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_sb.h"
15#include "xfs_mount.h"
16#include "xfs_defer.h"
17#include "xfs_inode.h"
18#include "xfs_dir2.h"
19#include "xfs_attr.h"
20#include "xfs_trans_space.h"
21#include "xfs_trans.h"
22#include "xfs_buf_item.h"
23#include "xfs_inode_item.h"
24#include "xfs_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
39kmem_zone_t *xfs_inode_zone;
40
41/*
42 * Used in xfs_itruncate_extents(). This is the maximum number of extents
43 * freed from a file in a single transaction.
44 */
45#define XFS_ITRUNC_MAX_EXTENTS 2
46
47STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
48STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *);
49
50/*
51 * helper function to extract extent size hint from inode
52 */
53xfs_extlen_t
54xfs_get_extsz_hint(
55 struct xfs_inode *ip)
56{
57 /*
58 * No point in aligning allocations if we need to COW to actually
59 * write to them.
60 */
61 if (xfs_is_always_cow_inode(ip))
62 return 0;
63 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
64 return ip->i_d.di_extsize;
65 if (XFS_IS_REALTIME_INODE(ip))
66 return ip->i_mount->m_sb.sb_rextsize;
67 return 0;
68}
69
70/*
71 * Helper function to extract CoW extent size hint from inode.
72 * Between the extent size hint and the CoW extent size hint, we
73 * return the greater of the two. If the value is zero (automatic),
74 * use the default size.
75 */
76xfs_extlen_t
77xfs_get_cowextsz_hint(
78 struct xfs_inode *ip)
79{
80 xfs_extlen_t a, b;
81
82 a = 0;
83 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
84 a = ip->i_d.di_cowextsize;
85 b = xfs_get_extsz_hint(ip);
86
87 a = max(a, b);
88 if (a == 0)
89 return XFS_DEFAULT_COWEXTSZ_HINT;
90 return a;
91}
92
93/*
94 * These two are wrapper routines around the xfs_ilock() routine used to
95 * centralize some grungy code. They are used in places that wish to lock the
96 * inode solely for reading the extents. The reason these places can't just
97 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
98 * bringing in of the extents from disk for a file in b-tree format. If the
99 * inode is in b-tree format, then we need to lock the inode exclusively until
100 * the extents are read in. Locking it exclusively all the time would limit
101 * our parallelism unnecessarily, though. What we do instead is check to see
102 * if the extents have been read in yet, and only lock the inode exclusively
103 * if they have not.
104 *
105 * The functions return a value which should be given to the corresponding
106 * xfs_iunlock() call.
107 */
108uint
109xfs_ilock_data_map_shared(
110 struct xfs_inode *ip)
111{
112 uint lock_mode = XFS_ILOCK_SHARED;
113
114 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE &&
115 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0)
116 lock_mode = XFS_ILOCK_EXCL;
117 xfs_ilock(ip, lock_mode);
118 return lock_mode;
119}
120
121uint
122xfs_ilock_attr_map_shared(
123 struct xfs_inode *ip)
124{
125 uint lock_mode = XFS_ILOCK_SHARED;
126
127 if (ip->i_afp &&
128 ip->i_afp->if_format == XFS_DINODE_FMT_BTREE &&
129 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0)
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
132 return lock_mode;
133}
134
135/*
136 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
137 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows
138 * various combinations of the locks to be obtained.
139 *
140 * The 3 locks should always be ordered so that the IO lock is obtained first,
141 * the mmap lock second and the ilock last in order to prevent deadlock.
142 *
143 * Basic locking order:
144 *
145 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
146 *
147 * mmap_lock locking order:
148 *
149 * i_rwsem -> page lock -> mmap_lock
150 * mmap_lock -> i_mmap_lock -> page_lock
151 *
152 * The difference in mmap_lock locking order mean that we cannot hold the
153 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
154 * fault in pages during copy in/out (for buffered IO) or require the mmap_lock
155 * in get_user_pages() to map the user pages into the kernel address space for
156 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
157 * page faults already hold the mmap_lock.
158 *
159 * Hence to serialise fully against both syscall and mmap based IO, we need to
160 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
161 * taken in places where we need to invalidate the page cache in a race
162 * free manner (e.g. truncate, hole punch and other extent manipulation
163 * functions).
164 */
165void
166xfs_ilock(
167 xfs_inode_t *ip,
168 uint lock_flags)
169{
170 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
171
172 /*
173 * You can't set both SHARED and EXCL for the same lock,
174 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
175 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
176 */
177 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
178 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
179 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
180 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
181 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
182 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
183 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
184
185 if (lock_flags & XFS_IOLOCK_EXCL) {
186 down_write_nested(&VFS_I(ip)->i_rwsem,
187 XFS_IOLOCK_DEP(lock_flags));
188 } else if (lock_flags & XFS_IOLOCK_SHARED) {
189 down_read_nested(&VFS_I(ip)->i_rwsem,
190 XFS_IOLOCK_DEP(lock_flags));
191 }
192
193 if (lock_flags & XFS_MMAPLOCK_EXCL)
194 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
195 else if (lock_flags & XFS_MMAPLOCK_SHARED)
196 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
197
198 if (lock_flags & XFS_ILOCK_EXCL)
199 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
200 else if (lock_flags & XFS_ILOCK_SHARED)
201 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
202}
203
204/*
205 * This is just like xfs_ilock(), except that the caller
206 * is guaranteed not to sleep. It returns 1 if it gets
207 * the requested locks and 0 otherwise. If the IO lock is
208 * obtained but the inode lock cannot be, then the IO lock
209 * is dropped before returning.
210 *
211 * ip -- the inode being locked
212 * lock_flags -- this parameter indicates the inode's locks to be
213 * to be locked. See the comment for xfs_ilock() for a list
214 * of valid values.
215 */
216int
217xfs_ilock_nowait(
218 xfs_inode_t *ip,
219 uint lock_flags)
220{
221 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
222
223 /*
224 * You can't set both SHARED and EXCL for the same lock,
225 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
226 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
227 */
228 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
229 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
230 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
231 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
232 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
233 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
234 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
235
236 if (lock_flags & XFS_IOLOCK_EXCL) {
237 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
238 goto out;
239 } else if (lock_flags & XFS_IOLOCK_SHARED) {
240 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
241 goto out;
242 }
243
244 if (lock_flags & XFS_MMAPLOCK_EXCL) {
245 if (!mrtryupdate(&ip->i_mmaplock))
246 goto out_undo_iolock;
247 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
248 if (!mrtryaccess(&ip->i_mmaplock))
249 goto out_undo_iolock;
250 }
251
252 if (lock_flags & XFS_ILOCK_EXCL) {
253 if (!mrtryupdate(&ip->i_lock))
254 goto out_undo_mmaplock;
255 } else if (lock_flags & XFS_ILOCK_SHARED) {
256 if (!mrtryaccess(&ip->i_lock))
257 goto out_undo_mmaplock;
258 }
259 return 1;
260
261out_undo_mmaplock:
262 if (lock_flags & XFS_MMAPLOCK_EXCL)
263 mrunlock_excl(&ip->i_mmaplock);
264 else if (lock_flags & XFS_MMAPLOCK_SHARED)
265 mrunlock_shared(&ip->i_mmaplock);
266out_undo_iolock:
267 if (lock_flags & XFS_IOLOCK_EXCL)
268 up_write(&VFS_I(ip)->i_rwsem);
269 else if (lock_flags & XFS_IOLOCK_SHARED)
270 up_read(&VFS_I(ip)->i_rwsem);
271out:
272 return 0;
273}
274
275/*
276 * xfs_iunlock() is used to drop the inode locks acquired with
277 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
278 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
279 * that we know which locks to drop.
280 *
281 * ip -- the inode being unlocked
282 * lock_flags -- this parameter indicates the inode's locks to be
283 * to be unlocked. See the comment for xfs_ilock() for a list
284 * of valid values for this parameter.
