Loading...
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_trans_space.h"
20#include "xfs_trans.h"
21#include "xfs_buf_item.h"
22#include "xfs_inode_item.h"
23#include "xfs_iunlink_item.h"
24#include "xfs_ialloc.h"
25#include "xfs_bmap.h"
26#include "xfs_bmap_util.h"
27#include "xfs_errortag.h"
28#include "xfs_error.h"
29#include "xfs_quota.h"
30#include "xfs_filestream.h"
31#include "xfs_trace.h"
32#include "xfs_icache.h"
33#include "xfs_symlink.h"
34#include "xfs_trans_priv.h"
35#include "xfs_log.h"
36#include "xfs_bmap_btree.h"
37#include "xfs_reflink.h"
38#include "xfs_ag.h"
39#include "xfs_log_priv.h"
40
41struct kmem_cache *xfs_inode_cache;
42
43/*
44 * Used in xfs_itruncate_extents(). This is the maximum number of extents
45 * freed from a file in a single transaction.
46 */
47#define XFS_ITRUNC_MAX_EXTENTS 2
48
49STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
50STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
51 struct xfs_inode *);
52
53/*
54 * helper function to extract extent size hint from inode
55 */
56xfs_extlen_t
57xfs_get_extsz_hint(
58 struct xfs_inode *ip)
59{
60 /*
61 * No point in aligning allocations if we need to COW to actually
62 * write to them.
63 */
64 if (xfs_is_always_cow_inode(ip))
65 return 0;
66 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
67 return ip->i_extsize;
68 if (XFS_IS_REALTIME_INODE(ip))
69 return ip->i_mount->m_sb.sb_rextsize;
70 return 0;
71}
72
73/*
74 * Helper function to extract CoW extent size hint from inode.
75 * Between the extent size hint and the CoW extent size hint, we
76 * return the greater of the two. If the value is zero (automatic),
77 * use the default size.
78 */
79xfs_extlen_t
80xfs_get_cowextsz_hint(
81 struct xfs_inode *ip)
82{
83 xfs_extlen_t a, b;
84
85 a = 0;
86 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
87 a = ip->i_cowextsize;
88 b = xfs_get_extsz_hint(ip);
89
90 a = max(a, b);
91 if (a == 0)
92 return XFS_DEFAULT_COWEXTSZ_HINT;
93 return a;
94}
95
96/*
97 * These two are wrapper routines around the xfs_ilock() routine used to
98 * centralize some grungy code. They are used in places that wish to lock the
99 * inode solely for reading the extents. The reason these places can't just
100 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
101 * bringing in of the extents from disk for a file in b-tree format. If the
102 * inode is in b-tree format, then we need to lock the inode exclusively until
103 * the extents are read in. Locking it exclusively all the time would limit
104 * our parallelism unnecessarily, though. What we do instead is check to see
105 * if the extents have been read in yet, and only lock the inode exclusively
106 * if they have not.
107 *
108 * The functions return a value which should be given to the corresponding
109 * xfs_iunlock() call.
110 */
111uint
112xfs_ilock_data_map_shared(
113 struct xfs_inode *ip)
114{
115 uint lock_mode = XFS_ILOCK_SHARED;
116
117 if (xfs_need_iread_extents(&ip->i_df))
118 lock_mode = XFS_ILOCK_EXCL;
119 xfs_ilock(ip, lock_mode);
120 return lock_mode;
121}
122
123uint
124xfs_ilock_attr_map_shared(
125 struct xfs_inode *ip)
126{
127 uint lock_mode = XFS_ILOCK_SHARED;
128
129 if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
132 return lock_mode;
133}
134
135/*
136 * You can't set both SHARED and EXCL for the same lock,
137 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
138 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
139 * to set in lock_flags.
140 */
141static inline void
142xfs_lock_flags_assert(
143 uint lock_flags)
144{
145 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
146 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
147 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
148 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
149 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
150 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
151 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
152 ASSERT(lock_flags != 0);
153}
154
155/*
156 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
157 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
158 * various combinations of the locks to be obtained.
159 *
160 * The 3 locks should always be ordered so that the IO lock is obtained first,
161 * the mmap lock second and the ilock last in order to prevent deadlock.
162 *
163 * Basic locking order:
164 *
165 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
166 *
167 * mmap_lock locking order:
168 *
169 * i_rwsem -> page lock -> mmap_lock
170 * mmap_lock -> invalidate_lock -> page_lock
171 *
172 * The difference in mmap_lock locking order mean that we cannot hold the
173 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
174 * can fault in pages during copy in/out (for buffered IO) or require the
175 * mmap_lock in get_user_pages() to map the user pages into the kernel address
176 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
177 * fault because page faults already hold the mmap_lock.
178 *
179 * Hence to serialise fully against both syscall and mmap based IO, we need to
180 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
181 * both taken in places where we need to invalidate the page cache in a race
182 * free manner (e.g. truncate, hole punch and other extent manipulation
183 * functions).
184 */
185void
186xfs_ilock(
187 xfs_inode_t *ip,
188 uint lock_flags)
189{
190 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
191
192 xfs_lock_flags_assert(lock_flags);
193
194 if (lock_flags & XFS_IOLOCK_EXCL) {
195 down_write_nested(&VFS_I(ip)->i_rwsem,
196 XFS_IOLOCK_DEP(lock_flags));
197 } else if (lock_flags & XFS_IOLOCK_SHARED) {
198 down_read_nested(&VFS_I(ip)->i_rwsem,
199 XFS_IOLOCK_DEP(lock_flags));
200 }
201
202 if (lock_flags & XFS_MMAPLOCK_EXCL) {
203 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
204 XFS_MMAPLOCK_DEP(lock_flags));
205 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
206 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
207 XFS_MMAPLOCK_DEP(lock_flags));
208 }
209
210 if (lock_flags & XFS_ILOCK_EXCL)
211 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 else if (lock_flags & XFS_ILOCK_SHARED)
213 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
214}
215
216/*
217 * This is just like xfs_ilock(), except that the caller
218 * is guaranteed not to sleep. It returns 1 if it gets
219 * the requested locks and 0 otherwise. If the IO lock is
220 * obtained but the inode lock cannot be, then the IO lock
221 * is dropped before returning.
222 *
223 * ip -- the inode being locked
224 * lock_flags -- this parameter indicates the inode's locks to be
225 * to be locked. See the comment for xfs_ilock() for a list
226 * of valid values.
227 */
228int
229xfs_ilock_nowait(
230 xfs_inode_t *ip,
231 uint lock_flags)
232{
233 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
234
235 xfs_lock_flags_assert(lock_flags);
236
237 if (lock_flags & XFS_IOLOCK_EXCL) {
238 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
239 goto out;
240 } else if (lock_flags & XFS_IOLOCK_SHARED) {
241 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
242 goto out;
243 }
244
245 if (lock_flags & XFS_MMAPLOCK_EXCL) {
246 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
247 goto out_undo_iolock;
248 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
249 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
250 goto out_undo_iolock;
251 }
252
253 if (lock_flags & XFS_ILOCK_EXCL) {
254 if (!mrtryupdate(&ip->i_lock))
255 goto out_undo_mmaplock;
256 } else if (lock_flags & XFS_ILOCK_SHARED) {
257 if (!mrtryaccess(&ip->i_lock))
258 goto out_undo_mmaplock;
259 }
260 return 1;
261
262out_undo_mmaplock:
263 if (lock_flags & XFS_MMAPLOCK_EXCL)
264 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
265 else if (lock_flags & XFS_MMAPLOCK_SHARED)
266 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
267out_undo_iolock:
268 if (lock_flags & XFS_IOLOCK_EXCL)
269 up_write(&VFS_I(ip)->i_rwsem);
270 else if (lock_flags & XFS_IOLOCK_SHARED)
271 up_read(&VFS_I(ip)->i_rwsem);
272out:
273 return 0;
274}
275
276/*
277 * xfs_iunlock() is used to drop the inode locks acquired with
278 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
279 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
280 * that we know which locks to drop.
281 *
282 * ip -- the inode being unlocked
283 * lock_flags -- this parameter indicates the inode's locks to be
284 * to be unlocked. See the comment for xfs_ilock() for a list
285 * of valid values for this parameter.
286 *
287 */
288void
289xfs_iunlock(
290 xfs_inode_t *ip,
291 uint lock_flags)
292{
293 xfs_lock_flags_assert(lock_flags);
294
295 if (lock_flags & XFS_IOLOCK_EXCL)
296 up_write(&VFS_I(ip)->i_rwsem);
297 else if (lock_flags & XFS_IOLOCK_SHARED)
298 up_read(&VFS_I(ip)->i_rwsem);
299
300 if (lock_flags & XFS_MMAPLOCK_EXCL)
301 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
302 else if (lock_flags & XFS_MMAPLOCK_SHARED)
303 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
304
305 if (lock_flags & XFS_ILOCK_EXCL)
306 mrunlock_excl(&ip->i_lock);
307 else if (lock_flags & XFS_ILOCK_SHARED)
308 mrunlock_shared(&ip->i_lock);
309
310 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
311}
312
313/*
314 * give up write locks. the i/o lock cannot be held nested
315 * if it is being demoted.
316 */
317void
318xfs_ilock_demote(
319 xfs_inode_t *ip,
320 uint lock_flags)
321{
322 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
323 ASSERT((lock_flags &
324 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
325
326 if (lock_flags & XFS_ILOCK_EXCL)
327 mrdemote(&ip->i_lock);
328 if (lock_flags & XFS_MMAPLOCK_EXCL)
329 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
330 if (lock_flags & XFS_IOLOCK_EXCL)
331 downgrade_write(&VFS_I(ip)->i_rwsem);
332
333 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
334}
335
336#if defined(DEBUG) || defined(XFS_WARN)
337static inline bool
338__xfs_rwsem_islocked(
339 struct rw_semaphore *rwsem,
340 bool shared)
341{
342 if (!debug_locks)
343 return rwsem_is_locked(rwsem);
344
345 if (!shared)
346 return lockdep_is_held_type(rwsem, 0);
347
348 /*
349 * We are checking that the lock is held at least in shared
350 * mode but don't care that it might be held exclusively
351 * (i.e. shared | excl). Hence we check if the lock is held
352 * in any mode rather than an explicit shared mode.
353 */
354 return lockdep_is_held_type(rwsem, -1);
355}
356
357bool
358xfs_isilocked(
359 struct xfs_inode *ip,
360 uint lock_flags)
361{
362 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
363 if (!(lock_flags & XFS_ILOCK_SHARED))
364 return !!ip->i_lock.mr_writer;
365 return rwsem_is_locked(&ip->i_lock.mr_lock);
366 }
367
368 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
369 return __xfs_rwsem_islocked(&VFS_I(ip)->i_mapping->invalidate_lock,
370 (lock_flags & XFS_MMAPLOCK_SHARED));
371 }
372
373 if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) {
374 return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
375 (lock_flags & XFS_IOLOCK_SHARED));
376 }
377
378 ASSERT(0);
379 return false;
380}
381#endif
382
383/*
384 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
385 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
386 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
387 * errors and warnings.
388 */
389#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
390static bool
391xfs_lockdep_subclass_ok(
392 int subclass)
393{
394 return subclass < MAX_LOCKDEP_SUBCLASSES;
395}
396#else
397#define xfs_lockdep_subclass_ok(subclass) (true)
398#endif
399
400/*
401 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
402 * value. This can be called for any type of inode lock combination, including
403 * parent locking. Care must be taken to ensure we don't overrun the subclass
404 * storage fields in the class mask we build.
405 */
406static inline uint
407xfs_lock_inumorder(
408 uint lock_mode,
409 uint subclass)
410{
411 uint class = 0;
412
413 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
414 XFS_ILOCK_RTSUM)));
415 ASSERT(xfs_lockdep_subclass_ok(subclass));
416
417 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
418 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
419 class += subclass << XFS_IOLOCK_SHIFT;
420 }
421
422 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
423 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
424 class += subclass << XFS_MMAPLOCK_SHIFT;
425 }
426
427 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
428 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
429 class += subclass << XFS_ILOCK_SHIFT;
430 }
431
432 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
433}
434
435/*
436 * The following routine will lock n inodes in exclusive mode. We assume the
437 * caller calls us with the inodes in i_ino order.
438 *
439 * We need to detect deadlock where an inode that we lock is in the AIL and we
440 * start waiting for another inode that is locked by a thread in a long running
441 * transaction (such as truncate). This can result in deadlock since the long
442 * running trans might need to wait for the inode we just locked in order to
443 * push the tail and free space in the log.
444 *
445 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
446 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
447 * lock more than one at a time, lockdep will report false positives saying we
448 * have violated locking orders.
449 */
450static void
451xfs_lock_inodes(
452 struct xfs_inode **ips,
453 int inodes,
454 uint lock_mode)
455{
456 int attempts = 0;
457 uint i;
458 int j;
459 bool try_lock;
460 struct xfs_log_item *lp;
461
462 /*
463 * Currently supports between 2 and 5 inodes with exclusive locking. We
464 * support an arbitrary depth of locking here, but absolute limits on
465 * inodes depend on the type of locking and the limits placed by
466 * lockdep annotations in xfs_lock_inumorder. These are all checked by
467 * the asserts.
468 */
469 ASSERT(ips && inodes >= 2 && inodes <= 5);
470 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
471 XFS_ILOCK_EXCL));
472 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
473 XFS_ILOCK_SHARED)));
474 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
475 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
476 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
477 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
478
479 if (lock_mode & XFS_IOLOCK_EXCL) {
480 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
481 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
482 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
483
484again:
485 try_lock = false;
486 i = 0;
487 for (; i < inodes; i++) {
488 ASSERT(ips[i]);
489
490 if (i && (ips[i] == ips[i - 1])) /* Already locked */
491 continue;
492
493 /*
494 * If try_lock is not set yet, make sure all locked inodes are
495 * not in the AIL. If any are, set try_lock to be used later.
496 */
497 if (!try_lock) {
498 for (j = (i - 1); j >= 0 && !try_lock; j--) {
499 lp = &ips[j]->i_itemp->ili_item;
500 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
501 try_lock = true;
502 }
503 }
504
505 /*
506 * If any of the previous locks we have locked is in the AIL,
507 * we must TRY to get the second and subsequent locks. If
508 * we can't get any, we must release all we have
509 * and try again.
510 */
511 if (!try_lock) {
512 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
513 continue;
514 }
515
516 /* try_lock means we have an inode locked that is in the AIL. */
517 ASSERT(i != 0);
518 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
519 continue;
520
521 /*
522 * Unlock all previous guys and try again. xfs_iunlock will try
523 * to push the tail if the inode is in the AIL.
524 */
525 attempts++;
526 for (j = i - 1; j >= 0; j--) {
527 /*
528 * Check to see if we've already unlocked this one. Not
529 * the first one going back, and the inode ptr is the
530 * same.
531 */
532 if (j != (i - 1) && ips[j] == ips[j + 1])
533 continue;
534
535 xfs_iunlock(ips[j], lock_mode);
536 }
537
538 if ((attempts % 5) == 0) {
539 delay(1); /* Don't just spin the CPU */
540 }
541 goto again;
542 }
543}
544
545/*
546 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
547 * mmaplock must be double-locked separately since we use i_rwsem and
548 * invalidate_lock for that. We now support taking one lock EXCL and the
549 * other SHARED.
550 */
551void
552xfs_lock_two_inodes(
553 struct xfs_inode *ip0,
554 uint ip0_mode,
555 struct xfs_inode *ip1,
556 uint ip1_mode)
557{
558 int attempts = 0;
559 struct xfs_log_item *lp;
560
561 ASSERT(hweight32(ip0_mode) == 1);
562 ASSERT(hweight32(ip1_mode) == 1);
563 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
565 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
566 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
567 ASSERT(ip0->i_ino != ip1->i_ino);
568
569 if (ip0->i_ino > ip1->i_ino) {
570 swap(ip0, ip1);
571 swap(ip0_mode, ip1_mode);
572 }
573
574 again:
575 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
576
577 /*
578 * If the first lock we have locked is in the AIL, we must TRY to get
579 * the second lock. If we can't get it, we must release the first one
580 * and try again.
581 */
582 lp = &ip0->i_itemp->ili_item;
583 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
584 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
585 xfs_iunlock(ip0, ip0_mode);
586 if ((++attempts % 5) == 0)
587 delay(1); /* Don't just spin the CPU */
588 goto again;
589 }
590 } else {
591 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
592 }
593}
594
595uint
596xfs_ip2xflags(
597 struct xfs_inode *ip)
598{
599 uint flags = 0;
600
601 if (ip->i_diflags & XFS_DIFLAG_ANY) {
602 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
603 flags |= FS_XFLAG_REALTIME;
604 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
605 flags |= FS_XFLAG_PREALLOC;
606 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
607 flags |= FS_XFLAG_IMMUTABLE;
608 if (ip->i_diflags & XFS_DIFLAG_APPEND)
609 flags |= FS_XFLAG_APPEND;
610 if (ip->i_diflags & XFS_DIFLAG_SYNC)
611 flags |= FS_XFLAG_SYNC;
612 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
613 flags |= FS_XFLAG_NOATIME;
614 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
615 flags |= FS_XFLAG_NODUMP;
616 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
617 flags |= FS_XFLAG_RTINHERIT;
618 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
619 flags |= FS_XFLAG_PROJINHERIT;
620 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
621 flags |= FS_XFLAG_NOSYMLINKS;
622 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
623 flags |= FS_XFLAG_EXTSIZE;
624 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
625 flags |= FS_XFLAG_EXTSZINHERIT;
626 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
627 flags |= FS_XFLAG_NODEFRAG;
628 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
629 flags |= FS_XFLAG_FILESTREAM;
630 }
631
632 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
633 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
634 flags |= FS_XFLAG_DAX;
635 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
636 flags |= FS_XFLAG_COWEXTSIZE;
637 }
638
639 if (xfs_inode_has_attr_fork(ip))
640 flags |= FS_XFLAG_HASATTR;
641 return flags;
642}
643
644/*
645 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
646 * is allowed, otherwise it has to be an exact match. If a CI match is found,
647 * ci_name->name will point to a the actual name (caller must free) or
648 * will be set to NULL if an exact match is found.
649 */
650int
651xfs_lookup(
652 struct xfs_inode *dp,
653 const struct xfs_name *name,
654 struct xfs_inode **ipp,
655 struct xfs_name *ci_name)
656{
657 xfs_ino_t inum;
658 int error;
659
660 trace_xfs_lookup(dp, name);
661
662 if (xfs_is_shutdown(dp->i_mount))
663 return -EIO;
664
665 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
666 if (error)
667 goto out_unlock;
668
669 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
670 if (error)
671 goto out_free_name;
672
673 return 0;
674
675out_free_name:
676 if (ci_name)
677 kmem_free(ci_name->name);
678out_unlock:
679 *ipp = NULL;
680 return error;
681}
682
683/* Propagate di_flags from a parent inode to a child inode. */
684static void
685xfs_inode_inherit_flags(
686 struct xfs_inode *ip,
687 const struct xfs_inode *pip)
688{
689 unsigned int di_flags = 0;
690 xfs_failaddr_t failaddr;
691 umode_t mode = VFS_I(ip)->i_mode;
692
693 if (S_ISDIR(mode)) {
694 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
695 di_flags |= XFS_DIFLAG_RTINHERIT;
696 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
697 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
698 ip->i_extsize = pip->i_extsize;
699 }
700 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
701 di_flags |= XFS_DIFLAG_PROJINHERIT;
702 } else if (S_ISREG(mode)) {
703 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
704 xfs_has_realtime(ip->i_mount))
705 di_flags |= XFS_DIFLAG_REALTIME;
706 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
707 di_flags |= XFS_DIFLAG_EXTSIZE;
708 ip->i_extsize = pip->i_extsize;
709 }
710 }
711 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
712 xfs_inherit_noatime)
713 di_flags |= XFS_DIFLAG_NOATIME;
714 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
715 xfs_inherit_nodump)
716 di_flags |= XFS_DIFLAG_NODUMP;
717 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
718 xfs_inherit_sync)
719 di_flags |= XFS_DIFLAG_SYNC;
720 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
721 xfs_inherit_nosymlinks)
722 di_flags |= XFS_DIFLAG_NOSYMLINKS;
723 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
724 xfs_inherit_nodefrag)
725 di_flags |= XFS_DIFLAG_NODEFRAG;
726 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
727 di_flags |= XFS_DIFLAG_FILESTREAM;
728
729 ip->i_diflags |= di_flags;
730
731 /*
732 * Inode verifiers on older kernels only check that the extent size
733 * hint is an integer multiple of the rt extent size on realtime files.
734 * They did not check the hint alignment on a directory with both
735 * rtinherit and extszinherit flags set. If the misaligned hint is
736 * propagated from a directory into a new realtime file, new file
737 * allocations will fail due to math errors in the rt allocator and/or
738 * trip the verifiers. Validate the hint settings in the new file so
739 * that we don't let broken hints propagate.
740 */
741 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
742 VFS_I(ip)->i_mode, ip->i_diflags);
743 if (failaddr) {
744 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
745 XFS_DIFLAG_EXTSZINHERIT);
746 ip->i_extsize = 0;
747 }
748}
749
750/* Propagate di_flags2 from a parent inode to a child inode. */
751static void
752xfs_inode_inherit_flags2(
753 struct xfs_inode *ip,
754 const struct xfs_inode *pip)
755{
756 xfs_failaddr_t failaddr;
757
758 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
759 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
760 ip->i_cowextsize = pip->i_cowextsize;
761 }
762 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
763 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
764
765 /* Don't let invalid cowextsize hints propagate. */
766 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
767 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
768 if (failaddr) {
769 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
770 ip->i_cowextsize = 0;
771 }
772}
773
774/*
775 * Initialise a newly allocated inode and return the in-core inode to the
776 * caller locked exclusively.
777 */
778int
779xfs_init_new_inode(
780 struct user_namespace *mnt_userns,
781 struct xfs_trans *tp,
782 struct xfs_inode *pip,
783 xfs_ino_t ino,
784 umode_t mode,
785 xfs_nlink_t nlink,
786 dev_t rdev,
787 prid_t prid,
788 bool init_xattrs,
789 struct xfs_inode **ipp)
790{
791 struct inode *dir = pip ? VFS_I(pip) : NULL;
792 struct xfs_mount *mp = tp->t_mountp;
793 struct xfs_inode *ip;
794 unsigned int flags;
795 int error;
796 struct timespec64 tv;
797 struct inode *inode;
798
799 /*
800 * Protect against obviously corrupt allocation btree records. Later
801 * xfs_iget checks will catch re-allocation of other active in-memory
802 * and on-disk inodes. If we don't catch reallocating the parent inode
803 * here we will deadlock in xfs_iget() so we have to do these checks
804 * first.
805 */
806 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
807 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
808 return -EFSCORRUPTED;
809 }
810
811 /*
812 * Get the in-core inode with the lock held exclusively to prevent
813 * others from looking at until we're done.
814 */
815 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
816 if (error)
817 return error;
818
819 ASSERT(ip != NULL);
820 inode = VFS_I(ip);
821 set_nlink(inode, nlink);
822 inode->i_rdev = rdev;
823 ip->i_projid = prid;
824
825 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
826 inode_fsuid_set(inode, mnt_userns);
827 inode->i_gid = dir->i_gid;
828 inode->i_mode = mode;
829 } else {
830 inode_init_owner(mnt_userns, inode, dir, mode);
831 }
832
833 /*
834 * If the group ID of the new file does not match the effective group
835 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
836 * (and only if the irix_sgid_inherit compatibility variable is set).
837 */
838 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
839 !vfsgid_in_group_p(i_gid_into_vfsgid(mnt_userns, inode)))
840 inode->i_mode &= ~S_ISGID;
841
842 ip->i_disk_size = 0;
843 ip->i_df.if_nextents = 0;
844 ASSERT(ip->i_nblocks == 0);
845
846 tv = current_time(inode);
847 inode->i_mtime = tv;
848 inode->i_atime = tv;
849 inode->i_ctime = tv;
850
851 ip->i_extsize = 0;
852 ip->i_diflags = 0;
853
854 if (xfs_has_v3inodes(mp)) {
855 inode_set_iversion(inode, 1);
856 ip->i_cowextsize = 0;
857 ip->i_crtime = tv;
858 }
859
860 flags = XFS_ILOG_CORE;
861 switch (mode & S_IFMT) {
862 case S_IFIFO:
863 case S_IFCHR:
864 case S_IFBLK:
865 case S_IFSOCK:
866 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
867 flags |= XFS_ILOG_DEV;
868 break;
869 case S_IFREG:
870 case S_IFDIR:
871 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
872 xfs_inode_inherit_flags(ip, pip);
873 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
874 xfs_inode_inherit_flags2(ip, pip);
875 fallthrough;
876 case S_IFLNK:
877 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
878 ip->i_df.if_bytes = 0;
879 ip->i_df.if_u1.if_root = NULL;
880 break;
881 default:
882 ASSERT(0);
883 }
884
885 /*
886 * If we need to create attributes immediately after allocating the
887 * inode, initialise an empty attribute fork right now. We use the
888 * default fork offset for attributes here as we don't know exactly what
889 * size or how many attributes we might be adding. We can do this
890 * safely here because we know the data fork is completely empty and
891 * this saves us from needing to run a separate transaction to set the
892 * fork offset in the immediate future.
893 */
894 if (init_xattrs && xfs_has_attr(mp)) {
895 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
896 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
897 }
898
899 /*
900 * Log the new values stuffed into the inode.
901 */
902 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
903 xfs_trans_log_inode(tp, ip, flags);
904
905 /* now that we have an i_mode we can setup the inode structure */
906 xfs_setup_inode(ip);
907
908 *ipp = ip;
909 return 0;
910}
911
912/*
913 * Decrement the link count on an inode & log the change. If this causes the
914 * link count to go to zero, move the inode to AGI unlinked list so that it can
915 * be freed when the last active reference goes away via xfs_inactive().
916 */
917static int /* error */
918xfs_droplink(
919 xfs_trans_t *tp,
920 xfs_inode_t *ip)
921{
922 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
923
924 drop_nlink(VFS_I(ip));
925 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
926
927 if (VFS_I(ip)->i_nlink)
928 return 0;
929
930 return xfs_iunlink(tp, ip);
931}
932
933/*
934 * Increment the link count on an inode & log the change.
935 */
936static void
937xfs_bumplink(
938 xfs_trans_t *tp,
939 xfs_inode_t *ip)
940{
941 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
942
943 inc_nlink(VFS_I(ip));
944 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
945}
946
947int
948xfs_create(
949 struct user_namespace *mnt_userns,
950 xfs_inode_t *dp,
951 struct xfs_name *name,
952 umode_t mode,
953 dev_t rdev,
954 bool init_xattrs,
955 xfs_inode_t **ipp)
956{
957 int is_dir = S_ISDIR(mode);
958 struct xfs_mount *mp = dp->i_mount;
959 struct xfs_inode *ip = NULL;
960 struct xfs_trans *tp = NULL;
961 int error;
962 bool unlock_dp_on_error = false;
963 prid_t prid;
964 struct xfs_dquot *udqp = NULL;
965 struct xfs_dquot *gdqp = NULL;
966 struct xfs_dquot *pdqp = NULL;
967 struct xfs_trans_res *tres;
968 uint resblks;
969 xfs_ino_t ino;
970
971 trace_xfs_create(dp, name);
972
973 if (xfs_is_shutdown(mp))
974 return -EIO;
975
976 prid = xfs_get_initial_prid(dp);
977
978 /*
979 * Make sure that we have allocated dquot(s) on disk.
