Loading...
1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include <linux/log2.h>
19
20#include "xfs.h"
21#include "xfs_fs.h"
22#include "xfs_shared.h"
23#include "xfs_format.h"
24#include "xfs_log_format.h"
25#include "xfs_trans_resv.h"
26#include "xfs_inum.h"
27#include "xfs_sb.h"
28#include "xfs_ag.h"
29#include "xfs_mount.h"
30#include "xfs_inode.h"
31#include "xfs_da_format.h"
32#include "xfs_da_btree.h"
33#include "xfs_dir2.h"
34#include "xfs_attr_sf.h"
35#include "xfs_attr.h"
36#include "xfs_trans_space.h"
37#include "xfs_trans.h"
38#include "xfs_buf_item.h"
39#include "xfs_inode_item.h"
40#include "xfs_ialloc.h"
41#include "xfs_bmap.h"
42#include "xfs_bmap_util.h"
43#include "xfs_error.h"
44#include "xfs_quota.h"
45#include "xfs_filestream.h"
46#include "xfs_cksum.h"
47#include "xfs_trace.h"
48#include "xfs_icache.h"
49#include "xfs_symlink.h"
50#include "xfs_trans_priv.h"
51#include "xfs_log.h"
52#include "xfs_bmap_btree.h"
53
54kmem_zone_t *xfs_inode_zone;
55
56/*
57 * Used in xfs_itruncate_extents(). This is the maximum number of extents
58 * freed from a file in a single transaction.
59 */
60#define XFS_ITRUNC_MAX_EXTENTS 2
61
62STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
63
64STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *);
65
66/*
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
80/*
81 * These two are wrapper routines around the xfs_ilock() routine used to
82 * centralize some grungy code. They are used in places that wish to lock the
83 * inode solely for reading the extents. The reason these places can't just
84 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
85 * bringing in of the extents from disk for a file in b-tree format. If the
86 * inode is in b-tree format, then we need to lock the inode exclusively until
87 * the extents are read in. Locking it exclusively all the time would limit
88 * our parallelism unnecessarily, though. What we do instead is check to see
89 * if the extents have been read in yet, and only lock the inode exclusively
90 * if they have not.
91 *
92 * The functions return a value which should be given to the corresponding
93 * xfs_iunlock() call.
94 */
95uint
96xfs_ilock_data_map_shared(
97 struct xfs_inode *ip)
98{
99 uint lock_mode = XFS_ILOCK_SHARED;
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 xfs_ilock(ip, lock_mode);
105 return lock_mode;
106}
107
108uint
109xfs_ilock_attr_map_shared(
110 struct xfs_inode *ip)
111{
112 uint lock_mode = XFS_ILOCK_SHARED;
113
114 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE &&
115 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
116 lock_mode = XFS_ILOCK_EXCL;
117 xfs_ilock(ip, lock_mode);
118 return lock_mode;
119}
120
121/*
122 * The xfs inode contains 2 locks: a multi-reader lock called the
123 * i_iolock and a multi-reader lock called the i_lock. This routine
124 * allows either or both of the locks to be obtained.
125 *
126 * The 2 locks should always be ordered so that the IO lock is
127 * obtained first in order to prevent deadlock.
128 *
129 * ip -- the inode being locked
130 * lock_flags -- this parameter indicates the inode's locks
131 * to be locked. It can be:
132 * XFS_IOLOCK_SHARED,
133 * XFS_IOLOCK_EXCL,
134 * XFS_ILOCK_SHARED,
135 * XFS_ILOCK_EXCL,
136 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
137 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
138 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
139 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
140 */
141void
142xfs_ilock(
143 xfs_inode_t *ip,
144 uint lock_flags)
145{
146 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
147
148 /*
149 * You can't set both SHARED and EXCL for the same lock,
150 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
151 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
152 */
153 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
154 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
155 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
156 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
157 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
158
159 if (lock_flags & XFS_IOLOCK_EXCL)
160 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
161 else if (lock_flags & XFS_IOLOCK_SHARED)
162 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
163
164 if (lock_flags & XFS_ILOCK_EXCL)
165 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
166 else if (lock_flags & XFS_ILOCK_SHARED)
167 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
168}
169
170/*
171 * This is just like xfs_ilock(), except that the caller
172 * is guaranteed not to sleep. It returns 1 if it gets
173 * the requested locks and 0 otherwise. If the IO lock is
174 * obtained but the inode lock cannot be, then the IO lock
175 * is dropped before returning.
176 *
177 * ip -- the inode being locked
178 * lock_flags -- this parameter indicates the inode's locks to be
179 * to be locked. See the comment for xfs_ilock() for a list
180 * of valid values.
181 */
182int
183xfs_ilock_nowait(
184 xfs_inode_t *ip,
185 uint lock_flags)
186{
187 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
188
189 /*
190 * You can't set both SHARED and EXCL for the same lock,
191 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
192 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
193 */
194 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
195 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
196 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
197 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
198 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
199
200 if (lock_flags & XFS_IOLOCK_EXCL) {
201 if (!mrtryupdate(&ip->i_iolock))
202 goto out;
203 } else if (lock_flags & XFS_IOLOCK_SHARED) {
204 if (!mrtryaccess(&ip->i_iolock))
205 goto out;
206 }
207 if (lock_flags & XFS_ILOCK_EXCL) {
208 if (!mrtryupdate(&ip->i_lock))
209 goto out_undo_iolock;
210 } else if (lock_flags & XFS_ILOCK_SHARED) {
211 if (!mrtryaccess(&ip->i_lock))
212 goto out_undo_iolock;
213 }
214 return 1;
215
216 out_undo_iolock:
217 if (lock_flags & XFS_IOLOCK_EXCL)
218 mrunlock_excl(&ip->i_iolock);
219 else if (lock_flags & XFS_IOLOCK_SHARED)
220 mrunlock_shared(&ip->i_iolock);
221 out:
222 return 0;
223}
224
225/*
226 * xfs_iunlock() is used to drop the inode locks acquired with
227 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
228 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
229 * that we know which locks to drop.
230 *
231 * ip -- the inode being unlocked
232 * lock_flags -- this parameter indicates the inode's locks to be
233 * to be unlocked. See the comment for xfs_ilock() for a list
234 * of valid values for this parameter.
235 *
236 */
237void
238xfs_iunlock(
239 xfs_inode_t *ip,
240 uint lock_flags)
241{
242 /*
243 * You can't set both SHARED and EXCL for the same lock,
244 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
245 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
246 */
247 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
248 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
249 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
250 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
251 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
252 ASSERT(lock_flags != 0);
253
254 if (lock_flags & XFS_IOLOCK_EXCL)
255 mrunlock_excl(&ip->i_iolock);
256 else if (lock_flags & XFS_IOLOCK_SHARED)
257 mrunlock_shared(&ip->i_iolock);
258
259 if (lock_flags & XFS_ILOCK_EXCL)
260 mrunlock_excl(&ip->i_lock);
261 else if (lock_flags & XFS_ILOCK_SHARED)
262 mrunlock_shared(&ip->i_lock);
263
264 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
265}
266
267/*
268 * give up write locks. the i/o lock cannot be held nested
269 * if it is being demoted.
270 */
271void
272xfs_ilock_demote(
273 xfs_inode_t *ip,
274 uint lock_flags)
275{
276 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
277 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
278
279 if (lock_flags & XFS_ILOCK_EXCL)
280 mrdemote(&ip->i_lock);
281 if (lock_flags & XFS_IOLOCK_EXCL)
282 mrdemote(&ip->i_iolock);
283
284 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
285}
286
287#if defined(DEBUG) || defined(XFS_WARN)
288int
289xfs_isilocked(
290 xfs_inode_t *ip,
291 uint lock_flags)
292{
293 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
294 if (!(lock_flags & XFS_ILOCK_SHARED))
295 return !!ip->i_lock.mr_writer;
296 return rwsem_is_locked(&ip->i_lock.mr_lock);
297 }
298
299 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
300 if (!(lock_flags & XFS_IOLOCK_SHARED))
301 return !!ip->i_iolock.mr_writer;
302 return rwsem_is_locked(&ip->i_iolock.mr_lock);
303 }
304
305 ASSERT(0);
306 return 0;
307}
308#endif
309
310#ifdef DEBUG
311int xfs_locked_n;
312int xfs_small_retries;
313int xfs_middle_retries;
314int xfs_lots_retries;
315int xfs_lock_delays;
316#endif
317
318/*
319 * Bump the subclass so xfs_lock_inodes() acquires each lock with
320 * a different value
321 */
322static inline int
323xfs_lock_inumorder(int lock_mode, int subclass)
324{
325 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
326 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
327 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
328 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
329
330 return lock_mode;
331}
332
333/*
334 * The following routine will lock n inodes in exclusive mode.
335 * We assume the caller calls us with the inodes in i_ino order.
336 *
337 * We need to detect deadlock where an inode that we lock
338 * is in the AIL and we start waiting for another inode that is locked
339 * by a thread in a long running transaction (such as truncate). This can
340 * result in deadlock since the long running trans might need to wait
341 * for the inode we just locked in order to push the tail and free space
342 * in the log.
343 */
344void
345xfs_lock_inodes(
346 xfs_inode_t **ips,
347 int inodes,
348 uint lock_mode)
349{
350 int attempts = 0, i, j, try_lock;
351 xfs_log_item_t *lp;
352
353 ASSERT(ips && (inodes >= 2)); /* we need at least two */
354
355 try_lock = 0;
356 i = 0;
357
358again:
359 for (; i < inodes; i++) {
360 ASSERT(ips[i]);
361
362 if (i && (ips[i] == ips[i-1])) /* Already locked */
363 continue;
364
365 /*
366 * If try_lock is not set yet, make sure all locked inodes
367 * are not in the AIL.
368 * If any are, set try_lock to be used later.
369 */
370
371 if (!try_lock) {
372 for (j = (i - 1); j >= 0 && !try_lock; j--) {
373 lp = (xfs_log_item_t *)ips[j]->i_itemp;
374 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
375 try_lock++;
376 }
377 }
378 }
379
380 /*
381 * If any of the previous locks we have locked is in the AIL,
382 * we must TRY to get the second and subsequent locks. If
383 * we can't get any, we must release all we have
384 * and try again.
385 */
386
387 if (try_lock) {
388 /* try_lock must be 0 if i is 0. */
389 /*
390 * try_lock means we have an inode locked
391 * that is in the AIL.
392 */
393 ASSERT(i != 0);
394 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
395 attempts++;
396
397 /*
398 * Unlock all previous guys and try again.
399 * xfs_iunlock will try to push the tail
400 * if the inode is in the AIL.
401 */
402
403 for(j = i - 1; j >= 0; j--) {
404
405 /*
406 * Check to see if we've already
407 * unlocked this one.
408 * Not the first one going back,
409 * and the inode ptr is the same.
410 */
411 if ((j != (i - 1)) && ips[j] ==
412 ips[j+1])
413 continue;
414
415 xfs_iunlock(ips[j], lock_mode);
416 }
417
418 if ((attempts % 5) == 0) {
419 delay(1); /* Don't just spin the CPU */
420#ifdef DEBUG
421 xfs_lock_delays++;
422#endif
423 }
424 i = 0;
425 try_lock = 0;
426 goto again;
427 }
428 } else {
429 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
430 }
431 }
432
433#ifdef DEBUG
434 if (attempts) {
435 if (attempts < 5) xfs_small_retries++;
436 else if (attempts < 100) xfs_middle_retries++;
437 else xfs_lots_retries++;
438 } else {
439 xfs_locked_n++;
440 }
441#endif
442}
443
444/*
445 * xfs_lock_two_inodes() can only be used to lock one type of lock
446 * at a time - the iolock or the ilock, but not both at once. If
447 * we lock both at once, lockdep will report false positives saying
448 * we have violated locking orders.
449 */
450void
451xfs_lock_two_inodes(
452 xfs_inode_t *ip0,
453 xfs_inode_t *ip1,
454 uint lock_mode)
455{
456 xfs_inode_t *temp;
457 int attempts = 0;
458 xfs_log_item_t *lp;
459
460 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
461 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
462 ASSERT(ip0->i_ino != ip1->i_ino);
463
464 if (ip0->i_ino > ip1->i_ino) {
465 temp = ip0;
466 ip0 = ip1;
467 ip1 = temp;
468 }
469
470 again:
471 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
472
473 /*
474 * If the first lock we have locked is in the AIL, we must TRY to get
475 * the second lock. If we can't get it, we must release the first one
476 * and try again.
477 */
478 lp = (xfs_log_item_t *)ip0->i_itemp;
479 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
480 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
481 xfs_iunlock(ip0, lock_mode);
482 if ((++attempts % 5) == 0)
483 delay(1); /* Don't just spin the CPU */
484 goto again;
485 }
486 } else {
487 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
488 }
489}
490
491
492void
493__xfs_iflock(
494 struct xfs_inode *ip)
495{
496 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
497 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
498
499 do {
500 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
501 if (xfs_isiflocked(ip))
502 io_schedule();
503 } while (!xfs_iflock_nowait(ip));
504
505 finish_wait(wq, &wait.wait);
506}
507
508STATIC uint
509_xfs_dic2xflags(
510 __uint16_t di_flags)
511{
512 uint flags = 0;
513
514 if (di_flags & XFS_DIFLAG_ANY) {
515 if (di_flags & XFS_DIFLAG_REALTIME)
516 flags |= XFS_XFLAG_REALTIME;
517 if (di_flags & XFS_DIFLAG_PREALLOC)
518 flags |= XFS_XFLAG_PREALLOC;
519 if (di_flags & XFS_DIFLAG_IMMUTABLE)
520 flags |= XFS_XFLAG_IMMUTABLE;
521 if (di_flags & XFS_DIFLAG_APPEND)
522 flags |= XFS_XFLAG_APPEND;
523 if (di_flags & XFS_DIFLAG_SYNC)
524 flags |= XFS_XFLAG_SYNC;
525 if (di_flags & XFS_DIFLAG_NOATIME)
526 flags |= XFS_XFLAG_NOATIME;
527 if (di_flags & XFS_DIFLAG_NODUMP)
528 flags |= XFS_XFLAG_NODUMP;
529 if (di_flags & XFS_DIFLAG_RTINHERIT)
530 flags |= XFS_XFLAG_RTINHERIT;
531 if (di_flags & XFS_DIFLAG_PROJINHERIT)
532 flags |= XFS_XFLAG_PROJINHERIT;
533 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
534 flags |= XFS_XFLAG_NOSYMLINKS;
535 if (di_flags & XFS_DIFLAG_EXTSIZE)
536 flags |= XFS_XFLAG_EXTSIZE;
537 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
538 flags |= XFS_XFLAG_EXTSZINHERIT;
539 if (di_flags & XFS_DIFLAG_NODEFRAG)
540 flags |= XFS_XFLAG_NODEFRAG;
541 if (di_flags & XFS_DIFLAG_FILESTREAM)
542 flags |= XFS_XFLAG_FILESTREAM;
543 }
544
545 return flags;
546}
547
548uint
549xfs_ip2xflags(
550 xfs_inode_t *ip)
551{
552 xfs_icdinode_t *dic = &ip->i_d;
553
554 return _xfs_dic2xflags(dic->di_flags) |
555 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
556}
557
558uint
559xfs_dic2xflags(
560 xfs_dinode_t *dip)
561{
562 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
563 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
564}
565
566/*
567 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
568 * is allowed, otherwise it has to be an exact match. If a CI match is found,
569 * ci_name->name will point to a the actual name (caller must free) or
570 * will be set to NULL if an exact match is found.
571 */
572int
573xfs_lookup(
574 xfs_inode_t *dp,
575 struct xfs_name *name,
576 xfs_inode_t **ipp,
577 struct xfs_name *ci_name)
578{
579 xfs_ino_t inum;
580 int error;
581 uint lock_mode;
582
583 trace_xfs_lookup(dp, name);
584
585 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
586 return XFS_ERROR(EIO);
587
588 lock_mode = xfs_ilock_data_map_shared(dp);
589 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
590 xfs_iunlock(dp, lock_mode);
591
592 if (error)
593 goto out;
594
595 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
596 if (error)
597 goto out_free_name;
598
599 return 0;
600
601out_free_name:
602 if (ci_name)
603 kmem_free(ci_name->name);
604out:
605 *ipp = NULL;
606 return error;
607}
608
609/*
610 * Allocate an inode on disk and return a copy of its in-core version.
611 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
612 * appropriately within the inode. The uid and gid for the inode are
613 * set according to the contents of the given cred structure.
614 *
615 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
616 * has a free inode available, call xfs_iget() to obtain the in-core
617 * version of the allocated inode. Finally, fill in the inode and
618 * log its initial contents. In this case, ialloc_context would be
619 * set to NULL.
620 *
621 * If xfs_dialloc() does not have an available inode, it will replenish
622 * its supply by doing an allocation. Since we can only do one
623 * allocation within a transaction without deadlocks, we must commit
624 * the current transaction before returning the inode itself.
625 * In this case, therefore, we will set ialloc_context and return.
626 * The caller should then commit the current transaction, start a new
627 * transaction, and call xfs_ialloc() again to actually get the inode.
628 *
629 * To ensure that some other process does not grab the inode that
630 * was allocated during the first call to xfs_ialloc(), this routine
631 * also returns the [locked] bp pointing to the head of the freelist
632 * as ialloc_context. The caller should hold this buffer across
633 * the commit and pass it back into this routine on the second call.
634 *
635 * If we are allocating quota inodes, we do not have a parent inode
636 * to attach to or associate with (i.e. pip == NULL) because they
637 * are not linked into the directory structure - they are attached
638 * directly to the superblock - and so have no parent.
639 */
640int
641xfs_ialloc(
642 xfs_trans_t *tp,
643 xfs_inode_t *pip,
644 umode_t mode,
645 xfs_nlink_t nlink,
646 xfs_dev_t rdev,
647 prid_t prid,
648 int okalloc,
649 xfs_buf_t **ialloc_context,
650 xfs_inode_t **ipp)
651{
652 struct xfs_mount *mp = tp->t_mountp;
653 xfs_ino_t ino;
654 xfs_inode_t *ip;
655 uint flags;
656 int error;
657 timespec_t tv;
658 int filestreams = 0;
659
660 /*
661 * Call the space management code to pick
662 * the on-disk inode to be allocated.
663 */
664 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
665 ialloc_context, &ino);
666 if (error)
667 return error;
668 if (*ialloc_context || ino == NULLFSINO) {
669 *ipp = NULL;
670 return 0;
671 }
672 ASSERT(*ialloc_context == NULL);
673
674 /*
675 * Get the in-core inode with the lock held exclusively.
676 * This is because we're setting fields here we need
677 * to prevent others from looking at until we're done.
678 */
679 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
680 XFS_ILOCK_EXCL, &ip);
681 if (error)
682 return error;
683 ASSERT(ip != NULL);
684
685 ip->i_d.di_mode = mode;
686 ip->i_d.di_onlink = 0;
687 ip->i_d.di_nlink = nlink;
688 ASSERT(ip->i_d.di_nlink == nlink);
689 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
690 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
691 xfs_set_projid(ip, prid);
692 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
693
694 /*
695 * If the superblock version is up to where we support new format
696 * inodes and this is currently an old format inode, then change
697 * the inode version number now. This way we only do the conversion
698 * here rather than here and in the flush/logging code.
699 */
700 if (xfs_sb_version_hasnlink(&mp->m_sb) &&
701 ip->i_d.di_version == 1) {
702 ip->i_d.di_version = 2;
703 /*
704 * We've already zeroed the old link count, the projid field,
705 * and the pad field.
706 */
707 }
708
709 /*
710 * Project ids won't be stored on disk if we are using a version 1 inode.
711 */
712 if ((prid != 0) && (ip->i_d.di_version == 1))
713 xfs_bump_ino_vers2(tp, ip);
714
715 if (pip && XFS_INHERIT_GID(pip)) {
716 ip->i_d.di_gid = pip->i_d.di_gid;
717 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
718 ip->i_d.di_mode |= S_ISGID;
719 }
720 }
721
722 /*
723 * If the group ID of the new file does not match the effective group
724 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
725 * (and only if the irix_sgid_inherit compatibility variable is set).
726 */
727 if ((irix_sgid_inherit) &&
728 (ip->i_d.di_mode & S_ISGID) &&
729 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
730 ip->i_d.di_mode &= ~S_ISGID;
731 }
732
733 ip->i_d.di_size = 0;
734 ip->i_d.di_nextents = 0;
735 ASSERT(ip->i_d.di_nblocks == 0);
736
737 nanotime(&tv);
738 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
739 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
740 ip->i_d.di_atime = ip->i_d.di_mtime;
741 ip->i_d.di_ctime = ip->i_d.di_mtime;
742
743 /*
744 * di_gen will have been taken care of in xfs_iread.
745 */
746 ip->i_d.di_extsize = 0;
747 ip->i_d.di_dmevmask = 0;
748 ip->i_d.di_dmstate = 0;
749 ip->i_d.di_flags = 0;
750
751 if (ip->i_d.di_version == 3) {
752 ASSERT(ip->i_d.di_ino == ino);
753 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
754 ip->i_d.di_crc = 0;
755 ip->i_d.di_changecount = 1;
756 ip->i_d.di_lsn = 0;
757 ip->i_d.di_flags2 = 0;
758 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
759 ip->i_d.di_crtime = ip->i_d.di_mtime;
760 }
761
762
763 flags = XFS_ILOG_CORE;
764 switch (mode & S_IFMT) {
765 case S_IFIFO:
766 case S_IFCHR:
767 case S_IFBLK:
768 case S_IFSOCK:
769 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
770 ip->i_df.if_u2.if_rdev = rdev;
771 ip->i_df.if_flags = 0;
772 flags |= XFS_ILOG_DEV;
773 break;
774 case S_IFREG:
775 /*
776 * we can't set up filestreams until after the VFS inode
777 * is set up properly.
778 */
779 if (pip && xfs_inode_is_filestream(pip))
780 filestreams = 1;
781 /* fall through */
782 case S_IFDIR:
783 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
784 uint di_flags = 0;
785
786 if (S_ISDIR(mode)) {
787 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
788 di_flags |= XFS_DIFLAG_RTINHERIT;
789 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
790 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
791 ip->i_d.di_extsize = pip->i_d.di_extsize;
792 }
793 } else if (S_ISREG(mode)) {
794 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
795 di_flags |= XFS_DIFLAG_REALTIME;
796 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
797 di_flags |= XFS_DIFLAG_EXTSIZE;
798 ip->i_d.di_extsize = pip->i_d.di_extsize;
799 }
800 }
801 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
802 xfs_inherit_noatime)
803 di_flags |= XFS_DIFLAG_NOATIME;
804 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
805 xfs_inherit_nodump)
806 di_flags |= XFS_DIFLAG_NODUMP;
807 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
808 xfs_inherit_sync)
809 di_flags |= XFS_DIFLAG_SYNC;
810 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
811 xfs_inherit_nosymlinks)
812 di_flags |= XFS_DIFLAG_NOSYMLINKS;
813 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
814 di_flags |= XFS_DIFLAG_PROJINHERIT;
815 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
816 xfs_inherit_nodefrag)
817 di_flags |= XFS_DIFLAG_NODEFRAG;
818 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
819 di_flags |= XFS_DIFLAG_FILESTREAM;
820 ip->i_d.di_flags |= di_flags;
821 }
822 /* FALLTHROUGH */
823 case S_IFLNK:
824 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
825 ip->i_df.if_flags = XFS_IFEXTENTS;
826 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
827 ip->i_df.if_u1.if_extents = NULL;
828 break;
829 default:
830 ASSERT(0);
831 }
832 /*
833 * Attribute fork settings for new inode.
834 */
835 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
836 ip->i_d.di_anextents = 0;
837
838 /*
839 * Log the new values stuffed into the inode.
840 */
841 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
842 xfs_trans_log_inode(tp, ip, flags);
843
844 /* now that we have an i_mode we can setup inode ops and unlock */
845 xfs_setup_inode(ip);
846
847 /* now we have set up the vfs inode we can associate the filestream */
848 if (filestreams) {
849 error = xfs_filestream_associate(pip, ip);
850 if (error < 0)
851 return -error;
852 if (!error)
853 xfs_iflags_set(ip, XFS_IFILESTREAM);
854 }
855
856 *ipp = ip;
857 return 0;
858}
859
860/*
861 * Allocates a new inode from disk and return a pointer to the
862 * incore copy. This routine will internally commit the current
863 * transaction and allocate a new one if the Space Manager needed
864 * to do an allocation to replenish the inode free-list.
865 *
866 * This routine is designed to be called from xfs_create and
867 * xfs_create_dir.