285 *
286 */
287void
288xfs_iunlock(
289 xfs_inode_t *ip,
290 uint lock_flags)
291{
292 /*
293 * You can't set both SHARED and EXCL for the same lock,
294 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
295 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
296 */
297 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
298 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
299 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
300 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
301 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
302 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
303 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
304 ASSERT(lock_flags != 0);
305
306 if (lock_flags & XFS_IOLOCK_EXCL)
307 up_write(&VFS_I(ip)->i_rwsem);
308 else if (lock_flags & XFS_IOLOCK_SHARED)
309 up_read(&VFS_I(ip)->i_rwsem);
310
311 if (lock_flags & XFS_MMAPLOCK_EXCL)
312 mrunlock_excl(&ip->i_mmaplock);
313 else if (lock_flags & XFS_MMAPLOCK_SHARED)
314 mrunlock_shared(&ip->i_mmaplock);
315
316 if (lock_flags & XFS_ILOCK_EXCL)
317 mrunlock_excl(&ip->i_lock);
318 else if (lock_flags & XFS_ILOCK_SHARED)
319 mrunlock_shared(&ip->i_lock);
320
321 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
322}
323
324/*
325 * give up write locks. the i/o lock cannot be held nested
326 * if it is being demoted.
327 */
328void
329xfs_ilock_demote(
330 xfs_inode_t *ip,
331 uint lock_flags)
332{
333 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
334 ASSERT((lock_flags &
335 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
336
337 if (lock_flags & XFS_ILOCK_EXCL)
338 mrdemote(&ip->i_lock);
339 if (lock_flags & XFS_MMAPLOCK_EXCL)
340 mrdemote(&ip->i_mmaplock);
341 if (lock_flags & XFS_IOLOCK_EXCL)
342 downgrade_write(&VFS_I(ip)->i_rwsem);
343
344 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
345}
346
347#if defined(DEBUG) || defined(XFS_WARN)
348int
349xfs_isilocked(
350 xfs_inode_t *ip,
351 uint lock_flags)
352{
353 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
354 if (!(lock_flags & XFS_ILOCK_SHARED))
355 return !!ip->i_lock.mr_writer;
356 return rwsem_is_locked(&ip->i_lock.mr_lock);
357 }
358
359 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
360 if (!(lock_flags & XFS_MMAPLOCK_SHARED))
361 return !!ip->i_mmaplock.mr_writer;
362 return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
363 }
364
365 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
366 if (!(lock_flags & XFS_IOLOCK_SHARED))
367 return !debug_locks ||
368 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
369 return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
370 }
371
372 ASSERT(0);
373 return 0;
374}
375#endif
376
377/*
378 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
379 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
380 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
381 * errors and warnings.
382 */
383#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
384static bool
385xfs_lockdep_subclass_ok(
386 int subclass)
387{
388 return subclass < MAX_LOCKDEP_SUBCLASSES;
389}
390#else
391#define xfs_lockdep_subclass_ok(subclass) (true)
392#endif
393
394/*
395 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
396 * value. This can be called for any type of inode lock combination, including
397 * parent locking. Care must be taken to ensure we don't overrun the subclass
398 * storage fields in the class mask we build.
399 */
400static inline int
401xfs_lock_inumorder(int lock_mode, int subclass)
402{
403 int class = 0;
404
405 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
406 XFS_ILOCK_RTSUM)));
407 ASSERT(xfs_lockdep_subclass_ok(subclass));
408
409 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
410 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
411 class += subclass << XFS_IOLOCK_SHIFT;
412 }
413
414 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
415 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
416 class += subclass << XFS_MMAPLOCK_SHIFT;
417 }
418
419 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
420 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
421 class += subclass << XFS_ILOCK_SHIFT;
422 }
423
424 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
425}
426
427/*
428 * The following routine will lock n inodes in exclusive mode. We assume the
429 * caller calls us with the inodes in i_ino order.
430 *
431 * We need to detect deadlock where an inode that we lock is in the AIL and we
432 * start waiting for another inode that is locked by a thread in a long running
433 * transaction (such as truncate). This can result in deadlock since the long
434 * running trans might need to wait for the inode we just locked in order to
435 * push the tail and free space in the log.
436 *
437 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
438 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
439 * lock more than one at a time, lockdep will report false positives saying we
440 * have violated locking orders.
441 */
442static void
443xfs_lock_inodes(
444 struct xfs_inode **ips,
445 int inodes,
446 uint lock_mode)
447{
448 int attempts = 0, i, j, try_lock;
449 struct xfs_log_item *lp;
450
451 /*
452 * Currently supports between 2 and 5 inodes with exclusive locking. We
453 * support an arbitrary depth of locking here, but absolute limits on
454 * inodes depend on the type of locking and the limits placed by
455 * lockdep annotations in xfs_lock_inumorder. These are all checked by
456 * the asserts.
457 */
458 ASSERT(ips && inodes >= 2 && inodes <= 5);
459 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
460 XFS_ILOCK_EXCL));
461 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
462 XFS_ILOCK_SHARED)));
463 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
464 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
466 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
467
468 if (lock_mode & XFS_IOLOCK_EXCL) {
469 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
470 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
471 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
472
473 try_lock = 0;
474 i = 0;
475again:
476 for (; i < inodes; i++) {
477 ASSERT(ips[i]);
478
479 if (i && (ips[i] == ips[i - 1])) /* Already locked */
480 continue;
481
482 /*
483 * If try_lock is not set yet, make sure all locked inodes are
484 * not in the AIL. If any are, set try_lock to be used later.
485 */
486 if (!try_lock) {
487 for (j = (i - 1); j >= 0 && !try_lock; j--) {
488 lp = &ips[j]->i_itemp->ili_item;
489 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
490 try_lock++;
491 }
492 }
493
494 /*
495 * If any of the previous locks we have locked is in the AIL,
496 * we must TRY to get the second and subsequent locks. If
497 * we can't get any, we must release all we have
498 * and try again.
499 */
500 if (!try_lock) {
501 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
502 continue;
503 }
504
505 /* try_lock means we have an inode locked that is in the AIL. */
506 ASSERT(i != 0);
507 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
508 continue;
509
510 /*
511 * Unlock all previous guys and try again. xfs_iunlock will try
512 * to push the tail if the inode is in the AIL.
513 */
514 attempts++;
515 for (j = i - 1; j >= 0; j--) {
516 /*
517 * Check to see if we've already unlocked this one. Not
518 * the first one going back, and the inode ptr is the
519 * same.
520 */
521 if (j != (i - 1) && ips[j] == ips[j + 1])
522 continue;
523
524 xfs_iunlock(ips[j], lock_mode);
525 }
526
527 if ((attempts % 5) == 0) {
528 delay(1); /* Don't just spin the CPU */
529 }
530 i = 0;
531 try_lock = 0;
532 goto again;
533 }
534}
535
536/*
537 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
538 * the mmaplock or the ilock, but not more than one type at a time. If we lock
539 * more than one at a time, lockdep will report false positives saying we have
540 * violated locking orders. The iolock must be double-locked separately since
541 * we use i_rwsem for that. We now support taking one lock EXCL and the other
542 * SHARED.
543 */
544void
545xfs_lock_two_inodes(
546 struct xfs_inode *ip0,
547 uint ip0_mode,
548 struct xfs_inode *ip1,
549 uint ip1_mode)
550{
551 struct xfs_inode *temp;
552 uint mode_temp;
553 int attempts = 0;
554 struct xfs_log_item *lp;
555
556 ASSERT(hweight32(ip0_mode) == 1);
557 ASSERT(hweight32(ip1_mode) == 1);
558 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
559 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
560 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
561 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
562 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
563 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
565 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
566 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
567 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
568
569 ASSERT(ip0->i_ino != ip1->i_ino);
570
571 if (ip0->i_ino > ip1->i_ino) {
572 temp = ip0;
573 ip0 = ip1;
574 ip1 = temp;
575 mode_temp = ip0_mode;
576 ip0_mode = ip1_mode;
577 ip1_mode = mode_temp;
578 }
579
580 again:
581 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
582
583 /*
584 * If the first lock we have locked is in the AIL, we must TRY to get
585 * the second lock. If we can't get it, we must release the first one
586 * and try again.