980 */
981 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns),
982 mapped_fsgid(mnt_userns, &init_user_ns), prid,
983 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
984 &udqp, &gdqp, &pdqp);
985 if (error)
986 return error;
987
988 if (is_dir) {
989 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
990 tres = &M_RES(mp)->tr_mkdir;
991 } else {
992 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
993 tres = &M_RES(mp)->tr_create;
994 }
995
996 /*
997 * Initially assume that the file does not exist and
998 * reserve the resources for that case. If that is not
999 * the case we'll drop the one we have and get a more
1000 * appropriate transaction later.
1001 */
1002 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1003 &tp);
1004 if (error == -ENOSPC) {
1005 /* flush outstanding delalloc blocks and retry */
1006 xfs_flush_inodes(mp);
1007 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1008 resblks, &tp);
1009 }
1010 if (error)
1011 goto out_release_dquots;
1012
1013 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1014 unlock_dp_on_error = true;
1015
1016 /*
1017 * A newly created regular or special file just has one directory
1018 * entry pointing to them, but a directory also the "." entry
1019 * pointing to itself.
1020 */
1021 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1022 if (!error)
1023 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1024 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1025 if (error)
1026 goto out_trans_cancel;
1027
1028 /*
1029 * Now we join the directory inode to the transaction. We do not do it
1030 * earlier because xfs_dialloc might commit the previous transaction
1031 * (and release all the locks). An error from here on will result in
1032 * the transaction cancel unlocking dp so don't do it explicitly in the
1033 * error path.
1034 */
1035 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1036 unlock_dp_on_error = false;
1037
1038 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1039 resblks - XFS_IALLOC_SPACE_RES(mp));
1040 if (error) {
1041 ASSERT(error != -ENOSPC);
1042 goto out_trans_cancel;
1043 }
1044 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1045 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1046
1047 if (is_dir) {
1048 error = xfs_dir_init(tp, ip, dp);
1049 if (error)
1050 goto out_trans_cancel;
1051
1052 xfs_bumplink(tp, dp);
1053 }
1054
1055 /*
1056 * If this is a synchronous mount, make sure that the
1057 * create transaction goes to disk before returning to
1058 * the user.
1059 */
1060 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1061 xfs_trans_set_sync(tp);
1062
1063 /*
1064 * Attach the dquot(s) to the inodes and modify them incore.
1065 * These ids of the inode couldn't have changed since the new
1066 * inode has been locked ever since it was created.
1067 */
1068 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1069
1070 error = xfs_trans_commit(tp);
1071 if (error)
1072 goto out_release_inode;
1073
1074 xfs_qm_dqrele(udqp);
1075 xfs_qm_dqrele(gdqp);
1076 xfs_qm_dqrele(pdqp);
1077
1078 *ipp = ip;
1079 return 0;
1080
1081 out_trans_cancel:
1082 xfs_trans_cancel(tp);
1083 out_release_inode:
1084 /*
1085 * Wait until after the current transaction is aborted to finish the
1086 * setup of the inode and release the inode. This prevents recursive
1087 * transactions and deadlocks from xfs_inactive.
1088 */
1089 if (ip) {
1090 xfs_finish_inode_setup(ip);
1091 xfs_irele(ip);
1092 }
1093 out_release_dquots:
1094 xfs_qm_dqrele(udqp);
1095 xfs_qm_dqrele(gdqp);
1096 xfs_qm_dqrele(pdqp);
1097
1098 if (unlock_dp_on_error)
1099 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1100 return error;
1101}
1102
1103int
1104xfs_create_tmpfile(
1105 struct user_namespace *mnt_userns,
1106 struct xfs_inode *dp,
1107 umode_t mode,
1108 struct xfs_inode **ipp)
1109{
1110 struct xfs_mount *mp = dp->i_mount;
1111 struct xfs_inode *ip = NULL;
1112 struct xfs_trans *tp = NULL;
1113 int error;
1114 prid_t prid;
1115 struct xfs_dquot *udqp = NULL;
1116 struct xfs_dquot *gdqp = NULL;
1117 struct xfs_dquot *pdqp = NULL;
1118 struct xfs_trans_res *tres;
1119 uint resblks;
1120 xfs_ino_t ino;
1121
1122 if (xfs_is_shutdown(mp))
1123 return -EIO;
1124
1125 prid = xfs_get_initial_prid(dp);
1126
1127 /*
1128 * Make sure that we have allocated dquot(s) on disk.
1129 */
1130 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns),
1131 mapped_fsgid(mnt_userns, &init_user_ns), prid,
1132 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1133 &udqp, &gdqp, &pdqp);
1134 if (error)
1135 return error;
1136
1137 resblks = XFS_IALLOC_SPACE_RES(mp);
1138 tres = &M_RES(mp)->tr_create_tmpfile;
1139
1140 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1141 &tp);
1142 if (error)
1143 goto out_release_dquots;
1144
1145 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1146 if (!error)
1147 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1148 0, 0, prid, false, &ip);
1149 if (error)
1150 goto out_trans_cancel;
1151
1152 if (xfs_has_wsync(mp))
1153 xfs_trans_set_sync(tp);
1154
1155 /*
1156 * Attach the dquot(s) to the inodes and modify them incore.
1157 * These ids of the inode couldn't have changed since the new
1158 * inode has been locked ever since it was created.
1159 */
1160 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1161
1162 error = xfs_iunlink(tp, ip);
1163 if (error)
1164 goto out_trans_cancel;
1165
1166 error = xfs_trans_commit(tp);
1167 if (error)
1168 goto out_release_inode;
1169
1170 xfs_qm_dqrele(udqp);
1171 xfs_qm_dqrele(gdqp);
1172 xfs_qm_dqrele(pdqp);
1173
1174 *ipp = ip;
1175 return 0;
1176
1177 out_trans_cancel:
1178 xfs_trans_cancel(tp);
1179 out_release_inode:
1180 /*
1181 * Wait until after the current transaction is aborted to finish the
1182 * setup of the inode and release the inode. This prevents recursive
1183 * transactions and deadlocks from xfs_inactive.
1184 */
1185 if (ip) {
1186 xfs_finish_inode_setup(ip);
1187 xfs_irele(ip);
1188 }
1189 out_release_dquots:
1190 xfs_qm_dqrele(udqp);
1191 xfs_qm_dqrele(gdqp);
1192 xfs_qm_dqrele(pdqp);
1193
1194 return error;
1195}
1196
1197int
1198xfs_link(
1199 xfs_inode_t *tdp,
1200 xfs_inode_t *sip,
1201 struct xfs_name *target_name)
1202{
1203 xfs_mount_t *mp = tdp->i_mount;
1204 xfs_trans_t *tp;
1205 int error, nospace_error = 0;
1206 int resblks;
1207
1208 trace_xfs_link(tdp, target_name);
1209
1210 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1211
1212 if (xfs_is_shutdown(mp))
1213 return -EIO;
1214
1215 error = xfs_qm_dqattach(sip);
1216 if (error)
1217 goto std_return;
1218
1219 error = xfs_qm_dqattach(tdp);
1220 if (error)
1221 goto std_return;
1222
1223 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1224 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1225 &tp, &nospace_error);
1226 if (error)
1227 goto std_return;
1228
1229 /*
1230 * If we are using project inheritance, we only allow hard link
1231 * creation in our tree when the project IDs are the same; else
1232 * the tree quota mechanism could be circumvented.
1233 */
1234 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1235 tdp->i_projid != sip->i_projid)) {
1236 error = -EXDEV;
1237 goto error_return;
1238 }
1239
1240 if (!resblks) {
1241 error = xfs_dir_canenter(tp, tdp, target_name);
1242 if (error)
1243 goto error_return;
1244 }
1245
1246 /*
1247 * Handle initial link state of O_TMPFILE inode
1248 */
1249 if (VFS_I(sip)->i_nlink == 0) {
1250 struct xfs_perag *pag;
1251
1252 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1253 error = xfs_iunlink_remove(tp, pag, sip);
1254 xfs_perag_put(pag);
1255 if (error)
1256 goto error_return;
1257 }
1258
1259 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1260 resblks);
1261 if (error)
1262 goto error_return;
1263 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1264 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1265
1266 xfs_bumplink(tp, sip);
1267
1268 /*
1269 * If this is a synchronous mount, make sure that the
1270 * link transaction goes to disk before returning to
1271 * the user.
1272 */
1273 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1274 xfs_trans_set_sync(tp);
1275
1276 return xfs_trans_commit(tp);
1277
1278 error_return:
1279 xfs_trans_cancel(tp);
1280 std_return:
1281 if (error == -ENOSPC && nospace_error)
1282 error = nospace_error;
1283 return error;
1284}
1285
1286/* Clear the reflink flag and the cowblocks tag if possible. */
1287static void
1288xfs_itruncate_clear_reflink_flags(
1289 struct xfs_inode *ip)
1290{
1291 struct xfs_ifork *dfork;
1292 struct xfs_ifork *cfork;
1293
1294 if (!xfs_is_reflink_inode(ip))
1295 return;
1296 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1297 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1298 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1299 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1300 if (cfork->if_bytes == 0)
1301 xfs_inode_clear_cowblocks_tag(ip);
1302}
1303
1304/*
1305 * Free up the underlying blocks past new_size. The new size must be smaller
1306 * than the current size. This routine can be used both for the attribute and
1307 * data fork, and does not modify the inode size, which is left to the caller.
1308 *
1309 * The transaction passed to this routine must have made a permanent log
1310 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1311 * given transaction and start new ones, so make sure everything involved in
1312 * the transaction is tidy before calling here. Some transaction will be
1313 * returned to the caller to be committed. The incoming transaction must
1314 * already include the inode, and both inode locks must be held exclusively.
1315 * The inode must also be "held" within the transaction. On return the inode
1316 * will be "held" within the returned transaction. This routine does NOT
1317 * require any disk space to be reserved for it within the transaction.
1318 *
1319 * If we get an error, we must return with the inode locked and linked into the
1320 * current transaction. This keeps things simple for the higher level code,
1321 * because it always knows that the inode is locked and held in the transaction
1322 * that returns to it whether errors occur or not. We don't mark the inode
1323 * dirty on error so that transactions can be easily aborted if possible.
1324 */
1325int
1326xfs_itruncate_extents_flags(
1327 struct xfs_trans **tpp,
1328 struct xfs_inode *ip,
1329 int whichfork,
1330 xfs_fsize_t new_size,
1331 int flags)
1332{
1333 struct xfs_mount *mp = ip->i_mount;
1334 struct xfs_trans *tp = *tpp;
1335 xfs_fileoff_t first_unmap_block;
1336 xfs_filblks_t unmap_len;
1337 int error = 0;
1338
1339 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1340 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1341 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1342 ASSERT(new_size <= XFS_ISIZE(ip));
1343 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1344 ASSERT(ip->i_itemp != NULL);
1345 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1346 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1347
1348 trace_xfs_itruncate_extents_start(ip, new_size);
1349
1350 flags |= xfs_bmapi_aflag(whichfork);
1351
1352 /*
1353 * Since it is possible for space to become allocated beyond
1354 * the end of the file (in a crash where the space is allocated
1355 * but the inode size is not yet updated), simply remove any
1356 * blocks which show up between the new EOF and the maximum
1357 * possible file size.
1358 *
1359 * We have to free all the blocks to the bmbt maximum offset, even if
1360 * the page cache can't scale that far.
1361 */
1362 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1363 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1364 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1365 return 0;
1366 }
1367
1368 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1369 while (unmap_len > 0) {
1370 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1371 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1372 flags, XFS_ITRUNC_MAX_EXTENTS);
1373 if (error)
1374 goto out;
1375
1376 /* free the just unmapped extents */
1377 error = xfs_defer_finish(&tp);
1378 if (error)
1379 goto out;
1380 }
1381
1382 if (whichfork == XFS_DATA_FORK) {
1383 /* Remove all pending CoW reservations. */
1384 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1385 first_unmap_block, XFS_MAX_FILEOFF, true);
1386 if (error)
1387 goto out;
1388
1389 xfs_itruncate_clear_reflink_flags(ip);
1390 }
1391
1392 /*
1393 * Always re-log the inode so that our permanent transaction can keep
1394 * on rolling it forward in the log.
1395 */
1396 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1397
1398 trace_xfs_itruncate_extents_end(ip, new_size);
1399
1400out:
1401 *tpp = tp;
1402 return error;
1403}
1404
1405int
1406xfs_release(
1407 xfs_inode_t *ip)
1408{
1409 xfs_mount_t *mp = ip->i_mount;
1410 int error = 0;
1411
1412 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1413 return 0;
1414
1415 /* If this is a read-only mount, don't do this (would generate I/O) */
1416 if (xfs_is_readonly(mp))
1417 return 0;
1418
1419 if (!xfs_is_shutdown(mp)) {
1420 int truncated;
1421
1422 /*
1423 * If we previously truncated this file and removed old data
1424 * in the process, we want to initiate "early" writeout on
1425 * the last close. This is an attempt to combat the notorious
1426 * NULL files problem which is particularly noticeable from a
1427 * truncate down, buffered (re-)write (delalloc), followed by
1428 * a crash. What we are effectively doing here is
1429 * significantly reducing the time window where we'd otherwise
1430 * be exposed to that problem.
1431 */
1432 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1433 if (truncated) {
1434 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1435 if (ip->i_delayed_blks > 0) {
1436 error = filemap_flush(VFS_I(ip)->i_mapping);
1437 if (error)
1438 return error;
1439 }
1440 }
1441 }
1442
1443 if (VFS_I(ip)->i_nlink == 0)
1444 return 0;
1445
1446 /*
1447 * If we can't get the iolock just skip truncating the blocks past EOF
1448 * because we could deadlock with the mmap_lock otherwise. We'll get
1449 * another chance to drop them once the last reference to the inode is
1450 * dropped, so we'll never leak blocks permanently.
1451 */
1452 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1453 return 0;
1454
1455 if (xfs_can_free_eofblocks(ip, false)) {
1456 /*
1457 * Check if the inode is being opened, written and closed
1458 * frequently and we have delayed allocation blocks outstanding
1459 * (e.g. streaming writes from the NFS server), truncating the
1460 * blocks past EOF will cause fragmentation to occur.
1461 *
1462 * In this case don't do the truncation, but we have to be
1463 * careful how we detect this case. Blocks beyond EOF show up as
1464 * i_delayed_blks even when the inode is clean, so we need to
1465 * truncate them away first before checking for a dirty release.
1466 * Hence on the first dirty close we will still remove the
1467 * speculative allocation, but after that we will leave it in
1468 * place.
1469 */
1470 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1471 goto out_unlock;
1472
1473 error = xfs_free_eofblocks(ip);
1474 if (error)
1475 goto out_unlock;
1476
1477 /* delalloc blocks after truncation means it really is dirty */
1478 if (ip->i_delayed_blks)
1479 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1480 }
1481
1482out_unlock:
1483 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1484 return error;
1485}
1486
1487/*
1488 * xfs_inactive_truncate
1489 *
1490 * Called to perform a truncate when an inode becomes unlinked.
1491 */
1492STATIC int
1493xfs_inactive_truncate(
1494 struct xfs_inode *ip)
1495{
1496 struct xfs_mount *mp = ip->i_mount;
1497 struct xfs_trans *tp;
1498 int error;
1499
1500 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1501 if (error) {
1502 ASSERT(xfs_is_shutdown(mp));
1503 return error;
1504 }
1505 xfs_ilock(ip, XFS_ILOCK_EXCL);
1506 xfs_trans_ijoin(tp, ip, 0);
1507
1508 /*
1509 * Log the inode size first to prevent stale data exposure in the event
1510 * of a system crash before the truncate completes. See the related
1511 * comment in xfs_vn_setattr_size() for details.
1512 */
1513 ip->i_disk_size = 0;
1514 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1515
1516 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1517 if (error)
1518 goto error_trans_cancel;
1519
1520 ASSERT(ip->i_df.if_nextents == 0);
1521
1522 error = xfs_trans_commit(tp);
1523 if (error)
1524 goto error_unlock;
1525
1526 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527 return 0;
1528
1529error_trans_cancel:
1530 xfs_trans_cancel(tp);
1531error_unlock:
1532 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1533 return error;
1534}
1535
1536/*
1537 * xfs_inactive_ifree()
1538 *
1539 * Perform the inode free when an inode is unlinked.
1540 */
1541STATIC int
1542xfs_inactive_ifree(
1543 struct xfs_inode *ip)
1544{
1545 struct xfs_mount *mp = ip->i_mount;
1546 struct xfs_trans *tp;
1547 int error;
1548
1549 /*
1550 * We try to use a per-AG reservation for any block needed by the finobt
1551 * tree, but as the finobt feature predates the per-AG reservation
1552 * support a degraded file system might not have enough space for the
1553 * reservation at mount time. In that case try to dip into the reserved
1554 * pool and pray.
1555 *
1556 * Send a warning if the reservation does happen to fail, as the inode
1557 * now remains allocated and sits on the unlinked list until the fs is
1558 * repaired.
1559 */
1560 if (unlikely(mp->m_finobt_nores)) {
1561 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1562 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1563 &tp);
1564 } else {
1565 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1566 }
1567 if (error) {
1568 if (error == -ENOSPC) {
1569 xfs_warn_ratelimited(mp,
1570 "Failed to remove inode(s) from unlinked list. "
1571 "Please free space, unmount and run xfs_repair.");
1572 } else {
1573 ASSERT(xfs_is_shutdown(mp));
1574 }
1575 return error;
1576 }
1577
1578 /*
1579 * We do not hold the inode locked across the entire rolling transaction
1580 * here. We only need to hold it for the first transaction that
1581 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1582 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1583 * here breaks the relationship between cluster buffer invalidation and
1584 * stale inode invalidation on cluster buffer item journal commit
1585 * completion, and can result in leaving dirty stale inodes hanging
1586 * around in memory.
1587 *
1588 * We have no need for serialising this inode operation against other
1589 * operations - we freed the inode and hence reallocation is required
1590 * and that will serialise on reallocating the space the deferops need
1591 * to free. Hence we can unlock the inode on the first commit of
1592 * the transaction rather than roll it right through the deferops. This
1593 * avoids relogging the XFS_ISTALE inode.
1594 *
1595 * We check that xfs_ifree() hasn't grown an internal transaction roll
1596 * by asserting that the inode is still locked when it returns.
1597 */
1598 xfs_ilock(ip, XFS_ILOCK_EXCL);
1599 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1600
1601 error = xfs_ifree(tp, ip);
1602 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1603 if (error) {
1604 /*
1605 * If we fail to free the inode, shut down. The cancel
1606 * might do that, we need to make sure. Otherwise the
1607 * inode might be lost for a long time or forever.
1608 */
1609 if (!xfs_is_shutdown(mp)) {
1610 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1611 __func__, error);
1612 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1613 }
1614 xfs_trans_cancel(tp);
1615 return error;
1616 }
1617
1618 /*
1619 * Credit the quota account(s). The inode is gone.
1620 */
1621 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1622
1623 /*
1624 * Just ignore errors at this point. There is nothing we can do except
1625 * to try to keep going. Make sure it's not a silent error.
1626 */
1627 error = xfs_trans_commit(tp);
1628 if (error)
1629 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1630 __func__, error);
1631
1632 return 0;
1633}
1634
1635/*
1636 * Returns true if we need to update the on-disk metadata before we can free
1637 * the memory used by this inode. Updates include freeing post-eof
1638 * preallocations; freeing COW staging extents; and marking the inode free in
1639 * the inobt if it is on the unlinked list.
1640 */
1641bool
1642xfs_inode_needs_inactive(
1643 struct xfs_inode *ip)
1644{
1645 struct xfs_mount *mp = ip->i_mount;
1646 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1647
1648 /*
1649 * If the inode is already free, then there can be nothing
1650 * to clean up here.
1651 */
1652 if (VFS_I(ip)->i_mode == 0)
1653 return false;
1654
1655 /* If this is a read-only mount, don't do this (would generate I/O) */
1656 if (xfs_is_readonly(mp))
1657 return false;
1658
1659 /* If the log isn't running, push inodes straight to reclaim. */
1660 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1661 return false;
1662
1663 /* Metadata inodes require explicit resource cleanup. */
1664 if (xfs_is_metadata_inode(ip))
1665 return false;
1666
1667 /* Want to clean out the cow blocks if there are any. */
1668 if (cow_ifp && cow_ifp->if_bytes > 0)
1669 return true;
1670
1671 /* Unlinked files must be freed. */
1672 if (VFS_I(ip)->i_nlink == 0)
1673 return true;
1674
1675 /*
1676 * This file isn't being freed, so check if there are post-eof blocks
1677 * to free. @force is true because we are evicting an inode from the
1678 * cache. Post-eof blocks must be freed, lest we end up with broken
1679 * free space accounting.
1680 *
1681 * Note: don't bother with iolock here since lockdep complains about
1682 * acquiring it in reclaim context. We have the only reference to the
1683 * inode at this point anyways.
1684 */
1685 return xfs_can_free_eofblocks(ip, true);
1686}
1687
1688/*
1689 * xfs_inactive
1690 *
1691 * This is called when the vnode reference count for the vnode
1692 * goes to zero. If the file has been unlinked, then it must
1693 * now be truncated. Also, we clear all of the read-ahead state
1694 * kept for the inode here since the file is now closed.
1695 */
1696void
1697xfs_inactive(
1698 xfs_inode_t *ip)
1699{
1700 struct xfs_mount *mp;
1701 int error;
1702 int truncate = 0;
1703
1704 /*
1705 * If the inode is already free, then there can be nothing
1706 * to clean up here.
1707 */
1708 if (VFS_I(ip)->i_mode == 0) {
1709 ASSERT(ip->i_df.if_broot_bytes == 0);
1710 goto out;
1711 }
1712
1713 mp = ip->i_mount;
1714 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1715
1716 /* If this is a read-only mount, don't do this (would generate I/O) */
1717 if (xfs_is_readonly(mp))
1718 goto out;
1719
1720 /* Metadata inodes require explicit resource cleanup. */
1721 if (xfs_is_metadata_inode(ip))
1722 goto out;
1723
1724 /* Try to clean out the cow blocks if there are any. */
1725 if (xfs_inode_has_cow_data(ip))
1726 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1727
1728 if (VFS_I(ip)->i_nlink != 0) {
1729 /*
1730 * force is true because we are evicting an inode from the
1731 * cache. Post-eof blocks must be freed, lest we end up with
1732 * broken free space accounting.
1733 *
1734 * Note: don't bother with iolock here since lockdep complains
1735 * about acquiring it in reclaim context. We have the only
1736 * reference to the inode at this point anyways.
1737 */
1738 if (xfs_can_free_eofblocks(ip, true))
1739 xfs_free_eofblocks(ip);
1740
1741 goto out;
1742 }
1743
1744 if (S_ISREG(VFS_I(ip)->i_mode) &&
1745 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1746 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1747 truncate = 1;
1748
1749 error = xfs_qm_dqattach(ip);
1750 if (error)
1751 goto out;
1752
1753 if (S_ISLNK(VFS_I(ip)->i_mode))
1754 error = xfs_inactive_symlink(ip);
1755 else if (truncate)
1756 error = xfs_inactive_truncate(ip);
1757 if (error)
1758 goto out;
1759
1760 /*
1761 * If there are attributes associated with the file then blow them away
1762 * now. The code calls a routine that recursively deconstructs the
1763 * attribute fork. If also blows away the in-core attribute fork.
1764 */
1765 if (xfs_inode_has_attr_fork(ip)) {
1766 error = xfs_attr_inactive(ip);
1767 if (error)
1768 goto out;
1769 }
1770
1771 ASSERT(ip->i_forkoff == 0);
1772
1773 /*
1774 * Free the inode.
1775 */
1776 xfs_inactive_ifree(ip);
1777
1778out:
1779 /*
1780 * We're done making metadata updates for this inode, so we can release
1781 * the attached dquots.
1782 */
1783 xfs_qm_dqdetach(ip);
1784}
1785
1786/*
1787 * In-Core Unlinked List Lookups
1788 * =============================
1789 *
1790 * Every inode is supposed to be reachable from some other piece of metadata
1791 * with the exception of the root directory. Inodes with a connection to a
1792 * file descriptor but not linked from anywhere in the on-disk directory tree
1793 * are collectively known as unlinked inodes, though the filesystem itself
1794 * maintains links to these inodes so that on-disk metadata are consistent.
1795 *
1796 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1797 * header contains a number of buckets that point to an inode, and each inode
1798 * record has a pointer to the next inode in the hash chain. This
1799 * singly-linked list causes scaling problems in the iunlink remove function
1800 * because we must walk that list to find the inode that points to the inode
1801 * being removed from the unlinked hash bucket list.
1802 *
1803 * Hence we keep an in-memory double linked list to link each inode on an
1804 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1805 * based lists would require having 64 list heads in the perag, one for each
1806 * list. This is expensive in terms of memory (think millions of AGs) and cache
1807 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1808 * must be referenced at the VFS level to keep them on the list and hence we
1809 * have an existence guarantee for inodes on the unlinked list.
1810 *
1811 * Given we have an existence guarantee, we can use lockless inode cache lookups
1812 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1813 * for the double linked unlinked list, and we don't need any extra locking to
1814 * keep the list safe as all manipulations are done under the AGI buffer lock.
1815 * Keeping the list up to date does not require memory allocation, just finding
1816 * the XFS inode and updating the next/prev unlinked list aginos.
1817 */
1818
1819/*
1820 * Find an inode on the unlinked list. This does not take references to the
1821 * inode as we have existence guarantees by holding the AGI buffer lock and that
1822 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1823 * don't find the inode in cache, then let the caller handle the situation.
1824 */
1825static struct xfs_inode *
1826xfs_iunlink_lookup(
1827 struct xfs_perag *pag,
1828 xfs_agino_t agino)
1829{
1830 struct xfs_inode *ip;
1831
1832 rcu_read_lock();
1833 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1834
1835 /*
1836 * Inode not in memory or in RCU freeing limbo should not happen.
1837 * Warn about this and let the caller handle the failure.
1838 */
1839 if (WARN_ON_ONCE(!ip || !ip->i_ino)) {
1840 rcu_read_unlock();
1841 return NULL;
1842 }
1843 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1844 rcu_read_unlock();
1845 return ip;
1846}
1847
1848/* Update the prev pointer of the next agino. */
1849static int
1850xfs_iunlink_update_backref(
1851 struct xfs_perag *pag,
1852 xfs_agino_t prev_agino,
1853 xfs_agino_t next_agino)
1854{
1855 struct xfs_inode *ip;
1856
1857 /* No update necessary if we are at the end of the list. */
1858 if (next_agino == NULLAGINO)
1859 return 0;
1860
1861 ip = xfs_iunlink_lookup(pag, next_agino);
1862 if (!ip)
1863 return -EFSCORRUPTED;
1864 ip->i_prev_unlinked = prev_agino;
1865 return 0;
1866}
1867
1868/*
1869 * Point the AGI unlinked bucket at an inode and log the results. The caller
1870 * is responsible for validating the old value.
1871 */
1872STATIC int
1873xfs_iunlink_update_bucket(
1874 struct xfs_trans *tp,
1875 struct xfs_perag *pag,
1876 struct xfs_buf *agibp,
1877 unsigned int bucket_index,
1878 xfs_agino_t new_agino)
1879{
1880 struct xfs_agi *agi = agibp->b_addr;
1881 xfs_agino_t old_value;
1882 int offset;
1883
1884 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
1885
1886 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1887 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1888 old_value, new_agino);
1889
1890 /*
1891 * We should never find the head of the list already set to the value
1892 * passed in because either we're adding or removing ourselves from the
1893 * head of the list.