868 *
869 */
870int
871xfs_dir_ialloc(
872 xfs_trans_t **tpp, /* input: current transaction;
873 output: may be a new transaction. */
874 xfs_inode_t *dp, /* directory within whose allocate
875 the inode. */
876 umode_t mode,
877 xfs_nlink_t nlink,
878 xfs_dev_t rdev,
879 prid_t prid, /* project id */
880 int okalloc, /* ok to allocate new space */
881 xfs_inode_t **ipp, /* pointer to inode; it will be
882 locked. */
883 int *committed)
884
885{
886 xfs_trans_t *tp;
887 xfs_trans_t *ntp;
888 xfs_inode_t *ip;
889 xfs_buf_t *ialloc_context = NULL;
890 int code;
891 void *dqinfo;
892 uint tflags;
893
894 tp = *tpp;
895 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
896
897 /*
898 * xfs_ialloc will return a pointer to an incore inode if
899 * the Space Manager has an available inode on the free
900 * list. Otherwise, it will do an allocation and replenish
901 * the freelist. Since we can only do one allocation per
902 * transaction without deadlocks, we will need to commit the
903 * current transaction and start a new one. We will then
904 * need to call xfs_ialloc again to get the inode.
905 *
906 * If xfs_ialloc did an allocation to replenish the freelist,
907 * it returns the bp containing the head of the freelist as
908 * ialloc_context. We will hold a lock on it across the
909 * transaction commit so that no other process can steal
910 * the inode(s) that we've just allocated.
911 */
912 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
913 &ialloc_context, &ip);
914
915 /*
916 * Return an error if we were unable to allocate a new inode.
917 * This should only happen if we run out of space on disk or
918 * encounter a disk error.
919 */
920 if (code) {
921 *ipp = NULL;
922 return code;
923 }
924 if (!ialloc_context && !ip) {
925 *ipp = NULL;
926 return XFS_ERROR(ENOSPC);
927 }
928
929 /*
930 * If the AGI buffer is non-NULL, then we were unable to get an
931 * inode in one operation. We need to commit the current
932 * transaction and call xfs_ialloc() again. It is guaranteed
933 * to succeed the second time.
934 */
935 if (ialloc_context) {
936 struct xfs_trans_res tres;
937
938 /*
939 * Normally, xfs_trans_commit releases all the locks.
940 * We call bhold to hang on to the ialloc_context across
941 * the commit. Holding this buffer prevents any other
942 * processes from doing any allocations in this
943 * allocation group.
944 */
945 xfs_trans_bhold(tp, ialloc_context);
946 /*
947 * Save the log reservation so we can use
948 * them in the next transaction.
949 */
950 tres.tr_logres = xfs_trans_get_log_res(tp);
951 tres.tr_logcount = xfs_trans_get_log_count(tp);
952
953 /*
954 * We want the quota changes to be associated with the next
955 * transaction, NOT this one. So, detach the dqinfo from this
956 * and attach it to the next transaction.
957 */
958 dqinfo = NULL;
959 tflags = 0;
960 if (tp->t_dqinfo) {
961 dqinfo = (void *)tp->t_dqinfo;
962 tp->t_dqinfo = NULL;
963 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
964 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
965 }
966
967 ntp = xfs_trans_dup(tp);
968 code = xfs_trans_commit(tp, 0);
969 tp = ntp;
970 if (committed != NULL) {
971 *committed = 1;
972 }
973 /*
974 * If we get an error during the commit processing,
975 * release the buffer that is still held and return
976 * to the caller.
977 */
978 if (code) {
979 xfs_buf_relse(ialloc_context);
980 if (dqinfo) {
981 tp->t_dqinfo = dqinfo;
982 xfs_trans_free_dqinfo(tp);
983 }
984 *tpp = ntp;
985 *ipp = NULL;
986 return code;
987 }
988
989 /*
990 * transaction commit worked ok so we can drop the extra ticket
991 * reference that we gained in xfs_trans_dup()
992 */
993 xfs_log_ticket_put(tp->t_ticket);
994 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
995 code = xfs_trans_reserve(tp, &tres, 0, 0);
996
997 /*
998 * Re-attach the quota info that we detached from prev trx.
999 */
1000 if (dqinfo) {
1001 tp->t_dqinfo = dqinfo;
1002 tp->t_flags |= tflags;
1003 }
1004
1005 if (code) {
1006 xfs_buf_relse(ialloc_context);
1007 *tpp = ntp;
1008 *ipp = NULL;
1009 return code;
1010 }
1011 xfs_trans_bjoin(tp, ialloc_context);
1012
1013 /*
1014 * Call ialloc again. Since we've locked out all
1015 * other allocations in this allocation group,
1016 * this call should always succeed.
1017 */
1018 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1019 okalloc, &ialloc_context, &ip);
1020
1021 /*
1022 * If we get an error at this point, return to the caller
1023 * so that the current transaction can be aborted.
1024 */
1025 if (code) {
1026 *tpp = tp;
1027 *ipp = NULL;
1028 return code;
1029 }
1030 ASSERT(!ialloc_context && ip);
1031
1032 } else {
1033 if (committed != NULL)
1034 *committed = 0;
1035 }
1036
1037 *ipp = ip;
1038 *tpp = tp;
1039
1040 return 0;
1041}
1042
1043/*
1044 * Decrement the link count on an inode & log the change.
1045 * If this causes the link count to go to zero, initiate the
1046 * logging activity required to truncate a file.
1047 */
1048int /* error */
1049xfs_droplink(
1050 xfs_trans_t *tp,
1051 xfs_inode_t *ip)
1052{
1053 int error;
1054
1055 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1056
1057 ASSERT (ip->i_d.di_nlink > 0);
1058 ip->i_d.di_nlink--;
1059 drop_nlink(VFS_I(ip));
1060 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1061
1062 error = 0;
1063 if (ip->i_d.di_nlink == 0) {
1064 /*
1065 * We're dropping the last link to this file.
1066 * Move the on-disk inode to the AGI unlinked list.
1067 * From xfs_inactive() we will pull the inode from
1068 * the list and free it.
1069 */
1070 error = xfs_iunlink(tp, ip);
1071 }
1072 return error;
1073}
1074
1075/*
1076 * This gets called when the inode's version needs to be changed from 1 to 2.
1077 * Currently this happens when the nlink field overflows the old 16-bit value
1078 * or when chproj is called to change the project for the first time.
1079 * As a side effect the superblock version will also get rev'd
1080 * to contain the NLINK bit.
1081 */
1082void
1083xfs_bump_ino_vers2(
1084 xfs_trans_t *tp,
1085 xfs_inode_t *ip)
1086{
1087 xfs_mount_t *mp;
1088
1089 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1090 ASSERT(ip->i_d.di_version == 1);
1091
1092 ip->i_d.di_version = 2;
1093 ip->i_d.di_onlink = 0;
1094 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1095 mp = tp->t_mountp;
1096 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1097 spin_lock(&mp->m_sb_lock);
1098 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1099 xfs_sb_version_addnlink(&mp->m_sb);
1100 spin_unlock(&mp->m_sb_lock);
1101 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
1102 } else {
1103 spin_unlock(&mp->m_sb_lock);
1104 }
1105 }
1106 /* Caller must log the inode */
1107}
1108
1109/*
1110 * Increment the link count on an inode & log the change.
1111 */
1112int
1113xfs_bumplink(
1114 xfs_trans_t *tp,
1115 xfs_inode_t *ip)
1116{
1117 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1118
1119 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE));
1120 ip->i_d.di_nlink++;
1121 inc_nlink(VFS_I(ip));
1122 if ((ip->i_d.di_version == 1) &&
1123 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
1124 /*
1125 * The inode has increased its number of links beyond
1126 * what can fit in an old format inode. It now needs
1127 * to be converted to a version 2 inode with a 32 bit
1128 * link count. If this is the first inode in the file
1129 * system to do this, then we need to bump the superblock
1130 * version number as well.
1131 */
1132 xfs_bump_ino_vers2(tp, ip);
1133 }
1134
1135 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1136 return 0;
1137}
1138
1139int
1140xfs_create(
1141 xfs_inode_t *dp,
1142 struct xfs_name *name,
1143 umode_t mode,
1144 xfs_dev_t rdev,
1145 xfs_inode_t **ipp)
1146{
1147 int is_dir = S_ISDIR(mode);
1148 struct xfs_mount *mp = dp->i_mount;
1149 struct xfs_inode *ip = NULL;
1150 struct xfs_trans *tp = NULL;
1151 int error;
1152 xfs_bmap_free_t free_list;
1153 xfs_fsblock_t first_block;
1154 bool unlock_dp_on_error = false;
1155 uint cancel_flags;
1156 int committed;
1157 prid_t prid;
1158 struct xfs_dquot *udqp = NULL;
1159 struct xfs_dquot *gdqp = NULL;
1160 struct xfs_dquot *pdqp = NULL;
1161 struct xfs_trans_res tres;
1162 uint resblks;
1163
1164 trace_xfs_create(dp, name);
1165
1166 if (XFS_FORCED_SHUTDOWN(mp))
1167 return XFS_ERROR(EIO);
1168
1169 prid = xfs_get_initial_prid(dp);
1170
1171 /*
1172 * Make sure that we have allocated dquot(s) on disk.
1173 */
1174 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1175 xfs_kgid_to_gid(current_fsgid()), prid,
1176 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1177 &udqp, &gdqp, &pdqp);
1178 if (error)
1179 return error;
1180
1181 if (is_dir) {
1182 rdev = 0;
1183 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1184 tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1185 tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1186 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1187 } else {
1188 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1189 tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1190 tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1191 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1192 }
1193
1194 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1195
1196 /*
1197 * Initially assume that the file does not exist and
1198 * reserve the resources for that case. If that is not
1199 * the case we'll drop the one we have and get a more
1200 * appropriate transaction later.
1201 */
1202 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1203 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1204 if (error == ENOSPC) {
1205 /* flush outstanding delalloc blocks and retry */
1206 xfs_flush_inodes(mp);
1207 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1208 }
1209 if (error == ENOSPC) {
1210 /* No space at all so try a "no-allocation" reservation */
1211 resblks = 0;
1212 error = xfs_trans_reserve(tp, &tres, 0, 0);
1213 }
1214 if (error) {
1215 cancel_flags = 0;
1216 goto out_trans_cancel;
1217 }
1218
1219 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1220 unlock_dp_on_error = true;
1221
1222 xfs_bmap_init(&free_list, &first_block);
1223
1224 /*
1225 * Reserve disk quota and the inode.
1226 */
1227 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1228 pdqp, resblks, 1, 0);
1229 if (error)
1230 goto out_trans_cancel;
1231
1232 error = xfs_dir_canenter(tp, dp, name, resblks);
1233 if (error)
1234 goto out_trans_cancel;
1235
1236 /*
1237 * A newly created regular or special file just has one directory
1238 * entry pointing to them, but a directory also the "." entry
1239 * pointing to itself.
1240 */
1241 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1242 prid, resblks > 0, &ip, &committed);
1243 if (error) {
1244 if (error == ENOSPC)
1245 goto out_trans_cancel;
1246 goto out_trans_abort;
1247 }
1248
1249 /*
1250 * Now we join the directory inode to the transaction. We do not do it
1251 * earlier because xfs_dir_ialloc might commit the previous transaction
1252 * (and release all the locks). An error from here on will result in
1253 * the transaction cancel unlocking dp so don't do it explicitly in the
1254 * error path.
1255 */
1256 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1257 unlock_dp_on_error = false;
1258
1259 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1260 &first_block, &free_list, resblks ?
1261 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1262 if (error) {
1263 ASSERT(error != ENOSPC);
1264 goto out_trans_abort;
1265 }
1266 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1267 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1268
1269 if (is_dir) {
1270 error = xfs_dir_init(tp, ip, dp);
1271 if (error)
1272 goto out_bmap_cancel;
1273
1274 error = xfs_bumplink(tp, dp);
1275 if (error)
1276 goto out_bmap_cancel;
1277 }
1278
1279 /*
1280 * If this is a synchronous mount, make sure that the
1281 * create transaction goes to disk before returning to
1282 * the user.
1283 */
1284 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1285 xfs_trans_set_sync(tp);
1286
1287 /*
1288 * Attach the dquot(s) to the inodes and modify them incore.
1289 * These ids of the inode couldn't have changed since the new
1290 * inode has been locked ever since it was created.
1291 */
1292 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1293
1294 error = xfs_bmap_finish(&tp, &free_list, &committed);
1295 if (error)
1296 goto out_bmap_cancel;
1297
1298 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1299 if (error)
1300 goto out_release_inode;
1301
1302 xfs_qm_dqrele(udqp);
1303 xfs_qm_dqrele(gdqp);
1304 xfs_qm_dqrele(pdqp);
1305
1306 *ipp = ip;
1307 return 0;
1308
1309 out_bmap_cancel:
1310 xfs_bmap_cancel(&free_list);
1311 out_trans_abort:
1312 cancel_flags |= XFS_TRANS_ABORT;
1313 out_trans_cancel:
1314 xfs_trans_cancel(tp, cancel_flags);
1315 out_release_inode:
1316 /*
1317 * Wait until after the current transaction is aborted to
1318 * release the inode. This prevents recursive transactions
1319 * and deadlocks from xfs_inactive.
1320 */
1321 if (ip)
1322 IRELE(ip);
1323
1324 xfs_qm_dqrele(udqp);
1325 xfs_qm_dqrele(gdqp);
1326 xfs_qm_dqrele(pdqp);
1327
1328 if (unlock_dp_on_error)
1329 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1330 return error;
1331}
1332
1333int
1334xfs_create_tmpfile(
1335 struct xfs_inode *dp,
1336 struct dentry *dentry,
1337 umode_t mode,
1338 struct xfs_inode **ipp)
1339{
1340 struct xfs_mount *mp = dp->i_mount;
1341 struct xfs_inode *ip = NULL;
1342 struct xfs_trans *tp = NULL;
1343 int error;
1344 uint cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1345 prid_t prid;
1346 struct xfs_dquot *udqp = NULL;
1347 struct xfs_dquot *gdqp = NULL;
1348 struct xfs_dquot *pdqp = NULL;
1349 struct xfs_trans_res *tres;
1350 uint resblks;
1351
1352 if (XFS_FORCED_SHUTDOWN(mp))
1353 return XFS_ERROR(EIO);
1354
1355 prid = xfs_get_initial_prid(dp);
1356
1357 /*
1358 * Make sure that we have allocated dquot(s) on disk.
1359 */
1360 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1361 xfs_kgid_to_gid(current_fsgid()), prid,
1362 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1363 &udqp, &gdqp, &pdqp);
1364 if (error)
1365 return error;
1366
1367 resblks = XFS_IALLOC_SPACE_RES(mp);
1368 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE);
1369
1370 tres = &M_RES(mp)->tr_create_tmpfile;
1371 error = xfs_trans_reserve(tp, tres, resblks, 0);
1372 if (error == ENOSPC) {
1373 /* No space at all so try a "no-allocation" reservation */
1374 resblks = 0;
1375 error = xfs_trans_reserve(tp, tres, 0, 0);
1376 }
1377 if (error) {
1378 cancel_flags = 0;
1379 goto out_trans_cancel;
1380 }
1381
1382 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1383 pdqp, resblks, 1, 0);
1384 if (error)
1385 goto out_trans_cancel;
1386
1387 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0,
1388 prid, resblks > 0, &ip, NULL);
1389 if (error) {
1390 if (error == ENOSPC)
1391 goto out_trans_cancel;
1392 goto out_trans_abort;
1393 }
1394
1395 if (mp->m_flags & XFS_MOUNT_WSYNC)
1396 xfs_trans_set_sync(tp);
1397
1398 /*
1399 * Attach the dquot(s) to the inodes and modify them incore.
1400 * These ids of the inode couldn't have changed since the new
1401 * inode has been locked ever since it was created.
1402 */
1403 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1404
1405 ip->i_d.di_nlink--;
1406 error = xfs_iunlink(tp, ip);
1407 if (error)
1408 goto out_trans_abort;
1409
1410 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1411 if (error)
1412 goto out_release_inode;
1413
1414 xfs_qm_dqrele(udqp);
1415 xfs_qm_dqrele(gdqp);
1416 xfs_qm_dqrele(pdqp);
1417
1418 *ipp = ip;
1419 return 0;
1420
1421 out_trans_abort:
1422 cancel_flags |= XFS_TRANS_ABORT;
1423 out_trans_cancel:
1424 xfs_trans_cancel(tp, cancel_flags);
1425 out_release_inode:
1426 /*
1427 * Wait until after the current transaction is aborted to
1428 * release the inode. This prevents recursive transactions
1429 * and deadlocks from xfs_inactive.
1430 */
1431 if (ip)
1432 IRELE(ip);
1433
1434 xfs_qm_dqrele(udqp);
1435 xfs_qm_dqrele(gdqp);
1436 xfs_qm_dqrele(pdqp);
1437
1438 return error;
1439}
1440
1441int
1442xfs_link(
1443 xfs_inode_t *tdp,
1444 xfs_inode_t *sip,
1445 struct xfs_name *target_name)
1446{
1447 xfs_mount_t *mp = tdp->i_mount;
1448 xfs_trans_t *tp;
1449 int error;
1450 xfs_bmap_free_t free_list;
1451 xfs_fsblock_t first_block;
1452 int cancel_flags;
1453 int committed;
1454 int resblks;
1455
1456 trace_xfs_link(tdp, target_name);
1457
1458 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1459
1460 if (XFS_FORCED_SHUTDOWN(mp))
1461 return XFS_ERROR(EIO);
1462
1463 error = xfs_qm_dqattach(sip, 0);
1464 if (error)
1465 goto std_return;
1466
1467 error = xfs_qm_dqattach(tdp, 0);
1468 if (error)
1469 goto std_return;
1470
1471 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1472 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1473 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1474 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1475 if (error == ENOSPC) {
1476 resblks = 0;
1477 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1478 }
1479 if (error) {
1480 cancel_flags = 0;
1481 goto error_return;
1482 }
1483
1484 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1485
1486 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1487 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1488
1489 /*
1490 * If we are using project inheritance, we only allow hard link
1491 * creation in our tree when the project IDs are the same; else
1492 * the tree quota mechanism could be circumvented.
1493 */
1494 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1495 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1496 error = XFS_ERROR(EXDEV);
1497 goto error_return;
1498 }
1499
1500 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1501 if (error)
1502 goto error_return;
1503
1504 xfs_bmap_init(&free_list, &first_block);
1505
1506 if (sip->i_d.di_nlink == 0) {
1507 error = xfs_iunlink_remove(tp, sip);
1508 if (error)
1509 goto abort_return;
1510 }
1511
1512 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1513 &first_block, &free_list, resblks);
1514 if (error)
1515 goto abort_return;
1516 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1517 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1518
1519 error = xfs_bumplink(tp, sip);
1520 if (error)
1521 goto abort_return;
1522
1523 /*
1524 * If this is a synchronous mount, make sure that the
1525 * link transaction goes to disk before returning to
1526 * the user.
1527 */
1528 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1529 xfs_trans_set_sync(tp);
1530 }
1531
1532 error = xfs_bmap_finish (&tp, &free_list, &committed);
1533 if (error) {
1534 xfs_bmap_cancel(&free_list);
1535 goto abort_return;
1536 }
1537
1538 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1539
1540 abort_return:
1541 cancel_flags |= XFS_TRANS_ABORT;
1542 error_return:
1543 xfs_trans_cancel(tp, cancel_flags);
1544 std_return:
1545 return error;
1546}
1547
1548/*
1549 * Free up the underlying blocks past new_size. The new size must be smaller
1550 * than the current size. This routine can be used both for the attribute and
1551 * data fork, and does not modify the inode size, which is left to the caller.
1552 *
1553 * The transaction passed to this routine must have made a permanent log
1554 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1555 * given transaction and start new ones, so make sure everything involved in
1556 * the transaction is tidy before calling here. Some transaction will be
1557 * returned to the caller to be committed. The incoming transaction must
1558 * already include the inode, and both inode locks must be held exclusively.
1559 * The inode must also be "held" within the transaction. On return the inode
1560 * will be "held" within the returned transaction. This routine does NOT
1561 * require any disk space to be reserved for it within the transaction.
1562 *
1563 * If we get an error, we must return with the inode locked and linked into the
1564 * current transaction. This keeps things simple for the higher level code,
1565 * because it always knows that the inode is locked and held in the transaction
1566 * that returns to it whether errors occur or not. We don't mark the inode
1567 * dirty on error so that transactions can be easily aborted if possible.
1568 */
1569int
1570xfs_itruncate_extents(
1571 struct xfs_trans **tpp,
1572 struct xfs_inode *ip,
1573 int whichfork,
1574 xfs_fsize_t new_size)
1575{
1576 struct xfs_mount *mp = ip->i_mount;
1577 struct xfs_trans *tp = *tpp;
1578 struct xfs_trans *ntp;
1579 xfs_bmap_free_t free_list;
1580 xfs_fsblock_t first_block;
1581 xfs_fileoff_t first_unmap_block;
1582 xfs_fileoff_t last_block;
1583 xfs_filblks_t unmap_len;
1584 int committed;
1585 int error = 0;
1586 int done = 0;
1587
1588 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1589 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1590 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1591 ASSERT(new_size <= XFS_ISIZE(ip));
1592 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1593 ASSERT(ip->i_itemp != NULL);
1594 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1595 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1596
1597 trace_xfs_itruncate_extents_start(ip, new_size);
1598
1599 /*
1600 * Since it is possible for space to become allocated beyond
1601 * the end of the file (in a crash where the space is allocated
1602 * but the inode size is not yet updated), simply remove any
1603 * blocks which show up between the new EOF and the maximum
1604 * possible file size. If the first block to be removed is
1605 * beyond the maximum file size (ie it is the same as last_block),
1606 * then there is nothing to do.
1607 */
1608 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1609 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1610 if (first_unmap_block == last_block)
1611 return 0;
1612
1613 ASSERT(first_unmap_block < last_block);
1614 unmap_len = last_block - first_unmap_block + 1;
1615 while (!done) {
1616 xfs_bmap_init(&free_list, &first_block);
1617 error = xfs_bunmapi(tp, ip,
1618 first_unmap_block, unmap_len,
1619 xfs_bmapi_aflag(whichfork),
1620 XFS_ITRUNC_MAX_EXTENTS,
1621 &first_block, &free_list,
1622 &done);
1623 if (error)
1624 goto out_bmap_cancel;
1625
1626 /*
1627 * Duplicate the transaction that has the permanent
1628 * reservation and commit the old transaction.
1629 */
1630 error = xfs_bmap_finish(&tp, &free_list, &committed);
1631 if (committed)
1632 xfs_trans_ijoin(tp, ip, 0);
1633 if (error)
1634 goto out_bmap_cancel;
1635
1636 if (committed) {
1637 /*
1638 * Mark the inode dirty so it will be logged and
1639 * moved forward in the log as part of every commit.
1640 */
1641 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1642 }
1643
1644 ntp = xfs_trans_dup(tp);
1645 error = xfs_trans_commit(tp, 0);
1646 tp = ntp;
1647
1648 xfs_trans_ijoin(tp, ip, 0);
1649
1650 if (error)
1651 goto out;
1652
1653 /*
1654 * Transaction commit worked ok so we can drop the extra ticket
1655 * reference that we gained in xfs_trans_dup()
1656 */
1657 xfs_log_ticket_put(tp->t_ticket);
1658 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1659 if (error)
1660 goto out;
1661 }
1662
1663 /*
1664 * Always re-log the inode so that our permanent transaction can keep
1665 * on rolling it forward in the log.
1666 */
1667 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1668
1669 trace_xfs_itruncate_extents_end(ip, new_size);
1670
1671out:
1672 *tpp = tp;
1673 return error;
1674out_bmap_cancel:
1675 /*
1676 * If the bunmapi call encounters an error, return to the caller where
1677 * the transaction can be properly aborted. We just need to make sure
1678 * we're not holding any resources that we were not when we came in.
1679 */
1680 xfs_bmap_cancel(&free_list);
1681 goto out;
1682}
1683
1684int
1685xfs_release(
1686 xfs_inode_t *ip)
1687{
1688 xfs_mount_t *mp = ip->i_mount;
1689 int error;
1690
1691 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1692 return 0;
1693
1694 /* If this is a read-only mount, don't do this (would generate I/O) */
1695 if (mp->m_flags & XFS_MOUNT_RDONLY)
1696 return 0;
1697
1698 if (!XFS_FORCED_SHUTDOWN(mp)) {
1699 int truncated;
1700
1701 /*
1702 * If we are using filestreams, and we have an unlinked
1703 * file that we are processing the last close on, then nothing
1704 * will be able to reopen and write to this file. Purge this
1705 * inode from the filestreams cache so that it doesn't delay
1706 * teardown of the inode.
