2549 int i, j;
2550
2551 /*
2552 * i_tab contains a list of pointers to inodes. We initialize
2553 * the table here & we'll sort it. We will then use it to
2554 * order the acquisition of the inode locks.
2555 *
2556 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2557 */
2558 i_tab[0] = dp1;
2559 i_tab[1] = dp2;
2560 i_tab[2] = ip1;
2561 if (ip2) {
2562 *num_inodes = 4;
2563 i_tab[3] = ip2;
2564 } else {
2565 *num_inodes = 3;
2566 i_tab[3] = NULL;
2567 }
2568
2569 /*
2570 * Sort the elements via bubble sort. (Remember, there are at
2571 * most 4 elements to sort, so this is adequate.)
2572 */
2573 for (i = 0; i < *num_inodes; i++) {
2574 for (j = 1; j < *num_inodes; j++) {
2575 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2576 temp = i_tab[j];
2577 i_tab[j] = i_tab[j-1];
2578 i_tab[j-1] = temp;
2579 }
2580 }
2581 }
2582}
2583
2584/*
2585 * xfs_rename
2586 */
2587int
2588xfs_rename(
2589 xfs_inode_t *src_dp,
2590 struct xfs_name *src_name,
2591 xfs_inode_t *src_ip,
2592 xfs_inode_t *target_dp,
2593 struct xfs_name *target_name,
2594 xfs_inode_t *target_ip)
2595{
2596 xfs_trans_t *tp = NULL;
2597 xfs_mount_t *mp = src_dp->i_mount;
2598 int new_parent; /* moving to a new dir */
2599 int src_is_directory; /* src_name is a directory */
2600 int error;
2601 xfs_bmap_free_t free_list;
2602 xfs_fsblock_t first_block;
2603 int cancel_flags;
2604 int committed;
2605 xfs_inode_t *inodes[4];
2606 int spaceres;
2607 int num_inodes;
2608
2609 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2610
2611 new_parent = (src_dp != target_dp);
2612 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2613
2614 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2615 inodes, &num_inodes);
2616
2617 xfs_bmap_init(&free_list, &first_block);
2618 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2619 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2620 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
Jie Liu3d3c8b52013-08-12 20:49:59 +10002621 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002622 if (error == ENOSPC) {
2623 spaceres = 0;
Jie Liu3d3c8b52013-08-12 20:49:59 +10002624 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
Dave Chinnerf6bba202013-08-12 20:49:46 +10002625 }
2626 if (error) {
2627 xfs_trans_cancel(tp, 0);
2628 goto std_return;
2629 }
2630
2631 /*
2632 * Attach the dquots to the inodes
2633 */
2634 error = xfs_qm_vop_rename_dqattach(inodes);
2635 if (error) {
2636 xfs_trans_cancel(tp, cancel_flags);
2637 goto std_return;
2638 }
2639
2640 /*
2641 * Lock all the participating inodes. Depending upon whether
2642 * the target_name exists in the target directory, and
2643 * whether the target directory is the same as the source
2644 * directory, we can lock from 2 to 4 inodes.
2645 */
2646 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2647
2648 /*
2649 * Join all the inodes to the transaction. From this point on,
2650 * we can rely on either trans_commit or trans_cancel to unlock
2651 * them.
2652 */
2653 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2654 if (new_parent)
2655 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2656 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2657 if (target_ip)
2658 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2659
2660 /*
2661 * If we are using project inheritance, we only allow renames
2662 * into our tree when the project IDs are the same; else the
2663 * tree quota mechanism would be circumvented.
2664 */
2665 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2666 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2667 error = XFS_ERROR(EXDEV);
2668 goto error_return;
2669 }
2670
2671 /*
2672 * Set up the target.
2673 */
2674 if (target_ip == NULL) {
2675 /*
2676 * If there's no space reservation, check the entry will
2677 * fit before actually inserting it.