1894 */
1895 if (old_value == new_agino) {
1896 xfs_buf_mark_corrupt(agibp);
1897 return -EFSCORRUPTED;
1898 }
1899
1900 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
1901 offset = offsetof(struct xfs_agi, agi_unlinked) +
1902 (sizeof(xfs_agino_t) * bucket_index);
1903 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
1904 return 0;
1905}
1906
1907static int
1908xfs_iunlink_insert_inode(
1909 struct xfs_trans *tp,
1910 struct xfs_perag *pag,
1911 struct xfs_buf *agibp,
1912 struct xfs_inode *ip)
1913{
1914 struct xfs_mount *mp = tp->t_mountp;
1915 struct xfs_agi *agi = agibp->b_addr;
1916 xfs_agino_t next_agino;
1917 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1918 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1919 int error;
1920
1921 /*
1922 * Get the index into the agi hash table for the list this inode will
1923 * go on. Make sure the pointer isn't garbage and that this inode
1924 * isn't already on the list.
1925 */
1926 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1927 if (next_agino == agino ||
1928 !xfs_verify_agino_or_null(pag, next_agino)) {
1929 xfs_buf_mark_corrupt(agibp);
1930 return -EFSCORRUPTED;
1931 }
1932
1933 /*
1934 * Update the prev pointer in the next inode to point back to this
1935 * inode.
1936 */
1937 error = xfs_iunlink_update_backref(pag, agino, next_agino);
1938 if (error)
1939 return error;
1940
1941 if (next_agino != NULLAGINO) {
1942 /*
1943 * There is already another inode in the bucket, so point this
1944 * inode to the current head of the list.
1945 */
1946 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
1947 if (error)
1948 return error;
1949 ip->i_next_unlinked = next_agino;
1950 }
1951
1952 /* Point the head of the list to point to this inode. */
1953 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
1954}
1955
1956/*
1957 * This is called when the inode's link count has gone to 0 or we are creating
1958 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
1959 *
1960 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1961 * list when the inode is freed.
1962 */
1963STATIC int
1964xfs_iunlink(
1965 struct xfs_trans *tp,
1966 struct xfs_inode *ip)
1967{
1968 struct xfs_mount *mp = tp->t_mountp;
1969 struct xfs_perag *pag;
1970 struct xfs_buf *agibp;
1971 int error;
1972
1973 ASSERT(VFS_I(ip)->i_nlink == 0);
1974 ASSERT(VFS_I(ip)->i_mode != 0);
1975 trace_xfs_iunlink(ip);
1976
1977 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1978
1979 /* Get the agi buffer first. It ensures lock ordering on the list. */
1980 error = xfs_read_agi(pag, tp, &agibp);
1981 if (error)
1982 goto out;
1983
1984 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
1985out:
1986 xfs_perag_put(pag);
1987 return error;
1988}
1989
1990static int
1991xfs_iunlink_remove_inode(
1992 struct xfs_trans *tp,
1993 struct xfs_perag *pag,
1994 struct xfs_buf *agibp,
1995 struct xfs_inode *ip)
1996{
1997 struct xfs_mount *mp = tp->t_mountp;
1998 struct xfs_agi *agi = agibp->b_addr;
1999 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2000 xfs_agino_t head_agino;
2001 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2002 int error;
2003
2004 trace_xfs_iunlink_remove(ip);
2005
2006 /*
2007 * Get the index into the agi hash table for the list this inode will
2008 * go on. Make sure the head pointer isn't garbage.
2009 */
2010 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2011 if (!xfs_verify_agino(pag, head_agino)) {
2012 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2013 agi, sizeof(*agi));
2014 return -EFSCORRUPTED;
2015 }
2016
2017 /*
2018 * Set our inode's next_unlinked pointer to NULL and then return
2019 * the old pointer value so that we can update whatever was previous
2020 * to us in the list to point to whatever was next in the list.
2021 */
2022 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2023 if (error)
2024 return error;
2025
2026 /*
2027 * Update the prev pointer in the next inode to point back to previous
2028 * inode in the chain.
2029 */
2030 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2031 ip->i_next_unlinked);
2032 if (error)
2033 return error;
2034
2035 if (head_agino != agino) {
2036 struct xfs_inode *prev_ip;
2037
2038 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2039 if (!prev_ip)
2040 return -EFSCORRUPTED;
2041
2042 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2043 ip->i_next_unlinked);
2044 prev_ip->i_next_unlinked = ip->i_next_unlinked;
2045 } else {
2046 /* Point the head of the list to the next unlinked inode. */
2047 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2048 ip->i_next_unlinked);
2049 }
2050
2051 ip->i_next_unlinked = NULLAGINO;
2052 ip->i_prev_unlinked = NULLAGINO;
2053 return error;
2054}
2055
2056/*
2057 * Pull the on-disk inode from the AGI unlinked list.
2058 */
2059STATIC int
2060xfs_iunlink_remove(
2061 struct xfs_trans *tp,
2062 struct xfs_perag *pag,
2063 struct xfs_inode *ip)
2064{
2065 struct xfs_buf *agibp;
2066 int error;
2067
2068 trace_xfs_iunlink_remove(ip);
2069
2070 /* Get the agi buffer first. It ensures lock ordering on the list. */
2071 error = xfs_read_agi(pag, tp, &agibp);
2072 if (error)
2073 return error;
2074
2075 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2076}
2077
2078/*
2079 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2080 * mark it stale. We should only find clean inodes in this lookup that aren't
2081 * already stale.
2082 */
2083static void
2084xfs_ifree_mark_inode_stale(
2085 struct xfs_perag *pag,
2086 struct xfs_inode *free_ip,
2087 xfs_ino_t inum)
2088{
2089 struct xfs_mount *mp = pag->pag_mount;
2090 struct xfs_inode_log_item *iip;
2091 struct xfs_inode *ip;
2092
2093retry:
2094 rcu_read_lock();
2095 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2096
2097 /* Inode not in memory, nothing to do */
2098 if (!ip) {
2099 rcu_read_unlock();
2100 return;
2101 }
2102
2103 /*
2104 * because this is an RCU protected lookup, we could find a recently
2105 * freed or even reallocated inode during the lookup. We need to check
2106 * under the i_flags_lock for a valid inode here. Skip it if it is not
2107 * valid, the wrong inode or stale.
2108 */
2109 spin_lock(&ip->i_flags_lock);
2110 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2111 goto out_iflags_unlock;
2112
2113 /*
2114 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2115 * other inodes that we did not find in the list attached to the buffer
2116 * and are not already marked stale. If we can't lock it, back off and
2117 * retry.
2118 */
2119 if (ip != free_ip) {
2120 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2121 spin_unlock(&ip->i_flags_lock);
2122 rcu_read_unlock();
2123 delay(1);
2124 goto retry;
2125 }
2126 }
2127 ip->i_flags |= XFS_ISTALE;
2128
2129 /*
2130 * If the inode is flushing, it is already attached to the buffer. All
2131 * we needed to do here is mark the inode stale so buffer IO completion
2132 * will remove it from the AIL.
2133 */
2134 iip = ip->i_itemp;
2135 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2136 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2137 ASSERT(iip->ili_last_fields);
2138 goto out_iunlock;
2139 }
2140
2141 /*
2142 * Inodes not attached to the buffer can be released immediately.
2143 * Everything else has to go through xfs_iflush_abort() on journal
2144 * commit as the flock synchronises removal of the inode from the
2145 * cluster buffer against inode reclaim.
2146 */
2147 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2148 goto out_iunlock;
2149
2150 __xfs_iflags_set(ip, XFS_IFLUSHING);
2151 spin_unlock(&ip->i_flags_lock);
2152 rcu_read_unlock();
2153
2154 /* we have a dirty inode in memory that has not yet been flushed. */
2155 spin_lock(&iip->ili_lock);
2156 iip->ili_last_fields = iip->ili_fields;
2157 iip->ili_fields = 0;
2158 iip->ili_fsync_fields = 0;
2159 spin_unlock(&iip->ili_lock);
2160 ASSERT(iip->ili_last_fields);
2161
2162 if (ip != free_ip)
2163 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2164 return;
2165
2166out_iunlock:
2167 if (ip != free_ip)
2168 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2169out_iflags_unlock:
2170 spin_unlock(&ip->i_flags_lock);
2171 rcu_read_unlock();
2172}
2173
2174/*
2175 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2176 * inodes that are in memory - they all must be marked stale and attached to
2177 * the cluster buffer.
2178 */
2179static int
2180xfs_ifree_cluster(
2181 struct xfs_trans *tp,
2182 struct xfs_perag *pag,
2183 struct xfs_inode *free_ip,
2184 struct xfs_icluster *xic)
2185{
2186 struct xfs_mount *mp = free_ip->i_mount;
2187 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2188 struct xfs_buf *bp;
2189 xfs_daddr_t blkno;
2190 xfs_ino_t inum = xic->first_ino;
2191 int nbufs;
2192 int i, j;
2193 int ioffset;
2194 int error;
2195
2196 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2197
2198 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2199 /*
2200 * The allocation bitmap tells us which inodes of the chunk were
2201 * physically allocated. Skip the cluster if an inode falls into
2202 * a sparse region.
2203 */
2204 ioffset = inum - xic->first_ino;
2205 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2206 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2207 continue;
2208 }
2209
2210 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2211 XFS_INO_TO_AGBNO(mp, inum));
2212
2213 /*
2214 * We obtain and lock the backing buffer first in the process
2215 * here to ensure dirty inodes attached to the buffer remain in
2216 * the flushing state while we mark them stale.
2217 *
2218 * If we scan the in-memory inodes first, then buffer IO can
2219 * complete before we get a lock on it, and hence we may fail
2220 * to mark all the active inodes on the buffer stale.
2221 */
2222 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2223 mp->m_bsize * igeo->blocks_per_cluster,
2224 XBF_UNMAPPED, &bp);
2225 if (error)
2226 return error;
2227
2228 /*
2229 * This buffer may not have been correctly initialised as we
2230 * didn't read it from disk. That's not important because we are
2231 * only using to mark the buffer as stale in the log, and to
2232 * attach stale cached inodes on it. That means it will never be
2233 * dispatched for IO. If it is, we want to know about it, and we
2234 * want it to fail. We can acheive this by adding a write
2235 * verifier to the buffer.
2236 */
2237 bp->b_ops = &xfs_inode_buf_ops;
2238
2239 /*
2240 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2241 * too. This requires lookups, and will skip inodes that we've
2242 * already marked XFS_ISTALE.
2243 */
2244 for (i = 0; i < igeo->inodes_per_cluster; i++)
2245 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2246
2247 xfs_trans_stale_inode_buf(tp, bp);
2248 xfs_trans_binval(tp, bp);
2249 }
2250 return 0;
2251}
2252
2253/*
2254 * This is called to return an inode to the inode free list. The inode should
2255 * already be truncated to 0 length and have no pages associated with it. This
2256 * routine also assumes that the inode is already a part of the transaction.
2257 *
2258 * The on-disk copy of the inode will have been added to the list of unlinked
2259 * inodes in the AGI. We need to remove the inode from that list atomically with
2260 * respect to freeing it here.
2261 */
2262int
2263xfs_ifree(
2264 struct xfs_trans *tp,
2265 struct xfs_inode *ip)
2266{
2267 struct xfs_mount *mp = ip->i_mount;
2268 struct xfs_perag *pag;
2269 struct xfs_icluster xic = { 0 };
2270 struct xfs_inode_log_item *iip = ip->i_itemp;
2271 int error;
2272
2273 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2274 ASSERT(VFS_I(ip)->i_nlink == 0);
2275 ASSERT(ip->i_df.if_nextents == 0);
2276 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2277 ASSERT(ip->i_nblocks == 0);
2278
2279 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2280
2281 /*
2282 * Free the inode first so that we guarantee that the AGI lock is going
2283 * to be taken before we remove the inode from the unlinked list. This
2284 * makes the AGI lock -> unlinked list modification order the same as
2285 * used in O_TMPFILE creation.
2286 */
2287 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2288 if (error)
2289 goto out;
2290
2291 error = xfs_iunlink_remove(tp, pag, ip);
2292 if (error)
2293 goto out;
2294
2295 /*
2296 * Free any local-format data sitting around before we reset the
2297 * data fork to extents format. Note that the attr fork data has
2298 * already been freed by xfs_attr_inactive.
2299 */
2300 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2301 kmem_free(ip->i_df.if_u1.if_data);
2302 ip->i_df.if_u1.if_data = NULL;
2303 ip->i_df.if_bytes = 0;
2304 }
2305
2306 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2307 ip->i_diflags = 0;
2308 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2309 ip->i_forkoff = 0; /* mark the attr fork not in use */
2310 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2311 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2312 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2313
2314 /* Don't attempt to replay owner changes for a deleted inode */
2315 spin_lock(&iip->ili_lock);
2316 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2317 spin_unlock(&iip->ili_lock);
2318
2319 /*
2320 * Bump the generation count so no one will be confused
2321 * by reincarnations of this inode.
2322 */
2323 VFS_I(ip)->i_generation++;
2324 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2325
2326 if (xic.deleted)
2327 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2328out:
2329 xfs_perag_put(pag);
2330 return error;
2331}
2332
2333/*
2334 * This is called to unpin an inode. The caller must have the inode locked
2335 * in at least shared mode so that the buffer cannot be subsequently pinned
2336 * once someone is waiting for it to be unpinned.
2337 */
2338static void
2339xfs_iunpin(
2340 struct xfs_inode *ip)
2341{
2342 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2343
2344 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2345
2346 /* Give the log a push to start the unpinning I/O */
2347 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2348
2349}
2350
2351static void
2352__xfs_iunpin_wait(
2353 struct xfs_inode *ip)
2354{
2355 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2356 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2357
2358 xfs_iunpin(ip);
2359
2360 do {
2361 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2362 if (xfs_ipincount(ip))
2363 io_schedule();
2364 } while (xfs_ipincount(ip));
2365 finish_wait(wq, &wait.wq_entry);
2366}
2367
2368void
2369xfs_iunpin_wait(
2370 struct xfs_inode *ip)
2371{
2372 if (xfs_ipincount(ip))
2373 __xfs_iunpin_wait(ip);
2374}
2375
2376/*
2377 * Removing an inode from the namespace involves removing the directory entry
2378 * and dropping the link count on the inode. Removing the directory entry can
2379 * result in locking an AGF (directory blocks were freed) and removing a link
2380 * count can result in placing the inode on an unlinked list which results in
2381 * locking an AGI.
2382 *
2383 * The big problem here is that we have an ordering constraint on AGF and AGI
2384 * locking - inode allocation locks the AGI, then can allocate a new extent for
2385 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2386 * removes the inode from the unlinked list, requiring that we lock the AGI
2387 * first, and then freeing the inode can result in an inode chunk being freed
2388 * and hence freeing disk space requiring that we lock an AGF.
2389 *
2390 * Hence the ordering that is imposed by other parts of the code is AGI before
2391 * AGF. This means we cannot remove the directory entry before we drop the inode
2392 * reference count and put it on the unlinked list as this results in a lock
2393 * order of AGF then AGI, and this can deadlock against inode allocation and
2394 * freeing. Therefore we must drop the link counts before we remove the
2395 * directory entry.
2396 *
2397 * This is still safe from a transactional point of view - it is not until we
2398 * get to xfs_defer_finish() that we have the possibility of multiple
2399 * transactions in this operation. Hence as long as we remove the directory
2400 * entry and drop the link count in the first transaction of the remove
2401 * operation, there are no transactional constraints on the ordering here.
2402 */
2403int
2404xfs_remove(
2405 xfs_inode_t *dp,
2406 struct xfs_name *name,
2407 xfs_inode_t *ip)
2408{
2409 xfs_mount_t *mp = dp->i_mount;
2410 xfs_trans_t *tp = NULL;
2411 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2412 int dontcare;
2413 int error = 0;
2414 uint resblks;
2415
2416 trace_xfs_remove(dp, name);
2417
2418 if (xfs_is_shutdown(mp))
2419 return -EIO;
2420
2421 error = xfs_qm_dqattach(dp);
2422 if (error)
2423 goto std_return;
2424
2425 error = xfs_qm_dqattach(ip);
2426 if (error)
2427 goto std_return;
2428
2429 /*
2430 * We try to get the real space reservation first, allowing for
2431 * directory btree deletion(s) implying possible bmap insert(s). If we
2432 * can't get the space reservation then we use 0 instead, and avoid the
2433 * bmap btree insert(s) in the directory code by, if the bmap insert
2434 * tries to happen, instead trimming the LAST block from the directory.
2435 *
2436 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2437 * the directory code can handle a reservationless update and we don't
2438 * want to prevent a user from trying to free space by deleting things.
2439 */
2440 resblks = XFS_REMOVE_SPACE_RES(mp);
2441 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2442 &tp, &dontcare);
2443 if (error) {
2444 ASSERT(error != -ENOSPC);
2445 goto std_return;
2446 }
2447
2448 /*
2449 * If we're removing a directory perform some additional validation.
2450 */
2451 if (is_dir) {
2452 ASSERT(VFS_I(ip)->i_nlink >= 2);
2453 if (VFS_I(ip)->i_nlink != 2) {
2454 error = -ENOTEMPTY;
2455 goto out_trans_cancel;
2456 }
2457 if (!xfs_dir_isempty(ip)) {
2458 error = -ENOTEMPTY;
2459 goto out_trans_cancel;
2460 }
2461
2462 /* Drop the link from ip's "..". */
2463 error = xfs_droplink(tp, dp);
2464 if (error)
2465 goto out_trans_cancel;
2466
2467 /* Drop the "." link from ip to self. */
2468 error = xfs_droplink(tp, ip);
2469 if (error)
2470 goto out_trans_cancel;
2471
2472 /*
2473 * Point the unlinked child directory's ".." entry to the root
2474 * directory to eliminate back-references to inodes that may
2475 * get freed before the child directory is closed. If the fs
2476 * gets shrunk, this can lead to dirent inode validation errors.
2477 */
2478 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2479 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2480 tp->t_mountp->m_sb.sb_rootino, 0);
2481 if (error)
2482 goto out_trans_cancel;
2483 }
2484 } else {
2485 /*
2486 * When removing a non-directory we need to log the parent
2487 * inode here. For a directory this is done implicitly
2488 * by the xfs_droplink call for the ".." entry.
2489 */
2490 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2491 }
2492 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2493
2494 /* Drop the link from dp to ip. */
2495 error = xfs_droplink(tp, ip);
2496 if (error)
2497 goto out_trans_cancel;
2498
2499 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2500 if (error) {
2501 ASSERT(error != -ENOENT);
2502 goto out_trans_cancel;
2503 }
2504
2505 /*
2506 * If this is a synchronous mount, make sure that the
2507 * remove transaction goes to disk before returning to
2508 * the user.
2509 */
2510 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2511 xfs_trans_set_sync(tp);
2512
2513 error = xfs_trans_commit(tp);
2514 if (error)
2515 goto std_return;
2516
2517 if (is_dir && xfs_inode_is_filestream(ip))
2518 xfs_filestream_deassociate(ip);
2519
2520 return 0;
2521
2522 out_trans_cancel:
2523 xfs_trans_cancel(tp);
2524 std_return:
2525 return error;
2526}
2527
2528/*
2529 * Enter all inodes for a rename transaction into a sorted array.
2530 */
2531#define __XFS_SORT_INODES 5
2532STATIC void
2533xfs_sort_for_rename(
2534 struct xfs_inode *dp1, /* in: old (source) directory inode */
2535 struct xfs_inode *dp2, /* in: new (target) directory inode */
2536 struct xfs_inode *ip1, /* in: inode of old entry */
2537 struct xfs_inode *ip2, /* in: inode of new entry */
2538 struct xfs_inode *wip, /* in: whiteout inode */
2539 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2540 int *num_inodes) /* in/out: inodes in array */
2541{
2542 int i, j;
2543
2544 ASSERT(*num_inodes == __XFS_SORT_INODES);
2545 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2546
2547 /*
2548 * i_tab contains a list of pointers to inodes. We initialize
2549 * the table here & we'll sort it. We will then use it to
2550 * order the acquisition of the inode locks.
2551 *
2552 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2553 */
2554 i = 0;
2555 i_tab[i++] = dp1;
2556 i_tab[i++] = dp2;
2557 i_tab[i++] = ip1;
2558 if (ip2)
2559 i_tab[i++] = ip2;
2560 if (wip)
2561 i_tab[i++] = wip;
2562 *num_inodes = i;
2563
2564 /*
2565 * Sort the elements via bubble sort. (Remember, there are at
2566 * most 5 elements to sort, so this is adequate.)
2567 */
2568 for (i = 0; i < *num_inodes; i++) {
2569 for (j = 1; j < *num_inodes; j++) {
2570 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2571 struct xfs_inode *temp = i_tab[j];
2572 i_tab[j] = i_tab[j-1];
2573 i_tab[j-1] = temp;
2574 }
2575 }
2576 }
2577}
2578
2579static int
2580xfs_finish_rename(
2581 struct xfs_trans *tp)
2582{
2583 /*
2584 * If this is a synchronous mount, make sure that the rename transaction
2585 * goes to disk before returning to the user.
2586 */
2587 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2588 xfs_trans_set_sync(tp);
2589
2590 return xfs_trans_commit(tp);
2591}
2592
2593/*
2594 * xfs_cross_rename()
2595 *
2596 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2597 */
2598STATIC int
2599xfs_cross_rename(
2600 struct xfs_trans *tp,
2601 struct xfs_inode *dp1,
2602 struct xfs_name *name1,
2603 struct xfs_inode *ip1,
2604 struct xfs_inode *dp2,
2605 struct xfs_name *name2,
2606 struct xfs_inode *ip2,
2607 int spaceres)
2608{
2609 int error = 0;
2610 int ip1_flags = 0;
2611 int ip2_flags = 0;
2612 int dp2_flags = 0;
2613
2614 /* Swap inode number for dirent in first parent */
2615 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2616 if (error)
2617 goto out_trans_abort;
2618
2619 /* Swap inode number for dirent in second parent */
2620 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2621 if (error)
2622 goto out_trans_abort;
2623
2624 /*
2625 * If we're renaming one or more directories across different parents,
2626 * update the respective ".." entries (and link counts) to match the new
2627 * parents.
2628 */
2629 if (dp1 != dp2) {
2630 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2631
2632 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2633 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2634 dp1->i_ino, spaceres);
2635 if (error)
2636 goto out_trans_abort;
2637
2638 /* transfer ip2 ".." reference to dp1 */
2639 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2640 error = xfs_droplink(tp, dp2);
2641 if (error)
2642 goto out_trans_abort;
2643 xfs_bumplink(tp, dp1);
2644 }
2645
2646 /*
2647 * Although ip1 isn't changed here, userspace needs
2648 * to be warned about the change, so that applications
2649 * relying on it (like backup ones), will properly
2650 * notify the change
2651 */
2652 ip1_flags |= XFS_ICHGTIME_CHG;
2653 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2654 }
2655
2656 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2657 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2658 dp2->i_ino, spaceres);
2659 if (error)
2660 goto out_trans_abort;
2661
2662 /* transfer ip1 ".." reference to dp2 */
2663 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2664 error = xfs_droplink(tp, dp1);
2665 if (error)
2666 goto out_trans_abort;
2667 xfs_bumplink(tp, dp2);
2668 }
2669
2670 /*
2671 * Although ip2 isn't changed here, userspace needs
2672 * to be warned about the change, so that applications
2673 * relying on it (like backup ones), will properly
2674 * notify the change
2675 */
2676 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2677 ip2_flags |= XFS_ICHGTIME_CHG;
2678 }
2679 }
2680
2681 if (ip1_flags) {
2682 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2683 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2684 }
2685 if (ip2_flags) {
2686 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2687 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2688 }
2689 if (dp2_flags) {
2690 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2691 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2692 }
2693 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2694 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2695 return xfs_finish_rename(tp);
2696
2697out_trans_abort:
2698 xfs_trans_cancel(tp);
2699 return error;
2700}
2701
2702/*
2703 * xfs_rename_alloc_whiteout()
2704 *
2705 * Return a referenced, unlinked, unlocked inode that can be used as a
2706 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2707 * crash between allocating the inode and linking it into the rename transaction
2708 * recovery will free the inode and we won't leak it.
2709 */
2710static int
2711xfs_rename_alloc_whiteout(
2712 struct user_namespace *mnt_userns,
2713 struct xfs_name *src_name,
2714 struct xfs_inode *dp,
2715 struct xfs_inode **wip)
2716{
2717 struct xfs_inode *tmpfile;
2718 struct qstr name;
2719 int error;
2720
2721 error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
2722 &tmpfile);
2723 if (error)
2724 return error;
2725
2726 name.name = src_name->name;
2727 name.len = src_name->len;
2728 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2729 if (error) {
2730 xfs_finish_inode_setup(tmpfile);
2731 xfs_irele(tmpfile);
2732 return error;
2733 }
2734
2735 /*
2736 * Prepare the tmpfile inode as if it were created through the VFS.
2737 * Complete the inode setup and flag it as linkable. nlink is already
2738 * zero, so we can skip the drop_nlink.
2739 */
2740 xfs_setup_iops(tmpfile);
2741 xfs_finish_inode_setup(tmpfile);
2742 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2743
2744 *wip = tmpfile;
2745 return 0;
2746}
2747
2748/*
2749 * xfs_rename
2750 */
2751int
2752xfs_rename(
2753 struct user_namespace *mnt_userns,
2754 struct xfs_inode *src_dp,
2755 struct xfs_name *src_name,
2756 struct xfs_inode *src_ip,
2757 struct xfs_inode *target_dp,
2758 struct xfs_name *target_name,
2759 struct xfs_inode *target_ip,
2760 unsigned int flags)
2761{
2762 struct xfs_mount *mp = src_dp->i_mount;
2763 struct xfs_trans *tp;
2764 struct xfs_inode *wip = NULL; /* whiteout inode */
2765 struct xfs_inode *inodes[__XFS_SORT_INODES];
2766 int i;
2767 int num_inodes = __XFS_SORT_INODES;
2768 bool new_parent = (src_dp != target_dp);
2769 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2770 int spaceres;
2771 bool retried = false;
2772 int error, nospace_error = 0;
2773
2774 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2775
2776 if ((flags & RENAME_EXCHANGE) && !target_ip)
2777 return -EINVAL;
2778
2779 /*
2780 * If we are doing a whiteout operation, allocate the whiteout inode
2781 * we will be placing at the target and ensure the type is set
2782 * appropriately.
2783 */
2784 if (flags & RENAME_WHITEOUT) {
2785 error = xfs_rename_alloc_whiteout(mnt_userns, src_name,
2786 target_dp, &wip);
2787 if (error)
2788 return error;
2789
2790 /* setup target dirent info as whiteout */
2791 src_name->type = XFS_DIR3_FT_CHRDEV;
2792 }
2793
2794 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2795 inodes, &num_inodes);
2796
2797retry:
2798 nospace_error = 0;
2799 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2800 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2801 if (error == -ENOSPC) {
2802 nospace_error = error;
2803 spaceres = 0;
2804 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2805 &tp);
2806 }
2807 if (error)
2808 goto out_release_wip;
2809
2810 /*
2811 * Attach the dquots to the inodes
2812 */
2813 error = xfs_qm_vop_rename_dqattach(inodes);
2814 if (error)
2815 goto out_trans_cancel;
2816
2817 /*
2818 * Lock all the participating inodes. Depending upon whether
2819 * the target_name exists in the target directory, and
2820 * whether the target directory is the same as the source
2821 * directory, we can lock from 2 to 5 inodes.