1707 */
1708 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1709 xfs_filestream_deassociate(ip);
1710
1711 /*
1712 * If we previously truncated this file and removed old data
1713 * in the process, we want to initiate "early" writeout on
1714 * the last close. This is an attempt to combat the notorious
1715 * NULL files problem which is particularly noticeable from a
1716 * truncate down, buffered (re-)write (delalloc), followed by
1717 * a crash. What we are effectively doing here is
1718 * significantly reducing the time window where we'd otherwise
1719 * be exposed to that problem.
1720 */
1721 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1722 if (truncated) {
1723 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1724 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1725 error = -filemap_flush(VFS_I(ip)->i_mapping);
1726 if (error)
1727 return error;
1728 }
1729 }
1730 }
1731
1732 if (ip->i_d.di_nlink == 0)
1733 return 0;
1734
1735 if (xfs_can_free_eofblocks(ip, false)) {
1736
1737 /*
1738 * If we can't get the iolock just skip truncating the blocks
1739 * past EOF because we could deadlock with the mmap_sem
1740 * otherwise. We'll get another chance to drop them once the
1741 * last reference to the inode is dropped, so we'll never leak
1742 * blocks permanently.
1743 *
1744 * Further, check if the inode is being opened, written and
1745 * closed frequently and we have delayed allocation blocks
1746 * outstanding (e.g. streaming writes from the NFS server),
1747 * truncating the blocks past EOF will cause fragmentation to
1748 * occur.
1749 *
1750 * In this case don't do the truncation, either, but we have to
1751 * be careful how we detect this case. Blocks beyond EOF show
1752 * up as i_delayed_blks even when the inode is clean, so we
1753 * need to truncate them away first before checking for a dirty
1754 * release. Hence on the first dirty close we will still remove
1755 * the speculative allocation, but after that we will leave it
1756 * in place.
1757 */
1758 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1759 return 0;
1760
1761 error = xfs_free_eofblocks(mp, ip, true);
1762 if (error && error != EAGAIN)
1763 return error;
1764
1765 /* delalloc blocks after truncation means it really is dirty */
1766 if (ip->i_delayed_blks)
1767 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1768 }
1769 return 0;
1770}
1771
1772/*
1773 * xfs_inactive_truncate
1774 *
1775 * Called to perform a truncate when an inode becomes unlinked.
1776 */
1777STATIC int
1778xfs_inactive_truncate(
1779 struct xfs_inode *ip)
1780{
1781 struct xfs_mount *mp = ip->i_mount;
1782 struct xfs_trans *tp;
1783 int error;
1784
1785 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1786 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1787 if (error) {
1788 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1789 xfs_trans_cancel(tp, 0);
1790 return error;
1791 }
1792
1793 xfs_ilock(ip, XFS_ILOCK_EXCL);
1794 xfs_trans_ijoin(tp, ip, 0);
1795
1796 /*
1797 * Log the inode size first to prevent stale data exposure in the event
1798 * of a system crash before the truncate completes. See the related
1799 * comment in xfs_setattr_size() for details.
1800 */
1801 ip->i_d.di_size = 0;
1802 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1803
1804 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1805 if (error)
1806 goto error_trans_cancel;
1807
1808 ASSERT(ip->i_d.di_nextents == 0);
1809
1810 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1811 if (error)
1812 goto error_unlock;
1813
1814 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1815 return 0;
1816
1817error_trans_cancel:
1818 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1819error_unlock:
1820 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1821 return error;
1822}
1823
1824/*
1825 * xfs_inactive_ifree()
1826 *
1827 * Perform the inode free when an inode is unlinked.
1828 */
1829STATIC int
1830xfs_inactive_ifree(
1831 struct xfs_inode *ip)
1832{
1833 xfs_bmap_free_t free_list;
1834 xfs_fsblock_t first_block;
1835 int committed;
1836 struct xfs_mount *mp = ip->i_mount;
1837 struct xfs_trans *tp;
1838 int error;
1839
1840 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1841 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0);
1842 if (error) {
1843 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1844 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1845 return error;
1846 }
1847
1848 xfs_ilock(ip, XFS_ILOCK_EXCL);
1849 xfs_trans_ijoin(tp, ip, 0);
1850
1851 xfs_bmap_init(&free_list, &first_block);
1852 error = xfs_ifree(tp, ip, &free_list);
1853 if (error) {
1854 /*
1855 * If we fail to free the inode, shut down. The cancel
1856 * might do that, we need to make sure. Otherwise the
1857 * inode might be lost for a long time or forever.
1858 */
1859 if (!XFS_FORCED_SHUTDOWN(mp)) {
1860 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1861 __func__, error);
1862 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1863 }
1864 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1865 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1866 return error;
1867 }
1868
1869 /*
1870 * Credit the quota account(s). The inode is gone.
1871 */
1872 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1873
1874 /*
1875 * Just ignore errors at this point. There is nothing we can
1876 * do except to try to keep going. Make sure it's not a silent
1877 * error.
1878 */
1879 error = xfs_bmap_finish(&tp, &free_list, &committed);
1880 if (error)
1881 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1882 __func__, error);
1883 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1884 if (error)
1885 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1886 __func__, error);
1887
1888 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1889 return 0;
1890}
1891
1892/*
1893 * xfs_inactive
1894 *
1895 * This is called when the vnode reference count for the vnode
1896 * goes to zero. If the file has been unlinked, then it must
1897 * now be truncated. Also, we clear all of the read-ahead state
1898 * kept for the inode here since the file is now closed.
1899 */
1900void
1901xfs_inactive(
1902 xfs_inode_t *ip)
1903{
1904 struct xfs_mount *mp;
1905 int error;
1906 int truncate = 0;
1907
1908 /*
1909 * If the inode is already free, then there can be nothing
1910 * to clean up here.
1911 */
1912 if (ip->i_d.di_mode == 0) {
1913 ASSERT(ip->i_df.if_real_bytes == 0);
1914 ASSERT(ip->i_df.if_broot_bytes == 0);
1915 return;
1916 }
1917
1918 mp = ip->i_mount;
1919
1920 /* If this is a read-only mount, don't do this (would generate I/O) */
1921 if (mp->m_flags & XFS_MOUNT_RDONLY)
1922 return;
1923
1924 if (ip->i_d.di_nlink != 0) {
1925 /*
1926 * force is true because we are evicting an inode from the
1927 * cache. Post-eof blocks must be freed, lest we end up with
1928 * broken free space accounting.
1929 */
1930 if (xfs_can_free_eofblocks(ip, true))
1931 xfs_free_eofblocks(mp, ip, false);
1932
1933 return;
1934 }
1935
1936 if (S_ISREG(ip->i_d.di_mode) &&
1937 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1938 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1939 truncate = 1;
1940
1941 error = xfs_qm_dqattach(ip, 0);
1942 if (error)
1943 return;
1944
1945 if (S_ISLNK(ip->i_d.di_mode))
1946 error = xfs_inactive_symlink(ip);
1947 else if (truncate)
1948 error = xfs_inactive_truncate(ip);
1949 if (error)
1950 return;
1951
1952 /*
1953 * If there are attributes associated with the file then blow them away
1954 * now. The code calls a routine that recursively deconstructs the
1955 * attribute fork. We need to just commit the current transaction
1956 * because we can't use it for xfs_attr_inactive().
1957 */
1958 if (ip->i_d.di_anextents > 0) {
1959 ASSERT(ip->i_d.di_forkoff != 0);
1960
1961 error = xfs_attr_inactive(ip);
1962 if (error)
1963 return;
1964 }
1965
1966 if (ip->i_afp)
1967 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1968
1969 ASSERT(ip->i_d.di_anextents == 0);
1970
1971 /*
1972 * Free the inode.
1973 */
1974 error = xfs_inactive_ifree(ip);
1975 if (error)
1976 return;
1977
1978 /*
1979 * Release the dquots held by inode, if any.
1980 */
1981 xfs_qm_dqdetach(ip);
1982}
1983
1984/*
1985 * This is called when the inode's link count goes to 0.
1986 * We place the on-disk inode on a list in the AGI. It
1987 * will be pulled from this list when the inode is freed.
1988 */
1989int
1990xfs_iunlink(
1991 xfs_trans_t *tp,
1992 xfs_inode_t *ip)
1993{
1994 xfs_mount_t *mp;
1995 xfs_agi_t *agi;
1996 xfs_dinode_t *dip;
1997 xfs_buf_t *agibp;
1998 xfs_buf_t *ibp;
1999 xfs_agino_t agino;
2000 short bucket_index;
2001 int offset;
2002 int error;
2003
2004 ASSERT(ip->i_d.di_nlink == 0);
2005 ASSERT(ip->i_d.di_mode != 0);
2006
2007 mp = tp->t_mountp;
2008
2009 /*
2010 * Get the agi buffer first. It ensures lock ordering
2011 * on the list.
2012 */
2013 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
2014 if (error)
2015 return error;
2016 agi = XFS_BUF_TO_AGI(agibp);
2017
2018 /*
2019 * Get the index into the agi hash table for the
2020 * list this inode will go on.
2021 */
2022 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2023 ASSERT(agino != 0);
2024 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2025 ASSERT(agi->agi_unlinked[bucket_index]);
2026 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
2027
2028 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
2029 /*
2030 * There is already another inode in the bucket we need
2031 * to add ourselves to. Add us at the front of the list.
2032 * Here we put the head pointer into our next pointer,
2033 * and then we fall through to point the head at us.
2034 */
2035 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2036 0, 0);
2037 if (error)
2038 return error;
2039
2040 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
2041 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
2042 offset = ip->i_imap.im_boffset +
2043 offsetof(xfs_dinode_t, di_next_unlinked);
2044
2045 /* need to recalc the inode CRC if appropriate */
2046 xfs_dinode_calc_crc(mp, dip);
2047
2048 xfs_trans_inode_buf(tp, ibp);
2049 xfs_trans_log_buf(tp, ibp, offset,
2050 (offset + sizeof(xfs_agino_t) - 1));
2051 xfs_inobp_check(mp, ibp);
2052 }
2053
2054 /*
2055 * Point the bucket head pointer at the inode being inserted.
2056 */
2057 ASSERT(agino != 0);
2058 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
2059 offset = offsetof(xfs_agi_t, agi_unlinked) +
2060 (sizeof(xfs_agino_t) * bucket_index);
2061 xfs_trans_log_buf(tp, agibp, offset,
2062 (offset + sizeof(xfs_agino_t) - 1));
2063 return 0;
2064}
2065
2066/*
2067 * Pull the on-disk inode from the AGI unlinked list.
2068 */
2069STATIC int
2070xfs_iunlink_remove(
2071 xfs_trans_t *tp,
2072 xfs_inode_t *ip)
2073{
2074 xfs_ino_t next_ino;
2075 xfs_mount_t *mp;
2076 xfs_agi_t *agi;
2077 xfs_dinode_t *dip;
2078 xfs_buf_t *agibp;
2079 xfs_buf_t *ibp;
2080 xfs_agnumber_t agno;
2081 xfs_agino_t agino;
2082 xfs_agino_t next_agino;
2083 xfs_buf_t *last_ibp;
2084 xfs_dinode_t *last_dip = NULL;
2085 short bucket_index;
2086 int offset, last_offset = 0;
2087 int error;
2088
2089 mp = tp->t_mountp;
2090 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2091
2092 /*
2093 * Get the agi buffer first. It ensures lock ordering
2094 * on the list.
2095 */
2096 error = xfs_read_agi(mp, tp, agno, &agibp);
2097 if (error)
2098 return error;
2099
2100 agi = XFS_BUF_TO_AGI(agibp);
2101
2102 /*
2103 * Get the index into the agi hash table for the
2104 * list this inode will go on.
2105 */
2106 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2107 ASSERT(agino != 0);
2108 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2109 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2110 ASSERT(agi->agi_unlinked[bucket_index]);
2111
2112 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2113 /*
2114 * We're at the head of the list. Get the inode's on-disk
2115 * buffer to see if there is anyone after us on the list.
2116 * Only modify our next pointer if it is not already NULLAGINO.
2117 * This saves us the overhead of dealing with the buffer when
2118 * there is no need to change it.
2119 */
2120 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2121 0, 0);
2122 if (error) {
2123 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2124 __func__, error);
2125 return error;
2126 }
2127 next_agino = be32_to_cpu(dip->di_next_unlinked);
2128 ASSERT(next_agino != 0);
2129 if (next_agino != NULLAGINO) {
2130 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2131 offset = ip->i_imap.im_boffset +
2132 offsetof(xfs_dinode_t, di_next_unlinked);
2133
2134 /* need to recalc the inode CRC if appropriate */
2135 xfs_dinode_calc_crc(mp, dip);
2136
2137 xfs_trans_inode_buf(tp, ibp);
2138 xfs_trans_log_buf(tp, ibp, offset,
2139 (offset + sizeof(xfs_agino_t) - 1));
2140 xfs_inobp_check(mp, ibp);
2141 } else {
2142 xfs_trans_brelse(tp, ibp);
2143 }
2144 /*
2145 * Point the bucket head pointer at the next inode.
2146 */
2147 ASSERT(next_agino != 0);
2148 ASSERT(next_agino != agino);
2149 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2150 offset = offsetof(xfs_agi_t, agi_unlinked) +
2151 (sizeof(xfs_agino_t) * bucket_index);
2152 xfs_trans_log_buf(tp, agibp, offset,
2153 (offset + sizeof(xfs_agino_t) - 1));
2154 } else {
2155 /*
2156 * We need to search the list for the inode being freed.
2157 */
2158 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2159 last_ibp = NULL;
2160 while (next_agino != agino) {
2161 struct xfs_imap imap;
2162
2163 if (last_ibp)
2164 xfs_trans_brelse(tp, last_ibp);
2165
2166 imap.im_blkno = 0;
2167 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2168
2169 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2170 if (error) {
2171 xfs_warn(mp,
2172 "%s: xfs_imap returned error %d.",
2173 __func__, error);
2174 return error;
2175 }
2176
2177 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2178 &last_ibp, 0, 0);
2179 if (error) {
2180 xfs_warn(mp,
2181 "%s: xfs_imap_to_bp returned error %d.",
2182 __func__, error);
2183 return error;
2184 }
2185
2186 last_offset = imap.im_boffset;
2187 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2188 ASSERT(next_agino != NULLAGINO);
2189 ASSERT(next_agino != 0);
2190 }
2191
2192 /*
2193 * Now last_ibp points to the buffer previous to us on the
2194 * unlinked list. Pull us from the list.
2195 */
2196 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2197 0, 0);
2198 if (error) {
2199 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2200 __func__, error);
2201 return error;
2202 }
2203 next_agino = be32_to_cpu(dip->di_next_unlinked);
2204 ASSERT(next_agino != 0);
2205 ASSERT(next_agino != agino);
2206 if (next_agino != NULLAGINO) {
2207 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2208 offset = ip->i_imap.im_boffset +
2209 offsetof(xfs_dinode_t, di_next_unlinked);
2210
2211 /* need to recalc the inode CRC if appropriate */
2212 xfs_dinode_calc_crc(mp, dip);
2213
2214 xfs_trans_inode_buf(tp, ibp);
2215 xfs_trans_log_buf(tp, ibp, offset,
2216 (offset + sizeof(xfs_agino_t) - 1));
2217 xfs_inobp_check(mp, ibp);
2218 } else {
2219 xfs_trans_brelse(tp, ibp);
2220 }
2221 /*
2222 * Point the previous inode on the list to the next inode.
2223 */
2224 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2225 ASSERT(next_agino != 0);
2226 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2227
2228 /* need to recalc the inode CRC if appropriate */
2229 xfs_dinode_calc_crc(mp, last_dip);
2230
2231 xfs_trans_inode_buf(tp, last_ibp);
2232 xfs_trans_log_buf(tp, last_ibp, offset,
2233 (offset + sizeof(xfs_agino_t) - 1));
2234 xfs_inobp_check(mp, last_ibp);
2235 }
2236 return 0;
2237}
2238
2239/*
2240 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2241 * inodes that are in memory - they all must be marked stale and attached to
2242 * the cluster buffer.
2243 */
2244STATIC int
2245xfs_ifree_cluster(
2246 xfs_inode_t *free_ip,
2247 xfs_trans_t *tp,
2248 xfs_ino_t inum)
2249{
2250 xfs_mount_t *mp = free_ip->i_mount;
2251 int blks_per_cluster;
2252 int inodes_per_cluster;
2253 int nbufs;
2254 int i, j;
2255 xfs_daddr_t blkno;
2256 xfs_buf_t *bp;
2257 xfs_inode_t *ip;
2258 xfs_inode_log_item_t *iip;
2259 xfs_log_item_t *lip;
2260 struct xfs_perag *pag;
2261
2262 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2263 blks_per_cluster = xfs_icluster_size_fsb(mp);
2264 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
2265 nbufs = mp->m_ialloc_blks / blks_per_cluster;
2266
2267 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) {
2268 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2269 XFS_INO_TO_AGBNO(mp, inum));
2270
2271 /*
2272 * We obtain and lock the backing buffer first in the process
2273 * here, as we have to ensure that any dirty inode that we
2274 * can't get the flush lock on is attached to the buffer.
2275 * If we scan the in-memory inodes first, then buffer IO can
2276 * complete before we get a lock on it, and hence we may fail
2277 * to mark all the active inodes on the buffer stale.
2278 */
2279 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2280 mp->m_bsize * blks_per_cluster,
2281 XBF_UNMAPPED);
2282
2283 if (!bp)
2284 return ENOMEM;
2285
2286 /*
2287 * This buffer may not have been correctly initialised as we
2288 * didn't read it from disk. That's not important because we are
2289 * only using to mark the buffer as stale in the log, and to
2290 * attach stale cached inodes on it. That means it will never be
2291 * dispatched for IO. If it is, we want to know about it, and we
2292 * want it to fail. We can acheive this by adding a write
2293 * verifier to the buffer.
2294 */
2295 bp->b_ops = &xfs_inode_buf_ops;
2296
2297 /*
2298 * Walk the inodes already attached to the buffer and mark them
2299 * stale. These will all have the flush locks held, so an
2300 * in-memory inode walk can't lock them. By marking them all
2301 * stale first, we will not attempt to lock them in the loop
2302 * below as the XFS_ISTALE flag will be set.
2303 */
2304 lip = bp->b_fspriv;
2305 while (lip) {
2306 if (lip->li_type == XFS_LI_INODE) {
2307 iip = (xfs_inode_log_item_t *)lip;
2308 ASSERT(iip->ili_logged == 1);
2309 lip->li_cb = xfs_istale_done;
2310 xfs_trans_ail_copy_lsn(mp->m_ail,
2311 &iip->ili_flush_lsn,
2312 &iip->ili_item.li_lsn);
2313 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2314 }
2315 lip = lip->li_bio_list;
2316 }
2317
2318
2319 /*
2320 * For each inode in memory attempt to add it to the inode
2321 * buffer and set it up for being staled on buffer IO
2322 * completion. This is safe as we've locked out tail pushing
2323 * and flushing by locking the buffer.
2324 *
2325 * We have already marked every inode that was part of a
2326 * transaction stale above, which means there is no point in
2327 * even trying to lock them.
2328 */
2329 for (i = 0; i < inodes_per_cluster; i++) {
2330retry:
2331 rcu_read_lock();
2332 ip = radix_tree_lookup(&pag->pag_ici_root,
2333 XFS_INO_TO_AGINO(mp, (inum + i)));
2334
2335 /* Inode not in memory, nothing to do */
2336 if (!ip) {
2337 rcu_read_unlock();
2338 continue;
2339 }
2340
2341 /*
2342 * because this is an RCU protected lookup, we could
2343 * find a recently freed or even reallocated inode
2344 * during the lookup. We need to check under the
2345 * i_flags_lock for a valid inode here. Skip it if it
2346 * is not valid, the wrong inode or stale.
2347 */
2348 spin_lock(&ip->i_flags_lock);
2349 if (ip->i_ino != inum + i ||
2350 __xfs_iflags_test(ip, XFS_ISTALE)) {
2351 spin_unlock(&ip->i_flags_lock);
2352 rcu_read_unlock();
2353 continue;
2354 }
2355 spin_unlock(&ip->i_flags_lock);
2356
2357 /*
2358 * Don't try to lock/unlock the current inode, but we
2359 * _cannot_ skip the other inodes that we did not find
2360 * in the list attached to the buffer and are not
2361 * already marked stale. If we can't lock it, back off
2362 * and retry.
2363 */
2364 if (ip != free_ip &&
2365 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2366 rcu_read_unlock();
2367 delay(1);
2368 goto retry;
2369 }
2370 rcu_read_unlock();
2371
2372 xfs_iflock(ip);
2373 xfs_iflags_set(ip, XFS_ISTALE);
2374
2375 /*
2376 * we don't need to attach clean inodes or those only
2377 * with unlogged changes (which we throw away, anyway).
2378 */
2379 iip = ip->i_itemp;
2380 if (!iip || xfs_inode_clean(ip)) {
2381 ASSERT(ip != free_ip);
2382 xfs_ifunlock(ip);
2383 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2384 continue;
2385 }
2386
2387 iip->ili_last_fields = iip->ili_fields;
2388 iip->ili_fields = 0;
2389 iip->ili_logged = 1;
2390 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2391 &iip->ili_item.li_lsn);
2392
2393 xfs_buf_attach_iodone(bp, xfs_istale_done,
2394 &iip->ili_item);
2395
2396 if (ip != free_ip)
2397 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2398 }
2399
2400 xfs_trans_stale_inode_buf(tp, bp);
2401 xfs_trans_binval(tp, bp);
2402 }
2403
2404 xfs_perag_put(pag);
2405 return 0;
2406}
2407
2408/*
2409 * This is called to return an inode to the inode free list.
2410 * The inode should already be truncated to 0 length and have
2411 * no pages associated with it. This routine also assumes that
2412 * the inode is already a part of the transaction.
2413 *
2414 * The on-disk copy of the inode will have been added to the list
2415 * of unlinked inodes in the AGI. We need to remove the inode from
2416 * that list atomically with respect to freeing it here.
2417 */
2418int
2419xfs_ifree(
2420 xfs_trans_t *tp,
2421 xfs_inode_t *ip,
2422 xfs_bmap_free_t *flist)
2423{
2424 int error;
2425 int delete;
2426 xfs_ino_t first_ino;
2427
2428 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2429 ASSERT(ip->i_d.di_nlink == 0);
2430 ASSERT(ip->i_d.di_nextents == 0);
2431 ASSERT(ip->i_d.di_anextents == 0);
2432 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2433 ASSERT(ip->i_d.di_nblocks == 0);
2434
2435 /*
2436 * Pull the on-disk inode from the AGI unlinked list.
2437 */
2438 error = xfs_iunlink_remove(tp, ip);
2439 if (error)
2440 return error;
2441
2442 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2443 if (error)
2444 return error;
2445
2446 ip->i_d.di_mode = 0; /* mark incore inode as free */
2447 ip->i_d.di_flags = 0;
2448 ip->i_d.di_dmevmask = 0;
2449 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2450 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2451 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2452 /*
2453 * Bump the generation count so no one will be confused
2454 * by reincarnations of this inode.
2455 */
2456 ip->i_d.di_gen++;
2457 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2458
2459 if (delete)
2460 error = xfs_ifree_cluster(ip, tp, first_ino);
2461
2462 return error;
2463}
2464
2465/*
2466 * This is called to unpin an inode. The caller must have the inode locked
2467 * in at least shared mode so that the buffer cannot be subsequently pinned
2468 * once someone is waiting for it to be unpinned.
2469 */
2470static void
2471xfs_iunpin(
2472 struct xfs_inode *ip)
2473{
2474 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2475
2476 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2477
2478 /* Give the log a push to start the unpinning I/O */
2479 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2480
2481}
2482
2483static void
2484__xfs_iunpin_wait(
2485 struct xfs_inode *ip)
2486{
2487 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2488 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2489
2490 xfs_iunpin(ip);
2491
2492 do {
2493 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2494 if (xfs_ipincount(ip))
2495 io_schedule();
2496 } while (xfs_ipincount(ip));
2497 finish_wait(wq, &wait.wait);
2498}
2499
2500void
2501xfs_iunpin_wait(
2502 struct xfs_inode *ip)
2503{
2504 if (xfs_ipincount(ip))
2505 __xfs_iunpin_wait(ip);
2506}
2507
2508/*
2509 * Removing an inode from the namespace involves removing the directory entry
2510 * and dropping the link count on the inode. Removing the directory entry can
2511 * result in locking an AGF (directory blocks were freed) and removing a link
2512 * count can result in placing the inode on an unlinked list which results in
2513 * locking an AGI.