2678 */
2679 error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2680 if (error)
2681 goto error_return;
2682 /*
2683 * If target does not exist and the rename crosses
2684 * directories, adjust the target directory link count
2685 * to account for the ".." reference from the new entry.
2686 */
2687 error = xfs_dir_createname(tp, target_dp, target_name,
2688 src_ip->i_ino, &first_block,
2689 &free_list, spaceres);
2690 if (error == ENOSPC)
2691 goto error_return;
2692 if (error)
2693 goto abort_return;
2694
2695 xfs_trans_ichgtime(tp, target_dp,
2696 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2697
2698 if (new_parent && src_is_directory) {
2699 error = xfs_bumplink(tp, target_dp);
2700 if (error)
2701 goto abort_return;
2702 }
2703 } else { /* target_ip != NULL */
2704 /*
2705 * If target exists and it's a directory, check that both
2706 * target and source are directories and that target can be
2707 * destroyed, or that neither is a directory.
2708 */
2709 if (S_ISDIR(target_ip->i_d.di_mode)) {
2710 /*
2711 * Make sure target dir is empty.
2712 */
2713 if (!(xfs_dir_isempty(target_ip)) ||
2714 (target_ip->i_d.di_nlink > 2)) {
2715 error = XFS_ERROR(EEXIST);
2716 goto error_return;
2717 }
2718 }
2719
2720 /*
2721 * Link the source inode under the target name.
2722 * If the source inode is a directory and we are moving
2723 * it across directories, its ".." entry will be
2724 * inconsistent until we replace that down below.
2725 *
2726 * In case there is already an entry with the same
2727 * name at the destination directory, remove it first.
2728 */
2729 error = xfs_dir_replace(tp, target_dp, target_name,
2730 src_ip->i_ino,
2731 &first_block, &free_list, spaceres);
2732 if (error)
2733 goto abort_return;
2734
2735 xfs_trans_ichgtime(tp, target_dp,
2736 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2737
2738 /*
2739 * Decrement the link count on the target since the target
2740 * dir no longer points to it.
2741 */
2742 error = xfs_droplink(tp, target_ip);
2743 if (error)
2744 goto abort_return;
2745
2746 if (src_is_directory) {
2747 /*
2748 * Drop the link from the old "." entry.
2749 */
2750 error = xfs_droplink(tp, target_ip);
2751 if (error)
2752 goto abort_return;
2753 }
2754 } /* target_ip != NULL */
2755
2756 /*
2757 * Remove the source.
2758 */
2759 if (new_parent && src_is_directory) {
2760 /*
2761 * Rewrite the ".." entry to point to the new
2762 * directory.
2763 */
2764 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2765 target_dp->i_ino,
2766 &first_block, &free_list, spaceres);
2767 ASSERT(error != EEXIST);
2768 if (error)
2769 goto abort_return;
2770 }
2771
2772 /*
2773 * We always want to hit the ctime on the source inode.
2774 *
2775 * This isn't strictly required by the standards since the source
2776 * inode isn't really being changed, but old unix file systems did
2777 * it and some incremental backup programs won't work without it.
2778 */
2779 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2780 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2781
2782 /*
2783 * Adjust the link count on src_dp. This is necessary when
2784 * renaming a directory, either within one parent when
2785 * the target existed, or across two parent directories.
2786 */
2787 if (src_is_directory && (new_parent || target_ip != NULL)) {
2788
2789 /*
2790 * Decrement link count on src_directory since the
2791 * entry that's moved no longer points to it.
2792 */
2793 error = xfs_droplink(tp, src_dp);
2794 if (error)
2795 goto abort_return;
2796 }
2797
2798 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2799 &first_block, &free_list, spaceres);
2800 if (error)
2801 goto abort_return;
2802
2803 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2804 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2805 if (new_parent)
2806 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2807
2808 /*
2809 * If this is a synchronous mount, make sure that the
2810 * rename transaction goes to disk before returning to
2811 * the user.