587 */
588 lp = &ip0->i_itemp->ili_item;
589 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
590 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
591 xfs_iunlock(ip0, ip0_mode);
592 if ((++attempts % 5) == 0)
593 delay(1); /* Don't just spin the CPU */
594 goto again;
595 }
596 } else {
597 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
598 }
599}
600
601void
602__xfs_iflock(
603 struct xfs_inode *ip)
604{
605 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
606 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
607
608 do {
609 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
610 if (xfs_isiflocked(ip))
611 io_schedule();
612 } while (!xfs_iflock_nowait(ip));
613
614 finish_wait(wq, &wait.wq_entry);
615}
616
617STATIC uint
618_xfs_dic2xflags(
619 uint16_t di_flags,
620 uint64_t di_flags2,
621 bool has_attr)
622{
623 uint flags = 0;
624
625 if (di_flags & XFS_DIFLAG_ANY) {
626 if (di_flags & XFS_DIFLAG_REALTIME)
627 flags |= FS_XFLAG_REALTIME;
628 if (di_flags & XFS_DIFLAG_PREALLOC)
629 flags |= FS_XFLAG_PREALLOC;
630 if (di_flags & XFS_DIFLAG_IMMUTABLE)
631 flags |= FS_XFLAG_IMMUTABLE;
632 if (di_flags & XFS_DIFLAG_APPEND)
633 flags |= FS_XFLAG_APPEND;
634 if (di_flags & XFS_DIFLAG_SYNC)
635 flags |= FS_XFLAG_SYNC;
636 if (di_flags & XFS_DIFLAG_NOATIME)
637 flags |= FS_XFLAG_NOATIME;
638 if (di_flags & XFS_DIFLAG_NODUMP)
639 flags |= FS_XFLAG_NODUMP;
640 if (di_flags & XFS_DIFLAG_RTINHERIT)
641 flags |= FS_XFLAG_RTINHERIT;
642 if (di_flags & XFS_DIFLAG_PROJINHERIT)
643 flags |= FS_XFLAG_PROJINHERIT;
644 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
645 flags |= FS_XFLAG_NOSYMLINKS;
646 if (di_flags & XFS_DIFLAG_EXTSIZE)
647 flags |= FS_XFLAG_EXTSIZE;
648 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
649 flags |= FS_XFLAG_EXTSZINHERIT;
650 if (di_flags & XFS_DIFLAG_NODEFRAG)
651 flags |= FS_XFLAG_NODEFRAG;
652 if (di_flags & XFS_DIFLAG_FILESTREAM)
653 flags |= FS_XFLAG_FILESTREAM;
654 }
655
656 if (di_flags2 & XFS_DIFLAG2_ANY) {
657 if (di_flags2 & XFS_DIFLAG2_DAX)
658 flags |= FS_XFLAG_DAX;
659 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE)
660 flags |= FS_XFLAG_COWEXTSIZE;
661 }
662
663 if (has_attr)
664 flags |= FS_XFLAG_HASATTR;
665
666 return flags;
667}
668
669uint
670xfs_ip2xflags(
671 struct xfs_inode *ip)
672{
673 struct xfs_icdinode *dic = &ip->i_d;
674
675 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip));
676}
677
678/*
679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
680 * is allowed, otherwise it has to be an exact match. If a CI match is found,
681 * ci_name->name will point to a the actual name (caller must free) or
682 * will be set to NULL if an exact match is found.
683 */
684int
685xfs_lookup(
686 xfs_inode_t *dp,
687 struct xfs_name *name,
688 xfs_inode_t **ipp,
689 struct xfs_name *ci_name)
690{
691 xfs_ino_t inum;
692 int error;
693
694 trace_xfs_lookup(dp, name);
695
696 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
697 return -EIO;
698
699 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
700 if (error)
701 goto out_unlock;
702
703 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
704 if (error)
705 goto out_free_name;
706
707 return 0;
708
709out_free_name:
710 if (ci_name)
711 kmem_free(ci_name->name);
712out_unlock:
713 *ipp = NULL;
714 return error;
715}
716
717/*
718 * Allocate an inode on disk and return a copy of its in-core version.
719 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
720 * appropriately within the inode. The uid and gid for the inode are
721 * set according to the contents of the given cred structure.
722 *
723 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
724 * has a free inode available, call xfs_iget() to obtain the in-core
725 * version of the allocated inode. Finally, fill in the inode and
726 * log its initial contents. In this case, ialloc_context would be
727 * set to NULL.
728 *
729 * If xfs_dialloc() does not have an available inode, it will replenish
730 * its supply by doing an allocation. Since we can only do one
731 * allocation within a transaction without deadlocks, we must commit
732 * the current transaction before returning the inode itself.
733 * In this case, therefore, we will set ialloc_context and return.
734 * The caller should then commit the current transaction, start a new
735 * transaction, and call xfs_ialloc() again to actually get the inode.
736 *
737 * To ensure that some other process does not grab the inode that
738 * was allocated during the first call to xfs_ialloc(), this routine
739 * also returns the [locked] bp pointing to the head of the freelist
740 * as ialloc_context. The caller should hold this buffer across
741 * the commit and pass it back into this routine on the second call.
742 *
743 * If we are allocating quota inodes, we do not have a parent inode
744 * to attach to or associate with (i.e. pip == NULL) because they
745 * are not linked into the directory structure - they are attached
746 * directly to the superblock - and so have no parent.
747 */
748static int
749xfs_ialloc(
750 xfs_trans_t *tp,
751 xfs_inode_t *pip,
752 umode_t mode,
753 xfs_nlink_t nlink,
754 dev_t rdev,
755 prid_t prid,
756 xfs_buf_t **ialloc_context,
757 xfs_inode_t **ipp)
758{
759 struct xfs_mount *mp = tp->t_mountp;
760 xfs_ino_t ino;
761 xfs_inode_t *ip;
762 uint flags;
763 int error;
764 struct timespec64 tv;
765 struct inode *inode;
766
767 /*
768 * Call the space management code to pick
769 * the on-disk inode to be allocated.
770 */
771 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode,
772 ialloc_context, &ino);
773 if (error)
774 return error;
775 if (*ialloc_context || ino == NULLFSINO) {
776 *ipp = NULL;
777 return 0;
778 }
779 ASSERT(*ialloc_context == NULL);
780
781 /*
782 * Protect against obviously corrupt allocation btree records. Later
783 * xfs_iget checks will catch re-allocation of other active in-memory
784 * and on-disk inodes. If we don't catch reallocating the parent inode
785 * here we will deadlock in xfs_iget() so we have to do these checks
786 * first.
787 */
788 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
789 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
790 return -EFSCORRUPTED;
791 }
792
793 /*
794 * Get the in-core inode with the lock held exclusively.
795 * This is because we're setting fields here we need
796 * to prevent others from looking at until we're done.
797 */
798 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
799 XFS_ILOCK_EXCL, &ip);
800 if (error)
801 return error;
802 ASSERT(ip != NULL);
803 inode = VFS_I(ip);
804 inode->i_mode = mode;
805 set_nlink(inode, nlink);
806 inode->i_uid = current_fsuid();
807 inode->i_rdev = rdev;
808 ip->i_d.di_projid = prid;
809
810 if (pip && XFS_INHERIT_GID(pip)) {
811 inode->i_gid = VFS_I(pip)->i_gid;
812 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode))
813 inode->i_mode |= S_ISGID;
814 } else {
815 inode->i_gid = current_fsgid();
816 }
817
818 /*
819 * If the group ID of the new file does not match the effective group
820 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
821 * (and only if the irix_sgid_inherit compatibility variable is set).