2514 *
2515 * The big problem here is that we have an ordering constraint on AGF and AGI
2516 * locking - inode allocation locks the AGI, then can allocate a new extent for
2517 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2518 * removes the inode from the unlinked list, requiring that we lock the AGI
2519 * first, and then freeing the inode can result in an inode chunk being freed
2520 * and hence freeing disk space requiring that we lock an AGF.
2521 *
2522 * Hence the ordering that is imposed by other parts of the code is AGI before
2523 * AGF. This means we cannot remove the directory entry before we drop the inode
2524 * reference count and put it on the unlinked list as this results in a lock
2525 * order of AGF then AGI, and this can deadlock against inode allocation and
2526 * freeing. Therefore we must drop the link counts before we remove the
2527 * directory entry.
2528 *
2529 * This is still safe from a transactional point of view - it is not until we
2530 * get to xfs_bmap_finish() that we have the possibility of multiple
2531 * transactions in this operation. Hence as long as we remove the directory
2532 * entry and drop the link count in the first transaction of the remove
2533 * operation, there are no transactional constraints on the ordering here.
2534 */
2535int
2536xfs_remove(
2537 xfs_inode_t *dp,
2538 struct xfs_name *name,
2539 xfs_inode_t *ip)
2540{
2541 xfs_mount_t *mp = dp->i_mount;
2542 xfs_trans_t *tp = NULL;
2543 int is_dir = S_ISDIR(ip->i_d.di_mode);
2544 int error = 0;
2545 xfs_bmap_free_t free_list;
2546 xfs_fsblock_t first_block;
2547 int cancel_flags;
2548 int committed;
2549 int link_zero;
2550 uint resblks;
2551 uint log_count;
2552
2553 trace_xfs_remove(dp, name);
2554
2555 if (XFS_FORCED_SHUTDOWN(mp))
2556 return XFS_ERROR(EIO);
2557
2558 error = xfs_qm_dqattach(dp, 0);
2559 if (error)
2560 goto std_return;
2561
2562 error = xfs_qm_dqattach(ip, 0);
2563 if (error)
2564 goto std_return;
2565
2566 if (is_dir) {
2567 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2568 log_count = XFS_DEFAULT_LOG_COUNT;
2569 } else {
2570 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2571 log_count = XFS_REMOVE_LOG_COUNT;
2572 }
2573 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2574
2575 /*
2576 * We try to get the real space reservation first,
2577 * allowing for directory btree deletion(s) implying
2578 * possible bmap insert(s). If we can't get the space
2579 * reservation then we use 0 instead, and avoid the bmap
2580 * btree insert(s) in the directory code by, if the bmap
2581 * insert tries to happen, instead trimming the LAST
2582 * block from the directory.
2583 */
2584 resblks = XFS_REMOVE_SPACE_RES(mp);
2585 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2586 if (error == ENOSPC) {
2587 resblks = 0;
2588 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2589 }
2590 if (error) {
2591 ASSERT(error != ENOSPC);
2592 cancel_flags = 0;
2593 goto out_trans_cancel;
2594 }
2595
2596 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2597
2598 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2599 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2600
2601 /*
2602 * If we're removing a directory perform some additional validation.
2603 */
2604 cancel_flags |= XFS_TRANS_ABORT;
2605 if (is_dir) {
2606 ASSERT(ip->i_d.di_nlink >= 2);
2607 if (ip->i_d.di_nlink != 2) {
2608 error = XFS_ERROR(ENOTEMPTY);
2609 goto out_trans_cancel;
2610 }
2611 if (!xfs_dir_isempty(ip)) {
2612 error = XFS_ERROR(ENOTEMPTY);
2613 goto out_trans_cancel;
2614 }
2615
2616 /* Drop the link from ip's "..". */
2617 error = xfs_droplink(tp, dp);
2618 if (error)
2619 goto out_trans_cancel;
2620
2621 /* Drop the "." link from ip to self. */
2622 error = xfs_droplink(tp, ip);
2623 if (error)
2624 goto out_trans_cancel;
2625 } else {
2626 /*
2627 * When removing a non-directory we need to log the parent
2628 * inode here. For a directory this is done implicitly
2629 * by the xfs_droplink call for the ".." entry.
2630 */
2631 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2632 }
2633 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2634
2635 /* Drop the link from dp to ip. */
2636 error = xfs_droplink(tp, ip);
2637 if (error)
2638 goto out_trans_cancel;
2639
2640 /* Determine if this is the last link while the inode is locked */
2641 link_zero = (ip->i_d.di_nlink == 0);
2642
2643 xfs_bmap_init(&free_list, &first_block);
2644 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2645 &first_block, &free_list, resblks);
2646 if (error) {
2647 ASSERT(error != ENOENT);
2648 goto out_bmap_cancel;
2649 }
2650
2651 /*
2652 * If this is a synchronous mount, make sure that the
2653 * remove transaction goes to disk before returning to
2654 * the user.
2655 */
2656 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2657 xfs_trans_set_sync(tp);
2658
2659 error = xfs_bmap_finish(&tp, &free_list, &committed);
2660 if (error)
2661 goto out_bmap_cancel;
2662
2663 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2664 if (error)
2665 goto std_return;
2666
2667 /*
2668 * If we are using filestreams, kill the stream association.
2669 * If the file is still open it may get a new one but that
2670 * will get killed on last close in xfs_close() so we don't
2671 * have to worry about that.
2672 */
2673 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2674 xfs_filestream_deassociate(ip);
2675
2676 return 0;
2677
2678 out_bmap_cancel:
2679 xfs_bmap_cancel(&free_list);
2680 out_trans_cancel:
2681 xfs_trans_cancel(tp, cancel_flags);
2682 std_return:
2683 return error;
2684}
2685
2686/*
2687 * Enter all inodes for a rename transaction into a sorted array.
2688 */
2689STATIC void
2690xfs_sort_for_rename(
2691 xfs_inode_t *dp1, /* in: old (source) directory inode */
2692 xfs_inode_t *dp2, /* in: new (target) directory inode */
2693 xfs_inode_t *ip1, /* in: inode of old entry */
2694 xfs_inode_t *ip2, /* in: inode of new entry, if it
2695 already exists, NULL otherwise. */
2696 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2697 int *num_inodes) /* out: number of inodes in array */
2698{
2699 xfs_inode_t *temp;
2700 int i, j;
2701
2702 /*
2703 * i_tab contains a list of pointers to inodes. We initialize
2704 * the table here & we'll sort it. We will then use it to
2705 * order the acquisition of the inode locks.
2706 *
2707 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2708 */
2709 i_tab[0] = dp1;
2710 i_tab[1] = dp2;
2711 i_tab[2] = ip1;
2712 if (ip2) {
2713 *num_inodes = 4;
2714 i_tab[3] = ip2;
2715 } else {
2716 *num_inodes = 3;
2717 i_tab[3] = NULL;
2718 }
2719
2720 /*
2721 * Sort the elements via bubble sort. (Remember, there are at
2722 * most 4 elements to sort, so this is adequate.)
2723 */
2724 for (i = 0; i < *num_inodes; i++) {
2725 for (j = 1; j < *num_inodes; j++) {
2726 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2727 temp = i_tab[j];
2728 i_tab[j] = i_tab[j-1];
2729 i_tab[j-1] = temp;
2730 }
2731 }
2732 }
2733}
2734
2735/*
2736 * xfs_rename
2737 */
2738int
2739xfs_rename(
2740 xfs_inode_t *src_dp,
2741 struct xfs_name *src_name,
2742 xfs_inode_t *src_ip,
2743 xfs_inode_t *target_dp,
2744 struct xfs_name *target_name,
2745 xfs_inode_t *target_ip)
2746{
2747 xfs_trans_t *tp = NULL;
2748 xfs_mount_t *mp = src_dp->i_mount;
2749 int new_parent; /* moving to a new dir */
2750 int src_is_directory; /* src_name is a directory */
2751 int error;
2752 xfs_bmap_free_t free_list;
2753 xfs_fsblock_t first_block;
2754 int cancel_flags;
2755 int committed;
2756 xfs_inode_t *inodes[4];
2757 int spaceres;
2758 int num_inodes;
2759
2760 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2761
2762 new_parent = (src_dp != target_dp);
2763 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2764
2765 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2766 inodes, &num_inodes);
2767
2768 xfs_bmap_init(&free_list, &first_block);
2769 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2770 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2771 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2772 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2773 if (error == ENOSPC) {
2774 spaceres = 0;
2775 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2776 }
2777 if (error) {
2778 xfs_trans_cancel(tp, 0);
2779 goto std_return;
2780 }
2781
2782 /*
2783 * Attach the dquots to the inodes
2784 */
2785 error = xfs_qm_vop_rename_dqattach(inodes);
2786 if (error) {
2787 xfs_trans_cancel(tp, cancel_flags);
2788 goto std_return;
2789 }
2790
2791 /*
2792 * Lock all the participating inodes. Depending upon whether
2793 * the target_name exists in the target directory, and
2794 * whether the target directory is the same as the source
2795 * directory, we can lock from 2 to 4 inodes.
2796 */
2797 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2798
2799 /*
2800 * Join all the inodes to the transaction. From this point on,
2801 * we can rely on either trans_commit or trans_cancel to unlock
2802 * them.
2803 */
2804 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2805 if (new_parent)
2806 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2807 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2808 if (target_ip)
2809 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2810
2811 /*
2812 * If we are using project inheritance, we only allow renames
2813 * into our tree when the project IDs are the same; else the
2814 * tree quota mechanism would be circumvented.
2815 */
2816 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2817 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2818 error = XFS_ERROR(EXDEV);
2819 goto error_return;
2820 }
2821
2822 /*
2823 * Set up the target.
2824 */
2825 if (target_ip == NULL) {
2826 /*
2827 * If there's no space reservation, check the entry will
2828 * fit before actually inserting it.
2829 */
2830 error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2831 if (error)
2832 goto error_return;
2833 /*
2834 * If target does not exist and the rename crosses
2835 * directories, adjust the target directory link count
2836 * to account for the ".." reference from the new entry.
2837 */
2838 error = xfs_dir_createname(tp, target_dp, target_name,
2839 src_ip->i_ino, &first_block,
2840 &free_list, spaceres);
2841 if (error == ENOSPC)
2842 goto error_return;
2843 if (error)
2844 goto abort_return;
2845
2846 xfs_trans_ichgtime(tp, target_dp,
2847 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2848
2849 if (new_parent && src_is_directory) {
2850 error = xfs_bumplink(tp, target_dp);
2851 if (error)
2852 goto abort_return;
2853 }
2854 } else { /* target_ip != NULL */
2855 /*
2856 * If target exists and it's a directory, check that both
2857 * target and source are directories and that target can be
2858 * destroyed, or that neither is a directory.
2859 */
2860 if (S_ISDIR(target_ip->i_d.di_mode)) {
2861 /*
2862 * Make sure target dir is empty.
2863 */
2864 if (!(xfs_dir_isempty(target_ip)) ||
2865 (target_ip->i_d.di_nlink > 2)) {
2866 error = XFS_ERROR(EEXIST);
2867 goto error_return;
2868 }
2869 }
2870
2871 /*
2872 * Link the source inode under the target name.
2873 * If the source inode is a directory and we are moving
2874 * it across directories, its ".." entry will be
2875 * inconsistent until we replace that down below.
2876 *
2877 * In case there is already an entry with the same
2878 * name at the destination directory, remove it first.
2879 */
2880 error = xfs_dir_replace(tp, target_dp, target_name,
2881 src_ip->i_ino,
2882 &first_block, &free_list, spaceres);
2883 if (error)
2884 goto abort_return;
2885
2886 xfs_trans_ichgtime(tp, target_dp,
2887 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2888
2889 /*
2890 * Decrement the link count on the target since the target
2891 * dir no longer points to it.
2892 */
2893 error = xfs_droplink(tp, target_ip);
2894 if (error)
2895 goto abort_return;
2896
2897 if (src_is_directory) {
2898 /*
2899 * Drop the link from the old "." entry.
2900 */
2901 error = xfs_droplink(tp, target_ip);
2902 if (error)
2903 goto abort_return;
2904 }
2905 } /* target_ip != NULL */
2906
2907 /*
2908 * Remove the source.
2909 */
2910 if (new_parent && src_is_directory) {
2911 /*
2912 * Rewrite the ".." entry to point to the new
2913 * directory.
2914 */
2915 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2916 target_dp->i_ino,
2917 &first_block, &free_list, spaceres);
2918 ASSERT(error != EEXIST);
2919 if (error)
2920 goto abort_return;
2921 }
2922
2923 /*
2924 * We always want to hit the ctime on the source inode.
2925 *
2926 * This isn't strictly required by the standards since the source
2927 * inode isn't really being changed, but old unix file systems did
2928 * it and some incremental backup programs won't work without it.
2929 */
2930 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2931 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2932
2933 /*
2934 * Adjust the link count on src_dp. This is necessary when
2935 * renaming a directory, either within one parent when
2936 * the target existed, or across two parent directories.
2937 */
2938 if (src_is_directory && (new_parent || target_ip != NULL)) {
2939
2940 /*
2941 * Decrement link count on src_directory since the
2942 * entry that's moved no longer points to it.
2943 */
2944 error = xfs_droplink(tp, src_dp);
2945 if (error)
2946 goto abort_return;
2947 }
2948
2949 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2950 &first_block, &free_list, spaceres);
2951 if (error)
2952 goto abort_return;
2953
2954 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2955 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2956 if (new_parent)
2957 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2958
2959 /*
2960 * If this is a synchronous mount, make sure that the
2961 * rename transaction goes to disk before returning to
2962 * the user.
2963 */
2964 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2965 xfs_trans_set_sync(tp);
2966 }
2967
2968 error = xfs_bmap_finish(&tp, &free_list, &committed);
2969 if (error) {
2970 xfs_bmap_cancel(&free_list);
2971 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2972 XFS_TRANS_ABORT));
2973 goto std_return;
2974 }
2975
2976 /*
2977 * trans_commit will unlock src_ip, target_ip & decrement
2978 * the vnode references.
2979 */
2980 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2981
2982 abort_return:
2983 cancel_flags |= XFS_TRANS_ABORT;
2984 error_return:
2985 xfs_bmap_cancel(&free_list);
2986 xfs_trans_cancel(tp, cancel_flags);
2987 std_return:
2988 return error;
2989}
2990
2991STATIC int
2992xfs_iflush_cluster(
2993 xfs_inode_t *ip,
2994 xfs_buf_t *bp)
2995{
2996 xfs_mount_t *mp = ip->i_mount;
2997 struct xfs_perag *pag;
2998 unsigned long first_index, mask;
2999 unsigned long inodes_per_cluster;
3000 int ilist_size;
3001 xfs_inode_t **ilist;
3002 xfs_inode_t *iq;
3003 int nr_found;
3004 int clcount = 0;
3005 int bufwasdelwri;
3006 int i;
3007
3008 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
3009
3010 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
3011 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
3012 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
3013 if (!ilist)
3014 goto out_put;
3015
3016 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1);
3017 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
3018 rcu_read_lock();
3019 /* really need a gang lookup range call here */
3020 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
3021 first_index, inodes_per_cluster);
3022 if (nr_found == 0)
3023 goto out_free;
3024
3025 for (i = 0; i < nr_found; i++) {
3026 iq = ilist[i];
3027 if (iq == ip)
3028 continue;
3029
3030 /*
3031 * because this is an RCU protected lookup, we could find a
3032 * recently freed or even reallocated inode during the lookup.
3033 * We need to check under the i_flags_lock for a valid inode
3034 * here. Skip it if it is not valid or the wrong inode.
3035 */
3036 spin_lock(&ip->i_flags_lock);
3037 if (!ip->i_ino ||
3038 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
3039 spin_unlock(&ip->i_flags_lock);
3040 continue;
3041 }
3042 spin_unlock(&ip->i_flags_lock);
3043
3044 /*
3045 * Do an un-protected check to see if the inode is dirty and
3046 * is a candidate for flushing. These checks will be repeated
3047 * later after the appropriate locks are acquired.
3048 */
3049 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
3050 continue;
3051
3052 /*
3053 * Try to get locks. If any are unavailable or it is pinned,
3054 * then this inode cannot be flushed and is skipped.
3055 */
3056
3057 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3058 continue;
3059 if (!xfs_iflock_nowait(iq)) {
3060 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3061 continue;
3062 }
3063 if (xfs_ipincount(iq)) {
3064 xfs_ifunlock(iq);
3065 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3066 continue;
3067 }
3068
3069 /*
3070 * arriving here means that this inode can be flushed. First
3071 * re-check that it's dirty before flushing.
3072 */
3073 if (!xfs_inode_clean(iq)) {
3074 int error;
3075 error = xfs_iflush_int(iq, bp);
3076 if (error) {
3077 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3078 goto cluster_corrupt_out;
3079 }
3080 clcount++;
3081 } else {
3082 xfs_ifunlock(iq);
3083 }
3084 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3085 }
3086
3087 if (clcount) {
3088 XFS_STATS_INC(xs_icluster_flushcnt);
3089 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3090 }
3091
3092out_free:
3093 rcu_read_unlock();
3094 kmem_free(ilist);
3095out_put:
3096 xfs_perag_put(pag);
3097 return 0;
3098
3099
3100cluster_corrupt_out:
3101 /*
3102 * Corruption detected in the clustering loop. Invalidate the
3103 * inode buffer and shut down the filesystem.
3104 */
3105 rcu_read_unlock();
3106 /*
3107 * Clean up the buffer. If it was delwri, just release it --
3108 * brelse can handle it with no problems. If not, shut down the
3109 * filesystem before releasing the buffer.
3110 */
3111 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3112 if (bufwasdelwri)
3113 xfs_buf_relse(bp);
3114
3115 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3116
3117 if (!bufwasdelwri) {
3118 /*
3119 * Just like incore_relse: if we have b_iodone functions,
3120 * mark the buffer as an error and call them. Otherwise
3121 * mark it as stale and brelse.
3122 */
3123 if (bp->b_iodone) {
3124 XFS_BUF_UNDONE(bp);
3125 xfs_buf_stale(bp);
3126 xfs_buf_ioerror(bp, EIO);
3127 xfs_buf_ioend(bp, 0);
3128 } else {
3129 xfs_buf_stale(bp);
3130 xfs_buf_relse(bp);
3131 }
3132 }
3133
3134 /*
3135 * Unlocks the flush lock
3136 */
3137 xfs_iflush_abort(iq, false);
3138 kmem_free(ilist);
3139 xfs_perag_put(pag);
3140 return XFS_ERROR(EFSCORRUPTED);
3141}
3142
3143/*
3144 * Flush dirty inode metadata into the backing buffer.
3145 *
3146 * The caller must have the inode lock and the inode flush lock held. The
3147 * inode lock will still be held upon return to the caller, and the inode
3148 * flush lock will be released after the inode has reached the disk.
3149 *
3150 * The caller must write out the buffer returned in *bpp and release it.
3151 */
3152int
3153xfs_iflush(
3154 struct xfs_inode *ip,
3155 struct xfs_buf **bpp)
3156{
3157 struct xfs_mount *mp = ip->i_mount;
3158 struct xfs_buf *bp;
3159 struct xfs_dinode *dip;
3160 int error;
3161
3162 XFS_STATS_INC(xs_iflush_count);
3163
3164 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3165 ASSERT(xfs_isiflocked(ip));
3166 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3167 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3168
3169 *bpp = NULL;
3170
3171 xfs_iunpin_wait(ip);
3172
3173 /*
3174 * For stale inodes we cannot rely on the backing buffer remaining
3175 * stale in cache for the remaining life of the stale inode and so
3176 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3177 * inodes below. We have to check this after ensuring the inode is
3178 * unpinned so that it is safe to reclaim the stale inode after the
3179 * flush call.
3180 */
3181 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3182 xfs_ifunlock(ip);
3183 return 0;
3184 }
3185
3186 /*
3187 * This may have been unpinned because the filesystem is shutting
3188 * down forcibly. If that's the case we must not write this inode
3189 * to disk, because the log record didn't make it to disk.
3190 *
3191 * We also have to remove the log item from the AIL in this case,
3192 * as we wait for an empty AIL as part of the unmount process.
3193 */
3194 if (XFS_FORCED_SHUTDOWN(mp)) {
3195 error = XFS_ERROR(EIO);
3196 goto abort_out;
3197 }
3198
3199 /*
3200 * Get the buffer containing the on-disk inode.
3201 */
3202 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3203 0);
3204 if (error || !bp) {
3205 xfs_ifunlock(ip);
3206 return error;
3207 }
3208
3209 /*
3210 * First flush out the inode that xfs_iflush was called with.
3211 */
3212 error = xfs_iflush_int(ip, bp);
3213 if (error)
3214 goto corrupt_out;
3215
3216 /*
3217 * If the buffer is pinned then push on the log now so we won't
3218 * get stuck waiting in the write for too long.
3219 */
3220 if (xfs_buf_ispinned(bp))
3221 xfs_log_force(mp, 0);
3222
3223 /*
3224 * inode clustering:
3225 * see if other inodes can be gathered into this write
3226 */
3227 error = xfs_iflush_cluster(ip, bp);
3228 if (error)
3229 goto cluster_corrupt_out;
3230
3231 *bpp = bp;
3232 return 0;
3233
3234corrupt_out:
3235 xfs_buf_relse(bp);
3236 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3237cluster_corrupt_out:
3238 error = XFS_ERROR(EFSCORRUPTED);
3239abort_out:
3240 /*
3241 * Unlocks the flush lock
3242 */
3243 xfs_iflush_abort(ip, false);
3244 return error;
3245}
3246
3247STATIC int
3248xfs_iflush_int(
3249 struct xfs_inode *ip,
3250 struct xfs_buf *bp)
3251{
3252 struct xfs_inode_log_item *iip = ip->i_itemp;
3253 struct xfs_dinode *dip;
3254 struct xfs_mount *mp = ip->i_mount;
3255
3256 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3257 ASSERT(xfs_isiflocked(ip));
3258 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3259 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3260 ASSERT(iip != NULL && iip->ili_fields != 0);
3261
3262 /* set *dip = inode's place in the buffer */
3263 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3264
3265 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3266 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3267 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3268 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3269 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3270 goto corrupt_out;
3271 }
3272 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3273 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3274 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3275 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3276 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3277 goto corrupt_out;
3278 }
3279 if (S_ISREG(ip->i_d.di_mode)) {
3280 if (XFS_TEST_ERROR(
3281 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3282 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3283 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3284 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3285 "%s: Bad regular inode %Lu, ptr 0x%p",
3286 __func__, ip->i_ino, ip);
3287 goto corrupt_out;
3288 }
3289 } else if (S_ISDIR(ip->i_d.di_mode)) {
3290 if (XFS_TEST_ERROR(
3291 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3292 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3293 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3294 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3295 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3296 "%s: Bad directory inode %Lu, ptr 0x%p",
3297 __func__, ip->i_ino, ip);
3298 goto corrupt_out;
3299 }
3300 }
3301 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3302 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3303 XFS_RANDOM_IFLUSH_5)) {
3304 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3305 "%s: detected corrupt incore inode %Lu, "
3306 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3307 __func__, ip->i_ino,
3308 ip->i_d.di_nextents + ip->i_d.di_anextents,
3309 ip->i_d.di_nblocks, ip);
3310 goto corrupt_out;
3311 }
3312 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3313 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3314 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3315 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3316 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3317 goto corrupt_out;
3318 }
3319
3320 /*
3321 * Inode item log recovery for v1/v2 inodes are dependent on the
3322 * di_flushiter count for correct sequencing. We bump the flush
3323 * iteration count so we can detect flushes which postdate a log record
3324 * during recovery. This is redundant as we now log every change and
3325 * hence this can't happen but we need to still do it to ensure
3326 * backwards compatibility with old kernels that predate logging all
3327 * inode changes.
3328 */
3329 if (ip->i_d.di_version < 3)
3330 ip->i_d.di_flushiter++;
3331
3332 /*
3333 * Copy the dirty parts of the inode into the on-disk
3334 * inode. We always copy out the core of the inode,
3335 * because if the inode is dirty at all the core must
3336 * be.
3337 */
3338 xfs_dinode_to_disk(dip, &ip->i_d);
3339
3340 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3341 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3342 ip->i_d.di_flushiter = 0;
3343
3344 /*
3345 * If this is really an old format inode and the superblock version
3346 * has not been updated to support only new format inodes, then
3347 * convert back to the old inode format. If the superblock version
3348 * has been updated, then make the conversion permanent.
3349 */
3350 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3351 if (ip->i_d.di_version == 1) {
3352 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3353 /*
3354 * Convert it back.