2812 */
2813 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2814 xfs_trans_set_sync(tp);
2815 }
2816
2817 error = xfs_bmap_finish(&tp, &free_list, &committed);
2818 if (error) {
2819 xfs_bmap_cancel(&free_list);
2820 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2821 XFS_TRANS_ABORT));
2822 goto std_return;
2823 }
2824
2825 /*
2826 * trans_commit will unlock src_ip, target_ip & decrement
2827 * the vnode references.
2828 */
2829 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2830
2831 abort_return:
2832 cancel_flags |= XFS_TRANS_ABORT;
2833 error_return:
2834 xfs_bmap_cancel(&free_list);
2835 xfs_trans_cancel(tp, cancel_flags);
2836 std_return:
2837 return error;
2838}
2839
David Chinnerbad55842008-03-06 13:43:49 +11002840STATIC int
2841xfs_iflush_cluster(
2842 xfs_inode_t *ip,
2843 xfs_buf_t *bp)
2844{
2845 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00002846 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11002847 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10002848 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11002849 int ilist_size;
2850 xfs_inode_t **ilist;
2851 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11002852 int nr_found;
2853 int clcount = 0;
2854 int bufwasdelwri;
2855 int i;
2856
Dave Chinner5017e972010-01-11 11:47:40 +00002857 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11002858
David Chinnerc8f5f122008-05-20 11:30:15 +10002859 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2860 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10002861 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11002862 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00002863 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11002864
2865 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2866 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002867 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11002868 /* really need a gang lookup range call here */
2869 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10002870 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11002871 if (nr_found == 0)
2872 goto out_free;
2873
2874 for (i = 0; i < nr_found; i++) {
2875 iq = ilist[i];
2876 if (iq == ip)
2877 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002878
2879 /*
2880 * because this is an RCU protected lookup, we could find a
2881 * recently freed or even reallocated inode during the lookup.
2882 * We need to check under the i_flags_lock for a valid inode
2883 * here. Skip it if it is not valid or the wrong inode.
2884 */
2885 spin_lock(&ip->i_flags_lock);
2886 if (!ip->i_ino ||
2887 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2888 spin_unlock(&ip->i_flags_lock);
2889 continue;
2890 }
2891 spin_unlock(&ip->i_flags_lock);
2892
David Chinnerbad55842008-03-06 13:43:49 +11002893 /*
2894 * Do an un-protected check to see if the inode is dirty and
2895 * is a candidate for flushing. These checks will be repeated
2896 * later after the appropriate locks are acquired.
2897 */
David Chinner33540402008-03-06 13:43:59 +11002898 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11002899 continue;
David Chinnerbad55842008-03-06 13:43:49 +11002900
2901 /*
2902 * Try to get locks. If any are unavailable or it is pinned,
2903 * then this inode cannot be flushed and is skipped.
2904 */
2905
2906 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2907 continue;
2908 if (!xfs_iflock_nowait(iq)) {
2909 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2910 continue;
2911 }
2912 if (xfs_ipincount(iq)) {
2913 xfs_ifunlock(iq);
2914 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2915 continue;
2916 }
2917
2918 /*
2919 * arriving here means that this inode can be flushed. First
2920 * re-check that it's dirty before flushing.
2921 */
David Chinner33540402008-03-06 13:43:59 +11002922 if (!xfs_inode_clean(iq)) {
2923 int error;
David Chinnerbad55842008-03-06 13:43:49 +11002924 error = xfs_iflush_int(iq, bp);
2925 if (error) {
2926 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2927 goto cluster_corrupt_out;
2928 }
2929 clcount++;
2930 } else {
2931 xfs_ifunlock(iq);
2932 }
2933 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2934 }
2935
2936 if (clcount) {
2937 XFS_STATS_INC(xs_icluster_flushcnt);
2938 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2939 }
2940
2941out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002942 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002943 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002944out_put:
2945 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002946 return 0;
2947
2948
2949cluster_corrupt_out:
2950 /*
2951 * Corruption detected in the clustering loop. Invalidate the
2952 * inode buffer and shut down the filesystem.