822 */
823 if (irix_sgid_inherit &&
824 (inode->i_mode & S_ISGID) && !in_group_p(inode->i_gid))
825 inode->i_mode &= ~S_ISGID;
826
827 ip->i_d.di_size = 0;
828 ip->i_df.if_nextents = 0;
829 ASSERT(ip->i_d.di_nblocks == 0);
830
831 tv = current_time(inode);
832 inode->i_mtime = tv;
833 inode->i_atime = tv;
834 inode->i_ctime = tv;
835
836 ip->i_d.di_extsize = 0;
837 ip->i_d.di_dmevmask = 0;
838 ip->i_d.di_dmstate = 0;
839 ip->i_d.di_flags = 0;
840
841 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
842 inode_set_iversion(inode, 1);
843 ip->i_d.di_flags2 = 0;
844 ip->i_d.di_cowextsize = 0;
845 ip->i_d.di_crtime = tv;
846 }
847
848 flags = XFS_ILOG_CORE;
849 switch (mode & S_IFMT) {
850 case S_IFIFO:
851 case S_IFCHR:
852 case S_IFBLK:
853 case S_IFSOCK:
854 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
855 ip->i_df.if_flags = 0;
856 flags |= XFS_ILOG_DEV;
857 break;
858 case S_IFREG:
859 case S_IFDIR:
860 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
861 uint di_flags = 0;
862
863 if (S_ISDIR(mode)) {
864 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
865 di_flags |= XFS_DIFLAG_RTINHERIT;
866 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
867 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
868 ip->i_d.di_extsize = pip->i_d.di_extsize;
869 }
870 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
871 di_flags |= XFS_DIFLAG_PROJINHERIT;
872 } else if (S_ISREG(mode)) {
873 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
874 di_flags |= XFS_DIFLAG_REALTIME;
875 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
876 di_flags |= XFS_DIFLAG_EXTSIZE;
877 ip->i_d.di_extsize = pip->i_d.di_extsize;
878 }
879 }
880 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
881 xfs_inherit_noatime)
882 di_flags |= XFS_DIFLAG_NOATIME;
883 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
884 xfs_inherit_nodump)
885 di_flags |= XFS_DIFLAG_NODUMP;
886 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
887 xfs_inherit_sync)
888 di_flags |= XFS_DIFLAG_SYNC;
889 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
890 xfs_inherit_nosymlinks)
891 di_flags |= XFS_DIFLAG_NOSYMLINKS;
892 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
893 xfs_inherit_nodefrag)
894 di_flags |= XFS_DIFLAG_NODEFRAG;
895 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
896 di_flags |= XFS_DIFLAG_FILESTREAM;
897
898 ip->i_d.di_flags |= di_flags;
899 }
900 if (pip && (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY)) {
901 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) {
902 ip->i_d.di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
903 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize;
904 }
905 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
906 ip->i_d.di_flags2 |= XFS_DIFLAG2_DAX;
907 }
908 /* FALLTHROUGH */
909 case S_IFLNK:
910 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
911 ip->i_df.if_flags = XFS_IFEXTENTS;
912 ip->i_df.if_bytes = 0;
913 ip->i_df.if_u1.if_root = NULL;
914 break;
915 default:
916 ASSERT(0);
917 }
918
919 /*
920 * Log the new values stuffed into the inode.
921 */
922 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
923 xfs_trans_log_inode(tp, ip, flags);
924
925 /* now that we have an i_mode we can setup the inode structure */
926 xfs_setup_inode(ip);
927
928 *ipp = ip;
929 return 0;
930}
931
932/*
933 * Allocates a new inode from disk and return a pointer to the
934 * incore copy. This routine will internally commit the current
935 * transaction and allocate a new one if the Space Manager needed
936 * to do an allocation to replenish the inode free-list.
937 *
938 * This routine is designed to be called from xfs_create and
939 * xfs_create_dir.
940 *
941 */
942int
943xfs_dir_ialloc(
944 xfs_trans_t **tpp, /* input: current transaction;
945 output: may be a new transaction. */
946 xfs_inode_t *dp, /* directory within whose allocate
947 the inode. */
948 umode_t mode,
949 xfs_nlink_t nlink,
950 dev_t rdev,
951 prid_t prid, /* project id */
952 xfs_inode_t **ipp) /* pointer to inode; it will be
953 locked. */
954{
955 xfs_trans_t *tp;
956 xfs_inode_t *ip;
957 xfs_buf_t *ialloc_context = NULL;
958 int code;
959 void *dqinfo;
960 uint tflags;
961
962 tp = *tpp;
963 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
964
965 /*
966 * xfs_ialloc will return a pointer to an incore inode if
967 * the Space Manager has an available inode on the free
968 * list. Otherwise, it will do an allocation and replenish
969 * the freelist. Since we can only do one allocation per
970 * transaction without deadlocks, we will need to commit the
971 * current transaction and start a new one. We will then
972 * need to call xfs_ialloc again to get the inode.
973 *
974 * If xfs_ialloc did an allocation to replenish the freelist,
975 * it returns the bp containing the head of the freelist as
976 * ialloc_context. We will hold a lock on it across the
977 * transaction commit so that no other process can steal
978 * the inode(s) that we've just allocated.
979 */
980 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context,
981 &ip);
982
983 /*
984 * Return an error if we were unable to allocate a new inode.
985 * This should only happen if we run out of space on disk or
986 * encounter a disk error.
987 */
988 if (code) {
989 *ipp = NULL;
990 return code;
991 }
992 if (!ialloc_context && !ip) {
993 *ipp = NULL;
994 return -ENOSPC;
995 }
996
997 /*
998 * If the AGI buffer is non-NULL, then we were unable to get an
999 * inode in one operation. We need to commit the current
1000 * transaction and call xfs_ialloc() again. It is guaranteed
1001 * to succeed the second time.
1002 */
1003 if (ialloc_context) {
1004 /*
1005 * Normally, xfs_trans_commit releases all the locks.
1006 * We call bhold to hang on to the ialloc_context across
1007 * the commit. Holding this buffer prevents any other
1008 * processes from doing any allocations in this
1009 * allocation group.
1010 */
1011 xfs_trans_bhold(tp, ialloc_context);
1012
1013 /*
1014 * We want the quota changes to be associated with the next
1015 * transaction, NOT this one. So, detach the dqinfo from this
1016 * and attach it to the next transaction.
1017 */
1018 dqinfo = NULL;
1019 tflags = 0;
1020 if (tp->t_dqinfo) {
1021 dqinfo = (void *)tp->t_dqinfo;
1022 tp->t_dqinfo = NULL;
1023 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
1024 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
1025 }
1026
1027 code = xfs_trans_roll(&tp);
1028
1029 /*
1030 * Re-attach the quota info that we detached from prev trx.
1031 */
1032 if (dqinfo) {
1033 tp->t_dqinfo = dqinfo;
1034 tp->t_flags |= tflags;
1035 }
1036
1037 if (code) {
1038 xfs_buf_relse(ialloc_context);
1039 *tpp = tp;
1040 *ipp = NULL;
1041 return code;
1042 }
1043 xfs_trans_bjoin(tp, ialloc_context);
1044
1045 /*
1046 * Call ialloc again. Since we've locked out all
1047 * other allocations in this allocation group,
1048 * this call should always succeed.
1049 */
1050 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1051 &ialloc_context, &ip);
1052
1053 /*
1054 * If we get an error at this point, return to the caller
1055 * so that the current transaction can be aborted.
1056 */
1057 if (code) {
1058 *tpp = tp;
1059 *ipp = NULL;
1060 return code;
1061 }
1062 ASSERT(!ialloc_context && ip);
1063
1064 }
1065
1066 *ipp = ip;
1067 *tpp = tp;
1068
1069 return 0;
1070}
1071
1072/*
1073 * Decrement the link count on an inode & log the change. If this causes the
1074 * link count to go to zero, move the inode to AGI unlinked list so that it can
1075 * be freed when the last active reference goes away via xfs_inactive().
1076 */
1077static int /* error */
1078xfs_droplink(
1079 xfs_trans_t *tp,
1080 xfs_inode_t *ip)
1081{
1082 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1083
1084 drop_nlink(VFS_I(ip));
1085 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1086
1087 if (VFS_I(ip)->i_nlink)
1088 return 0;
1089
1090 return xfs_iunlink(tp, ip);
1091}
1092
1093/*
1094 * Increment the link count on an inode & log the change.
1095 */
1096static void
1097xfs_bumplink(
1098 xfs_trans_t *tp,
1099 xfs_inode_t *ip)
1100{
1101 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1102
1103 inc_nlink(VFS_I(ip));
1104 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1105}
1106
1107int
1108xfs_create(
1109 xfs_inode_t *dp,
1110 struct xfs_name *name,
1111 umode_t mode,
1112 dev_t rdev,
1113 xfs_inode_t **ipp)
1114{
1115 int is_dir = S_ISDIR(mode);
1116 struct xfs_mount *mp = dp->i_mount;
1117 struct xfs_inode *ip = NULL;
1118 struct xfs_trans *tp = NULL;
1119 int error;
1120 bool unlock_dp_on_error = false;
1121 prid_t prid;
1122 struct xfs_dquot *udqp = NULL;
1123 struct xfs_dquot *gdqp = NULL;
1124 struct xfs_dquot *pdqp = NULL;
1125 struct xfs_trans_res *tres;
1126 uint resblks;
1127
1128 trace_xfs_create(dp, name);
1129
1130 if (XFS_FORCED_SHUTDOWN(mp))
1131 return -EIO;
1132
1133 prid = xfs_get_initial_prid(dp);
1134
1135 /*
1136 * Make sure that we have allocated dquot(s) on disk.