3355 */
3356 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3357 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3358 } else {
3359 /*
3360 * The superblock version has already been bumped,
3361 * so just make the conversion to the new inode
3362 * format permanent.
3363 */
3364 ip->i_d.di_version = 2;
3365 dip->di_version = 2;
3366 ip->i_d.di_onlink = 0;
3367 dip->di_onlink = 0;
3368 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3369 memset(&(dip->di_pad[0]), 0,
3370 sizeof(dip->di_pad));
3371 ASSERT(xfs_get_projid(ip) == 0);
3372 }
3373 }
3374
3375 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3376 if (XFS_IFORK_Q(ip))
3377 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3378 xfs_inobp_check(mp, bp);
3379
3380 /*
3381 * We've recorded everything logged in the inode, so we'd like to clear
3382 * the ili_fields bits so we don't log and flush things unnecessarily.
3383 * However, we can't stop logging all this information until the data
3384 * we've copied into the disk buffer is written to disk. If we did we
3385 * might overwrite the copy of the inode in the log with all the data
3386 * after re-logging only part of it, and in the face of a crash we
3387 * wouldn't have all the data we need to recover.
3388 *
3389 * What we do is move the bits to the ili_last_fields field. When
3390 * logging the inode, these bits are moved back to the ili_fields field.
3391 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3392 * know that the information those bits represent is permanently on
3393 * disk. As long as the flush completes before the inode is logged
3394 * again, then both ili_fields and ili_last_fields will be cleared.
3395 *
3396 * We can play with the ili_fields bits here, because the inode lock
3397 * must be held exclusively in order to set bits there and the flush
3398 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3399 * done routine can tell whether or not to look in the AIL. Also, store
3400 * the current LSN of the inode so that we can tell whether the item has
3401 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3402 * need the AIL lock, because it is a 64 bit value that cannot be read
3403 * atomically.
3404 */
3405 iip->ili_last_fields = iip->ili_fields;
3406 iip->ili_fields = 0;
3407 iip->ili_logged = 1;
3408
3409 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3410 &iip->ili_item.li_lsn);
3411
3412 /*
3413 * Attach the function xfs_iflush_done to the inode's
3414 * buffer. This will remove the inode from the AIL
3415 * and unlock the inode's flush lock when the inode is
3416 * completely written to disk.
3417 */
3418 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3419
3420 /* update the lsn in the on disk inode if required */
3421 if (ip->i_d.di_version == 3)
3422 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3423
3424 /* generate the checksum. */
3425 xfs_dinode_calc_crc(mp, dip);
3426
3427 ASSERT(bp->b_fspriv != NULL);
3428 ASSERT(bp->b_iodone != NULL);
3429 return 0;
3430
3431corrupt_out:
3432 return XFS_ERROR(EFSCORRUPTED);
3433}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include <linux/iversion.h>
7
8#include "xfs.h"
9#include "xfs_fs.h"
10#include "xfs_shared.h"
11#include "xfs_format.h"
12#include "xfs_log_format.h"
13#include "xfs_trans_resv.h"
14#include "xfs_mount.h"
15#include "xfs_defer.h"
16#include "xfs_inode.h"
17#include "xfs_dir2.h"
18#include "xfs_attr.h"
19#include "xfs_bit.h"
20#include "xfs_trans_space.h"
21#include "xfs_trans.h"
22#include "xfs_buf_item.h"
23#include "xfs_inode_item.h"
24#include "xfs_iunlink_item.h"
25#include "xfs_ialloc.h"
26#include "xfs_bmap.h"
27#include "xfs_bmap_util.h"
28#include "xfs_errortag.h"
29#include "xfs_error.h"
30#include "xfs_quota.h"
31#include "xfs_filestream.h"
32#include "xfs_trace.h"
33#include "xfs_icache.h"
34#include "xfs_symlink.h"
35#include "xfs_trans_priv.h"
36#include "xfs_log.h"
37#include "xfs_bmap_btree.h"
38#include "xfs_reflink.h"
39#include "xfs_ag.h"
40#include "xfs_log_priv.h"
41#include "xfs_health.h"
42#include "xfs_pnfs.h"
43#include "xfs_parent.h"
44#include "xfs_xattr.h"
45#include "xfs_inode_util.h"
46#include "xfs_metafile.h"
47
48struct kmem_cache *xfs_inode_cache;
49
50/*
51 * These two are wrapper routines around the xfs_ilock() routine used to
52 * centralize some grungy code. They are used in places that wish to lock the
53 * inode solely for reading the extents. The reason these places can't just
54 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
55 * bringing in of the extents from disk for a file in b-tree format. If the
56 * inode is in b-tree format, then we need to lock the inode exclusively until
57 * the extents are read in. Locking it exclusively all the time would limit
58 * our parallelism unnecessarily, though. What we do instead is check to see
59 * if the extents have been read in yet, and only lock the inode exclusively
60 * if they have not.
61 *
62 * The functions return a value which should be given to the corresponding
63 * xfs_iunlock() call.
64 */
65uint
66xfs_ilock_data_map_shared(
67 struct xfs_inode *ip)
68{
69 uint lock_mode = XFS_ILOCK_SHARED;
70
71 if (xfs_need_iread_extents(&ip->i_df))
72 lock_mode = XFS_ILOCK_EXCL;
73 xfs_ilock(ip, lock_mode);
74 return lock_mode;
75}
76
77uint
78xfs_ilock_attr_map_shared(
79 struct xfs_inode *ip)
80{
81 uint lock_mode = XFS_ILOCK_SHARED;
82
83 if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
84 lock_mode = XFS_ILOCK_EXCL;
85 xfs_ilock(ip, lock_mode);
86 return lock_mode;
87}
88
89/*
90 * You can't set both SHARED and EXCL for the same lock,
91 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
92 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
93 * to set in lock_flags.
94 */
95static inline void
96xfs_lock_flags_assert(
97 uint lock_flags)
98{
99 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
100 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
101 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
102 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
103 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
104 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
105 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
106 ASSERT(lock_flags != 0);
107}
108
109/*
110 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
111 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
112 * various combinations of the locks to be obtained.
113 *
114 * The 3 locks should always be ordered so that the IO lock is obtained first,
115 * the mmap lock second and the ilock last in order to prevent deadlock.
116 *
117 * Basic locking order:
118 *
119 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
120 *
121 * mmap_lock locking order:
122 *
123 * i_rwsem -> page lock -> mmap_lock
124 * mmap_lock -> invalidate_lock -> page_lock
125 *
126 * The difference in mmap_lock locking order mean that we cannot hold the
127 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
128 * can fault in pages during copy in/out (for buffered IO) or require the
129 * mmap_lock in get_user_pages() to map the user pages into the kernel address
130 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
131 * fault because page faults already hold the mmap_lock.
132 *
133 * Hence to serialise fully against both syscall and mmap based IO, we need to
134 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
135 * both taken in places where we need to invalidate the page cache in a race
136 * free manner (e.g. truncate, hole punch and other extent manipulation
137 * functions).
138 */
139void
140xfs_ilock(
141 xfs_inode_t *ip,
142 uint lock_flags)
143{
144 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
145
146 xfs_lock_flags_assert(lock_flags);
147
148 if (lock_flags & XFS_IOLOCK_EXCL) {
149 down_write_nested(&VFS_I(ip)->i_rwsem,
150 XFS_IOLOCK_DEP(lock_flags));
151 } else if (lock_flags & XFS_IOLOCK_SHARED) {
152 down_read_nested(&VFS_I(ip)->i_rwsem,
153 XFS_IOLOCK_DEP(lock_flags));
154 }
155
156 if (lock_flags & XFS_MMAPLOCK_EXCL) {
157 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
158 XFS_MMAPLOCK_DEP(lock_flags));
159 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
160 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
161 XFS_MMAPLOCK_DEP(lock_flags));
162 }
163
164 if (lock_flags & XFS_ILOCK_EXCL)
165 down_write_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
166 else if (lock_flags & XFS_ILOCK_SHARED)
167 down_read_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
168}
169
170/*
171 * This is just like xfs_ilock(), except that the caller
172 * is guaranteed not to sleep. It returns 1 if it gets
173 * the requested locks and 0 otherwise. If the IO lock is
174 * obtained but the inode lock cannot be, then the IO lock
175 * is dropped before returning.
176 *
177 * ip -- the inode being locked
178 * lock_flags -- this parameter indicates the inode's locks to be
179 * to be locked. See the comment for xfs_ilock() for a list
180 * of valid values.
181 */
182int
183xfs_ilock_nowait(
184 xfs_inode_t *ip,
185 uint lock_flags)
186{
187 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
188
189 xfs_lock_flags_assert(lock_flags);
190
191 if (lock_flags & XFS_IOLOCK_EXCL) {
192 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
193 goto out;
194 } else if (lock_flags & XFS_IOLOCK_SHARED) {
195 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
196 goto out;
197 }
198
199 if (lock_flags & XFS_MMAPLOCK_EXCL) {
200 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
201 goto out_undo_iolock;
202 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
203 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
204 goto out_undo_iolock;
205 }
206
207 if (lock_flags & XFS_ILOCK_EXCL) {
208 if (!down_write_trylock(&ip->i_lock))
209 goto out_undo_mmaplock;
210 } else if (lock_flags & XFS_ILOCK_SHARED) {
211 if (!down_read_trylock(&ip->i_lock))
212 goto out_undo_mmaplock;
213 }
214 return 1;
215
216out_undo_mmaplock:
217 if (lock_flags & XFS_MMAPLOCK_EXCL)
218 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
219 else if (lock_flags & XFS_MMAPLOCK_SHARED)
220 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
221out_undo_iolock:
222 if (lock_flags & XFS_IOLOCK_EXCL)
223 up_write(&VFS_I(ip)->i_rwsem);
224 else if (lock_flags & XFS_IOLOCK_SHARED)
225 up_read(&VFS_I(ip)->i_rwsem);
226out:
227 return 0;
228}
229
230/*
231 * xfs_iunlock() is used to drop the inode locks acquired with
232 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
233 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
234 * that we know which locks to drop.
235 *
236 * ip -- the inode being unlocked
237 * lock_flags -- this parameter indicates the inode's locks to be
238 * to be unlocked. See the comment for xfs_ilock() for a list
239 * of valid values for this parameter.
240 *
241 */
242void
243xfs_iunlock(
244 xfs_inode_t *ip,
245 uint lock_flags)
246{
247 xfs_lock_flags_assert(lock_flags);
248
249 if (lock_flags & XFS_IOLOCK_EXCL)
250 up_write(&VFS_I(ip)->i_rwsem);
251 else if (lock_flags & XFS_IOLOCK_SHARED)
252 up_read(&VFS_I(ip)->i_rwsem);
253
254 if (lock_flags & XFS_MMAPLOCK_EXCL)
255 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
256 else if (lock_flags & XFS_MMAPLOCK_SHARED)
257 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
258
259 if (lock_flags & XFS_ILOCK_EXCL)
260 up_write(&ip->i_lock);
261 else if (lock_flags & XFS_ILOCK_SHARED)
262 up_read(&ip->i_lock);
263
264 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
265}
266
267/*
268 * give up write locks. the i/o lock cannot be held nested
269 * if it is being demoted.
270 */
271void
272xfs_ilock_demote(
273 xfs_inode_t *ip,
274 uint lock_flags)
275{
276 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
277 ASSERT((lock_flags &
278 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
279
280 if (lock_flags & XFS_ILOCK_EXCL)
281 downgrade_write(&ip->i_lock);
282 if (lock_flags & XFS_MMAPLOCK_EXCL)
283 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
284 if (lock_flags & XFS_IOLOCK_EXCL)
285 downgrade_write(&VFS_I(ip)->i_rwsem);
286
287 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
288}
289
290void
291xfs_assert_ilocked(
292 struct xfs_inode *ip,
293 uint lock_flags)
294{
295 /*
296 * Sometimes we assert the ILOCK is held exclusively, but we're in
297 * a workqueue, so lockdep doesn't know we're the owner.
298 */
299 if (lock_flags & XFS_ILOCK_SHARED)
300 rwsem_assert_held(&ip->i_lock);
301 else if (lock_flags & XFS_ILOCK_EXCL)
302 rwsem_assert_held_write_nolockdep(&ip->i_lock);
303
304 if (lock_flags & XFS_MMAPLOCK_SHARED)
305 rwsem_assert_held(&VFS_I(ip)->i_mapping->invalidate_lock);
306 else if (lock_flags & XFS_MMAPLOCK_EXCL)
307 rwsem_assert_held_write(&VFS_I(ip)->i_mapping->invalidate_lock);
308
309 if (lock_flags & XFS_IOLOCK_SHARED)
310 rwsem_assert_held(&VFS_I(ip)->i_rwsem);
311 else if (lock_flags & XFS_IOLOCK_EXCL)
312 rwsem_assert_held_write(&VFS_I(ip)->i_rwsem);
313}
314
315/*
316 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
317 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
318 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
319 * errors and warnings.
320 */
321#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
322static bool
323xfs_lockdep_subclass_ok(
324 int subclass)
325{
326 return subclass < MAX_LOCKDEP_SUBCLASSES;
327}
328#else
329#define xfs_lockdep_subclass_ok(subclass) (true)
330#endif
331
332/*
333 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
334 * value. This can be called for any type of inode lock combination, including
335 * parent locking. Care must be taken to ensure we don't overrun the subclass
336 * storage fields in the class mask we build.
337 */
338static inline uint
339xfs_lock_inumorder(
340 uint lock_mode,
341 uint subclass)
342{
343 uint class = 0;
344
345 ASSERT(!(lock_mode & XFS_ILOCK_PARENT));
346 ASSERT(xfs_lockdep_subclass_ok(subclass));
347
348 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
349 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
350 class += subclass << XFS_IOLOCK_SHIFT;
351 }
352
353 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
354 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
355 class += subclass << XFS_MMAPLOCK_SHIFT;
356 }
357
358 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
359 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
360 class += subclass << XFS_ILOCK_SHIFT;
361 }
362
363 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
364}
365
366/*
367 * The following routine will lock n inodes in exclusive mode. We assume the
368 * caller calls us with the inodes in i_ino order.
369 *
370 * We need to detect deadlock where an inode that we lock is in the AIL and we
371 * start waiting for another inode that is locked by a thread in a long running
372 * transaction (such as truncate). This can result in deadlock since the long
373 * running trans might need to wait for the inode we just locked in order to
374 * push the tail and free space in the log.
375 *
376 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
377 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
378 * lock more than one at a time, lockdep will report false positives saying we
379 * have violated locking orders.
380 */
381void
382xfs_lock_inodes(
383 struct xfs_inode **ips,
384 int inodes,
385 uint lock_mode)
386{
387 int attempts = 0;
388 uint i;
389 int j;
390 bool try_lock;
391 struct xfs_log_item *lp;
392
393 /*
394 * Currently supports between 2 and 5 inodes with exclusive locking. We
395 * support an arbitrary depth of locking here, but absolute limits on
396 * inodes depend on the type of locking and the limits placed by
397 * lockdep annotations in xfs_lock_inumorder. These are all checked by
398 * the asserts.
399 */
400 ASSERT(ips && inodes >= 2 && inodes <= 5);
401 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
402 XFS_ILOCK_EXCL));
403 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
404 XFS_ILOCK_SHARED)));
405 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
406 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
407 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
408 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
409
410 if (lock_mode & XFS_IOLOCK_EXCL) {
411 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
412 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
413 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
414
415again:
416 try_lock = false;
417 i = 0;
418 for (; i < inodes; i++) {
419 ASSERT(ips[i]);
420
421 if (i && (ips[i] == ips[i - 1])) /* Already locked */
422 continue;
423
424 /*
425 * If try_lock is not set yet, make sure all locked inodes are
426 * not in the AIL. If any are, set try_lock to be used later.
427 */
428 if (!try_lock) {
429 for (j = (i - 1); j >= 0 && !try_lock; j--) {
430 lp = &ips[j]->i_itemp->ili_item;
431 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
432 try_lock = true;
433 }
434 }
435
436 /*
437 * If any of the previous locks we have locked is in the AIL,
438 * we must TRY to get the second and subsequent locks. If
439 * we can't get any, we must release all we have
440 * and try again.
441 */
442 if (!try_lock) {
443 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
444 continue;
445 }
446
447 /* try_lock means we have an inode locked that is in the AIL. */
448 ASSERT(i != 0);
449 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
450 continue;
451
452 /*
453 * Unlock all previous guys and try again. xfs_iunlock will try
454 * to push the tail if the inode is in the AIL.
455 */
456 attempts++;
457 for (j = i - 1; j >= 0; j--) {
458 /*
459 * Check to see if we've already unlocked this one. Not
460 * the first one going back, and the inode ptr is the
461 * same.
462 */
463 if (j != (i - 1) && ips[j] == ips[j + 1])
464 continue;
465
466 xfs_iunlock(ips[j], lock_mode);
467 }
468
469 if ((attempts % 5) == 0) {
470 delay(1); /* Don't just spin the CPU */
471 }
472 goto again;
473 }
474}
475
476/*
477 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
478 * mmaplock must be double-locked separately since we use i_rwsem and
479 * invalidate_lock for that. We now support taking one lock EXCL and the
480 * other SHARED.
481 */
482void
483xfs_lock_two_inodes(
484 struct xfs_inode *ip0,
485 uint ip0_mode,
486 struct xfs_inode *ip1,
487 uint ip1_mode)
488{
489 int attempts = 0;
490 struct xfs_log_item *lp;
491
492 ASSERT(hweight32(ip0_mode) == 1);
493 ASSERT(hweight32(ip1_mode) == 1);
494 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
495 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
496 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
497 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
498 ASSERT(ip0->i_ino != ip1->i_ino);
499
500 if (ip0->i_ino > ip1->i_ino) {
501 swap(ip0, ip1);
502 swap(ip0_mode, ip1_mode);
503 }
504
505 again:
506 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
507
508 /*
509 * If the first lock we have locked is in the AIL, we must TRY to get
510 * the second lock. If we can't get it, we must release the first one
511 * and try again.
512 */
513 lp = &ip0->i_itemp->ili_item;
514 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
515 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
516 xfs_iunlock(ip0, ip0_mode);
517 if ((++attempts % 5) == 0)
518 delay(1); /* Don't just spin the CPU */
519 goto again;
520 }
521 } else {
522 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
523 }
524}
525
526/*
527 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
528 * is allowed, otherwise it has to be an exact match. If a CI match is found,
529 * ci_name->name will point to a the actual name (caller must free) or
530 * will be set to NULL if an exact match is found.
531 */
532int
533xfs_lookup(
534 struct xfs_inode *dp,
535 const struct xfs_name *name,
536 struct xfs_inode **ipp,
537 struct xfs_name *ci_name)
538{
539 xfs_ino_t inum;
540 int error;
541
542 trace_xfs_lookup(dp, name);
543
544 if (xfs_is_shutdown(dp->i_mount))
545 return -EIO;
546 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
547 return -EIO;
548
549 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
550 if (error)
551 goto out_unlock;
552
553 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
554 if (error)
555 goto out_free_name;
556
557 /*
558 * Fail if a directory entry in the regular directory tree points to
559 * a metadata file.
560 */
561 if (XFS_IS_CORRUPT(dp->i_mount, xfs_is_metadir_inode(*ipp))) {
562 xfs_fs_mark_sick(dp->i_mount, XFS_SICK_FS_METADIR);
563 error = -EFSCORRUPTED;
564 goto out_irele;
565 }
566
567 return 0;
568
569out_irele:
570 xfs_irele(*ipp);
571out_free_name:
572 if (ci_name)
573 kfree(ci_name->name);
574out_unlock:
575 *ipp = NULL;
576 return error;
577}
578
579/*
580 * Initialise a newly allocated inode and return the in-core inode to the
581 * caller locked exclusively.
582 *
583 * Caller is responsible for unlocking the inode manually upon return
584 */
585int
586xfs_icreate(
587 struct xfs_trans *tp,
588 xfs_ino_t ino,
589 const struct xfs_icreate_args *args,
590 struct xfs_inode **ipp)
591{
592 struct xfs_mount *mp = tp->t_mountp;
593 struct xfs_inode *ip = NULL;
594 int error;
595
596 /*
597 * Get the in-core inode with the lock held exclusively to prevent
598 * others from looking at until we're done.
599 */
600 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
601 if (error)
602 return error;
603
604 ASSERT(ip != NULL);
605 xfs_trans_ijoin(tp, ip, 0);
606 xfs_inode_init(tp, args, ip);
607
608 /* now that we have an i_mode we can setup the inode structure */
609 xfs_setup_inode(ip);
610
611 *ipp = ip;
612 return 0;
613}
614
615/* Return dquots for the ids that will be assigned to a new file. */
616int
617xfs_icreate_dqalloc(
618 const struct xfs_icreate_args *args,
619 struct xfs_dquot **udqpp,
620 struct xfs_dquot **gdqpp,
621 struct xfs_dquot **pdqpp)
622{
623 struct inode *dir = VFS_I(args->pip);
624 kuid_t uid = GLOBAL_ROOT_UID;
625 kgid_t gid = GLOBAL_ROOT_GID;
626 prid_t prid = 0;
627 unsigned int flags = XFS_QMOPT_QUOTALL;
628
629 if (args->idmap) {
630 /*
631 * The uid/gid computation code must match what the VFS uses to
632 * assign i_[ug]id. INHERIT adjusts the gid computation for
633 * setgid/grpid systems.
634 */
635 uid = mapped_fsuid(args->idmap, i_user_ns(dir));
636 gid = mapped_fsgid(args->idmap, i_user_ns(dir));
637 prid = xfs_get_initial_prid(args->pip);
638 flags |= XFS_QMOPT_INHERIT;
639 }
640
641 *udqpp = *gdqpp = *pdqpp = NULL;
642
643 return xfs_qm_vop_dqalloc(args->pip, uid, gid, prid, flags, udqpp,
644 gdqpp, pdqpp);
645}
646
647int
648xfs_create(
649 const struct xfs_icreate_args *args,
650 struct xfs_name *name,
651 struct xfs_inode **ipp)
652{
653 struct xfs_inode *dp = args->pip;
654 struct xfs_dir_update du = {
655 .dp = dp,
656 .name = name,
657 };
658 struct xfs_mount *mp = dp->i_mount;
659 struct xfs_trans *tp = NULL;
660 struct xfs_dquot *udqp;
661 struct xfs_dquot *gdqp;
662 struct xfs_dquot *pdqp;
663 struct xfs_trans_res *tres;
664 xfs_ino_t ino;
665 bool unlock_dp_on_error = false;
666 bool is_dir = S_ISDIR(args->mode);
667 uint resblks;
668 int error;
669
670 trace_xfs_create(dp, name);
671
672 if (xfs_is_shutdown(mp))
673 return -EIO;
674 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
675 return -EIO;
676
677 /* Make sure that we have allocated dquot(s) on disk. */
678 error = xfs_icreate_dqalloc(args, &udqp, &gdqp, &pdqp);
679 if (error)
680 return error;
681
682 if (is_dir) {
683 resblks = xfs_mkdir_space_res(mp, name->len);
684 tres = &M_RES(mp)->tr_mkdir;
685 } else {
686 resblks = xfs_create_space_res(mp, name->len);
687 tres = &M_RES(mp)->tr_create;
688 }
689
690 error = xfs_parent_start(mp, &du.ppargs);
691 if (error)
692 goto out_release_dquots;
693
694 /*
695 * Initially assume that the file does not exist and
696 * reserve the resources for that case. If that is not
697 * the case we'll drop the one we have and get a more
698 * appropriate transaction later.
699 */
700 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
701 &tp);
702 if (error == -ENOSPC) {
703 /* flush outstanding delalloc blocks and retry */
704 xfs_flush_inodes(mp);
705 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
706 resblks, &tp);
707 }
708 if (error)
709 goto out_parent;
710
711 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
712 unlock_dp_on_error = true;
713
714 /*
715 * A newly created regular or special file just has one directory
716 * entry pointing to them, but a directory also the "." entry
717 * pointing to itself.
718 */
719 error = xfs_dialloc(&tp, args, &ino);
720 if (!error)
721 error = xfs_icreate(tp, ino, args, &du.ip);
722 if (error)
723 goto out_trans_cancel;
724
725 /*
726 * Now we join the directory inode to the transaction. We do not do it
727 * earlier because xfs_dialloc might commit the previous transaction
728 * (and release all the locks). An error from here on will result in
729 * the transaction cancel unlocking dp so don't do it explicitly in the
730 * error path.
731 */
732 xfs_trans_ijoin(tp, dp, 0);
733
734 error = xfs_dir_create_child(tp, resblks, &du);
735 if (error)
736 goto out_trans_cancel;
737
738 /*
739 * If this is a synchronous mount, make sure that the
740 * create transaction goes to disk before returning to
741 * the user.