2822 */
2823 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2824
2825 /*
2826 * Join all the inodes to the transaction. From this point on,
2827 * we can rely on either trans_commit or trans_cancel to unlock
2828 * them.
2829 */
2830 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2831 if (new_parent)
2832 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2833 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2834 if (target_ip)
2835 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2836 if (wip)
2837 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2838
2839 /*
2840 * If we are using project inheritance, we only allow renames
2841 * into our tree when the project IDs are the same; else the
2842 * tree quota mechanism would be circumvented.
2843 */
2844 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2845 target_dp->i_projid != src_ip->i_projid)) {
2846 error = -EXDEV;
2847 goto out_trans_cancel;
2848 }
2849
2850 /* RENAME_EXCHANGE is unique from here on. */
2851 if (flags & RENAME_EXCHANGE)
2852 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2853 target_dp, target_name, target_ip,
2854 spaceres);
2855
2856 /*
2857 * Try to reserve quota to handle an expansion of the target directory.
2858 * We'll allow the rename to continue in reservationless mode if we hit
2859 * a space usage constraint. If we trigger reservationless mode, save
2860 * the errno if there isn't any free space in the target directory.
2861 */
2862 if (spaceres != 0) {
2863 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2864 0, false);
2865 if (error == -EDQUOT || error == -ENOSPC) {
2866 if (!retried) {
2867 xfs_trans_cancel(tp);
2868 xfs_blockgc_free_quota(target_dp, 0);
2869 retried = true;
2870 goto retry;
2871 }
2872
2873 nospace_error = error;
2874 spaceres = 0;
2875 error = 0;
2876 }
2877 if (error)
2878 goto out_trans_cancel;
2879 }
2880
2881 /*
2882 * Check for expected errors before we dirty the transaction
2883 * so we can return an error without a transaction abort.
2884 */
2885 if (target_ip == NULL) {
2886 /*
2887 * If there's no space reservation, check the entry will
2888 * fit before actually inserting it.
2889 */
2890 if (!spaceres) {
2891 error = xfs_dir_canenter(tp, target_dp, target_name);
2892 if (error)
2893 goto out_trans_cancel;
2894 }
2895 } else {
2896 /*
2897 * If target exists and it's a directory, check that whether
2898 * it can be destroyed.
2899 */
2900 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
2901 (!xfs_dir_isempty(target_ip) ||
2902 (VFS_I(target_ip)->i_nlink > 2))) {
2903 error = -EEXIST;
2904 goto out_trans_cancel;
2905 }
2906 }
2907
2908 /*
2909 * Lock the AGI buffers we need to handle bumping the nlink of the
2910 * whiteout inode off the unlinked list and to handle dropping the
2911 * nlink of the target inode. Per locking order rules, do this in
2912 * increasing AG order and before directory block allocation tries to
2913 * grab AGFs because we grab AGIs before AGFs.
2914 *
2915 * The (vfs) caller must ensure that if src is a directory then
2916 * target_ip is either null or an empty directory.
2917 */
2918 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2919 if (inodes[i] == wip ||
2920 (inodes[i] == target_ip &&
2921 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2922 struct xfs_perag *pag;
2923 struct xfs_buf *bp;
2924
2925 pag = xfs_perag_get(mp,
2926 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2927 error = xfs_read_agi(pag, tp, &bp);
2928 xfs_perag_put(pag);
2929 if (error)
2930 goto out_trans_cancel;
2931 }
2932 }
2933
2934 /*
2935 * Directory entry creation below may acquire the AGF. Remove
2936 * the whiteout from the unlinked list first to preserve correct
2937 * AGI/AGF locking order. This dirties the transaction so failures
2938 * after this point will abort and log recovery will clean up the
2939 * mess.
2940 *
2941 * For whiteouts, we need to bump the link count on the whiteout
2942 * inode. After this point, we have a real link, clear the tmpfile
2943 * state flag from the inode so it doesn't accidentally get misused
2944 * in future.
2945 */
2946 if (wip) {
2947 struct xfs_perag *pag;
2948
2949 ASSERT(VFS_I(wip)->i_nlink == 0);
2950
2951 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
2952 error = xfs_iunlink_remove(tp, pag, wip);
2953 xfs_perag_put(pag);
2954 if (error)
2955 goto out_trans_cancel;
2956
2957 xfs_bumplink(tp, wip);
2958 VFS_I(wip)->i_state &= ~I_LINKABLE;
2959 }
2960
2961 /*
2962 * Set up the target.
2963 */
2964 if (target_ip == NULL) {
2965 /*
2966 * If target does not exist and the rename crosses
2967 * directories, adjust the target directory link count
2968 * to account for the ".." reference from the new entry.
2969 */
2970 error = xfs_dir_createname(tp, target_dp, target_name,
2971 src_ip->i_ino, spaceres);
2972 if (error)
2973 goto out_trans_cancel;
2974
2975 xfs_trans_ichgtime(tp, target_dp,
2976 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2977
2978 if (new_parent && src_is_directory) {
2979 xfs_bumplink(tp, target_dp);
2980 }
2981 } else { /* target_ip != NULL */
2982 /*
2983 * Link the source inode under the target name.
2984 * If the source inode is a directory and we are moving
2985 * it across directories, its ".." entry will be
2986 * inconsistent until we replace that down below.
2987 *
2988 * In case there is already an entry with the same
2989 * name at the destination directory, remove it first.
2990 */
2991 error = xfs_dir_replace(tp, target_dp, target_name,
2992 src_ip->i_ino, spaceres);
2993 if (error)
2994 goto out_trans_cancel;
2995
2996 xfs_trans_ichgtime(tp, target_dp,
2997 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2998
2999 /*
3000 * Decrement the link count on the target since the target
3001 * dir no longer points to it.
3002 */
3003 error = xfs_droplink(tp, target_ip);
3004 if (error)
3005 goto out_trans_cancel;
3006
3007 if (src_is_directory) {
3008 /*
3009 * Drop the link from the old "." entry.
3010 */
3011 error = xfs_droplink(tp, target_ip);
3012 if (error)
3013 goto out_trans_cancel;
3014 }
3015 } /* target_ip != NULL */
3016
3017 /*
3018 * Remove the source.
3019 */
3020 if (new_parent && src_is_directory) {
3021 /*
3022 * Rewrite the ".." entry to point to the new
3023 * directory.
3024 */
3025 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3026 target_dp->i_ino, spaceres);
3027 ASSERT(error != -EEXIST);
3028 if (error)
3029 goto out_trans_cancel;
3030 }
3031
3032 /*
3033 * We always want to hit the ctime on the source inode.
3034 *
3035 * This isn't strictly required by the standards since the source
3036 * inode isn't really being changed, but old unix file systems did
3037 * it and some incremental backup programs won't work without it.
3038 */
3039 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3040 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3041
3042 /*
3043 * Adjust the link count on src_dp. This is necessary when
3044 * renaming a directory, either within one parent when
3045 * the target existed, or across two parent directories.
3046 */
3047 if (src_is_directory && (new_parent || target_ip != NULL)) {
3048
3049 /*
3050 * Decrement link count on src_directory since the
3051 * entry that's moved no longer points to it.
3052 */
3053 error = xfs_droplink(tp, src_dp);
3054 if (error)
3055 goto out_trans_cancel;
3056 }
3057
3058 /*
3059 * For whiteouts, we only need to update the source dirent with the
3060 * inode number of the whiteout inode rather than removing it
3061 * altogether.
3062 */
3063 if (wip)
3064 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3065 spaceres);
3066 else
3067 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3068 spaceres);
3069
3070 if (error)
3071 goto out_trans_cancel;
3072
3073 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3074 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3075 if (new_parent)
3076 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3077
3078 error = xfs_finish_rename(tp);
3079 if (wip)
3080 xfs_irele(wip);
3081 return error;
3082
3083out_trans_cancel:
3084 xfs_trans_cancel(tp);
3085out_release_wip:
3086 if (wip)
3087 xfs_irele(wip);
3088 if (error == -ENOSPC && nospace_error)
3089 error = nospace_error;
3090 return error;
3091}
3092
3093static int
3094xfs_iflush(
3095 struct xfs_inode *ip,
3096 struct xfs_buf *bp)
3097{
3098 struct xfs_inode_log_item *iip = ip->i_itemp;
3099 struct xfs_dinode *dip;
3100 struct xfs_mount *mp = ip->i_mount;
3101 int error;
3102
3103 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3104 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3105 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3106 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3107 ASSERT(iip->ili_item.li_buf == bp);
3108
3109 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3110
3111 /*
3112 * We don't flush the inode if any of the following checks fail, but we
3113 * do still update the log item and attach to the backing buffer as if
3114 * the flush happened. This is a formality to facilitate predictable
3115 * error handling as the caller will shutdown and fail the buffer.
3116 */
3117 error = -EFSCORRUPTED;
3118 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3119 mp, XFS_ERRTAG_IFLUSH_1)) {
3120 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3121 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3122 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3123 goto flush_out;
3124 }
3125 if (S_ISREG(VFS_I(ip)->i_mode)) {
3126 if (XFS_TEST_ERROR(
3127 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3128 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3129 mp, XFS_ERRTAG_IFLUSH_3)) {
3130 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3131 "%s: Bad regular inode %llu, ptr "PTR_FMT,
3132 __func__, ip->i_ino, ip);
3133 goto flush_out;
3134 }
3135 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3136 if (XFS_TEST_ERROR(
3137 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3138 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3139 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3140 mp, XFS_ERRTAG_IFLUSH_4)) {
3141 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3142 "%s: Bad directory inode %llu, ptr "PTR_FMT,
3143 __func__, ip->i_ino, ip);
3144 goto flush_out;
3145 }
3146 }
3147 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3148 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3149 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3150 "%s: detected corrupt incore inode %llu, "
3151 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3152 __func__, ip->i_ino,
3153 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3154 ip->i_nblocks, ip);
3155 goto flush_out;
3156 }
3157 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3158 mp, XFS_ERRTAG_IFLUSH_6)) {
3159 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3160 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3161 __func__, ip->i_ino, ip->i_forkoff, ip);
3162 goto flush_out;
3163 }
3164
3165 /*
3166 * Inode item log recovery for v2 inodes are dependent on the flushiter
3167 * count for correct sequencing. We bump the flush iteration count so
3168 * we can detect flushes which postdate a log record during recovery.
3169 * This is redundant as we now log every change and hence this can't
3170 * happen but we need to still do it to ensure backwards compatibility
3171 * with old kernels that predate logging all inode changes.
3172 */
3173 if (!xfs_has_v3inodes(mp))
3174 ip->i_flushiter++;
3175
3176 /*
3177 * If there are inline format data / attr forks attached to this inode,
3178 * make sure they are not corrupt.
3179 */
3180 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3181 xfs_ifork_verify_local_data(ip))
3182 goto flush_out;
3183 if (xfs_inode_has_attr_fork(ip) &&
3184 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3185 xfs_ifork_verify_local_attr(ip))
3186 goto flush_out;
3187
3188 /*
3189 * Copy the dirty parts of the inode into the on-disk inode. We always
3190 * copy out the core of the inode, because if the inode is dirty at all
3191 * the core must be.
3192 */
3193 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3194
3195 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3196 if (!xfs_has_v3inodes(mp)) {
3197 if (ip->i_flushiter == DI_MAX_FLUSH)
3198 ip->i_flushiter = 0;
3199 }
3200
3201 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3202 if (xfs_inode_has_attr_fork(ip))
3203 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3204
3205 /*
3206 * We've recorded everything logged in the inode, so we'd like to clear
3207 * the ili_fields bits so we don't log and flush things unnecessarily.
3208 * However, we can't stop logging all this information until the data
3209 * we've copied into the disk buffer is written to disk. If we did we
3210 * might overwrite the copy of the inode in the log with all the data
3211 * after re-logging only part of it, and in the face of a crash we
3212 * wouldn't have all the data we need to recover.
3213 *
3214 * What we do is move the bits to the ili_last_fields field. When
3215 * logging the inode, these bits are moved back to the ili_fields field.
3216 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3217 * we know that the information those bits represent is permanently on
3218 * disk. As long as the flush completes before the inode is logged
3219 * again, then both ili_fields and ili_last_fields will be cleared.
3220 */
3221 error = 0;
3222flush_out:
3223 spin_lock(&iip->ili_lock);
3224 iip->ili_last_fields = iip->ili_fields;
3225 iip->ili_fields = 0;
3226 iip->ili_fsync_fields = 0;
3227 spin_unlock(&iip->ili_lock);
3228
3229 /*
3230 * Store the current LSN of the inode so that we can tell whether the
3231 * item has moved in the AIL from xfs_buf_inode_iodone().
3232 */
3233 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3234 &iip->ili_item.li_lsn);
3235
3236 /* generate the checksum. */
3237 xfs_dinode_calc_crc(mp, dip);
3238 return error;
3239}
3240
3241/*
3242 * Non-blocking flush of dirty inode metadata into the backing buffer.
3243 *
3244 * The caller must have a reference to the inode and hold the cluster buffer
3245 * locked. The function will walk across all the inodes on the cluster buffer it
3246 * can find and lock without blocking, and flush them to the cluster buffer.
3247 *
3248 * On successful flushing of at least one inode, the caller must write out the
3249 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3250 * the caller needs to release the buffer. On failure, the filesystem will be
3251 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3252 * will be returned.
3253 */
3254int
3255xfs_iflush_cluster(
3256 struct xfs_buf *bp)
3257{
3258 struct xfs_mount *mp = bp->b_mount;
3259 struct xfs_log_item *lip, *n;
3260 struct xfs_inode *ip;
3261 struct xfs_inode_log_item *iip;
3262 int clcount = 0;
3263 int error = 0;
3264
3265 /*
3266 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3267 * will remove itself from the list.
3268 */
3269 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3270 iip = (struct xfs_inode_log_item *)lip;
3271 ip = iip->ili_inode;
3272
3273 /*
3274 * Quick and dirty check to avoid locks if possible.
3275 */
3276 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3277 continue;
3278 if (xfs_ipincount(ip))
3279 continue;
3280
3281 /*
3282 * The inode is still attached to the buffer, which means it is
3283 * dirty but reclaim might try to grab it. Check carefully for
3284 * that, and grab the ilock while still holding the i_flags_lock
3285 * to guarantee reclaim will not be able to reclaim this inode
3286 * once we drop the i_flags_lock.
3287 */
3288 spin_lock(&ip->i_flags_lock);
3289 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3290 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3291 spin_unlock(&ip->i_flags_lock);
3292 continue;
3293 }
3294
3295 /*
3296 * ILOCK will pin the inode against reclaim and prevent
3297 * concurrent transactions modifying the inode while we are
3298 * flushing the inode. If we get the lock, set the flushing
3299 * state before we drop the i_flags_lock.
3300 */
3301 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3302 spin_unlock(&ip->i_flags_lock);
3303 continue;
3304 }
3305 __xfs_iflags_set(ip, XFS_IFLUSHING);
3306 spin_unlock(&ip->i_flags_lock);
3307
3308 /*
3309 * Abort flushing this inode if we are shut down because the
3310 * inode may not currently be in the AIL. This can occur when
3311 * log I/O failure unpins the inode without inserting into the
3312 * AIL, leaving a dirty/unpinned inode attached to the buffer
3313 * that otherwise looks like it should be flushed.
3314 */
3315 if (xlog_is_shutdown(mp->m_log)) {
3316 xfs_iunpin_wait(ip);
3317 xfs_iflush_abort(ip);
3318 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3319 error = -EIO;
3320 continue;
3321 }
3322
3323 /* don't block waiting on a log force to unpin dirty inodes */
3324 if (xfs_ipincount(ip)) {
3325 xfs_iflags_clear(ip, XFS_IFLUSHING);
3326 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3327 continue;
3328 }
3329
3330 if (!xfs_inode_clean(ip))
3331 error = xfs_iflush(ip, bp);
3332 else
3333 xfs_iflags_clear(ip, XFS_IFLUSHING);
3334 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3335 if (error)
3336 break;
3337 clcount++;
3338 }
3339
3340 if (error) {
3341 /*
3342 * Shutdown first so we kill the log before we release this
3343 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3344 * of the log, failing it before the _log_ is shut down can
3345 * result in the log tail being moved forward in the journal
3346 * on disk because log writes can still be taking place. Hence
3347 * unpinning the tail will allow the ICREATE intent to be
3348 * removed from the log an recovery will fail with uninitialised
3349 * inode cluster buffers.
3350 */
3351 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3352 bp->b_flags |= XBF_ASYNC;
3353 xfs_buf_ioend_fail(bp);
3354 return error;
3355 }
3356
3357 if (!clcount)
3358 return -EAGAIN;
3359
3360 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3361 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3362 return 0;
3363
3364}
3365
3366/* Release an inode. */
3367void
3368xfs_irele(
3369 struct xfs_inode *ip)
3370{
3371 trace_xfs_irele(ip, _RET_IP_);
3372 iput(VFS_I(ip));
3373}
3374
3375/*
3376 * Ensure all commited transactions touching the inode are written to the log.
3377 */
3378int
3379xfs_log_force_inode(
3380 struct xfs_inode *ip)
3381{
3382 xfs_csn_t seq = 0;
3383
3384 xfs_ilock(ip, XFS_ILOCK_SHARED);
3385 if (xfs_ipincount(ip))
3386 seq = ip->i_itemp->ili_commit_seq;
3387 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3388
3389 if (!seq)
3390 return 0;
3391 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3392}
3393
3394/*
3395 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3396 * abide vfs locking order (lowest pointer value goes first) and breaking the
3397 * layout leases before proceeding. The loop is needed because we cannot call
3398 * the blocking break_layout() with the iolocks held, and therefore have to
3399 * back out both locks.
3400 */
3401static int
3402xfs_iolock_two_inodes_and_break_layout(
3403 struct inode *src,
3404 struct inode *dest)
3405{
3406 int error;
3407
3408 if (src > dest)
3409 swap(src, dest);
3410
3411retry:
3412 /* Wait to break both inodes' layouts before we start locking. */
3413 error = break_layout(src, true);
3414 if (error)
3415 return error;
3416 if (src != dest) {
3417 error = break_layout(dest, true);
3418 if (error)
3419 return error;
3420 }
3421
3422 /* Lock one inode and make sure nobody got in and leased it. */
3423 inode_lock(src);
3424 error = break_layout(src, false);
3425 if (error) {
3426 inode_unlock(src);
3427 if (error == -EWOULDBLOCK)
3428 goto retry;
3429 return error;
3430 }
3431
3432 if (src == dest)
3433 return 0;
3434
3435 /* Lock the other inode and make sure nobody got in and leased it. */
3436 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3437 error = break_layout(dest, false);
3438 if (error) {
3439 inode_unlock(src);
3440 inode_unlock(dest);
3441 if (error == -EWOULDBLOCK)
3442 goto retry;
3443 return error;
3444 }
3445
3446 return 0;
3447}
3448
3449static int
3450xfs_mmaplock_two_inodes_and_break_dax_layout(
3451 struct xfs_inode *ip1,
3452 struct xfs_inode *ip2)
3453{
3454 int error;
3455 bool retry;
3456 struct page *page;
3457
3458 if (ip1->i_ino > ip2->i_ino)
3459 swap(ip1, ip2);
3460
3461again:
3462 retry = false;
3463 /* Lock the first inode */
3464 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3465 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3466 if (error || retry) {
3467 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3468 if (error == 0 && retry)
3469 goto again;
3470 return error;
3471 }
3472
3473 if (ip1 == ip2)
3474 return 0;
3475
3476 /* Nested lock the second inode */
3477 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3478 /*
3479 * We cannot use xfs_break_dax_layouts() directly here because it may
3480 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3481 * for this nested lock case.
3482 */
3483 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3484 if (page && page_ref_count(page) != 1) {
3485 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3486 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3487 goto again;
3488 }
3489
3490 return 0;
3491}
3492
3493/*
3494 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3495 * mmap activity.
3496 */
3497int
3498xfs_ilock2_io_mmap(
3499 struct xfs_inode *ip1,
3500 struct xfs_inode *ip2)
3501{
3502 int ret;
3503
3504 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3505 if (ret)
3506 return ret;
3507
3508 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3509 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3510 if (ret) {
3511 inode_unlock(VFS_I(ip2));
3512 if (ip1 != ip2)
3513 inode_unlock(VFS_I(ip1));
3514 return ret;
3515 }
3516 } else
3517 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3518 VFS_I(ip2)->i_mapping);
3519
3520 return 0;
3521}
3522
3523/* Unlock both inodes to allow IO and mmap activity. */
3524void
3525xfs_iunlock2_io_mmap(
3526 struct xfs_inode *ip1,
3527 struct xfs_inode *ip2)
3528{
3529 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3530 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3531 if (ip1 != ip2)
3532 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3533 } else
3534 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3535 VFS_I(ip2)->i_mapping);
3536
3537 inode_unlock(VFS_I(ip2));
3538 if (ip1 != ip2)
3539 inode_unlock(VFS_I(ip1));
3540}
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_types.h"
23#include "xfs_log.h"
24#include "xfs_inum.h"
25#include "xfs_trans.h"
26#include "xfs_trans_priv.h"
27#include "xfs_sb.h"
28#include "xfs_ag.h"
29#include "xfs_mount.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
33#include "xfs_attr_sf.h"
34#include "xfs_dinode.h"
35#include "xfs_inode.h"
36#include "xfs_buf_item.h"
37#include "xfs_inode_item.h"
38#include "xfs_btree.h"
39#include "xfs_alloc.h"
40#include "xfs_ialloc.h"
41#include "xfs_bmap.h"
42#include "xfs_error.h"
43#include "xfs_utils.h"
44#include "xfs_quota.h"
45#include "xfs_filestream.h"
46#include "xfs_vnodeops.h"
47#include "xfs_trace.h"
48
49kmem_zone_t *xfs_ifork_zone;
50kmem_zone_t *xfs_inode_zone;
51
52/*
53 * Used in xfs_itruncate_extents(). This is the maximum number of extents
54 * freed from a file in a single transaction.
55 */
56#define XFS_ITRUNC_MAX_EXTENTS 2
57
58STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
59STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
60STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
61STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
62
63/*
64 * helper function to extract extent size hint from inode
65 */
66xfs_extlen_t
67xfs_get_extsz_hint(
68 struct xfs_inode *ip)
69{
70 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
71 return ip->i_d.di_extsize;
72 if (XFS_IS_REALTIME_INODE(ip))
73 return ip->i_mount->m_sb.sb_rextsize;
74 return 0;
75}
76
77#ifdef DEBUG
78/*
79 * Make sure that the extents in the given memory buffer
80 * are valid.
81 */
82STATIC void
83xfs_validate_extents(
84 xfs_ifork_t *ifp,
85 int nrecs,
86 xfs_exntfmt_t fmt)
87{
88 xfs_bmbt_irec_t irec;
89 xfs_bmbt_rec_host_t rec;
90 int i;
91
92 for (i = 0; i < nrecs; i++) {
93 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
94 rec.l0 = get_unaligned(&ep->l0);
95 rec.l1 = get_unaligned(&ep->l1);
96 xfs_bmbt_get_all(&rec, &irec);
97 if (fmt == XFS_EXTFMT_NOSTATE)
98 ASSERT(irec.br_state == XFS_EXT_NORM);
99 }
100}
101#else /* DEBUG */
102#define xfs_validate_extents(ifp, nrecs, fmt)
103#endif /* DEBUG */
104
105/*
106 * Check that none of the inode's in the buffer have a next
107 * unlinked field of 0.
108 */
109#if defined(DEBUG)
110void
111xfs_inobp_check(
112 xfs_mount_t *mp,
113 xfs_buf_t *bp)
114{
115 int i;
116 int j;
117 xfs_dinode_t *dip;
118
119 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
120
121 for (i = 0; i < j; i++) {
122 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
123 i * mp->m_sb.sb_inodesize);
124 if (!dip->di_next_unlinked) {
125 xfs_alert(mp,
126 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
127 bp);
128 ASSERT(dip->di_next_unlinked);
129 }
130 }
131}
132#endif
133
134/*
135 * Find the buffer associated with the given inode map
136 * We do basic validation checks on the buffer once it has been
137 * retrieved from disk.
138 */
139STATIC int
140xfs_imap_to_bp(
141 xfs_mount_t *mp,
142 xfs_trans_t *tp,
143 struct xfs_imap *imap,
144 xfs_buf_t **bpp,
145 uint buf_flags,
146 uint iget_flags)
147{
148 int error;
149 int i;
150 int ni;
151 xfs_buf_t *bp;
152
153 buf_flags |= XBF_UNMAPPED;
154 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
155 (int)imap->im_len, buf_flags, &bp);
156 if (error) {
157 if (error != EAGAIN) {
158 xfs_warn(mp,
159 "%s: xfs_trans_read_buf() returned error %d.",
160 __func__, error);
161 } else {
162 ASSERT(buf_flags & XBF_TRYLOCK);
163 }
164 return error;
165 }
166
167 /*
168 * Validate the magic number and version of every inode in the buffer
169 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
170 */
171#ifdef DEBUG
172 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
173#else /* usual case */
174 ni = 1;
175#endif
176
177 for (i = 0; i < ni; i++) {
178 int di_ok;
179 xfs_dinode_t *dip;
180
181 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
182 (i << mp->m_sb.sb_inodelog));
183 di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
184 XFS_DINODE_GOOD_VERSION(dip->di_version);
185 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
186 XFS_ERRTAG_ITOBP_INOTOBP,
187 XFS_RANDOM_ITOBP_INOTOBP))) {
188 if (iget_flags & XFS_IGET_UNTRUSTED) {
189 xfs_trans_brelse(tp, bp);
190 return XFS_ERROR(EINVAL);
191 }
192 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
193 XFS_ERRLEVEL_HIGH, mp, dip);
194#ifdef DEBUG
195 xfs_emerg(mp,
196 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
197 (unsigned long long)imap->im_blkno, i,
198 be16_to_cpu(dip->di_magic));
199 ASSERT(0);
200#endif
201 xfs_trans_brelse(tp, bp);
202 return XFS_ERROR(EFSCORRUPTED);
203 }
204 }
205
206 xfs_inobp_check(mp, bp);
207 *bpp = bp;
208 return 0;
209}
210
211/*
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
217 *
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
220 *
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
223 */
224int
225xfs_inotobp(
226 xfs_mount_t *mp,
227 xfs_trans_t *tp,
228 xfs_ino_t ino,
229 xfs_dinode_t **dipp,
230 xfs_buf_t **bpp,
231 int *offset,
232 uint imap_flags)
233{
234 struct xfs_imap imap;
235 xfs_buf_t *bp;
236 int error;
237
238 imap.im_blkno = 0;
239 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
240 if (error)
241 return error;
242
243 error = xfs_imap_to_bp(mp, tp, &imap, &bp, 0, imap_flags);
244 if (error)
245 return error;
246
247 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
248 *bpp = bp;
249 *offset = imap.im_boffset;
250 return 0;
251}
252
253
254/*
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
259 * that buffer.
260 *
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
263 *
264 * The inode is expected to already been mapped to its buffer and read
265 * in once, thus we can use the mapping information stored in the inode
266 * rather than calling xfs_imap(). This allows us to avoid the overhead
267 * of looking at the inode btree for small block file systems
268 * (see xfs_imap()).