1137 */
1138 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1139 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1140 &udqp, &gdqp, &pdqp);
1141 if (error)
1142 return error;
1143
1144 if (is_dir) {
1145 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1146 tres = &M_RES(mp)->tr_mkdir;
1147 } else {
1148 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1149 tres = &M_RES(mp)->tr_create;
1150 }
1151
1152 /*
1153 * Initially assume that the file does not exist and
1154 * reserve the resources for that case. If that is not
1155 * the case we'll drop the one we have and get a more
1156 * appropriate transaction later.
1157 */
1158 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1159 if (error == -ENOSPC) {
1160 /* flush outstanding delalloc blocks and retry */
1161 xfs_flush_inodes(mp);
1162 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1163 }
1164 if (error)
1165 goto out_release_inode;
1166
1167 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1168 unlock_dp_on_error = true;
1169
1170 /*
1171 * Reserve disk quota and the inode.
1172 */
1173 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1174 pdqp, resblks, 1, 0);
1175 if (error)
1176 goto out_trans_cancel;
1177
1178 /*
1179 * A newly created regular or special file just has one directory
1180 * entry pointing to them, but a directory also the "." entry
1181 * pointing to itself.
1182 */
1183 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip);
1184 if (error)
1185 goto out_trans_cancel;
1186
1187 /*
1188 * Now we join the directory inode to the transaction. We do not do it
1189 * earlier because xfs_dir_ialloc might commit the previous transaction
1190 * (and release all the locks). An error from here on will result in
1191 * the transaction cancel unlocking dp so don't do it explicitly in the
1192 * error path.
1193 */
1194 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1195 unlock_dp_on_error = false;
1196
1197 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1198 resblks - XFS_IALLOC_SPACE_RES(mp));
1199 if (error) {
1200 ASSERT(error != -ENOSPC);
1201 goto out_trans_cancel;
1202 }
1203 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1204 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1205
1206 if (is_dir) {
1207 error = xfs_dir_init(tp, ip, dp);
1208 if (error)
1209 goto out_trans_cancel;
1210
1211 xfs_bumplink(tp, dp);
1212 }
1213
1214 /*
1215 * If this is a synchronous mount, make sure that the
1216 * create transaction goes to disk before returning to
1217 * the user.
1218 */
1219 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1220 xfs_trans_set_sync(tp);
1221
1222 /*
1223 * Attach the dquot(s) to the inodes and modify them incore.
1224 * These ids of the inode couldn't have changed since the new
1225 * inode has been locked ever since it was created.
1226 */
1227 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1228
1229 error = xfs_trans_commit(tp);
1230 if (error)
1231 goto out_release_inode;
1232
1233 xfs_qm_dqrele(udqp);
1234 xfs_qm_dqrele(gdqp);
1235 xfs_qm_dqrele(pdqp);
1236
1237 *ipp = ip;
1238 return 0;
1239
1240 out_trans_cancel:
1241 xfs_trans_cancel(tp);
1242 out_release_inode:
1243 /*
1244 * Wait until after the current transaction is aborted to finish the
1245 * setup of the inode and release the inode. This prevents recursive
1246 * transactions and deadlocks from xfs_inactive.
1247 */
1248 if (ip) {
1249 xfs_finish_inode_setup(ip);
1250 xfs_irele(ip);
1251 }
1252
1253 xfs_qm_dqrele(udqp);
1254 xfs_qm_dqrele(gdqp);
1255 xfs_qm_dqrele(pdqp);
1256
1257 if (unlock_dp_on_error)
1258 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1259 return error;
1260}
1261
1262int
1263xfs_create_tmpfile(
1264 struct xfs_inode *dp,
1265 umode_t mode,
1266 struct xfs_inode **ipp)
1267{
1268 struct xfs_mount *mp = dp->i_mount;
1269 struct xfs_inode *ip = NULL;
1270 struct xfs_trans *tp = NULL;
1271 int error;
1272 prid_t prid;
1273 struct xfs_dquot *udqp = NULL;
1274 struct xfs_dquot *gdqp = NULL;
1275 struct xfs_dquot *pdqp = NULL;
1276 struct xfs_trans_res *tres;
1277 uint resblks;
1278
1279 if (XFS_FORCED_SHUTDOWN(mp))
1280 return -EIO;
1281
1282 prid = xfs_get_initial_prid(dp);
1283
1284 /*
1285 * Make sure that we have allocated dquot(s) on disk.
1286 */
1287 error = xfs_qm_vop_dqalloc(dp, current_fsuid(), current_fsgid(), prid,
1288 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1289 &udqp, &gdqp, &pdqp);
1290 if (error)
1291 return error;
1292
1293 resblks = XFS_IALLOC_SPACE_RES(mp);
1294 tres = &M_RES(mp)->tr_create_tmpfile;
1295
1296 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp);
1297 if (error)
1298 goto out_release_inode;
1299
1300 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1301 pdqp, resblks, 1, 0);
1302 if (error)
1303 goto out_trans_cancel;
1304
1305 error = xfs_dir_ialloc(&tp, dp, mode, 0, 0, prid, &ip);
1306 if (error)
1307 goto out_trans_cancel;
1308
1309 if (mp->m_flags & XFS_MOUNT_WSYNC)
1310 xfs_trans_set_sync(tp);
1311
1312 /*
1313 * Attach the dquot(s) to the inodes and modify them incore.
1314 * These ids of the inode couldn't have changed since the new
1315 * inode has been locked ever since it was created.
1316 */
1317 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1318
1319 error = xfs_iunlink(tp, ip);
1320 if (error)
1321 goto out_trans_cancel;
1322
1323 error = xfs_trans_commit(tp);
1324 if (error)
1325 goto out_release_inode;
1326
1327 xfs_qm_dqrele(udqp);
1328 xfs_qm_dqrele(gdqp);
1329 xfs_qm_dqrele(pdqp);
1330
1331 *ipp = ip;
1332 return 0;
1333
1334 out_trans_cancel:
1335 xfs_trans_cancel(tp);
1336 out_release_inode:
1337 /*
1338 * Wait until after the current transaction is aborted to finish the
1339 * setup of the inode and release the inode. This prevents recursive
1340 * transactions and deadlocks from xfs_inactive.
1341 */
1342 if (ip) {
1343 xfs_finish_inode_setup(ip);
1344 xfs_irele(ip);
1345 }
1346
1347 xfs_qm_dqrele(udqp);
1348 xfs_qm_dqrele(gdqp);
1349 xfs_qm_dqrele(pdqp);
1350
1351 return error;
1352}
1353
1354int
1355xfs_link(
1356 xfs_inode_t *tdp,
1357 xfs_inode_t *sip,
1358 struct xfs_name *target_name)
1359{
1360 xfs_mount_t *mp = tdp->i_mount;
1361 xfs_trans_t *tp;
1362 int error;
1363 int resblks;
1364
1365 trace_xfs_link(tdp, target_name);
1366
1367 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1368
1369 if (XFS_FORCED_SHUTDOWN(mp))
1370 return -EIO;
1371
1372 error = xfs_qm_dqattach(sip);
1373 if (error)
1374 goto std_return;
1375
1376 error = xfs_qm_dqattach(tdp);
1377 if (error)
1378 goto std_return;
1379
1380 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1381 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
1382 if (error == -ENOSPC) {
1383 resblks = 0;
1384 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
1385 }
1386 if (error)
1387 goto std_return;
1388
1389 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
1390
1391 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1392 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1393
1394 /*
1395 * If we are using project inheritance, we only allow hard link
1396 * creation in our tree when the project IDs are the same; else
1397 * the tree quota mechanism could be circumvented.