742 */
743 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
744 xfs_trans_set_sync(tp);
745
746 /*
747 * Attach the dquot(s) to the inodes and modify them incore.
748 * These ids of the inode couldn't have changed since the new
749 * inode has been locked ever since it was created.
750 */
751 xfs_qm_vop_create_dqattach(tp, du.ip, udqp, gdqp, pdqp);
752
753 error = xfs_trans_commit(tp);
754 if (error)
755 goto out_release_inode;
756
757 xfs_qm_dqrele(udqp);
758 xfs_qm_dqrele(gdqp);
759 xfs_qm_dqrele(pdqp);
760
761 *ipp = du.ip;
762 xfs_iunlock(du.ip, XFS_ILOCK_EXCL);
763 xfs_iunlock(dp, XFS_ILOCK_EXCL);
764 xfs_parent_finish(mp, du.ppargs);
765 return 0;
766
767 out_trans_cancel:
768 xfs_trans_cancel(tp);
769 out_release_inode:
770 /*
771 * Wait until after the current transaction is aborted to finish the
772 * setup of the inode and release the inode. This prevents recursive
773 * transactions and deadlocks from xfs_inactive.
774 */
775 if (du.ip) {
776 xfs_iunlock(du.ip, XFS_ILOCK_EXCL);
777 xfs_finish_inode_setup(du.ip);
778 xfs_irele(du.ip);
779 }
780 out_parent:
781 xfs_parent_finish(mp, du.ppargs);
782 out_release_dquots:
783 xfs_qm_dqrele(udqp);
784 xfs_qm_dqrele(gdqp);
785 xfs_qm_dqrele(pdqp);
786
787 if (unlock_dp_on_error)
788 xfs_iunlock(dp, XFS_ILOCK_EXCL);
789 return error;
790}
791
792int
793xfs_create_tmpfile(
794 const struct xfs_icreate_args *args,
795 struct xfs_inode **ipp)
796{
797 struct xfs_inode *dp = args->pip;
798 struct xfs_mount *mp = dp->i_mount;
799 struct xfs_inode *ip = NULL;
800 struct xfs_trans *tp = NULL;
801 struct xfs_dquot *udqp;
802 struct xfs_dquot *gdqp;
803 struct xfs_dquot *pdqp;
804 struct xfs_trans_res *tres;
805 xfs_ino_t ino;
806 uint resblks;
807 int error;
808
809 ASSERT(args->flags & XFS_ICREATE_TMPFILE);
810
811 if (xfs_is_shutdown(mp))
812 return -EIO;
813
814 /* Make sure that we have allocated dquot(s) on disk. */
815 error = xfs_icreate_dqalloc(args, &udqp, &gdqp, &pdqp);
816 if (error)
817 return error;
818
819 resblks = XFS_IALLOC_SPACE_RES(mp);
820 tres = &M_RES(mp)->tr_create_tmpfile;
821
822 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
823 &tp);
824 if (error)
825 goto out_release_dquots;
826
827 error = xfs_dialloc(&tp, args, &ino);
828 if (!error)
829 error = xfs_icreate(tp, ino, args, &ip);
830 if (error)
831 goto out_trans_cancel;
832
833 if (xfs_has_wsync(mp))
834 xfs_trans_set_sync(tp);
835
836 /*
837 * Attach the dquot(s) to the inodes and modify them incore.
838 * These ids of the inode couldn't have changed since the new
839 * inode has been locked ever since it was created.
840 */
841 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
842
843 error = xfs_iunlink(tp, ip);
844 if (error)
845 goto out_trans_cancel;
846
847 error = xfs_trans_commit(tp);
848 if (error)
849 goto out_release_inode;
850
851 xfs_qm_dqrele(udqp);
852 xfs_qm_dqrele(gdqp);
853 xfs_qm_dqrele(pdqp);
854
855 *ipp = ip;
856 xfs_iunlock(ip, XFS_ILOCK_EXCL);
857 return 0;
858
859 out_trans_cancel:
860 xfs_trans_cancel(tp);
861 out_release_inode:
862 /*
863 * Wait until after the current transaction is aborted to finish the
864 * setup of the inode and release the inode. This prevents recursive
865 * transactions and deadlocks from xfs_inactive.
866 */
867 if (ip) {
868 xfs_iunlock(ip, XFS_ILOCK_EXCL);
869 xfs_finish_inode_setup(ip);
870 xfs_irele(ip);
871 }
872 out_release_dquots:
873 xfs_qm_dqrele(udqp);
874 xfs_qm_dqrele(gdqp);
875 xfs_qm_dqrele(pdqp);
876
877 return error;
878}
879
880int
881xfs_link(
882 struct xfs_inode *tdp,
883 struct xfs_inode *sip,
884 struct xfs_name *target_name)
885{
886 struct xfs_dir_update du = {
887 .dp = tdp,
888 .name = target_name,
889 .ip = sip,
890 };
891 struct xfs_mount *mp = tdp->i_mount;
892 struct xfs_trans *tp;
893 int error, nospace_error = 0;
894 int resblks;
895
896 trace_xfs_link(tdp, target_name);
897
898 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
899
900 if (xfs_is_shutdown(mp))
901 return -EIO;
902 if (xfs_ifork_zapped(tdp, XFS_DATA_FORK))
903 return -EIO;
904
905 error = xfs_qm_dqattach(sip);
906 if (error)
907 goto std_return;
908
909 error = xfs_qm_dqattach(tdp);
910 if (error)
911 goto std_return;
912
913 error = xfs_parent_start(mp, &du.ppargs);
914 if (error)
915 goto std_return;
916
917 resblks = xfs_link_space_res(mp, target_name->len);
918 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
919 &tp, &nospace_error);
920 if (error)
921 goto out_parent;
922
923 /*
924 * We don't allow reservationless or quotaless hardlinking when parent
925 * pointers are enabled because we can't back out if the xattrs must
926 * grow.
927 */
928 if (du.ppargs && nospace_error) {
929 error = nospace_error;
930 goto error_return;
931 }
932
933 /*
934 * If we are using project inheritance, we only allow hard link
935 * creation in our tree when the project IDs are the same; else
936 * the tree quota mechanism could be circumvented.
937 */
938 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
939 tdp->i_projid != sip->i_projid)) {
940 /*
941 * Project quota setup skips special files which can
942 * leave inodes in a PROJINHERIT directory without a
943 * project ID set. We need to allow links to be made
944 * to these "project-less" inodes because userspace
945 * expects them to succeed after project ID setup,
946 * but everything else should be rejected.
947 */
948 if (!special_file(VFS_I(sip)->i_mode) ||
949 sip->i_projid != 0) {
950 error = -EXDEV;
951 goto error_return;
952 }
953 }
954
955 error = xfs_dir_add_child(tp, resblks, &du);
956 if (error)
957 goto error_return;
958
959 /*
960 * If this is a synchronous mount, make sure that the
961 * link transaction goes to disk before returning to
962 * the user.
963 */
964 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
965 xfs_trans_set_sync(tp);
966
967 error = xfs_trans_commit(tp);
968 xfs_iunlock(tdp, XFS_ILOCK_EXCL);
969 xfs_iunlock(sip, XFS_ILOCK_EXCL);
970 xfs_parent_finish(mp, du.ppargs);
971 return error;
972
973 error_return:
974 xfs_trans_cancel(tp);
975 xfs_iunlock(tdp, XFS_ILOCK_EXCL);
976 xfs_iunlock(sip, XFS_ILOCK_EXCL);
977 out_parent:
978 xfs_parent_finish(mp, du.ppargs);
979 std_return:
980 if (error == -ENOSPC && nospace_error)
981 error = nospace_error;
982 return error;
983}
984
985/* Clear the reflink flag and the cowblocks tag if possible. */
986static void
987xfs_itruncate_clear_reflink_flags(
988 struct xfs_inode *ip)
989{
990 struct xfs_ifork *dfork;
991 struct xfs_ifork *cfork;
992
993 if (!xfs_is_reflink_inode(ip))
994 return;
995 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
996 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
997 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
998 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
999 if (cfork->if_bytes == 0)
1000 xfs_inode_clear_cowblocks_tag(ip);
1001}
1002
1003/*
1004 * Free up the underlying blocks past new_size. The new size must be smaller
1005 * than the current size. This routine can be used both for the attribute and
1006 * data fork, and does not modify the inode size, which is left to the caller.
1007 *
1008 * The transaction passed to this routine must have made a permanent log
1009 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1010 * given transaction and start new ones, so make sure everything involved in
1011 * the transaction is tidy before calling here. Some transaction will be
1012 * returned to the caller to be committed. The incoming transaction must
1013 * already include the inode, and both inode locks must be held exclusively.
1014 * The inode must also be "held" within the transaction. On return the inode
1015 * will be "held" within the returned transaction. This routine does NOT
1016 * require any disk space to be reserved for it within the transaction.
1017 *
1018 * If we get an error, we must return with the inode locked and linked into the
1019 * current transaction. This keeps things simple for the higher level code,
1020 * because it always knows that the inode is locked and held in the transaction
1021 * that returns to it whether errors occur or not. We don't mark the inode
1022 * dirty on error so that transactions can be easily aborted if possible.
1023 */
1024int
1025xfs_itruncate_extents_flags(
1026 struct xfs_trans **tpp,
1027 struct xfs_inode *ip,
1028 int whichfork,
1029 xfs_fsize_t new_size,
1030 int flags)
1031{
1032 struct xfs_mount *mp = ip->i_mount;
1033 struct xfs_trans *tp = *tpp;
1034 xfs_fileoff_t first_unmap_block;
1035 int error = 0;
1036
1037 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1038 if (atomic_read(&VFS_I(ip)->i_count))
1039 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
1040 ASSERT(new_size <= XFS_ISIZE(ip));
1041 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1042 ASSERT(ip->i_itemp != NULL);
1043 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1044 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1045
1046 trace_xfs_itruncate_extents_start(ip, new_size);
1047
1048 flags |= xfs_bmapi_aflag(whichfork);
1049
1050 /*
1051 * Since it is possible for space to become allocated beyond
1052 * the end of the file (in a crash where the space is allocated
1053 * but the inode size is not yet updated), simply remove any
1054 * blocks which show up between the new EOF and the maximum
1055 * possible file size.
1056 *
1057 * We have to free all the blocks to the bmbt maximum offset, even if
1058 * the page cache can't scale that far.
1059 */
1060 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1061 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1062 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1063 return 0;
1064 }
1065
1066 error = xfs_bunmapi_range(&tp, ip, flags, first_unmap_block,
1067 XFS_MAX_FILEOFF);
1068 if (error)
1069 goto out;
1070
1071 if (whichfork == XFS_DATA_FORK) {
1072 /* Remove all pending CoW reservations. */
1073 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1074 first_unmap_block, XFS_MAX_FILEOFF, true);
1075 if (error)
1076 goto out;
1077
1078 xfs_itruncate_clear_reflink_flags(ip);
1079 }
1080
1081 /*
1082 * Always re-log the inode so that our permanent transaction can keep
1083 * on rolling it forward in the log.
1084 */
1085 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1086
1087 trace_xfs_itruncate_extents_end(ip, new_size);
1088
1089out:
1090 *tpp = tp;
1091 return error;
1092}
1093
1094/*
1095 * Mark all the buffers attached to this directory stale. In theory we should
1096 * never be freeing a directory with any blocks at all, but this covers the
1097 * case where we've recovered a directory swap with a "temporary" directory
1098 * created by online repair and now need to dump it.
1099 */
1100STATIC void
1101xfs_inactive_dir(
1102 struct xfs_inode *dp)
1103{
1104 struct xfs_iext_cursor icur;
1105 struct xfs_bmbt_irec got;
1106 struct xfs_mount *mp = dp->i_mount;
1107 struct xfs_da_geometry *geo = mp->m_dir_geo;
1108 struct xfs_ifork *ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);
1109 xfs_fileoff_t off;
1110
1111 /*
1112 * Invalidate each directory block. All directory blocks are of
1113 * fsbcount length and alignment, so we only need to walk those same
1114 * offsets. We hold the only reference to this inode, so we must wait
1115 * for the buffer locks.
1116 */
1117 for_each_xfs_iext(ifp, &icur, &got) {
1118 for (off = round_up(got.br_startoff, geo->fsbcount);
1119 off < got.br_startoff + got.br_blockcount;
1120 off += geo->fsbcount) {
1121 struct xfs_buf *bp = NULL;
1122 xfs_fsblock_t fsbno;
1123 int error;
1124
1125 fsbno = (off - got.br_startoff) + got.br_startblock;
1126 error = xfs_buf_incore(mp->m_ddev_targp,
1127 XFS_FSB_TO_DADDR(mp, fsbno),
1128 XFS_FSB_TO_BB(mp, geo->fsbcount),
1129 XBF_LIVESCAN, &bp);
1130 if (error)
1131 continue;
1132
1133 xfs_buf_stale(bp);
1134 xfs_buf_relse(bp);
1135 }
1136 }
1137}
1138
1139/*
1140 * xfs_inactive_truncate
1141 *
1142 * Called to perform a truncate when an inode becomes unlinked.
1143 */
1144STATIC int
1145xfs_inactive_truncate(
1146 struct xfs_inode *ip)
1147{
1148 struct xfs_mount *mp = ip->i_mount;
1149 struct xfs_trans *tp;
1150 int error;
1151
1152 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1153 if (error) {
1154 ASSERT(xfs_is_shutdown(mp));
1155 return error;
1156 }
1157 xfs_ilock(ip, XFS_ILOCK_EXCL);
1158 xfs_trans_ijoin(tp, ip, 0);
1159
1160 /*
1161 * Log the inode size first to prevent stale data exposure in the event
1162 * of a system crash before the truncate completes. See the related
1163 * comment in xfs_vn_setattr_size() for details.
1164 */
1165 ip->i_disk_size = 0;
1166 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1167
1168 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1169 if (error)
1170 goto error_trans_cancel;
1171
1172 ASSERT(ip->i_df.if_nextents == 0);
1173
1174 error = xfs_trans_commit(tp);
1175 if (error)
1176 goto error_unlock;
1177
1178 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1179 return 0;
1180
1181error_trans_cancel:
1182 xfs_trans_cancel(tp);
1183error_unlock:
1184 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1185 return error;
1186}
1187
1188/*
1189 * xfs_inactive_ifree()
1190 *
1191 * Perform the inode free when an inode is unlinked.
1192 */
1193STATIC int
1194xfs_inactive_ifree(
1195 struct xfs_inode *ip)
1196{
1197 struct xfs_mount *mp = ip->i_mount;
1198 struct xfs_trans *tp;
1199 int error;
1200
1201 /*
1202 * We try to use a per-AG reservation for any block needed by the finobt
1203 * tree, but as the finobt feature predates the per-AG reservation
1204 * support a degraded file system might not have enough space for the
1205 * reservation at mount time. In that case try to dip into the reserved
1206 * pool and pray.
1207 *
1208 * Send a warning if the reservation does happen to fail, as the inode
1209 * now remains allocated and sits on the unlinked list until the fs is
1210 * repaired.
1211 */
1212 if (unlikely(mp->m_finobt_nores)) {
1213 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1214 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1215 &tp);
1216 } else {
1217 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1218 }
1219 if (error) {
1220 if (error == -ENOSPC) {
1221 xfs_warn_ratelimited(mp,
1222 "Failed to remove inode(s) from unlinked list. "
1223 "Please free space, unmount and run xfs_repair.");
1224 } else {
1225 ASSERT(xfs_is_shutdown(mp));
1226 }
1227 return error;
1228 }
1229
1230 /*
1231 * We do not hold the inode locked across the entire rolling transaction
1232 * here. We only need to hold it for the first transaction that
1233 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1234 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1235 * here breaks the relationship between cluster buffer invalidation and
1236 * stale inode invalidation on cluster buffer item journal commit
1237 * completion, and can result in leaving dirty stale inodes hanging
1238 * around in memory.
1239 *
1240 * We have no need for serialising this inode operation against other
1241 * operations - we freed the inode and hence reallocation is required
1242 * and that will serialise on reallocating the space the deferops need
1243 * to free. Hence we can unlock the inode on the first commit of
1244 * the transaction rather than roll it right through the deferops. This
1245 * avoids relogging the XFS_ISTALE inode.
1246 *
1247 * We check that xfs_ifree() hasn't grown an internal transaction roll
1248 * by asserting that the inode is still locked when it returns.
1249 */
1250 xfs_ilock(ip, XFS_ILOCK_EXCL);
1251 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1252
1253 error = xfs_ifree(tp, ip);
1254 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1255 if (error) {
1256 /*
1257 * If we fail to free the inode, shut down. The cancel
1258 * might do that, we need to make sure. Otherwise the
1259 * inode might be lost for a long time or forever.
1260 */
1261 if (!xfs_is_shutdown(mp)) {
1262 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1263 __func__, error);
1264 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1265 }
1266 xfs_trans_cancel(tp);
1267 return error;
1268 }
1269
1270 /*
1271 * Credit the quota account(s). The inode is gone.
1272 */
1273 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1274
1275 return xfs_trans_commit(tp);
1276}
1277
1278/*
1279 * Returns true if we need to update the on-disk metadata before we can free
1280 * the memory used by this inode. Updates include freeing post-eof
1281 * preallocations; freeing COW staging extents; and marking the inode free in
1282 * the inobt if it is on the unlinked list.
1283 */
1284bool
1285xfs_inode_needs_inactive(
1286 struct xfs_inode *ip)
1287{
1288 struct xfs_mount *mp = ip->i_mount;
1289 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1290
1291 /*
1292 * If the inode is already free, then there can be nothing
1293 * to clean up here.
1294 */
1295 if (VFS_I(ip)->i_mode == 0)
1296 return false;
1297
1298 /*
1299 * If this is a read-only mount, don't do this (would generate I/O)
1300 * unless we're in log recovery and cleaning the iunlinked list.
1301 */
1302 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1303 return false;
1304
1305 /* If the log isn't running, push inodes straight to reclaim. */
1306 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1307 return false;
1308
1309 /* Metadata inodes require explicit resource cleanup. */
1310 if (xfs_is_internal_inode(ip))
1311 return false;
1312
1313 /* Want to clean out the cow blocks if there are any. */
1314 if (cow_ifp && cow_ifp->if_bytes > 0)
1315 return true;
1316
1317 /* Unlinked files must be freed. */
1318 if (VFS_I(ip)->i_nlink == 0)
1319 return true;
1320
1321 /*
1322 * This file isn't being freed, so check if there are post-eof blocks
1323 * to free.
1324 *
1325 * Note: don't bother with iolock here since lockdep complains about
1326 * acquiring it in reclaim context. We have the only reference to the
1327 * inode at this point anyways.
1328 */
1329 return xfs_can_free_eofblocks(ip);
1330}
1331
1332/*
1333 * Save health status somewhere, if we're dumping an inode with uncorrected
1334 * errors and online repair isn't running.
1335 */
1336static inline void
1337xfs_inactive_health(
1338 struct xfs_inode *ip)
1339{
1340 struct xfs_mount *mp = ip->i_mount;
1341 struct xfs_perag *pag;
1342 unsigned int sick;
1343 unsigned int checked;
1344
1345 xfs_inode_measure_sickness(ip, &sick, &checked);
1346 if (!sick)
1347 return;
1348
1349 trace_xfs_inode_unfixed_corruption(ip, sick);
1350
1351 if (sick & XFS_SICK_INO_FORGET)
1352 return;
1353
1354 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1355 if (!pag) {
1356 /* There had better still be a perag structure! */
1357 ASSERT(0);
1358 return;
1359 }
1360
1361 xfs_ag_mark_sick(pag, XFS_SICK_AG_INODES);
1362 xfs_perag_put(pag);
1363}
1364
1365/*
1366 * xfs_inactive
1367 *
1368 * This is called when the vnode reference count for the vnode
1369 * goes to zero. If the file has been unlinked, then it must
1370 * now be truncated. Also, we clear all of the read-ahead state
1371 * kept for the inode here since the file is now closed.
1372 */
1373int
1374xfs_inactive(
1375 xfs_inode_t *ip)
1376{
1377 struct xfs_mount *mp;
1378 int error = 0;
1379 int truncate = 0;
1380
1381 /*
1382 * If the inode is already free, then there can be nothing
1383 * to clean up here.
1384 */
1385 if (VFS_I(ip)->i_mode == 0) {
1386 ASSERT(ip->i_df.if_broot_bytes == 0);
1387 goto out;
1388 }
1389
1390 mp = ip->i_mount;
1391 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1392
1393 xfs_inactive_health(ip);
1394
1395 /*
1396 * If this is a read-only mount, don't do this (would generate I/O)
1397 * unless we're in log recovery and cleaning the iunlinked list.
1398 */
1399 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1400 goto out;
1401
1402 /* Metadata inodes require explicit resource cleanup. */
1403 if (xfs_is_internal_inode(ip))
1404 goto out;
1405
1406 /* Try to clean out the cow blocks if there are any. */
1407 if (xfs_inode_has_cow_data(ip)) {
1408 error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1409 if (error)
1410 goto out;
1411 }
1412
1413 if (VFS_I(ip)->i_nlink != 0) {
1414 /*
1415 * Note: don't bother with iolock here since lockdep complains
1416 * about acquiring it in reclaim context. We have the only
1417 * reference to the inode at this point anyways.
1418 */
1419 if (xfs_can_free_eofblocks(ip))
1420 error = xfs_free_eofblocks(ip);
1421
1422 goto out;
1423 }
1424
1425 if (S_ISREG(VFS_I(ip)->i_mode) &&
1426 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1427 xfs_inode_has_filedata(ip)))
1428 truncate = 1;
1429
1430 if (xfs_iflags_test(ip, XFS_IQUOTAUNCHECKED)) {
1431 /*
1432 * If this inode is being inactivated during a quotacheck and
1433 * has not yet been scanned by quotacheck, we /must/ remove
1434 * the dquots from the inode before inactivation changes the
1435 * block and inode counts. Most probably this is a result of
1436 * reloading the incore iunlinked list to purge unrecovered
1437 * unlinked inodes.
1438 */
1439 xfs_qm_dqdetach(ip);
1440 } else {
1441 error = xfs_qm_dqattach(ip);
1442 if (error)
1443 goto out;
1444 }
1445
1446 if (S_ISDIR(VFS_I(ip)->i_mode) && ip->i_df.if_nextents > 0) {
1447 xfs_inactive_dir(ip);
1448 truncate = 1;
1449 }
1450
1451 if (S_ISLNK(VFS_I(ip)->i_mode))
1452 error = xfs_inactive_symlink(ip);
1453 else if (truncate)
1454 error = xfs_inactive_truncate(ip);
1455 if (error)
1456 goto out;
1457
1458 /*
1459 * If there are attributes associated with the file then blow them away
1460 * now. The code calls a routine that recursively deconstructs the
1461 * attribute fork. If also blows away the in-core attribute fork.
1462 */
1463 if (xfs_inode_has_attr_fork(ip)) {
1464 error = xfs_attr_inactive(ip);
1465 if (error)
1466 goto out;
1467 }
1468
1469 ASSERT(ip->i_forkoff == 0);
1470
1471 /*
1472 * Free the inode.
1473 */
1474 error = xfs_inactive_ifree(ip);
1475
1476out:
1477 /*
1478 * We're done making metadata updates for this inode, so we can release
1479 * the attached dquots.
1480 */
1481 xfs_qm_dqdetach(ip);
1482 return error;
1483}
1484
1485/*
1486 * Find an inode on the unlinked list. This does not take references to the
1487 * inode as we have existence guarantees by holding the AGI buffer lock and that
1488 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1489 * don't find the inode in cache, then let the caller handle the situation.
1490 */
1491struct xfs_inode *
1492xfs_iunlink_lookup(
1493 struct xfs_perag *pag,
1494 xfs_agino_t agino)
1495{
1496 struct xfs_inode *ip;
1497
1498 rcu_read_lock();
1499 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1500 if (!ip) {
1501 /* Caller can handle inode not being in memory. */
1502 rcu_read_unlock();
1503 return NULL;
1504 }
1505
1506 /*
1507 * Inode in RCU freeing limbo should not happen. Warn about this and
1508 * let the caller handle the failure.
1509 */
1510 if (WARN_ON_ONCE(!ip->i_ino)) {
1511 rcu_read_unlock();
1512 return NULL;
1513 }
1514 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1515 rcu_read_unlock();
1516 return ip;
1517}
1518
1519/*
1520 * Load the inode @next_agino into the cache and set its prev_unlinked pointer
1521 * to @prev_agino. Caller must hold the AGI to synchronize with other changes
1522 * to the unlinked list.