269 */
270int
271xfs_itobp(
272 xfs_mount_t *mp,
273 xfs_trans_t *tp,
274 xfs_inode_t *ip,
275 xfs_dinode_t **dipp,
276 xfs_buf_t **bpp,
277 uint buf_flags)
278{
279 xfs_buf_t *bp;
280 int error;
281
282 ASSERT(ip->i_imap.im_blkno != 0);
283
284 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
285 if (error)
286 return error;
287
288 if (!bp) {
289 ASSERT(buf_flags & XBF_TRYLOCK);
290 ASSERT(tp == NULL);
291 *bpp = NULL;
292 return EAGAIN;
293 }
294
295 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
296 *bpp = bp;
297 return 0;
298}
299
300/*
301 * Move inode type and inode format specific information from the
302 * on-disk inode to the in-core inode. For fifos, devs, and sockets
303 * this means set if_rdev to the proper value. For files, directories,
304 * and symlinks this means to bring in the in-line data or extent
305 * pointers. For a file in B-tree format, only the root is immediately
306 * brought in-core. The rest will be in-lined in if_extents when it
307 * is first referenced (see xfs_iread_extents()).
308 */
309STATIC int
310xfs_iformat(
311 xfs_inode_t *ip,
312 xfs_dinode_t *dip)
313{
314 xfs_attr_shortform_t *atp;
315 int size;
316 int error = 0;
317 xfs_fsize_t di_size;
318
319 if (unlikely(be32_to_cpu(dip->di_nextents) +
320 be16_to_cpu(dip->di_anextents) >
321 be64_to_cpu(dip->di_nblocks))) {
322 xfs_warn(ip->i_mount,
323 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
324 (unsigned long long)ip->i_ino,
325 (int)(be32_to_cpu(dip->di_nextents) +
326 be16_to_cpu(dip->di_anextents)),
327 (unsigned long long)
328 be64_to_cpu(dip->di_nblocks));
329 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
330 ip->i_mount, dip);
331 return XFS_ERROR(EFSCORRUPTED);
332 }
333
334 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
335 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
336 (unsigned long long)ip->i_ino,
337 dip->di_forkoff);
338 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
339 ip->i_mount, dip);
340 return XFS_ERROR(EFSCORRUPTED);
341 }
342
343 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
344 !ip->i_mount->m_rtdev_targp)) {
345 xfs_warn(ip->i_mount,
346 "corrupt dinode %Lu, has realtime flag set.",
347 ip->i_ino);
348 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
349 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
350 return XFS_ERROR(EFSCORRUPTED);
351 }
352
353 switch (ip->i_d.di_mode & S_IFMT) {
354 case S_IFIFO:
355 case S_IFCHR:
356 case S_IFBLK:
357 case S_IFSOCK:
358 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
359 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
360 ip->i_mount, dip);
361 return XFS_ERROR(EFSCORRUPTED);
362 }
363 ip->i_d.di_size = 0;
364 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
365 break;
366
367 case S_IFREG:
368 case S_IFLNK:
369 case S_IFDIR:
370 switch (dip->di_format) {
371 case XFS_DINODE_FMT_LOCAL:
372 /*
373 * no local regular files yet
374 */
375 if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
376 xfs_warn(ip->i_mount,
377 "corrupt inode %Lu (local format for regular file).",
378 (unsigned long long) ip->i_ino);
379 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
380 XFS_ERRLEVEL_LOW,
381 ip->i_mount, dip);
382 return XFS_ERROR(EFSCORRUPTED);
383 }
384
385 di_size = be64_to_cpu(dip->di_size);
386 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
387 xfs_warn(ip->i_mount,
388 "corrupt inode %Lu (bad size %Ld for local inode).",
389 (unsigned long long) ip->i_ino,
390 (long long) di_size);
391 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
392 XFS_ERRLEVEL_LOW,
393 ip->i_mount, dip);
394 return XFS_ERROR(EFSCORRUPTED);
395 }
396
397 size = (int)di_size;
398 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
399 break;
400 case XFS_DINODE_FMT_EXTENTS:
401 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
402 break;
403 case XFS_DINODE_FMT_BTREE:
404 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
405 break;
406 default:
407 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
408 ip->i_mount);
409 return XFS_ERROR(EFSCORRUPTED);
410 }
411 break;
412
413 default:
414 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
415 return XFS_ERROR(EFSCORRUPTED);
416 }
417 if (error) {
418 return error;
419 }
420 if (!XFS_DFORK_Q(dip))
421 return 0;
422
423 ASSERT(ip->i_afp == NULL);
424 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
425
426 switch (dip->di_aformat) {
427 case XFS_DINODE_FMT_LOCAL:
428 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
429 size = be16_to_cpu(atp->hdr.totsize);
430
431 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
432 xfs_warn(ip->i_mount,
433 "corrupt inode %Lu (bad attr fork size %Ld).",
434 (unsigned long long) ip->i_ino,
435 (long long) size);
436 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
437 XFS_ERRLEVEL_LOW,
438 ip->i_mount, dip);
439 return XFS_ERROR(EFSCORRUPTED);
440 }
441
442 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
443 break;
444 case XFS_DINODE_FMT_EXTENTS:
445 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
446 break;
447 case XFS_DINODE_FMT_BTREE:
448 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
449 break;
450 default:
451 error = XFS_ERROR(EFSCORRUPTED);
452 break;
453 }
454 if (error) {
455 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
456 ip->i_afp = NULL;
457 xfs_idestroy_fork(ip, XFS_DATA_FORK);
458 }
459 return error;
460}
461
462/*
463 * The file is in-lined in the on-disk inode.
464 * If it fits into if_inline_data, then copy
465 * it there, otherwise allocate a buffer for it
466 * and copy the data there. Either way, set
467 * if_data to point at the data.
468 * If we allocate a buffer for the data, make
469 * sure that its size is a multiple of 4 and
470 * record the real size in i_real_bytes.
471 */
472STATIC int
473xfs_iformat_local(
474 xfs_inode_t *ip,
475 xfs_dinode_t *dip,
476 int whichfork,
477 int size)
478{
479 xfs_ifork_t *ifp;
480 int real_size;
481
482 /*
483 * If the size is unreasonable, then something
484 * is wrong and we just bail out rather than crash in
485 * kmem_alloc() or memcpy() below.
486 */
487 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
488 xfs_warn(ip->i_mount,
489 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
490 (unsigned long long) ip->i_ino, size,
491 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
492 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
493 ip->i_mount, dip);
494 return XFS_ERROR(EFSCORRUPTED);
495 }
496 ifp = XFS_IFORK_PTR(ip, whichfork);
497 real_size = 0;
498 if (size == 0)
499 ifp->if_u1.if_data = NULL;
500 else if (size <= sizeof(ifp->if_u2.if_inline_data))
501 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
502 else {
503 real_size = roundup(size, 4);
504 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
505 }
506 ifp->if_bytes = size;
507 ifp->if_real_bytes = real_size;
508 if (size)
509 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
510 ifp->if_flags &= ~XFS_IFEXTENTS;
511 ifp->if_flags |= XFS_IFINLINE;
512 return 0;
513}
514
515/*
516 * The file consists of a set of extents all
517 * of which fit into the on-disk inode.
518 * If there are few enough extents to fit into
519 * the if_inline_ext, then copy them there.
520 * Otherwise allocate a buffer for them and copy
521 * them into it. Either way, set if_extents
522 * to point at the extents.
523 */
524STATIC int
525xfs_iformat_extents(
526 xfs_inode_t *ip,
527 xfs_dinode_t *dip,
528 int whichfork)
529{
530 xfs_bmbt_rec_t *dp;
531 xfs_ifork_t *ifp;
532 int nex;
533 int size;
534 int i;
535
536 ifp = XFS_IFORK_PTR(ip, whichfork);
537 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
538 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
539
540 /*
541 * If the number of extents is unreasonable, then something
542 * is wrong and we just bail out rather than crash in
543 * kmem_alloc() or memcpy() below.
544 */
545 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
546 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
547 (unsigned long long) ip->i_ino, nex);
548 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
549 ip->i_mount, dip);
550 return XFS_ERROR(EFSCORRUPTED);
551 }
552
553 ifp->if_real_bytes = 0;
554 if (nex == 0)
555 ifp->if_u1.if_extents = NULL;
556 else if (nex <= XFS_INLINE_EXTS)
557 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
558 else
559 xfs_iext_add(ifp, 0, nex);
560
561 ifp->if_bytes = size;
562 if (size) {
563 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
564 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
565 for (i = 0; i < nex; i++, dp++) {
566 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
567 ep->l0 = get_unaligned_be64(&dp->l0);
568 ep->l1 = get_unaligned_be64(&dp->l1);
569 }
570 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
571 if (whichfork != XFS_DATA_FORK ||
572 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
573 if (unlikely(xfs_check_nostate_extents(
574 ifp, 0, nex))) {
575 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
576 XFS_ERRLEVEL_LOW,
577 ip->i_mount);
578 return XFS_ERROR(EFSCORRUPTED);
579 }
580 }
581 ifp->if_flags |= XFS_IFEXTENTS;
582 return 0;
583}
584
585/*
586 * The file has too many extents to fit into
587 * the inode, so they are in B-tree format.
588 * Allocate a buffer for the root of the B-tree
589 * and copy the root into it. The i_extents
590 * field will remain NULL until all of the
591 * extents are read in (when they are needed).
592 */
593STATIC int
594xfs_iformat_btree(
595 xfs_inode_t *ip,
596 xfs_dinode_t *dip,
597 int whichfork)
598{
599 xfs_bmdr_block_t *dfp;
600 xfs_ifork_t *ifp;
601 /* REFERENCED */
602 int nrecs;
603 int size;
604
605 ifp = XFS_IFORK_PTR(ip, whichfork);
606 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
607 size = XFS_BMAP_BROOT_SPACE(dfp);
608 nrecs = be16_to_cpu(dfp->bb_numrecs);
609
610 /*
611 * blow out if -- fork has less extents than can fit in
612 * fork (fork shouldn't be a btree format), root btree
613 * block has more records than can fit into the fork,
614 * or the number of extents is greater than the number of
615 * blocks.
616 */
617 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
618 XFS_IFORK_MAXEXT(ip, whichfork) ||
619 XFS_BMDR_SPACE_CALC(nrecs) >
620 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) ||
621 XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
622 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
623 (unsigned long long) ip->i_ino);
624 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
625 ip->i_mount, dip);
626 return XFS_ERROR(EFSCORRUPTED);
627 }
628
629 ifp->if_broot_bytes = size;
630 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
631 ASSERT(ifp->if_broot != NULL);
632 /*
633 * Copy and convert from the on-disk structure
634 * to the in-memory structure.
635 */
636 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
637 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
638 ifp->if_broot, size);
639 ifp->if_flags &= ~XFS_IFEXTENTS;
640 ifp->if_flags |= XFS_IFBROOT;
641
642 return 0;
643}
644
645STATIC void
646xfs_dinode_from_disk(
647 xfs_icdinode_t *to,
648 xfs_dinode_t *from)
649{
650 to->di_magic = be16_to_cpu(from->di_magic);
651 to->di_mode = be16_to_cpu(from->di_mode);
652 to->di_version = from ->di_version;
653 to->di_format = from->di_format;
654 to->di_onlink = be16_to_cpu(from->di_onlink);
655 to->di_uid = be32_to_cpu(from->di_uid);
656 to->di_gid = be32_to_cpu(from->di_gid);
657 to->di_nlink = be32_to_cpu(from->di_nlink);
658 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
659 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
660 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
661 to->di_flushiter = be16_to_cpu(from->di_flushiter);
662 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
663 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
664 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
665 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
666 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
667 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
668 to->di_size = be64_to_cpu(from->di_size);
669 to->di_nblocks = be64_to_cpu(from->di_nblocks);
670 to->di_extsize = be32_to_cpu(from->di_extsize);
671 to->di_nextents = be32_to_cpu(from->di_nextents);
672 to->di_anextents = be16_to_cpu(from->di_anextents);
673 to->di_forkoff = from->di_forkoff;
674 to->di_aformat = from->di_aformat;
675 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
676 to->di_dmstate = be16_to_cpu(from->di_dmstate);
677 to->di_flags = be16_to_cpu(from->di_flags);
678 to->di_gen = be32_to_cpu(from->di_gen);
679}
680
681void
682xfs_dinode_to_disk(
683 xfs_dinode_t *to,
684 xfs_icdinode_t *from)
685{
686 to->di_magic = cpu_to_be16(from->di_magic);
687 to->di_mode = cpu_to_be16(from->di_mode);
688 to->di_version = from ->di_version;
689 to->di_format = from->di_format;
690 to->di_onlink = cpu_to_be16(from->di_onlink);
691 to->di_uid = cpu_to_be32(from->di_uid);
692 to->di_gid = cpu_to_be32(from->di_gid);
693 to->di_nlink = cpu_to_be32(from->di_nlink);
694 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
695 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
696 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
697 to->di_flushiter = cpu_to_be16(from->di_flushiter);
698 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
699 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
700 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
701 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
702 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
703 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
704 to->di_size = cpu_to_be64(from->di_size);
705 to->di_nblocks = cpu_to_be64(from->di_nblocks);
706 to->di_extsize = cpu_to_be32(from->di_extsize);
707 to->di_nextents = cpu_to_be32(from->di_nextents);
708 to->di_anextents = cpu_to_be16(from->di_anextents);
709 to->di_forkoff = from->di_forkoff;
710 to->di_aformat = from->di_aformat;
711 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
712 to->di_dmstate = cpu_to_be16(from->di_dmstate);
713 to->di_flags = cpu_to_be16(from->di_flags);
714 to->di_gen = cpu_to_be32(from->di_gen);
715}
716
717STATIC uint
718_xfs_dic2xflags(
719 __uint16_t di_flags)
720{
721 uint flags = 0;
722
723 if (di_flags & XFS_DIFLAG_ANY) {
724 if (di_flags & XFS_DIFLAG_REALTIME)
725 flags |= XFS_XFLAG_REALTIME;
726 if (di_flags & XFS_DIFLAG_PREALLOC)
727 flags |= XFS_XFLAG_PREALLOC;
728 if (di_flags & XFS_DIFLAG_IMMUTABLE)
729 flags |= XFS_XFLAG_IMMUTABLE;
730 if (di_flags & XFS_DIFLAG_APPEND)
731 flags |= XFS_XFLAG_APPEND;
732 if (di_flags & XFS_DIFLAG_SYNC)
733 flags |= XFS_XFLAG_SYNC;
734 if (di_flags & XFS_DIFLAG_NOATIME)
735 flags |= XFS_XFLAG_NOATIME;
736 if (di_flags & XFS_DIFLAG_NODUMP)
737 flags |= XFS_XFLAG_NODUMP;
738 if (di_flags & XFS_DIFLAG_RTINHERIT)
739 flags |= XFS_XFLAG_RTINHERIT;
740 if (di_flags & XFS_DIFLAG_PROJINHERIT)
741 flags |= XFS_XFLAG_PROJINHERIT;
742 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
743 flags |= XFS_XFLAG_NOSYMLINKS;
744 if (di_flags & XFS_DIFLAG_EXTSIZE)
745 flags |= XFS_XFLAG_EXTSIZE;
746 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
747 flags |= XFS_XFLAG_EXTSZINHERIT;
748 if (di_flags & XFS_DIFLAG_NODEFRAG)
749 flags |= XFS_XFLAG_NODEFRAG;
750 if (di_flags & XFS_DIFLAG_FILESTREAM)
751 flags |= XFS_XFLAG_FILESTREAM;
752 }
753
754 return flags;
755}
756
757uint
758xfs_ip2xflags(
759 xfs_inode_t *ip)
760{
761 xfs_icdinode_t *dic = &ip->i_d;
762
763 return _xfs_dic2xflags(dic->di_flags) |
764 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
765}
766
767uint
768xfs_dic2xflags(
769 xfs_dinode_t *dip)
770{
771 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
772 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
773}
774
775/*
776 * Read the disk inode attributes into the in-core inode structure.
777 */
778int
779xfs_iread(
780 xfs_mount_t *mp,
781 xfs_trans_t *tp,
782 xfs_inode_t *ip,
783 uint iget_flags)
784{
785 xfs_buf_t *bp;
786 xfs_dinode_t *dip;
787 int error;
788
789 /*
790 * Fill in the location information in the in-core inode.
791 */
792 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
793 if (error)
794 return error;
795
796 /*
797 * Get pointers to the on-disk inode and the buffer containing it.
798 */
799 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, 0, iget_flags);
800 if (error)
801 return error;
802 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
803
804 /*
805 * If we got something that isn't an inode it means someone
806 * (nfs or dmi) has a stale handle.
807 */
808 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) {
809#ifdef DEBUG
810 xfs_alert(mp,
811 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
812 __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
813#endif /* DEBUG */
814 error = XFS_ERROR(EINVAL);
815 goto out_brelse;
816 }
817
818 /*
819 * If the on-disk inode is already linked to a directory
820 * entry, copy all of the inode into the in-core inode.
821 * xfs_iformat() handles copying in the inode format
822 * specific information.
823 * Otherwise, just get the truly permanent information.
824 */
825 if (dip->di_mode) {
826 xfs_dinode_from_disk(&ip->i_d, dip);
827 error = xfs_iformat(ip, dip);
828 if (error) {
829#ifdef DEBUG
830 xfs_alert(mp, "%s: xfs_iformat() returned error %d",
831 __func__, error);
832#endif /* DEBUG */
833 goto out_brelse;
834 }
835 } else {
836 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
837 ip->i_d.di_version = dip->di_version;
838 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
839 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
840 /*
841 * Make sure to pull in the mode here as well in
842 * case the inode is released without being used.
843 * This ensures that xfs_inactive() will see that
844 * the inode is already free and not try to mess
845 * with the uninitialized part of it.
846 */
847 ip->i_d.di_mode = 0;
848 }
849
850 /*
851 * The inode format changed when we moved the link count and
852 * made it 32 bits long. If this is an old format inode,
853 * convert it in memory to look like a new one. If it gets
854 * flushed to disk we will convert back before flushing or
855 * logging it. We zero out the new projid field and the old link
856 * count field. We'll handle clearing the pad field (the remains
857 * of the old uuid field) when we actually convert the inode to
858 * the new format. We don't change the version number so that we
859 * can distinguish this from a real new format inode.
860 */
861 if (ip->i_d.di_version == 1) {
862 ip->i_d.di_nlink = ip->i_d.di_onlink;
863 ip->i_d.di_onlink = 0;
864 xfs_set_projid(ip, 0);
865 }
866
867 ip->i_delayed_blks = 0;
868
869 /*
870 * Mark the buffer containing the inode as something to keep
871 * around for a while. This helps to keep recently accessed
872 * meta-data in-core longer.
873 */
874 xfs_buf_set_ref(bp, XFS_INO_REF);
875
876 /*
877 * Use xfs_trans_brelse() to release the buffer containing the
878 * on-disk inode, because it was acquired with xfs_trans_read_buf()
879 * in xfs_itobp() above. If tp is NULL, this is just a normal
880 * brelse(). If we're within a transaction, then xfs_trans_brelse()
881 * will only release the buffer if it is not dirty within the
882 * transaction. It will be OK to release the buffer in this case,
883 * because inodes on disk are never destroyed and we will be
884 * locking the new in-core inode before putting it in the hash
885 * table where other processes can find it. Thus we don't have
886 * to worry about the inode being changed just because we released
887 * the buffer.
888 */
889 out_brelse:
890 xfs_trans_brelse(tp, bp);
891 return error;
892}
893
894/*
895 * Read in extents from a btree-format inode.
896 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
897 */
898int
899xfs_iread_extents(
900 xfs_trans_t *tp,
901 xfs_inode_t *ip,
902 int whichfork)
903{
904 int error;
905 xfs_ifork_t *ifp;
906 xfs_extnum_t nextents;
907
908 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
909 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
910 ip->i_mount);
911 return XFS_ERROR(EFSCORRUPTED);
912 }
913 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
914 ifp = XFS_IFORK_PTR(ip, whichfork);
915
916 /*
917 * We know that the size is valid (it's checked in iformat_btree)
918 */
919 ifp->if_bytes = ifp->if_real_bytes = 0;
920 ifp->if_flags |= XFS_IFEXTENTS;
921 xfs_iext_add(ifp, 0, nextents);
922 error = xfs_bmap_read_extents(tp, ip, whichfork);
923 if (error) {
924 xfs_iext_destroy(ifp);
925 ifp->if_flags &= ~XFS_IFEXTENTS;
926 return error;
927 }
928 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
929 return 0;
930}
931
932/*
933 * Allocate an inode on disk and return a copy of its in-core version.
934 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
935 * appropriately within the inode. The uid and gid for the inode are
936 * set according to the contents of the given cred structure.
937 *
938 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
939 * has a free inode available, call xfs_iget()
940 * to obtain the in-core version of the allocated inode. Finally,
941 * fill in the inode and log its initial contents. In this case,
942 * ialloc_context would be set to NULL and call_again set to false.
943 *
944 * If xfs_dialloc() does not have an available inode,
945 * it will replenish its supply by doing an allocation. Since we can
946 * only do one allocation within a transaction without deadlocks, we
947 * must commit the current transaction before returning the inode itself.
948 * In this case, therefore, we will set call_again to true and return.
949 * The caller should then commit the current transaction, start a new
950 * transaction, and call xfs_ialloc() again to actually get the inode.
951 *
952 * To ensure that some other process does not grab the inode that
953 * was allocated during the first call to xfs_ialloc(), this routine
954 * also returns the [locked] bp pointing to the head of the freelist
955 * as ialloc_context. The caller should hold this buffer across
956 * the commit and pass it back into this routine on the second call.
957 *
958 * If we are allocating quota inodes, we do not have a parent inode
959 * to attach to or associate with (i.e. pip == NULL) because they
960 * are not linked into the directory structure - they are attached
961 * directly to the superblock - and so have no parent.
962 */
963int
964xfs_ialloc(
965 xfs_trans_t *tp,
966 xfs_inode_t *pip,
967 umode_t mode,
968 xfs_nlink_t nlink,
969 xfs_dev_t rdev,
970 prid_t prid,
971 int okalloc,
972 xfs_buf_t **ialloc_context,
973 boolean_t *call_again,
974 xfs_inode_t **ipp)
975{
976 xfs_ino_t ino;
977 xfs_inode_t *ip;
978 uint flags;
979 int error;
980 timespec_t tv;
981 int filestreams = 0;
982
983 /*
984 * Call the space management code to pick
985 * the on-disk inode to be allocated.
986 */
987 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
988 ialloc_context, call_again, &ino);
989 if (error)
990 return error;
991 if (*call_again || ino == NULLFSINO) {
992 *ipp = NULL;
993 return 0;
994 }
995 ASSERT(*ialloc_context == NULL);
996
997 /*
998 * Get the in-core inode with the lock held exclusively.
999 * This is because we're setting fields here we need
1000 * to prevent others from looking at until we're done.
1001 */
1002 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1003 XFS_ILOCK_EXCL, &ip);
1004 if (error)
1005 return error;
1006 ASSERT(ip != NULL);
1007
1008 ip->i_d.di_mode = mode;
1009 ip->i_d.di_onlink = 0;
1010 ip->i_d.di_nlink = nlink;
1011 ASSERT(ip->i_d.di_nlink == nlink);
1012 ip->i_d.di_uid = current_fsuid();
1013 ip->i_d.di_gid = current_fsgid();
1014 xfs_set_projid(ip, prid);
1015 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1016
1017 /*
1018 * If the superblock version is up to where we support new format
1019 * inodes and this is currently an old format inode, then change
1020 * the inode version number now. This way we only do the conversion
1021 * here rather than here and in the flush/logging code.
1022 */
1023 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1024 ip->i_d.di_version == 1) {
1025 ip->i_d.di_version = 2;
1026 /*
1027 * We've already zeroed the old link count, the projid field,
1028 * and the pad field.
1029 */
1030 }
1031
1032 /*
1033 * Project ids won't be stored on disk if we are using a version 1 inode.
1034 */
1035 if ((prid != 0) && (ip->i_d.di_version == 1))
1036 xfs_bump_ino_vers2(tp, ip);
1037
1038 if (pip && XFS_INHERIT_GID(pip)) {
1039 ip->i_d.di_gid = pip->i_d.di_gid;
1040 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
1041 ip->i_d.di_mode |= S_ISGID;
1042 }
1043 }
1044
1045 /*
1046 * If the group ID of the new file does not match the effective group
1047 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1048 * (and only if the irix_sgid_inherit compatibility variable is set).
1049 */
1050 if ((irix_sgid_inherit) &&
1051 (ip->i_d.di_mode & S_ISGID) &&
1052 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1053 ip->i_d.di_mode &= ~S_ISGID;
1054 }
1055
1056 ip->i_d.di_size = 0;
1057 ip->i_d.di_nextents = 0;
1058 ASSERT(ip->i_d.di_nblocks == 0);
1059
1060 nanotime(&tv);
1061 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1062 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1063 ip->i_d.di_atime = ip->i_d.di_mtime;
1064 ip->i_d.di_ctime = ip->i_d.di_mtime;
1065
1066 /*
1067 * di_gen will have been taken care of in xfs_iread.
1068 */
1069 ip->i_d.di_extsize = 0;
1070 ip->i_d.di_dmevmask = 0;
1071 ip->i_d.di_dmstate = 0;
1072 ip->i_d.di_flags = 0;
1073 flags = XFS_ILOG_CORE;
1074 switch (mode & S_IFMT) {
1075 case S_IFIFO:
1076 case S_IFCHR:
1077 case S_IFBLK:
1078 case S_IFSOCK:
1079 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1080 ip->i_df.if_u2.if_rdev = rdev;
1081 ip->i_df.if_flags = 0;
1082 flags |= XFS_ILOG_DEV;
1083 break;
1084 case S_IFREG:
1085 /*
1086 * we can't set up filestreams until after the VFS inode
1087 * is set up properly.
1088 */
1089 if (pip && xfs_inode_is_filestream(pip))
1090 filestreams = 1;
1091 /* fall through */
1092 case S_IFDIR:
1093 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1094 uint di_flags = 0;
1095
1096 if (S_ISDIR(mode)) {
1097 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1098 di_flags |= XFS_DIFLAG_RTINHERIT;
1099 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1100 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1101 ip->i_d.di_extsize = pip->i_d.di_extsize;
1102 }
1103 } else if (S_ISREG(mode)) {
1104 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1105 di_flags |= XFS_DIFLAG_REALTIME;
1106 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1107 di_flags |= XFS_DIFLAG_EXTSIZE;
1108 ip->i_d.di_extsize = pip->i_d.di_extsize;
1109 }
1110 }
1111 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1112 xfs_inherit_noatime)
1113 di_flags |= XFS_DIFLAG_NOATIME;
1114 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1115 xfs_inherit_nodump)
1116 di_flags |= XFS_DIFLAG_NODUMP;
1117 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1118 xfs_inherit_sync)
1119 di_flags |= XFS_DIFLAG_SYNC;
1120 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1121 xfs_inherit_nosymlinks)
1122 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1123 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1124 di_flags |= XFS_DIFLAG_PROJINHERIT;
1125 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1126 xfs_inherit_nodefrag)
1127 di_flags |= XFS_DIFLAG_NODEFRAG;
1128 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1129 di_flags |= XFS_DIFLAG_FILESTREAM;
1130 ip->i_d.di_flags |= di_flags;
1131 }
1132 /* FALLTHROUGH */
1133 case S_IFLNK:
1134 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1135 ip->i_df.if_flags = XFS_IFEXTENTS;
1136 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1137 ip->i_df.if_u1.if_extents = NULL;
1138 break;
1139 default:
1140 ASSERT(0);
1141 }
1142 /*
1143 * Attribute fork settings for new inode.