1398 */
1399 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1400 tdp->i_d.di_projid != sip->i_d.di_projid)) {
1401 error = -EXDEV;
1402 goto error_return;
1403 }
1404
1405 if (!resblks) {
1406 error = xfs_dir_canenter(tp, tdp, target_name);
1407 if (error)
1408 goto error_return;
1409 }
1410
1411 /*
1412 * Handle initial link state of O_TMPFILE inode
1413 */
1414 if (VFS_I(sip)->i_nlink == 0) {
1415 error = xfs_iunlink_remove(tp, sip);
1416 if (error)
1417 goto error_return;
1418 }
1419
1420 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1421 resblks);
1422 if (error)
1423 goto error_return;
1424 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1425 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1426
1427 xfs_bumplink(tp, sip);
1428
1429 /*
1430 * If this is a synchronous mount, make sure that the
1431 * link transaction goes to disk before returning to
1432 * the user.
1433 */
1434 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1435 xfs_trans_set_sync(tp);
1436
1437 return xfs_trans_commit(tp);
1438
1439 error_return:
1440 xfs_trans_cancel(tp);
1441 std_return:
1442 return error;
1443}
1444
1445/* Clear the reflink flag and the cowblocks tag if possible. */
1446static void
1447xfs_itruncate_clear_reflink_flags(
1448 struct xfs_inode *ip)
1449{
1450 struct xfs_ifork *dfork;
1451 struct xfs_ifork *cfork;
1452
1453 if (!xfs_is_reflink_inode(ip))
1454 return;
1455 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1456 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
1457 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1458 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1459 if (cfork->if_bytes == 0)
1460 xfs_inode_clear_cowblocks_tag(ip);
1461}
1462
1463/*
1464 * Free up the underlying blocks past new_size. The new size must be smaller
1465 * than the current size. This routine can be used both for the attribute and
1466 * data fork, and does not modify the inode size, which is left to the caller.
1467 *
1468 * The transaction passed to this routine must have made a permanent log
1469 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1470 * given transaction and start new ones, so make sure everything involved in
1471 * the transaction is tidy before calling here. Some transaction will be
1472 * returned to the caller to be committed. The incoming transaction must
1473 * already include the inode, and both inode locks must be held exclusively.
1474 * The inode must also be "held" within the transaction. On return the inode
1475 * will be "held" within the returned transaction. This routine does NOT
1476 * require any disk space to be reserved for it within the transaction.
1477 *
1478 * If we get an error, we must return with the inode locked and linked into the
1479 * current transaction. This keeps things simple for the higher level code,
1480 * because it always knows that the inode is locked and held in the transaction
1481 * that returns to it whether errors occur or not. We don't mark the inode
1482 * dirty on error so that transactions can be easily aborted if possible.
1483 */
1484int
1485xfs_itruncate_extents_flags(
1486 struct xfs_trans **tpp,
1487 struct xfs_inode *ip,
1488 int whichfork,
1489 xfs_fsize_t new_size,
1490 int flags)
1491{
1492 struct xfs_mount *mp = ip->i_mount;
1493 struct xfs_trans *tp = *tpp;
1494 xfs_fileoff_t first_unmap_block;
1495 xfs_filblks_t unmap_len;
1496 int error = 0;
1497
1498 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1499 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1500 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1501 ASSERT(new_size <= XFS_ISIZE(ip));
1502 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1503 ASSERT(ip->i_itemp != NULL);
1504 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1505 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1506
1507 trace_xfs_itruncate_extents_start(ip, new_size);
1508
1509 flags |= xfs_bmapi_aflag(whichfork);
1510
1511 /*
1512 * Since it is possible for space to become allocated beyond
1513 * the end of the file (in a crash where the space is allocated
1514 * but the inode size is not yet updated), simply remove any
1515 * blocks which show up between the new EOF and the maximum
1516 * possible file size.
1517 *
1518 * We have to free all the blocks to the bmbt maximum offset, even if
1519 * the page cache can't scale that far.
1520 */
1521 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1522 if (first_unmap_block >= XFS_MAX_FILEOFF) {
1523 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1524 return 0;
1525 }
1526
1527 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1528 while (unmap_len > 0) {
1529 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1530 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1531 flags, XFS_ITRUNC_MAX_EXTENTS);
1532 if (error)
1533 goto out;
1534
1535 /*
1536 * Duplicate the transaction that has the permanent
1537 * reservation and commit the old transaction.
1538 */
1539 error = xfs_defer_finish(&tp);
1540 if (error)
1541 goto out;
1542
1543 error = xfs_trans_roll_inode(&tp, ip);
1544 if (error)
1545 goto out;
1546 }
1547
1548 if (whichfork == XFS_DATA_FORK) {
1549 /* Remove all pending CoW reservations. */
1550 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1551 first_unmap_block, XFS_MAX_FILEOFF, true);
1552 if (error)
1553 goto out;
1554
1555 xfs_itruncate_clear_reflink_flags(ip);
1556 }
1557
1558 /*
1559 * Always re-log the inode so that our permanent transaction can keep
1560 * on rolling it forward in the log.
1561 */
1562 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1563
1564 trace_xfs_itruncate_extents_end(ip, new_size);
1565
1566out:
1567 *tpp = tp;
1568 return error;
1569}
1570
1571int
1572xfs_release(
1573 xfs_inode_t *ip)
1574{
1575 xfs_mount_t *mp = ip->i_mount;
1576 int error;
1577
1578 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1579 return 0;
1580
1581 /* If this is a read-only mount, don't do this (would generate I/O) */
1582 if (mp->m_flags & XFS_MOUNT_RDONLY)
1583 return 0;
1584
1585 if (!XFS_FORCED_SHUTDOWN(mp)) {
1586 int truncated;
1587
1588 /*
1589 * If we previously truncated this file and removed old data
1590 * in the process, we want to initiate "early" writeout on
1591 * the last close. This is an attempt to combat the notorious
1592 * NULL files problem which is particularly noticeable from a
1593 * truncate down, buffered (re-)write (delalloc), followed by
1594 * a crash. What we are effectively doing here is
1595 * significantly reducing the time window where we'd otherwise
1596 * be exposed to that problem.
1597 */
1598 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1599 if (truncated) {
1600 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1601 if (ip->i_delayed_blks > 0) {
1602 error = filemap_flush(VFS_I(ip)->i_mapping);
1603 if (error)
1604 return error;
1605 }
1606 }
1607 }
1608
1609 if (VFS_I(ip)->i_nlink == 0)
1610 return 0;
1611
1612 if (xfs_can_free_eofblocks(ip, false)) {
1613
1614 /*
1615 * Check if the inode is being opened, written and closed
1616 * frequently and we have delayed allocation blocks outstanding
1617 * (e.g. streaming writes from the NFS server), truncating the
1618 * blocks past EOF will cause fragmentation to occur.
1619 *
1620 * In this case don't do the truncation, but we have to be
1621 * careful how we detect this case. Blocks beyond EOF show up as
1622 * i_delayed_blks even when the inode is clean, so we need to
1623 * truncate them away first before checking for a dirty release.
1624 * Hence on the first dirty close we will still remove the
1625 * speculative allocation, but after that we will leave it in
1626 * place.
1627 */
1628 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1629 return 0;
1630 /*
1631 * If we can't get the iolock just skip truncating the blocks
1632 * past EOF because we could deadlock with the mmap_lock
1633 * otherwise. We'll get another chance to drop them once the
1634 * last reference to the inode is dropped, so we'll never leak
1635 * blocks permanently.
1636 */
1637 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
1638 error = xfs_free_eofblocks(ip);
1639 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1640 if (error)
1641 return error;
1642 }
1643
1644 /* delalloc blocks after truncation means it really is dirty */
1645 if (ip->i_delayed_blks)
1646 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1647 }
1648 return 0;
1649}
1650
1651/*
1652 * xfs_inactive_truncate
1653 *
1654 * Called to perform a truncate when an inode becomes unlinked.
1655 */
1656STATIC int
1657xfs_inactive_truncate(
1658 struct xfs_inode *ip)
1659{
1660 struct xfs_mount *mp = ip->i_mount;
1661 struct xfs_trans *tp;
1662 int error;
1663
1664 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1665 if (error) {
1666 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1667 return error;
1668 }
1669 xfs_ilock(ip, XFS_ILOCK_EXCL);
1670 xfs_trans_ijoin(tp, ip, 0);
1671
1672 /*
1673 * Log the inode size first to prevent stale data exposure in the event
1674 * of a system crash before the truncate completes. See the related
1675 * comment in xfs_vn_setattr_size() for details.