2953 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002954 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11002955 /*
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002956 * Clean up the buffer. If it was delwri, just release it --
David Chinnerbad55842008-03-06 13:43:49 +11002957 * brelse can handle it with no problems. If not, shut down the
2958 * filesystem before releasing the buffer.
2959 */
Christoph Hellwig43ff2122012-04-23 15:58:39 +10002960 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
David Chinnerbad55842008-03-06 13:43:49 +11002961 if (bufwasdelwri)
2962 xfs_buf_relse(bp);
2963
2964 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2965
2966 if (!bufwasdelwri) {
2967 /*
2968 * Just like incore_relse: if we have b_iodone functions,
2969 * mark the buffer as an error and call them. Otherwise
2970 * mark it as stale and brelse.
2971 */
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002972 if (bp->b_iodone) {
David Chinnerbad55842008-03-06 13:43:49 +11002973 XFS_BUF_UNDONE(bp);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002974 xfs_buf_stale(bp);
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00002975 xfs_buf_ioerror(bp, EIO);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002976 xfs_buf_ioend(bp, 0);
David Chinnerbad55842008-03-06 13:43:49 +11002977 } else {
Christoph Hellwigc867cb62011-10-10 16:52:46 +00002978 xfs_buf_stale(bp);
David Chinnerbad55842008-03-06 13:43:49 +11002979 xfs_buf_relse(bp);
2980 }
2981 }
2982
2983 /*
2984 * Unlocks the flush lock
2985 */
Dave Chinner04913fd2012-04-23 15:58:41 +10002986 xfs_iflush_abort(iq, false);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002987 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002988 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002989 return XFS_ERROR(EFSCORRUPTED);
2990}
2991
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992/*
Christoph Hellwig4c468192012-04-23 15:58:36 +10002993 * Flush dirty inode metadata into the backing buffer.
2994 *
2995 * The caller must have the inode lock and the inode flush lock held. The
2996 * inode lock will still be held upon return to the caller, and the inode
2997 * flush lock will be released after the inode has reached the disk.
2998 *
2999 * The caller must write out the buffer returned in *bpp and release it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 */
3001int
3002xfs_iflush(
Christoph Hellwig4c468192012-04-23 15:58:36 +10003003 struct xfs_inode *ip,
3004 struct xfs_buf **bpp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005{
Christoph Hellwig4c468192012-04-23 15:58:36 +10003006 struct xfs_mount *mp = ip->i_mount;
3007 struct xfs_buf *bp;
3008 struct xfs_dinode *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010
3011 XFS_STATS_INC(xs_iflush_count);
3012
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003013 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003014 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003016 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Christoph Hellwig4c468192012-04-23 15:58:36 +10003018 *bpp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020 xfs_iunpin_wait(ip);
3021
3022 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00003023 * For stale inodes we cannot rely on the backing buffer remaining
3024 * stale in cache for the remaining life of the stale inode and so
Christoph Hellwig475ee412012-07-03 12:21:22 -04003025 * xfs_imap_to_bp() below may give us a buffer that no longer contains
Dave Chinner4b6a4682010-01-11 11:45:21 +00003026 * inodes below. We have to check this after ensuring the inode is
3027 * unpinned so that it is safe to reclaim the stale inode after the
3028 * flush call.
3029 */
3030 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3031 xfs_ifunlock(ip);
3032 return 0;
3033 }
3034
3035 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 * This may have been unpinned because the filesystem is shutting
3037 * down forcibly. If that's the case we must not write this inode
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003038 * to disk, because the log record didn't make it to disk.
3039 *
3040 * We also have to remove the log item from the AIL in this case,
3041 * as we wait for an empty AIL as part of the unmount process.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 */
3043 if (XFS_FORCED_SHUTDOWN(mp)) {
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003044 error = XFS_ERROR(EIO);
3045 goto abort_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 }
3047
3048 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003049 * Get the buffer containing the on-disk inode.