1523 */
1524int
1525xfs_iunlink_reload_next(
1526 struct xfs_trans *tp,
1527 struct xfs_buf *agibp,
1528 xfs_agino_t prev_agino,
1529 xfs_agino_t next_agino)
1530{
1531 struct xfs_perag *pag = agibp->b_pag;
1532 struct xfs_mount *mp = pag_mount(pag);
1533 struct xfs_inode *next_ip = NULL;
1534 int error;
1535
1536 ASSERT(next_agino != NULLAGINO);
1537
1538#ifdef DEBUG
1539 rcu_read_lock();
1540 next_ip = radix_tree_lookup(&pag->pag_ici_root, next_agino);
1541 ASSERT(next_ip == NULL);
1542 rcu_read_unlock();
1543#endif
1544
1545 xfs_info_ratelimited(mp,
1546 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating recovery.",
1547 next_agino, pag_agno(pag));
1548
1549 /*
1550 * Use an untrusted lookup just to be cautious in case the AGI has been
1551 * corrupted and now points at a free inode. That shouldn't happen,
1552 * but we'd rather shut down now since we're already running in a weird
1553 * situation.
1554 */
1555 error = xfs_iget(mp, tp, xfs_agino_to_ino(pag, next_agino),
1556 XFS_IGET_UNTRUSTED, 0, &next_ip);
1557 if (error) {
1558 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
1559 return error;
1560 }
1561
1562 /* If this is not an unlinked inode, something is very wrong. */
1563 if (VFS_I(next_ip)->i_nlink != 0) {
1564 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
1565 error = -EFSCORRUPTED;
1566 goto rele;
1567 }
1568
1569 next_ip->i_prev_unlinked = prev_agino;
1570 trace_xfs_iunlink_reload_next(next_ip);
1571rele:
1572 ASSERT(!(VFS_I(next_ip)->i_state & I_DONTCACHE));
1573 if (xfs_is_quotacheck_running(mp) && next_ip)
1574 xfs_iflags_set(next_ip, XFS_IQUOTAUNCHECKED);
1575 xfs_irele(next_ip);
1576 return error;
1577}
1578
1579/*
1580 * Look up the inode number specified and if it is not already marked XFS_ISTALE
1581 * mark it stale. We should only find clean inodes in this lookup that aren't
1582 * already stale.
1583 */
1584static void
1585xfs_ifree_mark_inode_stale(
1586 struct xfs_perag *pag,
1587 struct xfs_inode *free_ip,
1588 xfs_ino_t inum)
1589{
1590 struct xfs_mount *mp = pag_mount(pag);
1591 struct xfs_inode_log_item *iip;
1592 struct xfs_inode *ip;
1593
1594retry:
1595 rcu_read_lock();
1596 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
1597
1598 /* Inode not in memory, nothing to do */
1599 if (!ip) {
1600 rcu_read_unlock();
1601 return;
1602 }
1603
1604 /*
1605 * because this is an RCU protected lookup, we could find a recently
1606 * freed or even reallocated inode during the lookup. We need to check
1607 * under the i_flags_lock for a valid inode here. Skip it if it is not
1608 * valid, the wrong inode or stale.
1609 */
1610 spin_lock(&ip->i_flags_lock);
1611 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
1612 goto out_iflags_unlock;
1613
1614 /*
1615 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
1616 * other inodes that we did not find in the list attached to the buffer
1617 * and are not already marked stale. If we can't lock it, back off and
1618 * retry.
1619 */
1620 if (ip != free_ip) {
1621 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1622 spin_unlock(&ip->i_flags_lock);
1623 rcu_read_unlock();
1624 delay(1);
1625 goto retry;
1626 }
1627 }
1628 ip->i_flags |= XFS_ISTALE;
1629
1630 /*
1631 * If the inode is flushing, it is already attached to the buffer. All
1632 * we needed to do here is mark the inode stale so buffer IO completion
1633 * will remove it from the AIL.
1634 */
1635 iip = ip->i_itemp;
1636 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
1637 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
1638 ASSERT(iip->ili_last_fields);
1639 goto out_iunlock;
1640 }
1641
1642 /*
1643 * Inodes not attached to the buffer can be released immediately.
1644 * Everything else has to go through xfs_iflush_abort() on journal
1645 * commit as the flock synchronises removal of the inode from the
1646 * cluster buffer against inode reclaim.
1647 */
1648 if (!iip || list_empty(&iip->ili_item.li_bio_list))
1649 goto out_iunlock;
1650
1651 __xfs_iflags_set(ip, XFS_IFLUSHING);
1652 spin_unlock(&ip->i_flags_lock);
1653 rcu_read_unlock();
1654
1655 /* we have a dirty inode in memory that has not yet been flushed. */
1656 spin_lock(&iip->ili_lock);
1657 iip->ili_last_fields = iip->ili_fields;
1658 iip->ili_fields = 0;
1659 iip->ili_fsync_fields = 0;
1660 spin_unlock(&iip->ili_lock);
1661 ASSERT(iip->ili_last_fields);
1662
1663 if (ip != free_ip)
1664 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1665 return;
1666
1667out_iunlock:
1668 if (ip != free_ip)
1669 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1670out_iflags_unlock:
1671 spin_unlock(&ip->i_flags_lock);
1672 rcu_read_unlock();
1673}
1674
1675/*
1676 * A big issue when freeing the inode cluster is that we _cannot_ skip any
1677 * inodes that are in memory - they all must be marked stale and attached to
1678 * the cluster buffer.
1679 */
1680static int
1681xfs_ifree_cluster(
1682 struct xfs_trans *tp,
1683 struct xfs_perag *pag,
1684 struct xfs_inode *free_ip,
1685 struct xfs_icluster *xic)
1686{
1687 struct xfs_mount *mp = free_ip->i_mount;
1688 struct xfs_ino_geometry *igeo = M_IGEO(mp);
1689 struct xfs_buf *bp;
1690 xfs_daddr_t blkno;
1691 xfs_ino_t inum = xic->first_ino;
1692 int nbufs;
1693 int i, j;
1694 int ioffset;
1695 int error;
1696
1697 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
1698
1699 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
1700 /*
1701 * The allocation bitmap tells us which inodes of the chunk were
1702 * physically allocated. Skip the cluster if an inode falls into
1703 * a sparse region.
1704 */
1705 ioffset = inum - xic->first_ino;
1706 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
1707 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
1708 continue;
1709 }
1710
1711 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1712 XFS_INO_TO_AGBNO(mp, inum));
1713
1714 /*
1715 * We obtain and lock the backing buffer first in the process
1716 * here to ensure dirty inodes attached to the buffer remain in
1717 * the flushing state while we mark them stale.
1718 *
1719 * If we scan the in-memory inodes first, then buffer IO can
1720 * complete before we get a lock on it, and hence we may fail
1721 * to mark all the active inodes on the buffer stale.
1722 */
1723 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1724 mp->m_bsize * igeo->blocks_per_cluster,
1725 XBF_UNMAPPED, &bp);
1726 if (error)
1727 return error;
1728
1729 /*
1730 * This buffer may not have been correctly initialised as we
1731 * didn't read it from disk. That's not important because we are
1732 * only using to mark the buffer as stale in the log, and to
1733 * attach stale cached inodes on it.
1734 *
1735 * For the inode that triggered the cluster freeing, this
1736 * attachment may occur in xfs_inode_item_precommit() after we
1737 * have marked this buffer stale. If this buffer was not in
1738 * memory before xfs_ifree_cluster() started, it will not be
1739 * marked XBF_DONE and this will cause problems later in
1740 * xfs_inode_item_precommit() when we trip over a (stale, !done)
1741 * buffer to attached to the transaction.
1742 *
1743 * Hence we have to mark the buffer as XFS_DONE here. This is
1744 * safe because we are also marking the buffer as XBF_STALE and
1745 * XFS_BLI_STALE. That means it will never be dispatched for
1746 * IO and it won't be unlocked until the cluster freeing has
1747 * been committed to the journal and the buffer unpinned. If it
1748 * is written, we want to know about it, and we want it to
1749 * fail. We can acheive this by adding a write verifier to the
1750 * buffer.
1751 */
1752 bp->b_flags |= XBF_DONE;
1753 bp->b_ops = &xfs_inode_buf_ops;
1754
1755 /*
1756 * Now we need to set all the cached clean inodes as XFS_ISTALE,
1757 * too. This requires lookups, and will skip inodes that we've
1758 * already marked XFS_ISTALE.
1759 */
1760 for (i = 0; i < igeo->inodes_per_cluster; i++)
1761 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
1762
1763 xfs_trans_stale_inode_buf(tp, bp);
1764 xfs_trans_binval(tp, bp);
1765 }
1766 return 0;
1767}
1768
1769/*
1770 * This is called to return an inode to the inode free list. The inode should
1771 * already be truncated to 0 length and have no pages associated with it. This
1772 * routine also assumes that the inode is already a part of the transaction.
1773 *
1774 * The on-disk copy of the inode will have been added to the list of unlinked
1775 * inodes in the AGI. We need to remove the inode from that list atomically with
1776 * respect to freeing it here.
1777 */
1778int
1779xfs_ifree(
1780 struct xfs_trans *tp,
1781 struct xfs_inode *ip)
1782{
1783 struct xfs_mount *mp = ip->i_mount;
1784 struct xfs_perag *pag;
1785 struct xfs_icluster xic = { 0 };
1786 struct xfs_inode_log_item *iip = ip->i_itemp;
1787 int error;
1788
1789 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1790 ASSERT(VFS_I(ip)->i_nlink == 0);
1791 ASSERT(ip->i_df.if_nextents == 0);
1792 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
1793 ASSERT(ip->i_nblocks == 0);
1794
1795 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1796
1797 error = xfs_inode_uninit(tp, pag, ip, &xic);
1798 if (error)
1799 goto out;
1800
1801 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
1802 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
1803
1804 /* Don't attempt to replay owner changes for a deleted inode */
1805 spin_lock(&iip->ili_lock);
1806 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
1807 spin_unlock(&iip->ili_lock);
1808
1809 if (xic.deleted)
1810 error = xfs_ifree_cluster(tp, pag, ip, &xic);
1811out:
1812 xfs_perag_put(pag);
1813 return error;
1814}
1815
1816/*
1817 * This is called to unpin an inode. The caller must have the inode locked
1818 * in at least shared mode so that the buffer cannot be subsequently pinned
1819 * once someone is waiting for it to be unpinned.
1820 */
1821static void
1822xfs_iunpin(
1823 struct xfs_inode *ip)
1824{
1825 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
1826
1827 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
1828
1829 /* Give the log a push to start the unpinning I/O */
1830 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
1831
1832}
1833
1834static void
1835__xfs_iunpin_wait(
1836 struct xfs_inode *ip)
1837{
1838 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
1839 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
1840
1841 xfs_iunpin(ip);
1842
1843 do {
1844 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1845 if (xfs_ipincount(ip))
1846 io_schedule();
1847 } while (xfs_ipincount(ip));
1848 finish_wait(wq, &wait.wq_entry);
1849}
1850
1851void
1852xfs_iunpin_wait(
1853 struct xfs_inode *ip)
1854{
1855 if (xfs_ipincount(ip))
1856 __xfs_iunpin_wait(ip);
1857}
1858
1859/*
1860 * Removing an inode from the namespace involves removing the directory entry
1861 * and dropping the link count on the inode. Removing the directory entry can
1862 * result in locking an AGF (directory blocks were freed) and removing a link
1863 * count can result in placing the inode on an unlinked list which results in
1864 * locking an AGI.
1865 *
1866 * The big problem here is that we have an ordering constraint on AGF and AGI
1867 * locking - inode allocation locks the AGI, then can allocate a new extent for
1868 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
1869 * removes the inode from the unlinked list, requiring that we lock the AGI
1870 * first, and then freeing the inode can result in an inode chunk being freed
1871 * and hence freeing disk space requiring that we lock an AGF.
1872 *
1873 * Hence the ordering that is imposed by other parts of the code is AGI before
1874 * AGF. This means we cannot remove the directory entry before we drop the inode
1875 * reference count and put it on the unlinked list as this results in a lock
1876 * order of AGF then AGI, and this can deadlock against inode allocation and
1877 * freeing. Therefore we must drop the link counts before we remove the
1878 * directory entry.
1879 *
1880 * This is still safe from a transactional point of view - it is not until we
1881 * get to xfs_defer_finish() that we have the possibility of multiple
1882 * transactions in this operation. Hence as long as we remove the directory
1883 * entry and drop the link count in the first transaction of the remove
1884 * operation, there are no transactional constraints on the ordering here.
1885 */
1886int
1887xfs_remove(
1888 struct xfs_inode *dp,
1889 struct xfs_name *name,
1890 struct xfs_inode *ip)
1891{
1892 struct xfs_dir_update du = {
1893 .dp = dp,
1894 .name = name,
1895 .ip = ip,
1896 };
1897 struct xfs_mount *mp = dp->i_mount;
1898 struct xfs_trans *tp = NULL;
1899 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
1900 int dontcare;
1901 int error = 0;
1902 uint resblks;
1903
1904 trace_xfs_remove(dp, name);
1905
1906 if (xfs_is_shutdown(mp))
1907 return -EIO;
1908 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
1909 return -EIO;
1910
1911 error = xfs_qm_dqattach(dp);
1912 if (error)
1913 goto std_return;
1914
1915 error = xfs_qm_dqattach(ip);
1916 if (error)
1917 goto std_return;
1918
1919 error = xfs_parent_start(mp, &du.ppargs);
1920 if (error)
1921 goto std_return;
1922
1923 /*
1924 * We try to get the real space reservation first, allowing for
1925 * directory btree deletion(s) implying possible bmap insert(s). If we
1926 * can't get the space reservation then we use 0 instead, and avoid the
1927 * bmap btree insert(s) in the directory code by, if the bmap insert
1928 * tries to happen, instead trimming the LAST block from the directory.
1929 *
1930 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
1931 * the directory code can handle a reservationless update and we don't
1932 * want to prevent a user from trying to free space by deleting things.
1933 */
1934 resblks = xfs_remove_space_res(mp, name->len);
1935 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
1936 &tp, &dontcare);
1937 if (error) {
1938 ASSERT(error != -ENOSPC);
1939 goto out_parent;
1940 }
1941
1942 error = xfs_dir_remove_child(tp, resblks, &du);
1943 if (error)
1944 goto out_trans_cancel;
1945
1946 /*
1947 * If this is a synchronous mount, make sure that the
1948 * remove transaction goes to disk before returning to
1949 * the user.
1950 */
1951 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1952 xfs_trans_set_sync(tp);
1953
1954 error = xfs_trans_commit(tp);
1955 if (error)
1956 goto out_unlock;
1957
1958 if (is_dir && xfs_inode_is_filestream(ip))
1959 xfs_filestream_deassociate(ip);
1960
1961 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1962 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1963 xfs_parent_finish(mp, du.ppargs);
1964 return 0;
1965
1966 out_trans_cancel:
1967 xfs_trans_cancel(tp);
1968 out_unlock:
1969 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1970 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1971 out_parent:
1972 xfs_parent_finish(mp, du.ppargs);
1973 std_return:
1974 return error;
1975}
1976
1977static inline void
1978xfs_iunlock_rename(
1979 struct xfs_inode **i_tab,
1980 int num_inodes)
1981{
1982 int i;
1983
1984 for (i = num_inodes - 1; i >= 0; i--) {
1985 /* Skip duplicate inodes if src and target dps are the same */
1986 if (!i_tab[i] || (i > 0 && i_tab[i] == i_tab[i - 1]))
1987 continue;
1988 xfs_iunlock(i_tab[i], XFS_ILOCK_EXCL);
1989 }
1990}
1991
1992/*
1993 * Enter all inodes for a rename transaction into a sorted array.
1994 */
1995#define __XFS_SORT_INODES 5
1996STATIC void
1997xfs_sort_for_rename(
1998 struct xfs_inode *dp1, /* in: old (source) directory inode */
1999 struct xfs_inode *dp2, /* in: new (target) directory inode */
2000 struct xfs_inode *ip1, /* in: inode of old entry */
2001 struct xfs_inode *ip2, /* in: inode of new entry */
2002 struct xfs_inode *wip, /* in: whiteout inode */
2003 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2004 int *num_inodes) /* in/out: inodes in array */
2005{
2006 int i;
2007
2008 ASSERT(*num_inodes == __XFS_SORT_INODES);
2009 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2010
2011 /*
2012 * i_tab contains a list of pointers to inodes. We initialize
2013 * the table here & we'll sort it. We will then use it to
2014 * order the acquisition of the inode locks.
2015 *
2016 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2017 */
2018 i = 0;
2019 i_tab[i++] = dp1;
2020 i_tab[i++] = dp2;
2021 i_tab[i++] = ip1;
2022 if (ip2)
2023 i_tab[i++] = ip2;
2024 if (wip)
2025 i_tab[i++] = wip;
2026 *num_inodes = i;
2027
2028 xfs_sort_inodes(i_tab, *num_inodes);
2029}
2030
2031void
2032xfs_sort_inodes(
2033 struct xfs_inode **i_tab,
2034 unsigned int num_inodes)
2035{
2036 int i, j;
2037
2038 ASSERT(num_inodes <= __XFS_SORT_INODES);
2039
2040 /*
2041 * Sort the elements via bubble sort. (Remember, there are at
2042 * most 5 elements to sort, so this is adequate.)
2043 */
2044 for (i = 0; i < num_inodes; i++) {
2045 for (j = 1; j < num_inodes; j++) {
2046 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino)
2047 swap(i_tab[j], i_tab[j - 1]);
2048 }
2049 }
2050}
2051
2052/*
2053 * xfs_rename_alloc_whiteout()
2054 *
2055 * Return a referenced, unlinked, unlocked inode that can be used as a
2056 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2057 * crash between allocating the inode and linking it into the rename transaction
2058 * recovery will free the inode and we won't leak it.
2059 */
2060static int
2061xfs_rename_alloc_whiteout(
2062 struct mnt_idmap *idmap,
2063 struct xfs_name *src_name,
2064 struct xfs_inode *dp,
2065 struct xfs_inode **wip)
2066{
2067 struct xfs_icreate_args args = {
2068 .idmap = idmap,
2069 .pip = dp,
2070 .mode = S_IFCHR | WHITEOUT_MODE,
2071 .flags = XFS_ICREATE_TMPFILE,
2072 };
2073 struct xfs_inode *tmpfile;
2074 struct qstr name;
2075 int error;
2076
2077 error = xfs_create_tmpfile(&args, &tmpfile);
2078 if (error)
2079 return error;
2080
2081 name.name = src_name->name;
2082 name.len = src_name->len;
2083 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2084 if (error) {
2085 xfs_finish_inode_setup(tmpfile);
2086 xfs_irele(tmpfile);
2087 return error;
2088 }
2089
2090 /*
2091 * Prepare the tmpfile inode as if it were created through the VFS.
2092 * Complete the inode setup and flag it as linkable. nlink is already
2093 * zero, so we can skip the drop_nlink.
2094 */
2095 xfs_setup_iops(tmpfile);
2096 xfs_finish_inode_setup(tmpfile);
2097 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2098
2099 *wip = tmpfile;
2100 return 0;
2101}
2102
2103/*
2104 * xfs_rename
2105 */
2106int
2107xfs_rename(
2108 struct mnt_idmap *idmap,
2109 struct xfs_inode *src_dp,
2110 struct xfs_name *src_name,
2111 struct xfs_inode *src_ip,
2112 struct xfs_inode *target_dp,
2113 struct xfs_name *target_name,
2114 struct xfs_inode *target_ip,
2115 unsigned int flags)
2116{
2117 struct xfs_dir_update du_src = {
2118 .dp = src_dp,
2119 .name = src_name,
2120 .ip = src_ip,
2121 };
2122 struct xfs_dir_update du_tgt = {
2123 .dp = target_dp,
2124 .name = target_name,
2125 .ip = target_ip,
2126 };
2127 struct xfs_dir_update du_wip = { };
2128 struct xfs_mount *mp = src_dp->i_mount;
2129 struct xfs_trans *tp;
2130 struct xfs_inode *inodes[__XFS_SORT_INODES];
2131 int i;
2132 int num_inodes = __XFS_SORT_INODES;
2133 bool new_parent = (src_dp != target_dp);
2134 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2135 int spaceres;
2136 bool retried = false;
2137 int error, nospace_error = 0;
2138
2139 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2140
2141 if ((flags & RENAME_EXCHANGE) && !target_ip)
2142 return -EINVAL;
2143
2144 /*
2145 * If we are doing a whiteout operation, allocate the whiteout inode
2146 * we will be placing at the target and ensure the type is set
2147 * appropriately.
2148 */
2149 if (flags & RENAME_WHITEOUT) {
2150 error = xfs_rename_alloc_whiteout(idmap, src_name, target_dp,
2151 &du_wip.ip);
2152 if (error)
2153 return error;
2154
2155 /* setup target dirent info as whiteout */
2156 src_name->type = XFS_DIR3_FT_CHRDEV;
2157 }
2158
2159 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, du_wip.ip,
2160 inodes, &num_inodes);
2161
2162 error = xfs_parent_start(mp, &du_src.ppargs);
2163 if (error)
2164 goto out_release_wip;
2165
2166 if (du_wip.ip) {
2167 error = xfs_parent_start(mp, &du_wip.ppargs);
2168 if (error)
2169 goto out_src_ppargs;
2170 }
2171
2172 if (target_ip) {
2173 error = xfs_parent_start(mp, &du_tgt.ppargs);
2174 if (error)
2175 goto out_wip_ppargs;
2176 }
2177
2178retry:
2179 nospace_error = 0;
2180 spaceres = xfs_rename_space_res(mp, src_name->len, target_ip != NULL,
2181 target_name->len, du_wip.ip != NULL);
2182 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2183 if (error == -ENOSPC) {
2184 nospace_error = error;
2185 spaceres = 0;
2186 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2187 &tp);
2188 }
2189 if (error)
2190 goto out_tgt_ppargs;
2191
2192 /*
2193 * We don't allow reservationless renaming when parent pointers are
2194 * enabled because we can't back out if the xattrs must grow.
2195 */
2196 if (du_src.ppargs && nospace_error) {
2197 error = nospace_error;
2198 xfs_trans_cancel(tp);
2199 goto out_tgt_ppargs;
2200 }
2201
2202 /*
2203 * Attach the dquots to the inodes
2204 */
2205 error = xfs_qm_vop_rename_dqattach(inodes);
2206 if (error) {
2207 xfs_trans_cancel(tp);
2208 goto out_tgt_ppargs;
2209 }
2210
2211 /*
2212 * Lock all the participating inodes. Depending upon whether
2213 * the target_name exists in the target directory, and
2214 * whether the target directory is the same as the source
2215 * directory, we can lock from 2 to 5 inodes.
2216 */
2217 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2218
2219 /*
2220 * Join all the inodes to the transaction.
2221 */
2222 xfs_trans_ijoin(tp, src_dp, 0);
2223 if (new_parent)
2224 xfs_trans_ijoin(tp, target_dp, 0);
2225 xfs_trans_ijoin(tp, src_ip, 0);
2226 if (target_ip)
2227 xfs_trans_ijoin(tp, target_ip, 0);
2228 if (du_wip.ip)
2229 xfs_trans_ijoin(tp, du_wip.ip, 0);
2230
2231 /*
2232 * If we are using project inheritance, we only allow renames
2233 * into our tree when the project IDs are the same; else the
2234 * tree quota mechanism would be circumvented.
2235 */
2236 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2237 target_dp->i_projid != src_ip->i_projid)) {
2238 error = -EXDEV;
2239 goto out_trans_cancel;
2240 }
2241
2242 /* RENAME_EXCHANGE is unique from here on. */
2243 if (flags & RENAME_EXCHANGE) {
2244 error = xfs_dir_exchange_children(tp, &du_src, &du_tgt,
2245 spaceres);
2246 if (error)
2247 goto out_trans_cancel;
2248 goto out_commit;
2249 }
2250
2251 /*
2252 * Try to reserve quota to handle an expansion of the target directory.
2253 * We'll allow the rename to continue in reservationless mode if we hit
2254 * a space usage constraint. If we trigger reservationless mode, save
2255 * the errno if there isn't any free space in the target directory.
2256 */
2257 if (spaceres != 0) {
2258 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2259 0, false);
2260 if (error == -EDQUOT || error == -ENOSPC) {
2261 if (!retried) {
2262 xfs_trans_cancel(tp);
2263 xfs_iunlock_rename(inodes, num_inodes);
2264 xfs_blockgc_free_quota(target_dp, 0);
2265 retried = true;
2266 goto retry;
2267 }
2268
2269 nospace_error = error;
2270 spaceres = 0;
2271 error = 0;
2272 }
2273 if (error)
2274 goto out_trans_cancel;
2275 }
2276
2277 /*
2278 * We don't allow quotaless renaming when parent pointers are enabled
2279 * because we can't back out if the xattrs must grow.