1144 */
1145 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1146 ip->i_d.di_anextents = 0;
1147
1148 /*
1149 * Log the new values stuffed into the inode.
1150 */
1151 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1152 xfs_trans_log_inode(tp, ip, flags);
1153
1154 /* now that we have an i_mode we can setup inode ops and unlock */
1155 xfs_setup_inode(ip);
1156
1157 /* now we have set up the vfs inode we can associate the filestream */
1158 if (filestreams) {
1159 error = xfs_filestream_associate(pip, ip);
1160 if (error < 0)
1161 return -error;
1162 if (!error)
1163 xfs_iflags_set(ip, XFS_IFILESTREAM);
1164 }
1165
1166 *ipp = ip;
1167 return 0;
1168}
1169
1170/*
1171 * Free up the underlying blocks past new_size. The new size must be smaller
1172 * than the current size. This routine can be used both for the attribute and
1173 * data fork, and does not modify the inode size, which is left to the caller.
1174 *
1175 * The transaction passed to this routine must have made a permanent log
1176 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1177 * given transaction and start new ones, so make sure everything involved in
1178 * the transaction is tidy before calling here. Some transaction will be
1179 * returned to the caller to be committed. The incoming transaction must
1180 * already include the inode, and both inode locks must be held exclusively.
1181 * The inode must also be "held" within the transaction. On return the inode
1182 * will be "held" within the returned transaction. This routine does NOT
1183 * require any disk space to be reserved for it within the transaction.
1184 *
1185 * If we get an error, we must return with the inode locked and linked into the
1186 * current transaction. This keeps things simple for the higher level code,
1187 * because it always knows that the inode is locked and held in the transaction
1188 * that returns to it whether errors occur or not. We don't mark the inode
1189 * dirty on error so that transactions can be easily aborted if possible.
1190 */
1191int
1192xfs_itruncate_extents(
1193 struct xfs_trans **tpp,
1194 struct xfs_inode *ip,
1195 int whichfork,
1196 xfs_fsize_t new_size)
1197{
1198 struct xfs_mount *mp = ip->i_mount;
1199 struct xfs_trans *tp = *tpp;
1200 struct xfs_trans *ntp;
1201 xfs_bmap_free_t free_list;
1202 xfs_fsblock_t first_block;
1203 xfs_fileoff_t first_unmap_block;
1204 xfs_fileoff_t last_block;
1205 xfs_filblks_t unmap_len;
1206 int committed;
1207 int error = 0;
1208 int done = 0;
1209
1210 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1211 ASSERT(new_size <= XFS_ISIZE(ip));
1212 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1213 ASSERT(ip->i_itemp != NULL);
1214 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1215 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1216
1217 trace_xfs_itruncate_extents_start(ip, new_size);
1218
1219 /*
1220 * Since it is possible for space to become allocated beyond
1221 * the end of the file (in a crash where the space is allocated
1222 * but the inode size is not yet updated), simply remove any
1223 * blocks which show up between the new EOF and the maximum
1224 * possible file size. If the first block to be removed is
1225 * beyond the maximum file size (ie it is the same as last_block),
1226 * then there is nothing to do.
1227 */
1228 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1229 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1230 if (first_unmap_block == last_block)
1231 return 0;
1232
1233 ASSERT(first_unmap_block < last_block);
1234 unmap_len = last_block - first_unmap_block + 1;
1235 while (!done) {
1236 xfs_bmap_init(&free_list, &first_block);
1237 error = xfs_bunmapi(tp, ip,
1238 first_unmap_block, unmap_len,
1239 xfs_bmapi_aflag(whichfork),
1240 XFS_ITRUNC_MAX_EXTENTS,
1241 &first_block, &free_list,
1242 &done);
1243 if (error)
1244 goto out_bmap_cancel;
1245
1246 /*
1247 * Duplicate the transaction that has the permanent
1248 * reservation and commit the old transaction.
1249 */
1250 error = xfs_bmap_finish(&tp, &free_list, &committed);
1251 if (committed)
1252 xfs_trans_ijoin(tp, ip, 0);
1253 if (error)
1254 goto out_bmap_cancel;
1255
1256 if (committed) {
1257 /*
1258 * Mark the inode dirty so it will be logged and
1259 * moved forward in the log as part of every commit.
1260 */
1261 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1262 }
1263
1264 ntp = xfs_trans_dup(tp);
1265 error = xfs_trans_commit(tp, 0);
1266 tp = ntp;
1267
1268 xfs_trans_ijoin(tp, ip, 0);
1269
1270 if (error)
1271 goto out;
1272
1273 /*
1274 * Transaction commit worked ok so we can drop the extra ticket
1275 * reference that we gained in xfs_trans_dup()
1276 */
1277 xfs_log_ticket_put(tp->t_ticket);
1278 error = xfs_trans_reserve(tp, 0,
1279 XFS_ITRUNCATE_LOG_RES(mp), 0,
1280 XFS_TRANS_PERM_LOG_RES,
1281 XFS_ITRUNCATE_LOG_COUNT);
1282 if (error)
1283 goto out;
1284 }
1285
1286 /*
1287 * Always re-log the inode so that our permanent transaction can keep
1288 * on rolling it forward in the log.
1289 */
1290 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1291
1292 trace_xfs_itruncate_extents_end(ip, new_size);
1293
1294out:
1295 *tpp = tp;
1296 return error;
1297out_bmap_cancel:
1298 /*
1299 * If the bunmapi call encounters an error, return to the caller where
1300 * the transaction can be properly aborted. We just need to make sure
1301 * we're not holding any resources that we were not when we came in.
1302 */
1303 xfs_bmap_cancel(&free_list);
1304 goto out;
1305}
1306
1307/*
1308 * This is called when the inode's link count goes to 0.
1309 * We place the on-disk inode on a list in the AGI. It
1310 * will be pulled from this list when the inode is freed.
1311 */
1312int
1313xfs_iunlink(
1314 xfs_trans_t *tp,
1315 xfs_inode_t *ip)
1316{
1317 xfs_mount_t *mp;
1318 xfs_agi_t *agi;
1319 xfs_dinode_t *dip;
1320 xfs_buf_t *agibp;
1321 xfs_buf_t *ibp;
1322 xfs_agino_t agino;
1323 short bucket_index;
1324 int offset;
1325 int error;
1326
1327 ASSERT(ip->i_d.di_nlink == 0);
1328 ASSERT(ip->i_d.di_mode != 0);
1329
1330 mp = tp->t_mountp;
1331
1332 /*
1333 * Get the agi buffer first. It ensures lock ordering
1334 * on the list.
1335 */
1336 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1337 if (error)
1338 return error;
1339 agi = XFS_BUF_TO_AGI(agibp);
1340
1341 /*
1342 * Get the index into the agi hash table for the
1343 * list this inode will go on.
1344 */
1345 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1346 ASSERT(agino != 0);
1347 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1348 ASSERT(agi->agi_unlinked[bucket_index]);
1349 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1350
1351 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1352 /*
1353 * There is already another inode in the bucket we need
1354 * to add ourselves to. Add us at the front of the list.
1355 * Here we put the head pointer into our next pointer,
1356 * and then we fall through to point the head at us.
1357 */
1358 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1359 if (error)
1360 return error;
1361
1362 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1363 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1364 offset = ip->i_imap.im_boffset +
1365 offsetof(xfs_dinode_t, di_next_unlinked);
1366 xfs_trans_inode_buf(tp, ibp);
1367 xfs_trans_log_buf(tp, ibp, offset,
1368 (offset + sizeof(xfs_agino_t) - 1));
1369 xfs_inobp_check(mp, ibp);
1370 }
1371
1372 /*
1373 * Point the bucket head pointer at the inode being inserted.
1374 */
1375 ASSERT(agino != 0);
1376 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1377 offset = offsetof(xfs_agi_t, agi_unlinked) +
1378 (sizeof(xfs_agino_t) * bucket_index);
1379 xfs_trans_log_buf(tp, agibp, offset,
1380 (offset + sizeof(xfs_agino_t) - 1));
1381 return 0;
1382}
1383
1384/*
1385 * Pull the on-disk inode from the AGI unlinked list.
1386 */
1387STATIC int
1388xfs_iunlink_remove(
1389 xfs_trans_t *tp,
1390 xfs_inode_t *ip)
1391{
1392 xfs_ino_t next_ino;
1393 xfs_mount_t *mp;
1394 xfs_agi_t *agi;
1395 xfs_dinode_t *dip;
1396 xfs_buf_t *agibp;
1397 xfs_buf_t *ibp;
1398 xfs_agnumber_t agno;
1399 xfs_agino_t agino;
1400 xfs_agino_t next_agino;
1401 xfs_buf_t *last_ibp;
1402 xfs_dinode_t *last_dip = NULL;
1403 short bucket_index;
1404 int offset, last_offset = 0;
1405 int error;
1406
1407 mp = tp->t_mountp;
1408 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1409
1410 /*
1411 * Get the agi buffer first. It ensures lock ordering
1412 * on the list.
1413 */
1414 error = xfs_read_agi(mp, tp, agno, &agibp);
1415 if (error)
1416 return error;
1417
1418 agi = XFS_BUF_TO_AGI(agibp);
1419
1420 /*
1421 * Get the index into the agi hash table for the
1422 * list this inode will go on.
1423 */
1424 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1425 ASSERT(agino != 0);
1426 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1427 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1428 ASSERT(agi->agi_unlinked[bucket_index]);
1429
1430 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1431 /*
1432 * We're at the head of the list. Get the inode's
1433 * on-disk buffer to see if there is anyone after us
1434 * on the list. Only modify our next pointer if it
1435 * is not already NULLAGINO. This saves us the overhead
1436 * of dealing with the buffer when there is no need to
1437 * change it.
1438 */
1439 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1440 if (error) {
1441 xfs_warn(mp, "%s: xfs_itobp() returned error %d.",
1442 __func__, error);
1443 return error;
1444 }
1445 next_agino = be32_to_cpu(dip->di_next_unlinked);
1446 ASSERT(next_agino != 0);
1447 if (next_agino != NULLAGINO) {
1448 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1449 offset = ip->i_imap.im_boffset +
1450 offsetof(xfs_dinode_t, di_next_unlinked);
1451 xfs_trans_inode_buf(tp, ibp);
1452 xfs_trans_log_buf(tp, ibp, offset,
1453 (offset + sizeof(xfs_agino_t) - 1));
1454 xfs_inobp_check(mp, ibp);
1455 } else {
1456 xfs_trans_brelse(tp, ibp);
1457 }
1458 /*
1459 * Point the bucket head pointer at the next inode.
1460 */
1461 ASSERT(next_agino != 0);
1462 ASSERT(next_agino != agino);
1463 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1464 offset = offsetof(xfs_agi_t, agi_unlinked) +
1465 (sizeof(xfs_agino_t) * bucket_index);
1466 xfs_trans_log_buf(tp, agibp, offset,
1467 (offset + sizeof(xfs_agino_t) - 1));
1468 } else {
1469 /*
1470 * We need to search the list for the inode being freed.
1471 */
1472 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1473 last_ibp = NULL;
1474 while (next_agino != agino) {
1475 /*
1476 * If the last inode wasn't the one pointing to
1477 * us, then release its buffer since we're not
1478 * going to do anything with it.
1479 */
1480 if (last_ibp != NULL) {
1481 xfs_trans_brelse(tp, last_ibp);
1482 }
1483 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1484 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1485 &last_ibp, &last_offset, 0);
1486 if (error) {
1487 xfs_warn(mp,
1488 "%s: xfs_inotobp() returned error %d.",
1489 __func__, error);
1490 return error;
1491 }
1492 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1493 ASSERT(next_agino != NULLAGINO);
1494 ASSERT(next_agino != 0);
1495 }
1496 /*
1497 * Now last_ibp points to the buffer previous to us on
1498 * the unlinked list. Pull us from the list.
1499 */
1500 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1501 if (error) {
1502 xfs_warn(mp, "%s: xfs_itobp(2) returned error %d.",
1503 __func__, error);
1504 return error;
1505 }
1506 next_agino = be32_to_cpu(dip->di_next_unlinked);
1507 ASSERT(next_agino != 0);
1508 ASSERT(next_agino != agino);
1509 if (next_agino != NULLAGINO) {
1510 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1511 offset = ip->i_imap.im_boffset +
1512 offsetof(xfs_dinode_t, di_next_unlinked);
1513 xfs_trans_inode_buf(tp, ibp);
1514 xfs_trans_log_buf(tp, ibp, offset,
1515 (offset + sizeof(xfs_agino_t) - 1));
1516 xfs_inobp_check(mp, ibp);
1517 } else {
1518 xfs_trans_brelse(tp, ibp);
1519 }
1520 /*
1521 * Point the previous inode on the list to the next inode.
1522 */
1523 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1524 ASSERT(next_agino != 0);
1525 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1526 xfs_trans_inode_buf(tp, last_ibp);
1527 xfs_trans_log_buf(tp, last_ibp, offset,
1528 (offset + sizeof(xfs_agino_t) - 1));
1529 xfs_inobp_check(mp, last_ibp);
1530 }
1531 return 0;
1532}
1533
1534/*
1535 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1536 * inodes that are in memory - they all must be marked stale and attached to
1537 * the cluster buffer.
1538 */
1539STATIC int
1540xfs_ifree_cluster(
1541 xfs_inode_t *free_ip,
1542 xfs_trans_t *tp,
1543 xfs_ino_t inum)
1544{
1545 xfs_mount_t *mp = free_ip->i_mount;
1546 int blks_per_cluster;
1547 int nbufs;
1548 int ninodes;
1549 int i, j;
1550 xfs_daddr_t blkno;
1551 xfs_buf_t *bp;
1552 xfs_inode_t *ip;
1553 xfs_inode_log_item_t *iip;
1554 xfs_log_item_t *lip;
1555 struct xfs_perag *pag;
1556
1557 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1558 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1559 blks_per_cluster = 1;
1560 ninodes = mp->m_sb.sb_inopblock;
1561 nbufs = XFS_IALLOC_BLOCKS(mp);
1562 } else {
1563 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1564 mp->m_sb.sb_blocksize;
1565 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1566 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1567 }
1568
1569 for (j = 0; j < nbufs; j++, inum += ninodes) {
1570 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1571 XFS_INO_TO_AGBNO(mp, inum));
1572
1573 /*
1574 * We obtain and lock the backing buffer first in the process
1575 * here, as we have to ensure that any dirty inode that we
1576 * can't get the flush lock on is attached to the buffer.
1577 * If we scan the in-memory inodes first, then buffer IO can
1578 * complete before we get a lock on it, and hence we may fail
1579 * to mark all the active inodes on the buffer stale.
1580 */
1581 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1582 mp->m_bsize * blks_per_cluster, 0);
1583
1584 if (!bp)
1585 return ENOMEM;
1586 /*
1587 * Walk the inodes already attached to the buffer and mark them
1588 * stale. These will all have the flush locks held, so an
1589 * in-memory inode walk can't lock them. By marking them all
1590 * stale first, we will not attempt to lock them in the loop
1591 * below as the XFS_ISTALE flag will be set.
1592 */
1593 lip = bp->b_fspriv;
1594 while (lip) {
1595 if (lip->li_type == XFS_LI_INODE) {
1596 iip = (xfs_inode_log_item_t *)lip;
1597 ASSERT(iip->ili_logged == 1);
1598 lip->li_cb = xfs_istale_done;
1599 xfs_trans_ail_copy_lsn(mp->m_ail,
1600 &iip->ili_flush_lsn,
1601 &iip->ili_item.li_lsn);
1602 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1603 }
1604 lip = lip->li_bio_list;
1605 }
1606
1607
1608 /*
1609 * For each inode in memory attempt to add it to the inode
1610 * buffer and set it up for being staled on buffer IO
1611 * completion. This is safe as we've locked out tail pushing
1612 * and flushing by locking the buffer.
1613 *
1614 * We have already marked every inode that was part of a
1615 * transaction stale above, which means there is no point in
1616 * even trying to lock them.
1617 */
1618 for (i = 0; i < ninodes; i++) {
1619retry:
1620 rcu_read_lock();
1621 ip = radix_tree_lookup(&pag->pag_ici_root,
1622 XFS_INO_TO_AGINO(mp, (inum + i)));
1623
1624 /* Inode not in memory, nothing to do */
1625 if (!ip) {
1626 rcu_read_unlock();
1627 continue;
1628 }
1629
1630 /*
1631 * because this is an RCU protected lookup, we could
1632 * find a recently freed or even reallocated inode
1633 * during the lookup. We need to check under the
1634 * i_flags_lock for a valid inode here. Skip it if it
1635 * is not valid, the wrong inode or stale.
1636 */
1637 spin_lock(&ip->i_flags_lock);
1638 if (ip->i_ino != inum + i ||
1639 __xfs_iflags_test(ip, XFS_ISTALE)) {
1640 spin_unlock(&ip->i_flags_lock);
1641 rcu_read_unlock();
1642 continue;
1643 }
1644 spin_unlock(&ip->i_flags_lock);
1645
1646 /*
1647 * Don't try to lock/unlock the current inode, but we
1648 * _cannot_ skip the other inodes that we did not find
1649 * in the list attached to the buffer and are not
1650 * already marked stale. If we can't lock it, back off
1651 * and retry.
1652 */
1653 if (ip != free_ip &&
1654 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1655 rcu_read_unlock();
1656 delay(1);
1657 goto retry;
1658 }
1659 rcu_read_unlock();
1660
1661 xfs_iflock(ip);
1662 xfs_iflags_set(ip, XFS_ISTALE);
1663
1664 /*
1665 * we don't need to attach clean inodes or those only
1666 * with unlogged changes (which we throw away, anyway).
1667 */
1668 iip = ip->i_itemp;
1669 if (!iip || xfs_inode_clean(ip)) {
1670 ASSERT(ip != free_ip);
1671 xfs_ifunlock(ip);
1672 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1673 continue;
1674 }
1675
1676 iip->ili_last_fields = iip->ili_fields;
1677 iip->ili_fields = 0;
1678 iip->ili_logged = 1;
1679 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1680 &iip->ili_item.li_lsn);
1681
1682 xfs_buf_attach_iodone(bp, xfs_istale_done,
1683 &iip->ili_item);
1684
1685 if (ip != free_ip)
1686 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1687 }
1688
1689 xfs_trans_stale_inode_buf(tp, bp);
1690 xfs_trans_binval(tp, bp);
1691 }
1692
1693 xfs_perag_put(pag);
1694 return 0;
1695}
1696
1697/*
1698 * This is called to return an inode to the inode free list.
1699 * The inode should already be truncated to 0 length and have
1700 * no pages associated with it. This routine also assumes that
1701 * the inode is already a part of the transaction.
1702 *
1703 * The on-disk copy of the inode will have been added to the list
1704 * of unlinked inodes in the AGI. We need to remove the inode from
1705 * that list atomically with respect to freeing it here.
1706 */
1707int
1708xfs_ifree(
1709 xfs_trans_t *tp,
1710 xfs_inode_t *ip,
1711 xfs_bmap_free_t *flist)
1712{
1713 int error;
1714 int delete;
1715 xfs_ino_t first_ino;
1716 xfs_dinode_t *dip;
1717 xfs_buf_t *ibp;
1718
1719 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1720 ASSERT(ip->i_d.di_nlink == 0);
1721 ASSERT(ip->i_d.di_nextents == 0);
1722 ASSERT(ip->i_d.di_anextents == 0);
1723 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
1724 ASSERT(ip->i_d.di_nblocks == 0);
1725
1726 /*
1727 * Pull the on-disk inode from the AGI unlinked list.
1728 */
1729 error = xfs_iunlink_remove(tp, ip);
1730 if (error != 0) {
1731 return error;
1732 }
1733
1734 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1735 if (error != 0) {
1736 return error;
1737 }
1738 ip->i_d.di_mode = 0; /* mark incore inode as free */
1739 ip->i_d.di_flags = 0;
1740 ip->i_d.di_dmevmask = 0;
1741 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
1742 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1743 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1744 /*
1745 * Bump the generation count so no one will be confused
1746 * by reincarnations of this inode.
1747 */
1748 ip->i_d.di_gen++;
1749
1750 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1751
1752 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0);
1753 if (error)
1754 return error;
1755
1756 /*
1757 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1758 * from picking up this inode when it is reclaimed (its incore state
1759 * initialzed but not flushed to disk yet). The in-core di_mode is
1760 * already cleared and a corresponding transaction logged.
1761 * The hack here just synchronizes the in-core to on-disk
1762 * di_mode value in advance before the actual inode sync to disk.
1763 * This is OK because the inode is already unlinked and would never
1764 * change its di_mode again for this inode generation.
1765 * This is a temporary hack that would require a proper fix
1766 * in the future.
1767 */
1768 dip->di_mode = 0;
1769
1770 if (delete) {
1771 error = xfs_ifree_cluster(ip, tp, first_ino);
1772 }
1773
1774 return error;
1775}
1776
1777/*
1778 * Reallocate the space for if_broot based on the number of records
1779 * being added or deleted as indicated in rec_diff. Move the records
1780 * and pointers in if_broot to fit the new size. When shrinking this
1781 * will eliminate holes between the records and pointers created by
1782 * the caller. When growing this will create holes to be filled in
1783 * by the caller.
1784 *
1785 * The caller must not request to add more records than would fit in
1786 * the on-disk inode root. If the if_broot is currently NULL, then
1787 * if we adding records one will be allocated. The caller must also
1788 * not request that the number of records go below zero, although
1789 * it can go to zero.
1790 *
1791 * ip -- the inode whose if_broot area is changing
1792 * ext_diff -- the change in the number of records, positive or negative,
1793 * requested for the if_broot array.
1794 */
1795void
1796xfs_iroot_realloc(
1797 xfs_inode_t *ip,
1798 int rec_diff,
1799 int whichfork)
1800{
1801 struct xfs_mount *mp = ip->i_mount;
1802 int cur_max;
1803 xfs_ifork_t *ifp;
1804 struct xfs_btree_block *new_broot;
1805 int new_max;
1806 size_t new_size;
1807 char *np;
1808 char *op;
1809
1810 /*
1811 * Handle the degenerate case quietly.
1812 */
1813 if (rec_diff == 0) {
1814 return;
1815 }
1816
1817 ifp = XFS_IFORK_PTR(ip, whichfork);
1818 if (rec_diff > 0) {
1819 /*
1820 * If there wasn't any memory allocated before, just
1821 * allocate it now and get out.
1822 */
1823 if (ifp->if_broot_bytes == 0) {
1824 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
1825 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1826 ifp->if_broot_bytes = (int)new_size;
1827 return;
1828 }
1829
1830 /*
1831 * If there is already an existing if_broot, then we need
1832 * to realloc() it and shift the pointers to their new
1833 * location. The records don't change location because
1834 * they are kept butted up against the btree block header.
1835 */
1836 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1837 new_max = cur_max + rec_diff;
1838 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1839 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
1840 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
1841 KM_SLEEP | KM_NOFS);
1842 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1843 ifp->if_broot_bytes);
1844 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1845 (int)new_size);
1846 ifp->if_broot_bytes = (int)new_size;
1847 ASSERT(ifp->if_broot_bytes <=
1848 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1849 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
1850 return;
1851 }
1852
1853 /*
1854 * rec_diff is less than 0. In this case, we are shrinking the
1855 * if_broot buffer. It must already exist. If we go to zero
1856 * records, just get rid of the root and clear the status bit.
1857 */
1858 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
1859 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1860 new_max = cur_max + rec_diff;
1861 ASSERT(new_max >= 0);
1862 if (new_max > 0)
1863 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1864 else
1865 new_size = 0;
1866 if (new_size > 0) {
1867 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1868 /*
1869 * First copy over the btree block header.
1870 */
1871 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
1872 } else {
1873 new_broot = NULL;
1874 ifp->if_flags &= ~XFS_IFBROOT;
1875 }
1876
1877 /*
1878 * Only copy the records and pointers if there are any.
1879 */
1880 if (new_max > 0) {
1881 /*
1882 * First copy the records.
1883 */
1884 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
1885 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
1886 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
1887
1888 /*
1889 * Then copy the pointers.
1890 */
1891 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1892 ifp->if_broot_bytes);
1893 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
1894 (int)new_size);
1895 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
1896 }
1897 kmem_free(ifp->if_broot);
1898 ifp->if_broot = new_broot;
1899 ifp->if_broot_bytes = (int)new_size;
1900 ASSERT(ifp->if_broot_bytes <=
1901 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1902 return;
1903}
1904
1905
1906/*
1907 * This is called when the amount of space needed for if_data
1908 * is increased or decreased. The change in size is indicated by
1909 * the number of bytes that need to be added or deleted in the
1910 * byte_diff parameter.
1911 *
1912 * If the amount of space needed has decreased below the size of the
1913 * inline buffer, then switch to using the inline buffer. Otherwise,
1914 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
1915 * to what is needed.
1916 *
1917 * ip -- the inode whose if_data area is changing
1918 * byte_diff -- the change in the number of bytes, positive or negative,
1919 * requested for the if_data array.
1920 */
1921void
1922xfs_idata_realloc(
1923 xfs_inode_t *ip,
1924 int byte_diff,
1925 int whichfork)
1926{
1927 xfs_ifork_t *ifp;
1928 int new_size;
1929 int real_size;
1930
1931 if (byte_diff == 0) {
1932 return;
1933 }
1934
1935 ifp = XFS_IFORK_PTR(ip, whichfork);
1936 new_size = (int)ifp->if_bytes + byte_diff;
1937 ASSERT(new_size >= 0);
1938
1939 if (new_size == 0) {
1940 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1941 kmem_free(ifp->if_u1.if_data);
1942 }
1943 ifp->if_u1.if_data = NULL;
1944 real_size = 0;
1945 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
1946 /*
1947 * If the valid extents/data can fit in if_inline_ext/data,
1948 * copy them from the malloc'd vector and free it.
1949 */
1950 if (ifp->if_u1.if_data == NULL) {
1951 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
1952 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1953 ASSERT(ifp->if_real_bytes != 0);
1954 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
1955 new_size);
1956 kmem_free(ifp->if_u1.if_data);
1957 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
1958 }
1959 real_size = 0;
1960 } else {
1961 /*
1962 * Stuck with malloc/realloc.
1963 * For inline data, the underlying buffer must be
1964 * a multiple of 4 bytes in size so that it can be
1965 * logged and stay on word boundaries. We enforce
1966 * that here.
1967 */
1968 real_size = roundup(new_size, 4);
1969 if (ifp->if_u1.if_data == NULL) {
1970 ASSERT(ifp->if_real_bytes == 0);
1971 ifp->if_u1.if_data = kmem_alloc(real_size,
1972 KM_SLEEP | KM_NOFS);
1973 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
1974 /*
1975 * Only do the realloc if the underlying size
1976 * is really changing.