1676 */
1677 ip->i_d.di_size = 0;
1678 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1679
1680 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1681 if (error)
1682 goto error_trans_cancel;
1683
1684 ASSERT(ip->i_df.if_nextents == 0);
1685
1686 error = xfs_trans_commit(tp);
1687 if (error)
1688 goto error_unlock;
1689
1690 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1691 return 0;
1692
1693error_trans_cancel:
1694 xfs_trans_cancel(tp);
1695error_unlock:
1696 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1697 return error;
1698}
1699
1700/*
1701 * xfs_inactive_ifree()
1702 *
1703 * Perform the inode free when an inode is unlinked.
1704 */
1705STATIC int
1706xfs_inactive_ifree(
1707 struct xfs_inode *ip)
1708{
1709 struct xfs_mount *mp = ip->i_mount;
1710 struct xfs_trans *tp;
1711 int error;
1712
1713 /*
1714 * We try to use a per-AG reservation for any block needed by the finobt
1715 * tree, but as the finobt feature predates the per-AG reservation
1716 * support a degraded file system might not have enough space for the
1717 * reservation at mount time. In that case try to dip into the reserved
1718 * pool and pray.
1719 *
1720 * Send a warning if the reservation does happen to fail, as the inode
1721 * now remains allocated and sits on the unlinked list until the fs is
1722 * repaired.
1723 */
1724 if (unlikely(mp->m_finobt_nores)) {
1725 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1726 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1727 &tp);
1728 } else {
1729 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1730 }
1731 if (error) {
1732 if (error == -ENOSPC) {
1733 xfs_warn_ratelimited(mp,
1734 "Failed to remove inode(s) from unlinked list. "
1735 "Please free space, unmount and run xfs_repair.");
1736 } else {
1737 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1738 }
1739 return error;
1740 }
1741
1742 /*
1743 * We do not hold the inode locked across the entire rolling transaction
1744 * here. We only need to hold it for the first transaction that
1745 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1746 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1747 * here breaks the relationship between cluster buffer invalidation and
1748 * stale inode invalidation on cluster buffer item journal commit
1749 * completion, and can result in leaving dirty stale inodes hanging
1750 * around in memory.
1751 *
1752 * We have no need for serialising this inode operation against other
1753 * operations - we freed the inode and hence reallocation is required
1754 * and that will serialise on reallocating the space the deferops need
1755 * to free. Hence we can unlock the inode on the first commit of
1756 * the transaction rather than roll it right through the deferops. This
1757 * avoids relogging the XFS_ISTALE inode.
1758 *
1759 * We check that xfs_ifree() hasn't grown an internal transaction roll
1760 * by asserting that the inode is still locked when it returns.
1761 */
1762 xfs_ilock(ip, XFS_ILOCK_EXCL);
1763 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1764
1765 error = xfs_ifree(tp, ip);
1766 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1767 if (error) {
1768 /*
1769 * If we fail to free the inode, shut down. The cancel
1770 * might do that, we need to make sure. Otherwise the
1771 * inode might be lost for a long time or forever.
1772 */
1773 if (!XFS_FORCED_SHUTDOWN(mp)) {
1774 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1775 __func__, error);
1776 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1777 }
1778 xfs_trans_cancel(tp);
1779 return error;
1780 }
1781
1782 /*
1783 * Credit the quota account(s). The inode is gone.
1784 */
1785 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1786
1787 /*
1788 * Just ignore errors at this point. There is nothing we can do except
1789 * to try to keep going. Make sure it's not a silent error.
1790 */
1791 error = xfs_trans_commit(tp);
1792 if (error)
1793 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1794 __func__, error);
1795
1796 return 0;
1797}
1798
1799/*
1800 * xfs_inactive
1801 *
1802 * This is called when the vnode reference count for the vnode
1803 * goes to zero. If the file has been unlinked, then it must
1804 * now be truncated. Also, we clear all of the read-ahead state
1805 * kept for the inode here since the file is now closed.
1806 */
1807void
1808xfs_inactive(
1809 xfs_inode_t *ip)
1810{
1811 struct xfs_mount *mp;
1812 int error;
1813 int truncate = 0;
1814
1815 /*
1816 * If the inode is already free, then there can be nothing
1817 * to clean up here.
1818 */
1819 if (VFS_I(ip)->i_mode == 0) {
1820 ASSERT(ip->i_df.if_broot_bytes == 0);
1821 return;
1822 }
1823
1824 mp = ip->i_mount;
1825 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1826
1827 /* If this is a read-only mount, don't do this (would generate I/O) */
1828 if (mp->m_flags & XFS_MOUNT_RDONLY)
1829 return;
1830
1831 /* Try to clean out the cow blocks if there are any. */
1832 if (xfs_inode_has_cow_data(ip))
1833 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1834
1835 if (VFS_I(ip)->i_nlink != 0) {
1836 /*
1837 * force is true because we are evicting an inode from the
1838 * cache. Post-eof blocks must be freed, lest we end up with
1839 * broken free space accounting.
1840 *
1841 * Note: don't bother with iolock here since lockdep complains
1842 * about acquiring it in reclaim context. We have the only
1843 * reference to the inode at this point anyways.
1844 */
1845 if (xfs_can_free_eofblocks(ip, true))
1846 xfs_free_eofblocks(ip);
1847
1848 return;
1849 }
1850
1851 if (S_ISREG(VFS_I(ip)->i_mode) &&
1852 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1853 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1854 truncate = 1;
1855
1856 error = xfs_qm_dqattach(ip);
1857 if (error)
1858 return;
1859
1860 if (S_ISLNK(VFS_I(ip)->i_mode))
1861 error = xfs_inactive_symlink(ip);
1862 else if (truncate)
1863 error = xfs_inactive_truncate(ip);
1864 if (error)
1865 return;
1866
1867 /*
1868 * If there are attributes associated with the file then blow them away
1869 * now. The code calls a routine that recursively deconstructs the
1870 * attribute fork. If also blows away the in-core attribute fork.
1871 */
1872 if (XFS_IFORK_Q(ip)) {
1873 error = xfs_attr_inactive(ip);
1874 if (error)
1875 return;
1876 }
1877
1878 ASSERT(!ip->i_afp);
1879 ASSERT(ip->i_d.di_forkoff == 0);
1880
1881 /*
1882 * Free the inode.
1883 */
1884 error = xfs_inactive_ifree(ip);
1885 if (error)
1886 return;
1887
1888 /*
1889 * Release the dquots held by inode, if any.
1890 */
1891 xfs_qm_dqdetach(ip);
1892}
1893
1894/*
1895 * In-Core Unlinked List Lookups
1896 * =============================
1897 *
1898 * Every inode is supposed to be reachable from some other piece of metadata
1899 * with the exception of the root directory. Inodes with a connection to a
1900 * file descriptor but not linked from anywhere in the on-disk directory tree
1901 * are collectively known as unlinked inodes, though the filesystem itself
1902 * maintains links to these inodes so that on-disk metadata are consistent.
1903 *
1904 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1905 * header contains a number of buckets that point to an inode, and each inode
1906 * record has a pointer to the next inode in the hash chain. This
1907 * singly-linked list causes scaling problems in the iunlink remove function
1908 * because we must walk that list to find the inode that points to the inode
1909 * being removed from the unlinked hash bucket list.
1910 *
1911 * What if we modelled the unlinked list as a collection of records capturing
1912 * "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
1913 * have a fast way to look up unlinked list predecessors, which avoids the
1914 * slow list walk. That's exactly what we do here (in-core) with a per-AG
1915 * rhashtable.
1916 *
1917 * Because this is a backref cache, we ignore operational failures since the
1918 * iunlink code can fall back to the slow bucket walk. The only errors that
1919 * should bubble out are for obviously incorrect situations.
1920 *
1921 * All users of the backref cache MUST hold the AGI buffer lock to serialize
1922 * access or have otherwise provided for concurrency control.