3050 */
Christoph Hellwig475ee412012-07-03 12:21:22 -04003051 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3052 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003053 if (error || !bp) {
3054 xfs_ifunlock(ip);
3055 return error;
3056 }
3057
3058 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 * First flush out the inode that xfs_iflush was called with.
3060 */
3061 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11003062 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
3065 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003066 * If the buffer is pinned then push on the log now so we won't
3067 * get stuck waiting in the write for too long.
3068 */
Chandra Seetharaman811e64c2011-07-22 23:40:27 +00003069 if (xfs_buf_ispinned(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003070 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11003071
3072 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 * inode clustering:
3074 * see if other inodes can be gathered into this write
3075 */
David Chinnerbad55842008-03-06 13:43:49 +11003076 error = xfs_iflush_cluster(ip, bp);
3077 if (error)
3078 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079
Christoph Hellwig4c468192012-04-23 15:58:36 +10003080 *bpp = bp;
3081 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
3083corrupt_out:
3084 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003085 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086cluster_corrupt_out:
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003087 error = XFS_ERROR(EFSCORRUPTED);
3088abort_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 /*
3090 * Unlocks the flush lock
3091 */
Dave Chinner04913fd2012-04-23 15:58:41 +10003092 xfs_iflush_abort(ip, false);
Christoph Hellwig32ce90a2012-04-23 15:58:32 +10003093 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094}
3095
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096STATIC int
3097xfs_iflush_int(
Christoph Hellwig93848a92013-04-03 16:11:17 +11003098 struct xfs_inode *ip,
3099 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100{
Christoph Hellwig93848a92013-04-03 16:11:17 +11003101 struct xfs_inode_log_item *iip = ip->i_itemp;
3102 struct xfs_dinode *dip;
3103 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003105 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Christoph Hellwig474fce02011-12-18 20:00:09 +00003106 ASSERT(xfs_isiflocked(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
Christoph Hellwig8096b1e2011-12-18 20:00:07 +00003108 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
Christoph Hellwig93848a92013-04-03 16:11:17 +11003109 ASSERT(iip != NULL && iip->ili_fields != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 /* set *dip = inode's place in the buffer */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11003112 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003114 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003116 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3117 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3118 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 goto corrupt_out;
3120 }
3121 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3122 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003123 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3124 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3125 __func__, ip->i_ino, ip, ip->i_d.di_magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 goto corrupt_out;
3127 }
Al Viroabbede12011-07-26 02:31:30 -04003128 if (S_ISREG(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 if (XFS_TEST_ERROR(
3130 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3131 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3132 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003133 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3134 "%s: Bad regular inode %Lu, ptr 0x%p",
3135 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 goto corrupt_out;
3137 }
Al Viroabbede12011-07-26 02:31:30 -04003138 } else if (S_ISDIR(ip->i_d.di_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 if (XFS_TEST_ERROR(
3140 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3141 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3142 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3143 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003144 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3145 "%s: Bad directory inode %Lu, ptr 0x%p",
3146 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 goto corrupt_out;
3148 }
3149 }
3150 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3151 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3152 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003153 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3154 "%s: detected corrupt incore inode %Lu, "
3155 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3156 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11003158 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 goto corrupt_out;
3160 }
3161 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3162 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11003163 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3164 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3165 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 goto corrupt_out;
3167 }
Dave Chinnere60896d2013-07-24 15:47:30 +10003168
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 /*
Dave Chinnere60896d2013-07-24 15:47:30 +10003170 * Inode item log recovery for v1/v2 inodes are dependent on the
3171 * di_flushiter count for correct sequencing. We bump the flush
3172 * iteration count so we can detect flushes which postdate a log record
3173 * during recovery. This is redundant as we now log every change and
3174 * hence this can't happen but we need to still do it to ensure
3175 * backwards compatibility with old kernels that predate logging all
3176 * inode changes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 */
Dave Chinnere60896d2013-07-24 15:47:30 +10003178 if (ip->i_d.di_version < 3)
3179 ip->i_d.di_flushiter++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180
3181 /*
3182 * Copy the dirty parts of the inode into the on-disk
3183 * inode. We always copy out the core of the inode,
3184 * because if the inode is dirty at all the core must
3185 * be.