1977 */
1978 if (ifp->if_real_bytes != real_size) {
1979 ifp->if_u1.if_data =
1980 kmem_realloc(ifp->if_u1.if_data,
1981 real_size,
1982 ifp->if_real_bytes,
1983 KM_SLEEP | KM_NOFS);
1984 }
1985 } else {
1986 ASSERT(ifp->if_real_bytes == 0);
1987 ifp->if_u1.if_data = kmem_alloc(real_size,
1988 KM_SLEEP | KM_NOFS);
1989 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
1990 ifp->if_bytes);
1991 }
1992 }
1993 ifp->if_real_bytes = real_size;
1994 ifp->if_bytes = new_size;
1995 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
1996}
1997
1998void
1999xfs_idestroy_fork(
2000 xfs_inode_t *ip,
2001 int whichfork)
2002{
2003 xfs_ifork_t *ifp;
2004
2005 ifp = XFS_IFORK_PTR(ip, whichfork);
2006 if (ifp->if_broot != NULL) {
2007 kmem_free(ifp->if_broot);
2008 ifp->if_broot = NULL;
2009 }
2010
2011 /*
2012 * If the format is local, then we can't have an extents
2013 * array so just look for an inline data array. If we're
2014 * not local then we may or may not have an extents list,
2015 * so check and free it up if we do.
2016 */
2017 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2018 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2019 (ifp->if_u1.if_data != NULL)) {
2020 ASSERT(ifp->if_real_bytes != 0);
2021 kmem_free(ifp->if_u1.if_data);
2022 ifp->if_u1.if_data = NULL;
2023 ifp->if_real_bytes = 0;
2024 }
2025 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2026 ((ifp->if_flags & XFS_IFEXTIREC) ||
2027 ((ifp->if_u1.if_extents != NULL) &&
2028 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2029 ASSERT(ifp->if_real_bytes != 0);
2030 xfs_iext_destroy(ifp);
2031 }
2032 ASSERT(ifp->if_u1.if_extents == NULL ||
2033 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2034 ASSERT(ifp->if_real_bytes == 0);
2035 if (whichfork == XFS_ATTR_FORK) {
2036 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2037 ip->i_afp = NULL;
2038 }
2039}
2040
2041/*
2042 * This is called to unpin an inode. The caller must have the inode locked
2043 * in at least shared mode so that the buffer cannot be subsequently pinned
2044 * once someone is waiting for it to be unpinned.
2045 */
2046static void
2047xfs_iunpin(
2048 struct xfs_inode *ip)
2049{
2050 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2051
2052 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2053
2054 /* Give the log a push to start the unpinning I/O */
2055 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2056
2057}
2058
2059static void
2060__xfs_iunpin_wait(
2061 struct xfs_inode *ip)
2062{
2063 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2064 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2065
2066 xfs_iunpin(ip);
2067
2068 do {
2069 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2070 if (xfs_ipincount(ip))
2071 io_schedule();
2072 } while (xfs_ipincount(ip));
2073 finish_wait(wq, &wait.wait);
2074}
2075
2076void
2077xfs_iunpin_wait(
2078 struct xfs_inode *ip)
2079{
2080 if (xfs_ipincount(ip))
2081 __xfs_iunpin_wait(ip);
2082}
2083
2084/*
2085 * xfs_iextents_copy()
2086 *
2087 * This is called to copy the REAL extents (as opposed to the delayed
2088 * allocation extents) from the inode into the given buffer. It
2089 * returns the number of bytes copied into the buffer.
2090 *
2091 * If there are no delayed allocation extents, then we can just
2092 * memcpy() the extents into the buffer. Otherwise, we need to
2093 * examine each extent in turn and skip those which are delayed.
2094 */
2095int
2096xfs_iextents_copy(
2097 xfs_inode_t *ip,
2098 xfs_bmbt_rec_t *dp,
2099 int whichfork)
2100{
2101 int copied;
2102 int i;
2103 xfs_ifork_t *ifp;
2104 int nrecs;
2105 xfs_fsblock_t start_block;
2106
2107 ifp = XFS_IFORK_PTR(ip, whichfork);
2108 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2109 ASSERT(ifp->if_bytes > 0);
2110
2111 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2112 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2113 ASSERT(nrecs > 0);
2114
2115 /*
2116 * There are some delayed allocation extents in the
2117 * inode, so copy the extents one at a time and skip
2118 * the delayed ones. There must be at least one
2119 * non-delayed extent.
2120 */
2121 copied = 0;
2122 for (i = 0; i < nrecs; i++) {
2123 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2124 start_block = xfs_bmbt_get_startblock(ep);
2125 if (isnullstartblock(start_block)) {
2126 /*
2127 * It's a delayed allocation extent, so skip it.
2128 */
2129 continue;
2130 }
2131
2132 /* Translate to on disk format */
2133 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2134 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2135 dp++;
2136 copied++;
2137 }
2138 ASSERT(copied != 0);
2139 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2140
2141 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2142}
2143
2144/*
2145 * Each of the following cases stores data into the same region
2146 * of the on-disk inode, so only one of them can be valid at
2147 * any given time. While it is possible to have conflicting formats
2148 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2149 * in EXTENTS format, this can only happen when the fork has
2150 * changed formats after being modified but before being flushed.
2151 * In these cases, the format always takes precedence, because the
2152 * format indicates the current state of the fork.
2153 */
2154/*ARGSUSED*/
2155STATIC void
2156xfs_iflush_fork(
2157 xfs_inode_t *ip,
2158 xfs_dinode_t *dip,
2159 xfs_inode_log_item_t *iip,
2160 int whichfork,
2161 xfs_buf_t *bp)
2162{
2163 char *cp;
2164 xfs_ifork_t *ifp;
2165 xfs_mount_t *mp;
2166#ifdef XFS_TRANS_DEBUG
2167 int first;
2168#endif
2169 static const short brootflag[2] =
2170 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2171 static const short dataflag[2] =
2172 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2173 static const short extflag[2] =
2174 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2175
2176 if (!iip)
2177 return;
2178 ifp = XFS_IFORK_PTR(ip, whichfork);
2179 /*
2180 * This can happen if we gave up in iformat in an error path,
2181 * for the attribute fork.
2182 */
2183 if (!ifp) {
2184 ASSERT(whichfork == XFS_ATTR_FORK);
2185 return;
2186 }
2187 cp = XFS_DFORK_PTR(dip, whichfork);
2188 mp = ip->i_mount;
2189 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2190 case XFS_DINODE_FMT_LOCAL:
2191 if ((iip->ili_fields & dataflag[whichfork]) &&
2192 (ifp->if_bytes > 0)) {
2193 ASSERT(ifp->if_u1.if_data != NULL);
2194 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2195 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2196 }
2197 break;
2198
2199 case XFS_DINODE_FMT_EXTENTS:
2200 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2201 !(iip->ili_fields & extflag[whichfork]));
2202 if ((iip->ili_fields & extflag[whichfork]) &&
2203 (ifp->if_bytes > 0)) {
2204 ASSERT(xfs_iext_get_ext(ifp, 0));
2205 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2206 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2207 whichfork);
2208 }
2209 break;
2210
2211 case XFS_DINODE_FMT_BTREE:
2212 if ((iip->ili_fields & brootflag[whichfork]) &&
2213 (ifp->if_broot_bytes > 0)) {
2214 ASSERT(ifp->if_broot != NULL);
2215 ASSERT(ifp->if_broot_bytes <=
2216 (XFS_IFORK_SIZE(ip, whichfork) +
2217 XFS_BROOT_SIZE_ADJ));
2218 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2219 (xfs_bmdr_block_t *)cp,
2220 XFS_DFORK_SIZE(dip, mp, whichfork));
2221 }
2222 break;
2223
2224 case XFS_DINODE_FMT_DEV:
2225 if (iip->ili_fields & XFS_ILOG_DEV) {
2226 ASSERT(whichfork == XFS_DATA_FORK);
2227 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2228 }
2229 break;
2230
2231 case XFS_DINODE_FMT_UUID:
2232 if (iip->ili_fields & XFS_ILOG_UUID) {
2233 ASSERT(whichfork == XFS_DATA_FORK);
2234 memcpy(XFS_DFORK_DPTR(dip),
2235 &ip->i_df.if_u2.if_uuid,
2236 sizeof(uuid_t));
2237 }
2238 break;
2239
2240 default:
2241 ASSERT(0);
2242 break;
2243 }
2244}
2245
2246STATIC int
2247xfs_iflush_cluster(
2248 xfs_inode_t *ip,
2249 xfs_buf_t *bp)
2250{
2251 xfs_mount_t *mp = ip->i_mount;
2252 struct xfs_perag *pag;
2253 unsigned long first_index, mask;
2254 unsigned long inodes_per_cluster;
2255 int ilist_size;
2256 xfs_inode_t **ilist;
2257 xfs_inode_t *iq;
2258 int nr_found;
2259 int clcount = 0;
2260 int bufwasdelwri;
2261 int i;
2262
2263 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2264
2265 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2266 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2267 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2268 if (!ilist)
2269 goto out_put;
2270
2271 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2272 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2273 rcu_read_lock();
2274 /* really need a gang lookup range call here */
2275 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2276 first_index, inodes_per_cluster);
2277 if (nr_found == 0)
2278 goto out_free;
2279
2280 for (i = 0; i < nr_found; i++) {
2281 iq = ilist[i];
2282 if (iq == ip)
2283 continue;
2284
2285 /*
2286 * because this is an RCU protected lookup, we could find a
2287 * recently freed or even reallocated inode during the lookup.
2288 * We need to check under the i_flags_lock for a valid inode
2289 * here. Skip it if it is not valid or the wrong inode.
2290 */
2291 spin_lock(&ip->i_flags_lock);
2292 if (!ip->i_ino ||
2293 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2294 spin_unlock(&ip->i_flags_lock);
2295 continue;
2296 }
2297 spin_unlock(&ip->i_flags_lock);
2298
2299 /*
2300 * Do an un-protected check to see if the inode is dirty and
2301 * is a candidate for flushing. These checks will be repeated
2302 * later after the appropriate locks are acquired.
2303 */
2304 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2305 continue;
2306
2307 /*
2308 * Try to get locks. If any are unavailable or it is pinned,
2309 * then this inode cannot be flushed and is skipped.
2310 */
2311
2312 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2313 continue;
2314 if (!xfs_iflock_nowait(iq)) {
2315 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2316 continue;
2317 }
2318 if (xfs_ipincount(iq)) {
2319 xfs_ifunlock(iq);
2320 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2321 continue;
2322 }
2323
2324 /*
2325 * arriving here means that this inode can be flushed. First
2326 * re-check that it's dirty before flushing.
2327 */
2328 if (!xfs_inode_clean(iq)) {
2329 int error;
2330 error = xfs_iflush_int(iq, bp);
2331 if (error) {
2332 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2333 goto cluster_corrupt_out;
2334 }
2335 clcount++;
2336 } else {
2337 xfs_ifunlock(iq);
2338 }
2339 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2340 }
2341
2342 if (clcount) {
2343 XFS_STATS_INC(xs_icluster_flushcnt);
2344 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2345 }
2346
2347out_free:
2348 rcu_read_unlock();
2349 kmem_free(ilist);
2350out_put:
2351 xfs_perag_put(pag);
2352 return 0;
2353
2354
2355cluster_corrupt_out:
2356 /*
2357 * Corruption detected in the clustering loop. Invalidate the
2358 * inode buffer and shut down the filesystem.
2359 */
2360 rcu_read_unlock();
2361 /*
2362 * Clean up the buffer. If it was delwri, just release it --
2363 * brelse can handle it with no problems. If not, shut down the
2364 * filesystem before releasing the buffer.
2365 */
2366 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
2367 if (bufwasdelwri)
2368 xfs_buf_relse(bp);
2369
2370 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2371
2372 if (!bufwasdelwri) {
2373 /*
2374 * Just like incore_relse: if we have b_iodone functions,
2375 * mark the buffer as an error and call them. Otherwise
2376 * mark it as stale and brelse.
2377 */
2378 if (bp->b_iodone) {
2379 XFS_BUF_UNDONE(bp);
2380 xfs_buf_stale(bp);
2381 xfs_buf_ioerror(bp, EIO);
2382 xfs_buf_ioend(bp, 0);
2383 } else {
2384 xfs_buf_stale(bp);
2385 xfs_buf_relse(bp);
2386 }
2387 }
2388
2389 /*
2390 * Unlocks the flush lock
2391 */
2392 xfs_iflush_abort(iq, false);
2393 kmem_free(ilist);
2394 xfs_perag_put(pag);
2395 return XFS_ERROR(EFSCORRUPTED);
2396}
2397
2398/*
2399 * Flush dirty inode metadata into the backing buffer.
2400 *
2401 * The caller must have the inode lock and the inode flush lock held. The
2402 * inode lock will still be held upon return to the caller, and the inode
2403 * flush lock will be released after the inode has reached the disk.
2404 *
2405 * The caller must write out the buffer returned in *bpp and release it.
2406 */
2407int
2408xfs_iflush(
2409 struct xfs_inode *ip,
2410 struct xfs_buf **bpp)
2411{
2412 struct xfs_mount *mp = ip->i_mount;
2413 struct xfs_buf *bp;
2414 struct xfs_dinode *dip;
2415 int error;
2416
2417 XFS_STATS_INC(xs_iflush_count);
2418
2419 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2420 ASSERT(xfs_isiflocked(ip));
2421 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2422 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2423
2424 *bpp = NULL;
2425
2426 xfs_iunpin_wait(ip);
2427
2428 /*
2429 * For stale inodes we cannot rely on the backing buffer remaining
2430 * stale in cache for the remaining life of the stale inode and so
2431 * xfs_itobp() below may give us a buffer that no longer contains
2432 * inodes below. We have to check this after ensuring the inode is
2433 * unpinned so that it is safe to reclaim the stale inode after the
2434 * flush call.
2435 */
2436 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2437 xfs_ifunlock(ip);
2438 return 0;
2439 }
2440
2441 /*
2442 * This may have been unpinned because the filesystem is shutting
2443 * down forcibly. If that's the case we must not write this inode
2444 * to disk, because the log record didn't make it to disk.
2445 *
2446 * We also have to remove the log item from the AIL in this case,
2447 * as we wait for an empty AIL as part of the unmount process.
2448 */
2449 if (XFS_FORCED_SHUTDOWN(mp)) {
2450 error = XFS_ERROR(EIO);
2451 goto abort_out;
2452 }
2453
2454 /*
2455 * Get the buffer containing the on-disk inode.
2456 */
2457 error = xfs_itobp(mp, NULL, ip, &dip, &bp, XBF_TRYLOCK);
2458 if (error || !bp) {
2459 xfs_ifunlock(ip);
2460 return error;
2461 }
2462
2463 /*
2464 * First flush out the inode that xfs_iflush was called with.
2465 */
2466 error = xfs_iflush_int(ip, bp);
2467 if (error)
2468 goto corrupt_out;
2469
2470 /*
2471 * If the buffer is pinned then push on the log now so we won't
2472 * get stuck waiting in the write for too long.
2473 */
2474 if (xfs_buf_ispinned(bp))
2475 xfs_log_force(mp, 0);
2476
2477 /*
2478 * inode clustering:
2479 * see if other inodes can be gathered into this write
2480 */
2481 error = xfs_iflush_cluster(ip, bp);
2482 if (error)
2483 goto cluster_corrupt_out;
2484
2485 *bpp = bp;
2486 return 0;
2487
2488corrupt_out:
2489 xfs_buf_relse(bp);
2490 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2491cluster_corrupt_out:
2492 error = XFS_ERROR(EFSCORRUPTED);
2493abort_out:
2494 /*
2495 * Unlocks the flush lock
2496 */
2497 xfs_iflush_abort(ip, false);
2498 return error;
2499}
2500
2501
2502STATIC int
2503xfs_iflush_int(
2504 xfs_inode_t *ip,
2505 xfs_buf_t *bp)
2506{
2507 xfs_inode_log_item_t *iip;
2508 xfs_dinode_t *dip;
2509 xfs_mount_t *mp;
2510#ifdef XFS_TRANS_DEBUG
2511 int first;
2512#endif
2513
2514 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2515 ASSERT(xfs_isiflocked(ip));
2516 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2517 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
2518
2519 iip = ip->i_itemp;
2520 mp = ip->i_mount;
2521
2522 /* set *dip = inode's place in the buffer */
2523 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2524
2525 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2526 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2527 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2528 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2529 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2530 goto corrupt_out;
2531 }
2532 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2533 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2534 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2535 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2536 __func__, ip->i_ino, ip, ip->i_d.di_magic);
2537 goto corrupt_out;
2538 }
2539 if (S_ISREG(ip->i_d.di_mode)) {
2540 if (XFS_TEST_ERROR(
2541 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2542 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2543 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2544 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2545 "%s: Bad regular inode %Lu, ptr 0x%p",
2546 __func__, ip->i_ino, ip);
2547 goto corrupt_out;
2548 }
2549 } else if (S_ISDIR(ip->i_d.di_mode)) {
2550 if (XFS_TEST_ERROR(
2551 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2552 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2553 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2554 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2555 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2556 "%s: Bad directory inode %Lu, ptr 0x%p",
2557 __func__, ip->i_ino, ip);
2558 goto corrupt_out;
2559 }
2560 }
2561 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2562 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2563 XFS_RANDOM_IFLUSH_5)) {
2564 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2565 "%s: detected corrupt incore inode %Lu, "
2566 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2567 __func__, ip->i_ino,
2568 ip->i_d.di_nextents + ip->i_d.di_anextents,
2569 ip->i_d.di_nblocks, ip);
2570 goto corrupt_out;
2571 }
2572 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2573 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2574 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2575 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2576 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2577 goto corrupt_out;
2578 }
2579 /*
2580 * bump the flush iteration count, used to detect flushes which
2581 * postdate a log record during recovery.
2582 */
2583
2584 ip->i_d.di_flushiter++;
2585
2586 /*
2587 * Copy the dirty parts of the inode into the on-disk
2588 * inode. We always copy out the core of the inode,
2589 * because if the inode is dirty at all the core must
2590 * be.
2591 */
2592 xfs_dinode_to_disk(dip, &ip->i_d);
2593
2594 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2595 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2596 ip->i_d.di_flushiter = 0;
2597
2598 /*
2599 * If this is really an old format inode and the superblock version
2600 * has not been updated to support only new format inodes, then
2601 * convert back to the old inode format. If the superblock version
2602 * has been updated, then make the conversion permanent.
2603 */
2604 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2605 if (ip->i_d.di_version == 1) {
2606 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2607 /*
2608 * Convert it back.
2609 */
2610 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2611 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2612 } else {
2613 /*
2614 * The superblock version has already been bumped,
2615 * so just make the conversion to the new inode
2616 * format permanent.
2617 */
2618 ip->i_d.di_version = 2;
2619 dip->di_version = 2;
2620 ip->i_d.di_onlink = 0;
2621 dip->di_onlink = 0;
2622 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2623 memset(&(dip->di_pad[0]), 0,
2624 sizeof(dip->di_pad));
2625 ASSERT(xfs_get_projid(ip) == 0);
2626 }
2627 }
2628
2629 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2630 if (XFS_IFORK_Q(ip))
2631 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2632 xfs_inobp_check(mp, bp);
2633
2634 /*
2635 * We've recorded everything logged in the inode, so we'd like to clear
2636 * the ili_fields bits so we don't log and flush things unnecessarily.
2637 * However, we can't stop logging all this information until the data
2638 * we've copied into the disk buffer is written to disk. If we did we
2639 * might overwrite the copy of the inode in the log with all the data
2640 * after re-logging only part of it, and in the face of a crash we
2641 * wouldn't have all the data we need to recover.
2642 *
2643 * What we do is move the bits to the ili_last_fields field. When
2644 * logging the inode, these bits are moved back to the ili_fields field.
2645 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
2646 * know that the information those bits represent is permanently on
2647 * disk. As long as the flush completes before the inode is logged
2648 * again, then both ili_fields and ili_last_fields will be cleared.
2649 *
2650 * We can play with the ili_fields bits here, because the inode lock
2651 * must be held exclusively in order to set bits there and the flush
2652 * lock protects the ili_last_fields bits. Set ili_logged so the flush
2653 * done routine can tell whether or not to look in the AIL. Also, store
2654 * the current LSN of the inode so that we can tell whether the item has
2655 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
2656 * need the AIL lock, because it is a 64 bit value that cannot be read
2657 * atomically.
2658 */
2659 if (iip != NULL && iip->ili_fields != 0) {
2660 iip->ili_last_fields = iip->ili_fields;
2661 iip->ili_fields = 0;
2662 iip->ili_logged = 1;
2663
2664 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2665 &iip->ili_item.li_lsn);
2666
2667 /*
2668 * Attach the function xfs_iflush_done to the inode's
2669 * buffer. This will remove the inode from the AIL
2670 * and unlock the inode's flush lock when the inode is
2671 * completely written to disk.
2672 */
2673 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
2674
2675 ASSERT(bp->b_fspriv != NULL);
2676 ASSERT(bp->b_iodone != NULL);
2677 } else {
2678 /*
2679 * We're flushing an inode which is not in the AIL and has
2680 * not been logged. For this case we can immediately drop
2681 * the inode flush lock because we can avoid the whole
2682 * AIL state thing. It's OK to drop the flush lock now,
2683 * because we've already locked the buffer and to do anything
2684 * you really need both.
2685 */
2686 if (iip != NULL) {
2687 ASSERT(iip->ili_logged == 0);
2688 ASSERT(iip->ili_last_fields == 0);
2689 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
2690 }
2691 xfs_ifunlock(ip);
2692 }
2693
2694 return 0;
2695
2696corrupt_out:
2697 return XFS_ERROR(EFSCORRUPTED);
2698}
2699
2700/*
2701 * Return a pointer to the extent record at file index idx.
2702 */
2703xfs_bmbt_rec_host_t *
2704xfs_iext_get_ext(
2705 xfs_ifork_t *ifp, /* inode fork pointer */
2706 xfs_extnum_t idx) /* index of target extent */
2707{
2708 ASSERT(idx >= 0);
2709 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2710
2711 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2712 return ifp->if_u1.if_ext_irec->er_extbuf;
2713 } else if (ifp->if_flags & XFS_IFEXTIREC) {
2714 xfs_ext_irec_t *erp; /* irec pointer */
2715 int erp_idx = 0; /* irec index */
2716 xfs_extnum_t page_idx = idx; /* ext index in target list */
2717
2718 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2719 return &erp->er_extbuf[page_idx];
2720 } else if (ifp->if_bytes) {
2721 return &ifp->if_u1.if_extents[idx];
2722 } else {
2723 return NULL;
2724 }
2725}
2726
2727/*
2728 * Insert new item(s) into the extent records for incore inode
2729 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2730 */
2731void
2732xfs_iext_insert(
2733 xfs_inode_t *ip, /* incore inode pointer */
2734 xfs_extnum_t idx, /* starting index of new items */
2735 xfs_extnum_t count, /* number of inserted items */
2736 xfs_bmbt_irec_t *new, /* items to insert */
2737 int state) /* type of extent conversion */
2738{
2739 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2740 xfs_extnum_t i; /* extent record index */
2741
2742 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2743
2744 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2745 xfs_iext_add(ifp, idx, count);
2746 for (i = idx; i < idx + count; i++, new++)
2747 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
2748}
2749
2750/*
2751 * This is called when the amount of space required for incore file
2752 * extents needs to be increased. The ext_diff parameter stores the
2753 * number of new extents being added and the idx parameter contains
2754 * the extent index where the new extents will be added. If the new
2755 * extents are being appended, then we just need to (re)allocate and
2756 * initialize the space. Otherwise, if the new extents are being
2757 * inserted into the middle of the existing entries, a bit more work
2758 * is required to make room for the new extents to be inserted. The
2759 * caller is responsible for filling in the new extent entries upon
2760 * return.
2761 */
2762void
2763xfs_iext_add(
2764 xfs_ifork_t *ifp, /* inode fork pointer */
2765 xfs_extnum_t idx, /* index to begin adding exts */
2766 int ext_diff) /* number of extents to add */
2767{
2768 int byte_diff; /* new bytes being added */
2769 int new_size; /* size of extents after adding */
2770 xfs_extnum_t nextents; /* number of extents in file */
2771
2772 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2773 ASSERT((idx >= 0) && (idx <= nextents));
2774 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2775 new_size = ifp->if_bytes + byte_diff;
2776 /*
2777 * If the new number of extents (nextents + ext_diff)
2778 * fits inside the inode, then continue to use the inline
2779 * extent buffer.
2780 */
2781 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2782 if (idx < nextents) {
2783 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2784 &ifp->if_u2.if_inline_ext[idx],
2785 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2786 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2787 }
2788 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2789 ifp->if_real_bytes = 0;
2790 }
2791 /*
2792 * Otherwise use a linear (direct) extent list.
2793 * If the extents are currently inside the inode,
2794 * xfs_iext_realloc_direct will switch us from
2795 * inline to direct extent allocation mode.
2796 */
2797 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
2798 xfs_iext_realloc_direct(ifp, new_size);
2799 if (idx < nextents) {
2800 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2801 &ifp->if_u1.if_extents[idx],
2802 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2803 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2804 }
2805 }
2806 /* Indirection array */
2807 else {
2808 xfs_ext_irec_t *erp;
2809 int erp_idx = 0;
2810 int page_idx = idx;
2811
2812 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2813 if (ifp->if_flags & XFS_IFEXTIREC) {
2814 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
2815 } else {
2816 xfs_iext_irec_init(ifp);
2817 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2818 erp = ifp->if_u1.if_ext_irec;
2819 }
2820 /* Extents fit in target extent page */
2821 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
2822 if (page_idx < erp->er_extcount) {
2823 memmove(&erp->er_extbuf[page_idx + ext_diff],
2824 &erp->er_extbuf[page_idx],
2825 (erp->er_extcount - page_idx) *
2826 sizeof(xfs_bmbt_rec_t));
2827 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
2828 }
2829 erp->er_extcount += ext_diff;
2830 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2831 }
2832 /* Insert a new extent page */
2833 else if (erp) {
2834 xfs_iext_add_indirect_multi(ifp,
2835 erp_idx, page_idx, ext_diff);
2836 }
2837 /*
2838 * If extent(s) are being appended to the last page in
2839 * the indirection array and the new extent(s) don't fit
2840 * in the page, then erp is NULL and erp_idx is set to
2841 * the next index needed in the indirection array.
2842 */
2843 else {
2844 int count = ext_diff;
2845
2846 while (count) {
2847 erp = xfs_iext_irec_new(ifp, erp_idx);
2848 erp->er_extcount = count;
2849 count -= MIN(count, (int)XFS_LINEAR_EXTS);
2850 if (count) {
2851 erp_idx++;
2852 }
2853 }
2854 }
2855 }
2856 ifp->if_bytes = new_size;
2857}
2858
2859/*
2860 * This is called when incore extents are being added to the indirection
2861 * array and the new extents do not fit in the target extent list. The
2862 * erp_idx parameter contains the irec index for the target extent list
2863 * in the indirection array, and the idx parameter contains the extent
2864 * index within the list. The number of extents being added is stored
2865 * in the count parameter.