2280 */
2281 if (du_src.ppargs && nospace_error) {
2282 error = nospace_error;
2283 goto out_trans_cancel;
2284 }
2285
2286 /*
2287 * Lock the AGI buffers we need to handle bumping the nlink of the
2288 * whiteout inode off the unlinked list and to handle dropping the
2289 * nlink of the target inode. Per locking order rules, do this in
2290 * increasing AG order and before directory block allocation tries to
2291 * grab AGFs because we grab AGIs before AGFs.
2292 *
2293 * The (vfs) caller must ensure that if src is a directory then
2294 * target_ip is either null or an empty directory.
2295 */
2296 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2297 if (inodes[i] == du_wip.ip ||
2298 (inodes[i] == target_ip &&
2299 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2300 struct xfs_perag *pag;
2301 struct xfs_buf *bp;
2302
2303 pag = xfs_perag_get(mp,
2304 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2305 error = xfs_read_agi(pag, tp, 0, &bp);
2306 xfs_perag_put(pag);
2307 if (error)
2308 goto out_trans_cancel;
2309 }
2310 }
2311
2312 error = xfs_dir_rename_children(tp, &du_src, &du_tgt, spaceres,
2313 &du_wip);
2314 if (error)
2315 goto out_trans_cancel;
2316
2317 if (du_wip.ip) {
2318 /*
2319 * Now we have a real link, clear the "I'm a tmpfile" state
2320 * flag from the inode so it doesn't accidentally get misused in
2321 * future.
2322 */
2323 VFS_I(du_wip.ip)->i_state &= ~I_LINKABLE;
2324 }
2325
2326out_commit:
2327 /*
2328 * If this is a synchronous mount, make sure that the rename
2329 * transaction goes to disk before returning to the user.
2330 */
2331 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2332 xfs_trans_set_sync(tp);
2333
2334 error = xfs_trans_commit(tp);
2335 nospace_error = 0;
2336 goto out_unlock;
2337
2338out_trans_cancel:
2339 xfs_trans_cancel(tp);
2340out_unlock:
2341 xfs_iunlock_rename(inodes, num_inodes);
2342out_tgt_ppargs:
2343 xfs_parent_finish(mp, du_tgt.ppargs);
2344out_wip_ppargs:
2345 xfs_parent_finish(mp, du_wip.ppargs);
2346out_src_ppargs:
2347 xfs_parent_finish(mp, du_src.ppargs);
2348out_release_wip:
2349 if (du_wip.ip)
2350 xfs_irele(du_wip.ip);
2351 if (error == -ENOSPC && nospace_error)
2352 error = nospace_error;
2353 return error;
2354}
2355
2356static int
2357xfs_iflush(
2358 struct xfs_inode *ip,
2359 struct xfs_buf *bp)
2360{
2361 struct xfs_inode_log_item *iip = ip->i_itemp;
2362 struct xfs_dinode *dip;
2363 struct xfs_mount *mp = ip->i_mount;
2364 int error;
2365
2366 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
2367 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
2368 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
2369 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2370 ASSERT(iip->ili_item.li_buf == bp);
2371
2372 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
2373
2374 /*
2375 * We don't flush the inode if any of the following checks fail, but we
2376 * do still update the log item and attach to the backing buffer as if
2377 * the flush happened. This is a formality to facilitate predictable
2378 * error handling as the caller will shutdown and fail the buffer.
2379 */
2380 error = -EFSCORRUPTED;
2381 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2382 mp, XFS_ERRTAG_IFLUSH_1)) {
2383 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2384 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
2385 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2386 goto flush_out;
2387 }
2388 if (S_ISREG(VFS_I(ip)->i_mode)) {
2389 if (XFS_TEST_ERROR(
2390 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
2391 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
2392 mp, XFS_ERRTAG_IFLUSH_3)) {
2393 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2394 "%s: Bad regular inode %llu, ptr "PTR_FMT,
2395 __func__, ip->i_ino, ip);
2396 goto flush_out;
2397 }
2398 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
2399 if (XFS_TEST_ERROR(
2400 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
2401 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
2402 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
2403 mp, XFS_ERRTAG_IFLUSH_4)) {
2404 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2405 "%s: Bad directory inode %llu, ptr "PTR_FMT,
2406 __func__, ip->i_ino, ip);
2407 goto flush_out;
2408 }
2409 }
2410 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
2411 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
2412 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2413 "%s: detected corrupt incore inode %llu, "
2414 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
2415 __func__, ip->i_ino,
2416 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
2417 ip->i_nblocks, ip);
2418 goto flush_out;
2419 }
2420 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
2421 mp, XFS_ERRTAG_IFLUSH_6)) {
2422 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2423 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
2424 __func__, ip->i_ino, ip->i_forkoff, ip);
2425 goto flush_out;
2426 }
2427
2428 /*
2429 * Inode item log recovery for v2 inodes are dependent on the flushiter
2430 * count for correct sequencing. We bump the flush iteration count so
2431 * we can detect flushes which postdate a log record during recovery.
2432 * This is redundant as we now log every change and hence this can't
2433 * happen but we need to still do it to ensure backwards compatibility
2434 * with old kernels that predate logging all inode changes.
2435 */
2436 if (!xfs_has_v3inodes(mp))
2437 ip->i_flushiter++;
2438
2439 /*
2440 * If there are inline format data / attr forks attached to this inode,
2441 * make sure they are not corrupt.
2442 */
2443 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
2444 xfs_ifork_verify_local_data(ip))
2445 goto flush_out;
2446 if (xfs_inode_has_attr_fork(ip) &&
2447 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
2448 xfs_ifork_verify_local_attr(ip))
2449 goto flush_out;
2450
2451 /*
2452 * Copy the dirty parts of the inode into the on-disk inode. We always
2453 * copy out the core of the inode, because if the inode is dirty at all
2454 * the core must be.
2455 */
2456 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
2457
2458 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2459 if (!xfs_has_v3inodes(mp)) {
2460 if (ip->i_flushiter == DI_MAX_FLUSH)
2461 ip->i_flushiter = 0;
2462 }
2463
2464 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
2465 if (xfs_inode_has_attr_fork(ip))
2466 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
2467
2468 /*
2469 * We've recorded everything logged in the inode, so we'd like to clear
2470 * the ili_fields bits so we don't log and flush things unnecessarily.
2471 * However, we can't stop logging all this information until the data
2472 * we've copied into the disk buffer is written to disk. If we did we
2473 * might overwrite the copy of the inode in the log with all the data
2474 * after re-logging only part of it, and in the face of a crash we
2475 * wouldn't have all the data we need to recover.
2476 *
2477 * What we do is move the bits to the ili_last_fields field. When
2478 * logging the inode, these bits are moved back to the ili_fields field.
2479 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
2480 * we know that the information those bits represent is permanently on
2481 * disk. As long as the flush completes before the inode is logged
2482 * again, then both ili_fields and ili_last_fields will be cleared.
2483 */
2484 error = 0;
2485flush_out:
2486 spin_lock(&iip->ili_lock);
2487 iip->ili_last_fields = iip->ili_fields;
2488 iip->ili_fields = 0;
2489 iip->ili_fsync_fields = 0;
2490 set_bit(XFS_LI_FLUSHING, &iip->ili_item.li_flags);
2491 spin_unlock(&iip->ili_lock);
2492
2493 /*
2494 * Store the current LSN of the inode so that we can tell whether the
2495 * item has moved in the AIL from xfs_buf_inode_iodone().
2496 */
2497 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2498 &iip->ili_item.li_lsn);
2499
2500 /* generate the checksum. */
2501 xfs_dinode_calc_crc(mp, dip);
2502 if (error)
2503 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
2504 return error;
2505}
2506
2507/*
2508 * Non-blocking flush of dirty inode metadata into the backing buffer.
2509 *
2510 * The caller must have a reference to the inode and hold the cluster buffer
2511 * locked. The function will walk across all the inodes on the cluster buffer it
2512 * can find and lock without blocking, and flush them to the cluster buffer.
2513 *
2514 * On successful flushing of at least one inode, the caller must write out the
2515 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
2516 * the caller needs to release the buffer. On failure, the filesystem will be
2517 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
2518 * will be returned.
2519 */
2520int
2521xfs_iflush_cluster(
2522 struct xfs_buf *bp)
2523{
2524 struct xfs_mount *mp = bp->b_mount;
2525 struct xfs_log_item *lip, *n;
2526 struct xfs_inode *ip;
2527 struct xfs_inode_log_item *iip;
2528 int clcount = 0;
2529 int error = 0;
2530
2531 /*
2532 * We must use the safe variant here as on shutdown xfs_iflush_abort()
2533 * will remove itself from the list.
2534 */
2535 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
2536 iip = (struct xfs_inode_log_item *)lip;
2537 ip = iip->ili_inode;
2538
2539 /*
2540 * Quick and dirty check to avoid locks if possible.
2541 */
2542 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
2543 continue;
2544 if (xfs_ipincount(ip))
2545 continue;
2546
2547 /*
2548 * The inode is still attached to the buffer, which means it is
2549 * dirty but reclaim might try to grab it. Check carefully for
2550 * that, and grab the ilock while still holding the i_flags_lock
2551 * to guarantee reclaim will not be able to reclaim this inode
2552 * once we drop the i_flags_lock.
2553 */
2554 spin_lock(&ip->i_flags_lock);
2555 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
2556 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
2557 spin_unlock(&ip->i_flags_lock);
2558 continue;
2559 }
2560
2561 /*
2562 * ILOCK will pin the inode against reclaim and prevent
2563 * concurrent transactions modifying the inode while we are
2564 * flushing the inode. If we get the lock, set the flushing
2565 * state before we drop the i_flags_lock.
2566 */
2567 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
2568 spin_unlock(&ip->i_flags_lock);
2569 continue;
2570 }
2571 __xfs_iflags_set(ip, XFS_IFLUSHING);
2572 spin_unlock(&ip->i_flags_lock);
2573
2574 /*
2575 * Abort flushing this inode if we are shut down because the
2576 * inode may not currently be in the AIL. This can occur when
2577 * log I/O failure unpins the inode without inserting into the
2578 * AIL, leaving a dirty/unpinned inode attached to the buffer
2579 * that otherwise looks like it should be flushed.
2580 */
2581 if (xlog_is_shutdown(mp->m_log)) {
2582 xfs_iunpin_wait(ip);
2583 xfs_iflush_abort(ip);
2584 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2585 error = -EIO;
2586 continue;
2587 }
2588
2589 /* don't block waiting on a log force to unpin dirty inodes */
2590 if (xfs_ipincount(ip)) {
2591 xfs_iflags_clear(ip, XFS_IFLUSHING);
2592 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2593 continue;
2594 }
2595
2596 if (!xfs_inode_clean(ip))
2597 error = xfs_iflush(ip, bp);
2598 else
2599 xfs_iflags_clear(ip, XFS_IFLUSHING);
2600 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2601 if (error)
2602 break;
2603 clcount++;
2604 }
2605
2606 if (error) {
2607 /*
2608 * Shutdown first so we kill the log before we release this
2609 * buffer. If it is an INODE_ALLOC buffer and pins the tail
2610 * of the log, failing it before the _log_ is shut down can
2611 * result in the log tail being moved forward in the journal
2612 * on disk because log writes can still be taking place. Hence
2613 * unpinning the tail will allow the ICREATE intent to be
2614 * removed from the log an recovery will fail with uninitialised
2615 * inode cluster buffers.
2616 */
2617 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2618 bp->b_flags |= XBF_ASYNC;
2619 xfs_buf_ioend_fail(bp);
2620 return error;
2621 }
2622
2623 if (!clcount)
2624 return -EAGAIN;
2625
2626 XFS_STATS_INC(mp, xs_icluster_flushcnt);
2627 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
2628 return 0;
2629
2630}
2631
2632/* Release an inode. */
2633void
2634xfs_irele(
2635 struct xfs_inode *ip)
2636{
2637 trace_xfs_irele(ip, _RET_IP_);
2638 iput(VFS_I(ip));
2639}
2640
2641/*
2642 * Ensure all commited transactions touching the inode are written to the log.
2643 */
2644int
2645xfs_log_force_inode(
2646 struct xfs_inode *ip)
2647{
2648 xfs_csn_t seq = 0;
2649
2650 xfs_ilock(ip, XFS_ILOCK_SHARED);
2651 if (xfs_ipincount(ip))
2652 seq = ip->i_itemp->ili_commit_seq;
2653 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2654
2655 if (!seq)
2656 return 0;
2657 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
2658}
2659
2660/*
2661 * Grab the exclusive iolock for a data copy from src to dest, making sure to
2662 * abide vfs locking order (lowest pointer value goes first) and breaking the
2663 * layout leases before proceeding. The loop is needed because we cannot call
2664 * the blocking break_layout() with the iolocks held, and therefore have to
2665 * back out both locks.
2666 */
2667static int
2668xfs_iolock_two_inodes_and_break_layout(
2669 struct inode *src,
2670 struct inode *dest)
2671{
2672 int error;
2673
2674 if (src > dest)
2675 swap(src, dest);
2676
2677retry:
2678 /* Wait to break both inodes' layouts before we start locking. */
2679 error = break_layout(src, true);
2680 if (error)
2681 return error;
2682 if (src != dest) {
2683 error = break_layout(dest, true);
2684 if (error)
2685 return error;
2686 }
2687
2688 /* Lock one inode and make sure nobody got in and leased it. */
2689 inode_lock(src);
2690 error = break_layout(src, false);
2691 if (error) {
2692 inode_unlock(src);
2693 if (error == -EWOULDBLOCK)
2694 goto retry;
2695 return error;
2696 }
2697
2698 if (src == dest)
2699 return 0;
2700
2701 /* Lock the other inode and make sure nobody got in and leased it. */
2702 inode_lock_nested(dest, I_MUTEX_NONDIR2);
2703 error = break_layout(dest, false);
2704 if (error) {
2705 inode_unlock(src);
2706 inode_unlock(dest);
2707 if (error == -EWOULDBLOCK)
2708 goto retry;
2709 return error;
2710 }
2711
2712 return 0;
2713}
2714
2715static int
2716xfs_mmaplock_two_inodes_and_break_dax_layout(
2717 struct xfs_inode *ip1,
2718 struct xfs_inode *ip2)
2719{
2720 int error;
2721 bool retry;
2722 struct page *page;
2723
2724 if (ip1->i_ino > ip2->i_ino)
2725 swap(ip1, ip2);
2726
2727again:
2728 retry = false;
2729 /* Lock the first inode */
2730 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
2731 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
2732 if (error || retry) {
2733 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
2734 if (error == 0 && retry)
2735 goto again;
2736 return error;
2737 }
2738
2739 if (ip1 == ip2)
2740 return 0;
2741
2742 /* Nested lock the second inode */
2743 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
2744 /*
2745 * We cannot use xfs_break_dax_layouts() directly here because it may
2746 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
2747 * for this nested lock case.
2748 */
2749 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
2750 if (page && page_ref_count(page) != 1) {
2751 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
2752 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
2753 goto again;
2754 }
2755
2756 return 0;
2757}
2758
2759/*
2760 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
2761 * mmap activity.
2762 */
2763int
2764xfs_ilock2_io_mmap(
2765 struct xfs_inode *ip1,
2766 struct xfs_inode *ip2)
2767{
2768 int ret;
2769
2770 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
2771 if (ret)
2772 return ret;
2773
2774 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
2775 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
2776 if (ret) {
2777 inode_unlock(VFS_I(ip2));
2778 if (ip1 != ip2)
2779 inode_unlock(VFS_I(ip1));
2780 return ret;
2781 }
2782 } else
2783 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
2784 VFS_I(ip2)->i_mapping);
2785
2786 return 0;
2787}
2788
2789/* Unlock both inodes to allow IO and mmap activity. */
2790void
2791xfs_iunlock2_io_mmap(
2792 struct xfs_inode *ip1,
2793 struct xfs_inode *ip2)
2794{
2795 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
2796 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
2797 if (ip1 != ip2)
2798 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
2799 } else
2800 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
2801 VFS_I(ip2)->i_mapping);
2802
2803 inode_unlock(VFS_I(ip2));
2804 if (ip1 != ip2)
2805 inode_unlock(VFS_I(ip1));
2806}
2807
2808/* Drop the MMAPLOCK and the IOLOCK after a remap completes. */
2809void
2810xfs_iunlock2_remapping(
2811 struct xfs_inode *ip1,
2812 struct xfs_inode *ip2)
2813{
2814 xfs_iflags_clear(ip1, XFS_IREMAPPING);
2815
2816 if (ip1 != ip2)
2817 xfs_iunlock(ip1, XFS_MMAPLOCK_SHARED);
2818 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
2819
2820 if (ip1 != ip2)
2821 inode_unlock_shared(VFS_I(ip1));
2822 inode_unlock(VFS_I(ip2));
2823}
2824
2825/*
2826 * Reload the incore inode list for this inode. Caller should ensure that
2827 * the link count cannot change, either by taking ILOCK_SHARED or otherwise
2828 * preventing other threads from executing.
2829 */
2830int
2831xfs_inode_reload_unlinked_bucket(
2832 struct xfs_trans *tp,
2833 struct xfs_inode *ip)
2834{
2835 struct xfs_mount *mp = tp->t_mountp;
2836 struct xfs_buf *agibp;
2837 struct xfs_agi *agi;
2838 struct xfs_perag *pag;
2839 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2840 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2841 xfs_agino_t prev_agino, next_agino;
2842 unsigned int bucket;
2843 bool foundit = false;
2844 int error;
2845
2846 /* Grab the first inode in the list */
2847 pag = xfs_perag_get(mp, agno);
2848 error = xfs_ialloc_read_agi(pag, tp, 0, &agibp);
2849 xfs_perag_put(pag);
2850 if (error)
2851 return error;
2852
2853 /*
2854 * We've taken ILOCK_SHARED and the AGI buffer lock to stabilize the
2855 * incore unlinked list pointers for this inode. Check once more to
2856 * see if we raced with anyone else to reload the unlinked list.
2857 */
2858 if (!xfs_inode_unlinked_incomplete(ip)) {
2859 foundit = true;
2860 goto out_agibp;
2861 }
2862
2863 bucket = agino % XFS_AGI_UNLINKED_BUCKETS;
2864 agi = agibp->b_addr;
2865
2866 trace_xfs_inode_reload_unlinked_bucket(ip);
2867
2868 xfs_info_ratelimited(mp,
2869 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating list recovery.",
2870 agino, agno);
2871
2872 prev_agino = NULLAGINO;
2873 next_agino = be32_to_cpu(agi->agi_unlinked[bucket]);
2874 while (next_agino != NULLAGINO) {
2875 struct xfs_inode *next_ip = NULL;
2876
2877 /* Found this caller's inode, set its backlink. */
2878 if (next_agino == agino) {
2879 next_ip = ip;
2880 next_ip->i_prev_unlinked = prev_agino;
2881 foundit = true;
2882 goto next_inode;
2883 }
2884
2885 /* Try in-memory lookup first. */
2886 next_ip = xfs_iunlink_lookup(pag, next_agino);
2887 if (next_ip)
2888 goto next_inode;
2889
2890 /* Inode not in memory, try reloading it. */
2891 error = xfs_iunlink_reload_next(tp, agibp, prev_agino,
2892 next_agino);
2893 if (error)
2894 break;
2895
2896 /* Grab the reloaded inode. */
2897 next_ip = xfs_iunlink_lookup(pag, next_agino);
2898 if (!next_ip) {
2899 /* No incore inode at all? We reloaded it... */
2900 ASSERT(next_ip != NULL);
2901 error = -EFSCORRUPTED;
2902 break;
2903 }
2904
2905next_inode:
2906 prev_agino = next_agino;
2907 next_agino = next_ip->i_next_unlinked;
2908 }
2909
2910out_agibp:
2911 xfs_trans_brelse(tp, agibp);
2912 /* Should have found this inode somewhere in the iunlinked bucket. */
2913 if (!error && !foundit)
2914 error = -EFSCORRUPTED;
2915 return error;
2916}
2917
2918/* Decide if this inode is missing its unlinked list and reload it. */
2919int
2920xfs_inode_reload_unlinked(
2921 struct xfs_inode *ip)
2922{
2923 struct xfs_trans *tp;
2924 int error;
2925
2926 error = xfs_trans_alloc_empty(ip->i_mount, &tp);
2927 if (error)
2928 return error;
2929
2930 xfs_ilock(ip, XFS_ILOCK_SHARED);
2931 if (xfs_inode_unlinked_incomplete(ip))
2932 error = xfs_inode_reload_unlinked_bucket(tp, ip);
2933 xfs_iunlock(ip, XFS_ILOCK_SHARED);
2934 xfs_trans_cancel(tp);
2935
2936 return error;
2937}
2938
2939/* Has this inode fork been zapped by repair? */
2940bool
2941xfs_ifork_zapped(
2942 const struct xfs_inode *ip,
2943 int whichfork)
2944{
2945 unsigned int datamask = 0;
2946
2947 switch (whichfork) {
2948 case XFS_DATA_FORK:
2949 switch (ip->i_vnode.i_mode & S_IFMT) {
2950 case S_IFDIR:
2951 datamask = XFS_SICK_INO_DIR_ZAPPED;
2952 break;
2953 case S_IFLNK:
2954 datamask = XFS_SICK_INO_SYMLINK_ZAPPED;
2955 break;
2956 }
2957 return ip->i_sick & (XFS_SICK_INO_BMBTD_ZAPPED | datamask);
2958 case XFS_ATTR_FORK:
2959 return ip->i_sick & XFS_SICK_INO_BMBTA_ZAPPED;
2960 default:
2961 return false;
2962 }
2963}
2964
2965/* Compute the number of data and realtime blocks used by a file. */
2966void
2967xfs_inode_count_blocks(
2968 struct xfs_trans *tp,
2969 struct xfs_inode *ip,
2970 xfs_filblks_t *dblocks,
2971 xfs_filblks_t *rblocks)
2972{
2973 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
2974
2975 *rblocks = 0;
2976 if (XFS_IS_REALTIME_INODE(ip))
2977 xfs_bmap_count_leaves(ifp, rblocks);
2978 *dblocks = ip->i_nblocks - *rblocks;
2979}
2980
2981static void
2982xfs_wait_dax_page(
2983 struct inode *inode)
2984{
2985 struct xfs_inode *ip = XFS_I(inode);
2986
2987 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
2988 schedule();
2989 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
2990}
2991
2992int
2993xfs_break_dax_layouts(
2994 struct inode *inode,
2995 bool *retry)
2996{
2997 struct page *page;
2998
2999 xfs_assert_ilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL);
3000
3001 page = dax_layout_busy_page(inode->i_mapping);
3002 if (!page)
3003 return 0;
3004
3005 *retry = true;
3006 return ___wait_var_event(&page->_refcount,
3007 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
3008 0, 0, xfs_wait_dax_page(inode));
3009}
3010
3011int
3012xfs_break_layouts(
3013 struct inode *inode,
3014 uint *iolock,
3015 enum layout_break_reason reason)
3016{
3017 bool retry;
3018 int error;
3019
3020 xfs_assert_ilocked(XFS_I(inode), XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL);
3021
3022 do {
3023 retry = false;
3024 switch (reason) {
3025 case BREAK_UNMAP:
3026 error = xfs_break_dax_layouts(inode, &retry);
3027 if (error || retry)
3028 break;
3029 fallthrough;
3030 case BREAK_WRITE:
3031 error = xfs_break_leased_layouts(inode, iolock, &retry);
3032 break;
3033 default:
3034 WARN_ON_ONCE(1);
3035 error = -EINVAL;
3036 }
3037 } while (error == 0 && retry);
3038
3039 return error;
3040}
3041
3042/* Returns the size of fundamental allocation unit for a file, in bytes. */
3043unsigned int
3044xfs_inode_alloc_unitsize(
3045 struct xfs_inode *ip)
3046{
3047 unsigned int blocks = 1;
3048
3049 if (XFS_IS_REALTIME_INODE(ip))
3050 blocks = ip->i_mount->m_sb.sb_rextsize;
3051
3052 return XFS_FSB_TO_B(ip->i_mount, blocks);
3053}
3054
3055/* Should we always be using copy on write for file writes? */
3056bool
3057xfs_is_always_cow_inode(
3058 const struct xfs_inode *ip)
3059{
3060 return ip->i_mount->m_always_cow && xfs_has_reflink(ip->i_mount);
3061}