3186 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003187 xfs_dinode_to_disk(dip, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188
3189 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3190 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3191 ip->i_d.di_flushiter = 0;
3192
3193 /*
3194 * If this is really an old format inode and the superblock version
3195 * has not been updated to support only new format inodes, then
3196 * convert back to the old inode format. If the superblock version
3197 * has been updated, then make the conversion permanent.
3198 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003199 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3200 if (ip->i_d.di_version == 1) {
Eric Sandeen62118702008-03-06 13:44:28 +11003201 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202 /*
3203 * Convert it back.
3204 */
3205 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003206 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 } else {
3208 /*
3209 * The superblock version has already been bumped,
3210 * so just make the conversion to the new inode
3211 * format permanent.
3212 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003213 ip->i_d.di_version = 2;
3214 dip->di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 ip->i_d.di_onlink = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003216 dip->di_onlink = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003218 memset(&(dip->di_pad[0]), 0,
3219 sizeof(dip->di_pad));
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00003220 ASSERT(xfs_get_projid(ip) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221 }
3222 }
3223
David Chinnere4ac9672008-04-10 12:23:58 +10003224 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3225 if (XFS_IFORK_Q(ip))
3226 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 xfs_inobp_check(mp, bp);
3228
3229 /*
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003230 * We've recorded everything logged in the inode, so we'd like to clear
3231 * the ili_fields bits so we don't log and flush things unnecessarily.
3232 * However, we can't stop logging all this information until the data
3233 * we've copied into the disk buffer is written to disk. If we did we
3234 * might overwrite the copy of the inode in the log with all the data
3235 * after re-logging only part of it, and in the face of a crash we
3236 * wouldn't have all the data we need to recover.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003238 * What we do is move the bits to the ili_last_fields field. When
3239 * logging the inode, these bits are moved back to the ili_fields field.
3240 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3241 * know that the information those bits represent is permanently on
3242 * disk. As long as the flush completes before the inode is logged
3243 * again, then both ili_fields and ili_last_fields will be cleared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 *
Christoph Hellwigf5d8d5c2012-02-29 09:53:54 +00003245 * We can play with the ili_fields bits here, because the inode lock
3246 * must be held exclusively in order to set bits there and the flush
3247 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3248 * done routine can tell whether or not to look in the AIL. Also, store
3249 * the current LSN of the inode so that we can tell whether the item has
3250 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3251 * need the AIL lock, because it is a 64 bit value that cannot be read
3252 * atomically.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 */
Christoph Hellwig93848a92013-04-03 16:11:17 +11003254 iip->ili_last_fields = iip->ili_fields;
3255 iip->ili_fields = 0;
3256 iip->ili_logged = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Christoph Hellwig93848a92013-04-03 16:11:17 +11003258 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3259 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260
Christoph Hellwig93848a92013-04-03 16:11:17 +11003261 /*
3262 * Attach the function xfs_iflush_done to the inode's
3263 * buffer. This will remove the inode from the AIL
3264 * and unlock the inode's flush lock when the inode is
3265 * completely written to disk.
3266 */
3267 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268
Christoph Hellwig93848a92013-04-03 16:11:17 +11003269 /* update the lsn in the on disk inode if required */
3270 if (ip->i_d.di_version == 3)
3271 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272
Christoph Hellwig93848a92013-04-03 16:11:17 +11003273 /* generate the checksum. */
3274 xfs_dinode_calc_crc(mp, dip);
3275
3276 ASSERT(bp->b_fspriv != NULL);
3277 ASSERT(bp->b_iodone != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 return 0;
3279
3280corrupt_out:
3281 return XFS_ERROR(EFSCORRUPTED);
3282}