2866 *
2867 * |-------| |-------|
2868 * | | | | idx - number of extents before idx
2869 * | idx | | count |
2870 * | | | | count - number of extents being inserted at idx
2871 * |-------| |-------|
2872 * | count | | nex2 | nex2 - number of extents after idx + count
2873 * |-------| |-------|
2874 */
2875void
2876xfs_iext_add_indirect_multi(
2877 xfs_ifork_t *ifp, /* inode fork pointer */
2878 int erp_idx, /* target extent irec index */
2879 xfs_extnum_t idx, /* index within target list */
2880 int count) /* new extents being added */
2881{
2882 int byte_diff; /* new bytes being added */
2883 xfs_ext_irec_t *erp; /* pointer to irec entry */
2884 xfs_extnum_t ext_diff; /* number of extents to add */
2885 xfs_extnum_t ext_cnt; /* new extents still needed */
2886 xfs_extnum_t nex2; /* extents after idx + count */
2887 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
2888 int nlists; /* number of irec's (lists) */
2889
2890 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2891 erp = &ifp->if_u1.if_ext_irec[erp_idx];
2892 nex2 = erp->er_extcount - idx;
2893 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
2894
2895 /*
2896 * Save second part of target extent list
2897 * (all extents past */
2898 if (nex2) {
2899 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
2900 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
2901 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
2902 erp->er_extcount -= nex2;
2903 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
2904 memset(&erp->er_extbuf[idx], 0, byte_diff);
2905 }
2906
2907 /*
2908 * Add the new extents to the end of the target
2909 * list, then allocate new irec record(s) and
2910 * extent buffer(s) as needed to store the rest
2911 * of the new extents.
2912 */
2913 ext_cnt = count;
2914 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
2915 if (ext_diff) {
2916 erp->er_extcount += ext_diff;
2917 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2918 ext_cnt -= ext_diff;
2919 }
2920 while (ext_cnt) {
2921 erp_idx++;
2922 erp = xfs_iext_irec_new(ifp, erp_idx);
2923 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
2924 erp->er_extcount = ext_diff;
2925 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2926 ext_cnt -= ext_diff;
2927 }
2928
2929 /* Add nex2 extents back to indirection array */
2930 if (nex2) {
2931 xfs_extnum_t ext_avail;
2932 int i;
2933
2934 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
2935 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
2936 i = 0;
2937 /*
2938 * If nex2 extents fit in the current page, append
2939 * nex2_ep after the new extents.
2940 */
2941 if (nex2 <= ext_avail) {
2942 i = erp->er_extcount;
2943 }
2944 /*
2945 * Otherwise, check if space is available in the
2946 * next page.
2947 */
2948 else if ((erp_idx < nlists - 1) &&
2949 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
2950 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
2951 erp_idx++;
2952 erp++;
2953 /* Create a hole for nex2 extents */
2954 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
2955 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
2956 }
2957 /*
2958 * Final choice, create a new extent page for
2959 * nex2 extents.
2960 */
2961 else {
2962 erp_idx++;
2963 erp = xfs_iext_irec_new(ifp, erp_idx);
2964 }
2965 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
2966 kmem_free(nex2_ep);
2967 erp->er_extcount += nex2;
2968 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
2969 }
2970}
2971
2972/*
2973 * This is called when the amount of space required for incore file
2974 * extents needs to be decreased. The ext_diff parameter stores the
2975 * number of extents to be removed and the idx parameter contains
2976 * the extent index where the extents will be removed from.
2977 *
2978 * If the amount of space needed has decreased below the linear
2979 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
2980 * extent array. Otherwise, use kmem_realloc() to adjust the
2981 * size to what is needed.
2982 */
2983void
2984xfs_iext_remove(
2985 xfs_inode_t *ip, /* incore inode pointer */
2986 xfs_extnum_t idx, /* index to begin removing exts */
2987 int ext_diff, /* number of extents to remove */
2988 int state) /* type of extent conversion */
2989{
2990 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2991 xfs_extnum_t nextents; /* number of extents in file */
2992 int new_size; /* size of extents after removal */
2993
2994 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
2995
2996 ASSERT(ext_diff > 0);
2997 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2998 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
2999
3000 if (new_size == 0) {
3001 xfs_iext_destroy(ifp);
3002 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3003 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3004 } else if (ifp->if_real_bytes) {
3005 xfs_iext_remove_direct(ifp, idx, ext_diff);
3006 } else {
3007 xfs_iext_remove_inline(ifp, idx, ext_diff);
3008 }
3009 ifp->if_bytes = new_size;
3010}
3011
3012/*
3013 * This removes ext_diff extents from the inline buffer, beginning
3014 * at extent index idx.
3015 */
3016void
3017xfs_iext_remove_inline(
3018 xfs_ifork_t *ifp, /* inode fork pointer */
3019 xfs_extnum_t idx, /* index to begin removing exts */
3020 int ext_diff) /* number of extents to remove */
3021{
3022 int nextents; /* number of extents in file */
3023
3024 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3025 ASSERT(idx < XFS_INLINE_EXTS);
3026 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3027 ASSERT(((nextents - ext_diff) > 0) &&
3028 (nextents - ext_diff) < XFS_INLINE_EXTS);
3029
3030 if (idx + ext_diff < nextents) {
3031 memmove(&ifp->if_u2.if_inline_ext[idx],
3032 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3033 (nextents - (idx + ext_diff)) *
3034 sizeof(xfs_bmbt_rec_t));
3035 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3036 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3037 } else {
3038 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3039 ext_diff * sizeof(xfs_bmbt_rec_t));
3040 }
3041}
3042
3043/*
3044 * This removes ext_diff extents from a linear (direct) extent list,
3045 * beginning at extent index idx. If the extents are being removed
3046 * from the end of the list (ie. truncate) then we just need to re-
3047 * allocate the list to remove the extra space. Otherwise, if the
3048 * extents are being removed from the middle of the existing extent
3049 * entries, then we first need to move the extent records beginning
3050 * at idx + ext_diff up in the list to overwrite the records being
3051 * removed, then remove the extra space via kmem_realloc.
3052 */
3053void
3054xfs_iext_remove_direct(
3055 xfs_ifork_t *ifp, /* inode fork pointer */
3056 xfs_extnum_t idx, /* index to begin removing exts */
3057 int ext_diff) /* number of extents to remove */
3058{
3059 xfs_extnum_t nextents; /* number of extents in file */
3060 int new_size; /* size of extents after removal */
3061
3062 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3063 new_size = ifp->if_bytes -
3064 (ext_diff * sizeof(xfs_bmbt_rec_t));
3065 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3066
3067 if (new_size == 0) {
3068 xfs_iext_destroy(ifp);
3069 return;
3070 }
3071 /* Move extents up in the list (if needed) */
3072 if (idx + ext_diff < nextents) {
3073 memmove(&ifp->if_u1.if_extents[idx],
3074 &ifp->if_u1.if_extents[idx + ext_diff],
3075 (nextents - (idx + ext_diff)) *
3076 sizeof(xfs_bmbt_rec_t));
3077 }
3078 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3079 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3080 /*
3081 * Reallocate the direct extent list. If the extents
3082 * will fit inside the inode then xfs_iext_realloc_direct
3083 * will switch from direct to inline extent allocation
3084 * mode for us.
3085 */
3086 xfs_iext_realloc_direct(ifp, new_size);
3087 ifp->if_bytes = new_size;
3088}
3089
3090/*
3091 * This is called when incore extents are being removed from the
3092 * indirection array and the extents being removed span multiple extent
3093 * buffers. The idx parameter contains the file extent index where we
3094 * want to begin removing extents, and the count parameter contains
3095 * how many extents need to be removed.
3096 *
3097 * |-------| |-------|
3098 * | nex1 | | | nex1 - number of extents before idx
3099 * |-------| | count |
3100 * | | | | count - number of extents being removed at idx
3101 * | count | |-------|
3102 * | | | nex2 | nex2 - number of extents after idx + count
3103 * |-------| |-------|
3104 */
3105void
3106xfs_iext_remove_indirect(
3107 xfs_ifork_t *ifp, /* inode fork pointer */
3108 xfs_extnum_t idx, /* index to begin removing extents */
3109 int count) /* number of extents to remove */
3110{
3111 xfs_ext_irec_t *erp; /* indirection array pointer */
3112 int erp_idx = 0; /* indirection array index */
3113 xfs_extnum_t ext_cnt; /* extents left to remove */
3114 xfs_extnum_t ext_diff; /* extents to remove in current list */
3115 xfs_extnum_t nex1; /* number of extents before idx */
3116 xfs_extnum_t nex2; /* extents after idx + count */
3117 int page_idx = idx; /* index in target extent list */
3118
3119 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3120 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3121 ASSERT(erp != NULL);
3122 nex1 = page_idx;
3123 ext_cnt = count;
3124 while (ext_cnt) {
3125 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3126 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3127 /*
3128 * Check for deletion of entire list;
3129 * xfs_iext_irec_remove() updates extent offsets.
3130 */
3131 if (ext_diff == erp->er_extcount) {
3132 xfs_iext_irec_remove(ifp, erp_idx);
3133 ext_cnt -= ext_diff;
3134 nex1 = 0;
3135 if (ext_cnt) {
3136 ASSERT(erp_idx < ifp->if_real_bytes /
3137 XFS_IEXT_BUFSZ);
3138 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3139 nex1 = 0;
3140 continue;
3141 } else {
3142 break;
3143 }
3144 }
3145 /* Move extents up (if needed) */
3146 if (nex2) {
3147 memmove(&erp->er_extbuf[nex1],
3148 &erp->er_extbuf[nex1 + ext_diff],
3149 nex2 * sizeof(xfs_bmbt_rec_t));
3150 }
3151 /* Zero out rest of page */
3152 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3153 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3154 /* Update remaining counters */
3155 erp->er_extcount -= ext_diff;
3156 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3157 ext_cnt -= ext_diff;
3158 nex1 = 0;
3159 erp_idx++;
3160 erp++;
3161 }
3162 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3163 xfs_iext_irec_compact(ifp);
3164}
3165
3166/*
3167 * Create, destroy, or resize a linear (direct) block of extents.
3168 */
3169void
3170xfs_iext_realloc_direct(
3171 xfs_ifork_t *ifp, /* inode fork pointer */
3172 int new_size) /* new size of extents */
3173{
3174 int rnew_size; /* real new size of extents */
3175
3176 rnew_size = new_size;
3177
3178 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3179 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3180 (new_size != ifp->if_real_bytes)));
3181
3182 /* Free extent records */
3183 if (new_size == 0) {
3184 xfs_iext_destroy(ifp);
3185 }
3186 /* Resize direct extent list and zero any new bytes */
3187 else if (ifp->if_real_bytes) {
3188 /* Check if extents will fit inside the inode */
3189 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3190 xfs_iext_direct_to_inline(ifp, new_size /
3191 (uint)sizeof(xfs_bmbt_rec_t));
3192 ifp->if_bytes = new_size;
3193 return;
3194 }
3195 if (!is_power_of_2(new_size)){
3196 rnew_size = roundup_pow_of_two(new_size);
3197 }
3198 if (rnew_size != ifp->if_real_bytes) {
3199 ifp->if_u1.if_extents =
3200 kmem_realloc(ifp->if_u1.if_extents,
3201 rnew_size,
3202 ifp->if_real_bytes, KM_NOFS);
3203 }
3204 if (rnew_size > ifp->if_real_bytes) {
3205 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3206 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3207 rnew_size - ifp->if_real_bytes);
3208 }
3209 }
3210 /*
3211 * Switch from the inline extent buffer to a direct
3212 * extent list. Be sure to include the inline extent
3213 * bytes in new_size.
3214 */
3215 else {
3216 new_size += ifp->if_bytes;
3217 if (!is_power_of_2(new_size)) {
3218 rnew_size = roundup_pow_of_two(new_size);
3219 }
3220 xfs_iext_inline_to_direct(ifp, rnew_size);
3221 }
3222 ifp->if_real_bytes = rnew_size;
3223 ifp->if_bytes = new_size;
3224}
3225
3226/*
3227 * Switch from linear (direct) extent records to inline buffer.
3228 */
3229void
3230xfs_iext_direct_to_inline(
3231 xfs_ifork_t *ifp, /* inode fork pointer */
3232 xfs_extnum_t nextents) /* number of extents in file */
3233{
3234 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3235 ASSERT(nextents <= XFS_INLINE_EXTS);
3236 /*
3237 * The inline buffer was zeroed when we switched
3238 * from inline to direct extent allocation mode,
3239 * so we don't need to clear it here.
3240 */
3241 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3242 nextents * sizeof(xfs_bmbt_rec_t));
3243 kmem_free(ifp->if_u1.if_extents);
3244 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3245 ifp->if_real_bytes = 0;
3246}
3247
3248/*
3249 * Switch from inline buffer to linear (direct) extent records.
3250 * new_size should already be rounded up to the next power of 2
3251 * by the caller (when appropriate), so use new_size as it is.
3252 * However, since new_size may be rounded up, we can't update
3253 * if_bytes here. It is the caller's responsibility to update
3254 * if_bytes upon return.
3255 */
3256void
3257xfs_iext_inline_to_direct(
3258 xfs_ifork_t *ifp, /* inode fork pointer */
3259 int new_size) /* number of extents in file */
3260{
3261 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3262 memset(ifp->if_u1.if_extents, 0, new_size);
3263 if (ifp->if_bytes) {
3264 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3265 ifp->if_bytes);
3266 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3267 sizeof(xfs_bmbt_rec_t));
3268 }
3269 ifp->if_real_bytes = new_size;
3270}
3271
3272/*
3273 * Resize an extent indirection array to new_size bytes.
3274 */
3275STATIC void
3276xfs_iext_realloc_indirect(
3277 xfs_ifork_t *ifp, /* inode fork pointer */
3278 int new_size) /* new indirection array size */
3279{
3280 int nlists; /* number of irec's (ex lists) */
3281 int size; /* current indirection array size */
3282
3283 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3284 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3285 size = nlists * sizeof(xfs_ext_irec_t);
3286 ASSERT(ifp->if_real_bytes);
3287 ASSERT((new_size >= 0) && (new_size != size));
3288 if (new_size == 0) {
3289 xfs_iext_destroy(ifp);
3290 } else {
3291 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3292 kmem_realloc(ifp->if_u1.if_ext_irec,
3293 new_size, size, KM_NOFS);
3294 }
3295}
3296
3297/*
3298 * Switch from indirection array to linear (direct) extent allocations.
3299 */
3300STATIC void
3301xfs_iext_indirect_to_direct(
3302 xfs_ifork_t *ifp) /* inode fork pointer */
3303{
3304 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3305 xfs_extnum_t nextents; /* number of extents in file */
3306 int size; /* size of file extents */
3307
3308 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3309 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3310 ASSERT(nextents <= XFS_LINEAR_EXTS);
3311 size = nextents * sizeof(xfs_bmbt_rec_t);
3312
3313 xfs_iext_irec_compact_pages(ifp);
3314 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3315
3316 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3317 kmem_free(ifp->if_u1.if_ext_irec);
3318 ifp->if_flags &= ~XFS_IFEXTIREC;
3319 ifp->if_u1.if_extents = ep;
3320 ifp->if_bytes = size;
3321 if (nextents < XFS_LINEAR_EXTS) {
3322 xfs_iext_realloc_direct(ifp, size);
3323 }
3324}
3325
3326/*
3327 * Free incore file extents.
3328 */
3329void
3330xfs_iext_destroy(
3331 xfs_ifork_t *ifp) /* inode fork pointer */
3332{
3333 if (ifp->if_flags & XFS_IFEXTIREC) {
3334 int erp_idx;
3335 int nlists;
3336
3337 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3338 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3339 xfs_iext_irec_remove(ifp, erp_idx);
3340 }
3341 ifp->if_flags &= ~XFS_IFEXTIREC;
3342 } else if (ifp->if_real_bytes) {
3343 kmem_free(ifp->if_u1.if_extents);
3344 } else if (ifp->if_bytes) {
3345 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3346 sizeof(xfs_bmbt_rec_t));
3347 }
3348 ifp->if_u1.if_extents = NULL;
3349 ifp->if_real_bytes = 0;
3350 ifp->if_bytes = 0;
3351}
3352
3353/*
3354 * Return a pointer to the extent record for file system block bno.
3355 */
3356xfs_bmbt_rec_host_t * /* pointer to found extent record */
3357xfs_iext_bno_to_ext(
3358 xfs_ifork_t *ifp, /* inode fork pointer */
3359 xfs_fileoff_t bno, /* block number to search for */
3360 xfs_extnum_t *idxp) /* index of target extent */
3361{
3362 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3363 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3364 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3365 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3366 int high; /* upper boundary in search */
3367 xfs_extnum_t idx = 0; /* index of target extent */
3368 int low; /* lower boundary in search */
3369 xfs_extnum_t nextents; /* number of file extents */
3370 xfs_fileoff_t startoff = 0; /* start offset of extent */
3371
3372 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3373 if (nextents == 0) {
3374 *idxp = 0;
3375 return NULL;
3376 }
3377 low = 0;
3378 if (ifp->if_flags & XFS_IFEXTIREC) {
3379 /* Find target extent list */
3380 int erp_idx = 0;
3381 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3382 base = erp->er_extbuf;
3383 high = erp->er_extcount - 1;
3384 } else {
3385 base = ifp->if_u1.if_extents;
3386 high = nextents - 1;
3387 }
3388 /* Binary search extent records */
3389 while (low <= high) {
3390 idx = (low + high) >> 1;
3391 ep = base + idx;
3392 startoff = xfs_bmbt_get_startoff(ep);
3393 blockcount = xfs_bmbt_get_blockcount(ep);
3394 if (bno < startoff) {
3395 high = idx - 1;
3396 } else if (bno >= startoff + blockcount) {
3397 low = idx + 1;
3398 } else {
3399 /* Convert back to file-based extent index */
3400 if (ifp->if_flags & XFS_IFEXTIREC) {
3401 idx += erp->er_extoff;
3402 }
3403 *idxp = idx;
3404 return ep;
3405 }
3406 }
3407 /* Convert back to file-based extent index */
3408 if (ifp->if_flags & XFS_IFEXTIREC) {
3409 idx += erp->er_extoff;
3410 }
3411 if (bno >= startoff + blockcount) {
3412 if (++idx == nextents) {
3413 ep = NULL;
3414 } else {
3415 ep = xfs_iext_get_ext(ifp, idx);
3416 }
3417 }
3418 *idxp = idx;
3419 return ep;
3420}
3421
3422/*
3423 * Return a pointer to the indirection array entry containing the
3424 * extent record for filesystem block bno. Store the index of the
3425 * target irec in *erp_idxp.
3426 */
3427xfs_ext_irec_t * /* pointer to found extent record */
3428xfs_iext_bno_to_irec(
3429 xfs_ifork_t *ifp, /* inode fork pointer */
3430 xfs_fileoff_t bno, /* block number to search for */
3431 int *erp_idxp) /* irec index of target ext list */
3432{
3433 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3434 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3435 int erp_idx; /* indirection array index */
3436 int nlists; /* number of extent irec's (lists) */
3437 int high; /* binary search upper limit */
3438 int low; /* binary search lower limit */
3439
3440 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3441 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3442 erp_idx = 0;
3443 low = 0;
3444 high = nlists - 1;
3445 while (low <= high) {
3446 erp_idx = (low + high) >> 1;
3447 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3448 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3449 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3450 high = erp_idx - 1;
3451 } else if (erp_next && bno >=
3452 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3453 low = erp_idx + 1;
3454 } else {
3455 break;
3456 }
3457 }
3458 *erp_idxp = erp_idx;
3459 return erp;
3460}
3461
3462/*
3463 * Return a pointer to the indirection array entry containing the
3464 * extent record at file extent index *idxp. Store the index of the
3465 * target irec in *erp_idxp and store the page index of the target
3466 * extent record in *idxp.
3467 */
3468xfs_ext_irec_t *
3469xfs_iext_idx_to_irec(
3470 xfs_ifork_t *ifp, /* inode fork pointer */
3471 xfs_extnum_t *idxp, /* extent index (file -> page) */
3472 int *erp_idxp, /* pointer to target irec */
3473 int realloc) /* new bytes were just added */
3474{
3475 xfs_ext_irec_t *prev; /* pointer to previous irec */
3476 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3477 int erp_idx; /* indirection array index */
3478 int nlists; /* number of irec's (ex lists) */
3479 int high; /* binary search upper limit */
3480 int low; /* binary search lower limit */
3481 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3482
3483 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3484 ASSERT(page_idx >= 0);
3485 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3486 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3487
3488 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3489 erp_idx = 0;
3490 low = 0;
3491 high = nlists - 1;
3492
3493 /* Binary search extent irec's */
3494 while (low <= high) {
3495 erp_idx = (low + high) >> 1;
3496 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3497 prev = erp_idx > 0 ? erp - 1 : NULL;
3498 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3499 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3500 high = erp_idx - 1;
3501 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3502 (page_idx == erp->er_extoff + erp->er_extcount &&
3503 !realloc)) {
3504 low = erp_idx + 1;
3505 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3506 erp->er_extcount == XFS_LINEAR_EXTS) {
3507 ASSERT(realloc);
3508 page_idx = 0;
3509 erp_idx++;
3510 erp = erp_idx < nlists ? erp + 1 : NULL;
3511 break;
3512 } else {
3513 page_idx -= erp->er_extoff;
3514 break;
3515 }
3516 }
3517 *idxp = page_idx;
3518 *erp_idxp = erp_idx;
3519 return(erp);
3520}
3521
3522/*
3523 * Allocate and initialize an indirection array once the space needed
3524 * for incore extents increases above XFS_IEXT_BUFSZ.
3525 */
3526void
3527xfs_iext_irec_init(
3528 xfs_ifork_t *ifp) /* inode fork pointer */
3529{
3530 xfs_ext_irec_t *erp; /* indirection array pointer */
3531 xfs_extnum_t nextents; /* number of extents in file */
3532
3533 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3534 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3535 ASSERT(nextents <= XFS_LINEAR_EXTS);
3536
3537 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3538
3539 if (nextents == 0) {
3540 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3541 } else if (!ifp->if_real_bytes) {
3542 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3543 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3544 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3545 }
3546 erp->er_extbuf = ifp->if_u1.if_extents;
3547 erp->er_extcount = nextents;
3548 erp->er_extoff = 0;
3549
3550 ifp->if_flags |= XFS_IFEXTIREC;
3551 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3552 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3553 ifp->if_u1.if_ext_irec = erp;
3554
3555 return;
3556}
3557
3558/*
3559 * Allocate and initialize a new entry in the indirection array.
3560 */
3561xfs_ext_irec_t *
3562xfs_iext_irec_new(
3563 xfs_ifork_t *ifp, /* inode fork pointer */
3564 int erp_idx) /* index for new irec */
3565{
3566 xfs_ext_irec_t *erp; /* indirection array pointer */
3567 int i; /* loop counter */
3568 int nlists; /* number of irec's (ex lists) */
3569
3570 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3571 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3572
3573 /* Resize indirection array */
3574 xfs_iext_realloc_indirect(ifp, ++nlists *
3575 sizeof(xfs_ext_irec_t));
3576 /*
3577 * Move records down in the array so the
3578 * new page can use erp_idx.
3579 */
3580 erp = ifp->if_u1.if_ext_irec;
3581 for (i = nlists - 1; i > erp_idx; i--) {
3582 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3583 }
3584 ASSERT(i == erp_idx);
3585
3586 /* Initialize new extent record */
3587 erp = ifp->if_u1.if_ext_irec;
3588 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3589 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3590 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3591 erp[erp_idx].er_extcount = 0;
3592 erp[erp_idx].er_extoff = erp_idx > 0 ?
3593 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3594 return (&erp[erp_idx]);
3595}
3596
3597/*
3598 * Remove a record from the indirection array.
3599 */
3600void
3601xfs_iext_irec_remove(
3602 xfs_ifork_t *ifp, /* inode fork pointer */
3603 int erp_idx) /* irec index to remove */
3604{
3605 xfs_ext_irec_t *erp; /* indirection array pointer */
3606 int i; /* loop counter */
3607 int nlists; /* number of irec's (ex lists) */
3608
3609 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3610 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3611 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3612 if (erp->er_extbuf) {
3613 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3614 -erp->er_extcount);
3615 kmem_free(erp->er_extbuf);
3616 }
3617 /* Compact extent records */
3618 erp = ifp->if_u1.if_ext_irec;
3619 for (i = erp_idx; i < nlists - 1; i++) {
3620 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3621 }
3622 /*
3623 * Manually free the last extent record from the indirection
3624 * array. A call to xfs_iext_realloc_indirect() with a size
3625 * of zero would result in a call to xfs_iext_destroy() which
3626 * would in turn call this function again, creating a nasty
3627 * infinite loop.
3628 */
3629 if (--nlists) {
3630 xfs_iext_realloc_indirect(ifp,
3631 nlists * sizeof(xfs_ext_irec_t));
3632 } else {
3633 kmem_free(ifp->if_u1.if_ext_irec);
3634 }
3635 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3636}
3637
3638/*
3639 * This is called to clean up large amounts of unused memory allocated
3640 * by the indirection array. Before compacting anything though, verify
3641 * that the indirection array is still needed and switch back to the
3642 * linear extent list (or even the inline buffer) if possible. The
3643 * compaction policy is as follows:
3644 *
3645 * Full Compaction: Extents fit into a single page (or inline buffer)
3646 * Partial Compaction: Extents occupy less than 50% of allocated space
3647 * No Compaction: Extents occupy at least 50% of allocated space
3648 */
3649void
3650xfs_iext_irec_compact(
3651 xfs_ifork_t *ifp) /* inode fork pointer */
3652{
3653 xfs_extnum_t nextents; /* number of extents in file */
3654 int nlists; /* number of irec's (ex lists) */
3655
3656 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3657 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3658 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3659
3660 if (nextents == 0) {
3661 xfs_iext_destroy(ifp);
3662 } else if (nextents <= XFS_INLINE_EXTS) {
3663 xfs_iext_indirect_to_direct(ifp);
3664 xfs_iext_direct_to_inline(ifp, nextents);
3665 } else if (nextents <= XFS_LINEAR_EXTS) {
3666 xfs_iext_indirect_to_direct(ifp);
3667 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3668 xfs_iext_irec_compact_pages(ifp);
3669 }
3670}
3671
3672/*
3673 * Combine extents from neighboring extent pages.
3674 */
3675void
3676xfs_iext_irec_compact_pages(
3677 xfs_ifork_t *ifp) /* inode fork pointer */
3678{
3679 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
3680 int erp_idx = 0; /* indirection array index */
3681 int nlists; /* number of irec's (ex lists) */
3682
3683 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3684 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3685 while (erp_idx < nlists - 1) {
3686 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3687 erp_next = erp + 1;
3688 if (erp_next->er_extcount <=
3689 (XFS_LINEAR_EXTS - erp->er_extcount)) {
3690 memcpy(&erp->er_extbuf[erp->er_extcount],
3691 erp_next->er_extbuf, erp_next->er_extcount *
3692 sizeof(xfs_bmbt_rec_t));
3693 erp->er_extcount += erp_next->er_extcount;
3694 /*
3695 * Free page before removing extent record
3696 * so er_extoffs don't get modified in
3697 * xfs_iext_irec_remove.
3698 */
3699 kmem_free(erp_next->er_extbuf);
3700 erp_next->er_extbuf = NULL;
3701 xfs_iext_irec_remove(ifp, erp_idx + 1);
3702 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3703 } else {
3704 erp_idx++;
3705 }
3706 }
3707}
3708
3709/*
3710 * This is called to update the er_extoff field in the indirection
3711 * array when extents have been added or removed from one of the
3712 * extent lists. erp_idx contains the irec index to begin updating
3713 * at and ext_diff contains the number of extents that were added
3714 * or removed.
3715 */
3716void
3717xfs_iext_irec_update_extoffs(
3718 xfs_ifork_t *ifp, /* inode fork pointer */
3719 int erp_idx, /* irec index to update */
3720 int ext_diff) /* number of new extents */
3721{
3722 int i; /* loop counter */
3723 int nlists; /* number of irec's (ex lists */
3724
3725 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3726 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3727 for (i = erp_idx; i < nlists; i++) {
3728 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
3729 }
3730}