1923 */
1924
1925/* Capture a "X.next_unlinked = Y" relationship. */
1926struct xfs_iunlink {
1927 struct rhash_head iu_rhash_head;
1928 xfs_agino_t iu_agino; /* X */
1929 xfs_agino_t iu_next_unlinked; /* Y */
1930};
1931
1932/* Unlinked list predecessor lookup hashtable construction */
1933static int
1934xfs_iunlink_obj_cmpfn(
1935 struct rhashtable_compare_arg *arg,
1936 const void *obj)
1937{
1938 const xfs_agino_t *key = arg->key;
1939 const struct xfs_iunlink *iu = obj;
1940
1941 if (iu->iu_next_unlinked != *key)
1942 return 1;
1943 return 0;
1944}
1945
1946static const struct rhashtable_params xfs_iunlink_hash_params = {
1947 .min_size = XFS_AGI_UNLINKED_BUCKETS,
1948 .key_len = sizeof(xfs_agino_t),
1949 .key_offset = offsetof(struct xfs_iunlink,
1950 iu_next_unlinked),
1951 .head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
1952 .automatic_shrinking = true,
1953 .obj_cmpfn = xfs_iunlink_obj_cmpfn,
1954};
1955
1956/*
1957 * Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
1958 * relation is found.
1959 */
1960static xfs_agino_t
1961xfs_iunlink_lookup_backref(
1962 struct xfs_perag *pag,
1963 xfs_agino_t agino)
1964{
1965 struct xfs_iunlink *iu;
1966
1967 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
1968 xfs_iunlink_hash_params);
1969 return iu ? iu->iu_agino : NULLAGINO;
1970}
1971
1972/*
1973 * Take ownership of an iunlink cache entry and insert it into the hash table.
1974 * If successful, the entry will be owned by the cache; if not, it is freed.
1975 * Either way, the caller does not own @iu after this call.
1976 */
1977static int
1978xfs_iunlink_insert_backref(
1979 struct xfs_perag *pag,
1980 struct xfs_iunlink *iu)
1981{
1982 int error;
1983
1984 error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
1985 &iu->iu_rhash_head, xfs_iunlink_hash_params);
1986 /*
1987 * Fail loudly if there already was an entry because that's a sign of
1988 * corruption of in-memory data. Also fail loudly if we see an error
1989 * code we didn't anticipate from the rhashtable code. Currently we
1990 * only anticipate ENOMEM.
1991 */
1992 if (error) {
1993 WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
1994 kmem_free(iu);
1995 }
1996 /*
1997 * Absorb any runtime errors that aren't a result of corruption because
1998 * this is a cache and we can always fall back to bucket list scanning.
1999 */
2000 if (error != 0 && error != -EEXIST)
2001 error = 0;
2002 return error;
2003}
2004
2005/* Remember that @prev_agino.next_unlinked = @this_agino. */
2006static int
2007xfs_iunlink_add_backref(
2008 struct xfs_perag *pag,
2009 xfs_agino_t prev_agino,
2010 xfs_agino_t this_agino)
2011{
2012 struct xfs_iunlink *iu;
2013
2014 if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
2015 return 0;
2016
2017 iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
2018 iu->iu_agino = prev_agino;
2019 iu->iu_next_unlinked = this_agino;
2020
2021 return xfs_iunlink_insert_backref(pag, iu);
2022}
2023
2024/*
2025 * Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
2026 * If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
2027 * wasn't any such entry then we don't bother.
2028 */
2029static int
2030xfs_iunlink_change_backref(
2031 struct xfs_perag *pag,
2032 xfs_agino_t agino,
2033 xfs_agino_t next_unlinked)
2034{
2035 struct xfs_iunlink *iu;
2036 int error;
2037
2038 /* Look up the old entry; if there wasn't one then exit. */
2039 iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
2040 xfs_iunlink_hash_params);
2041 if (!iu)
2042 return 0;
2043
2044 /*
2045 * Remove the entry. This shouldn't ever return an error, but if we
2046 * couldn't remove the old entry we don't want to add it again to the
2047 * hash table, and if the entry disappeared on us then someone's
2048 * violated the locking rules and we need to fail loudly. Either way
2049 * we cannot remove the inode because internal state is or would have
2050 * been corrupt.
2051 */
2052 error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
2053 &iu->iu_rhash_head, xfs_iunlink_hash_params);
2054 if (error)
2055 return error;
2056
2057 /* If there is no new next entry just free our item and return. */
2058 if (next_unlinked == NULLAGINO) {
2059 kmem_free(iu);
2060 return 0;
2061 }
2062
2063 /* Update the entry and re-add it to the hash table. */
2064 iu->iu_next_unlinked = next_unlinked;
2065 return xfs_iunlink_insert_backref(pag, iu);
2066}
2067
2068/* Set up the in-core predecessor structures. */
2069int
2070xfs_iunlink_init(
2071 struct xfs_perag *pag)
2072{
2073 return rhashtable_init(&pag->pagi_unlinked_hash,
2074 &xfs_iunlink_hash_params);
2075}
2076
2077/* Free the in-core predecessor structures. */
2078static void
2079xfs_iunlink_free_item(
2080 void *ptr,
2081 void *arg)
2082{
2083 struct xfs_iunlink *iu = ptr;
2084 bool *freed_anything = arg;
2085
2086 *freed_anything = true;
2087 kmem_free(iu);
2088}
2089
2090void
2091xfs_iunlink_destroy(
2092 struct xfs_perag *pag)
2093{
2094 bool freed_anything = false;
2095
2096 rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
2097 xfs_iunlink_free_item, &freed_anything);
2098
2099 ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
2100}
2101
2102/*
2103 * Point the AGI unlinked bucket at an inode and log the results. The caller
2104 * is responsible for validating the old value.
2105 */
2106STATIC int
2107xfs_iunlink_update_bucket(
2108 struct xfs_trans *tp,
2109 xfs_agnumber_t agno,
2110 struct xfs_buf *agibp,
2111 unsigned int bucket_index,
2112 xfs_agino_t new_agino)
2113{
2114 struct xfs_agi *agi = agibp->b_addr;
2115 xfs_agino_t old_value;
2116 int offset;
2117
2118 ASSERT(xfs_verify_agino_or_null(tp->t_mountp, agno, new_agino));
2119
2120 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2121 trace_xfs_iunlink_update_bucket(tp->t_mountp, agno, bucket_index,
2122 old_value, new_agino);
2123
2124 /*
2125 * We should never find the head of the list already set to the value
2126 * passed in because either we're adding or removing ourselves from the
2127 * head of the list.
2128 */
2129 if (old_value == new_agino) {
2130 xfs_buf_mark_corrupt(agibp);
2131 return -EFSCORRUPTED;
2132 }
2133
2134 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2135 offset = offsetof(struct xfs_agi, agi_unlinked) +
2136 (sizeof(xfs_agino_t) * bucket_index);
2137 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2138 return 0;
2139}
2140
2141/* Set an on-disk inode's next_unlinked pointer. */
2142STATIC void
2143xfs_iunlink_update_dinode(
2144 struct xfs_trans *tp,
2145 xfs_agnumber_t agno,
2146 xfs_agino_t agino,
2147 struct xfs_buf *ibp,
2148 struct xfs_dinode *dip,
2149 struct xfs_imap *imap,
2150 xfs_agino_t next_agino)
2151{
2152 struct xfs_mount *mp = tp->t_mountp;
2153 int offset;
2154
2155 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2156
2157 trace_xfs_iunlink_update_dinode(mp, agno, agino,
2158 be32_to_cpu(dip->di_next_unlinked), next_agino);
2159
2160 dip->di_next_unlinked = cpu_to_be32(next_agino);
2161 offset = imap->im_boffset +
2162 offsetof(struct xfs_dinode, di_next_unlinked);
2163
2164 /* need to recalc the inode CRC if appropriate */
2165 xfs_dinode_calc_crc(mp, dip);
2166 xfs_trans_inode_buf(tp, ibp);
2167 xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
2168}
2169
2170/* Set an in-core inode's unlinked pointer and return the old value. */
2171STATIC int
2172xfs_iunlink_update_inode(
2173 struct xfs_trans *tp,
2174 struct xfs_inode *ip,
2175 xfs_agnumber_t agno,
2176 xfs_agino_t next_agino,
2177 xfs_agino_t *old_next_agino)
2178{
2179 struct xfs_mount *mp = tp->t_mountp;
2180 struct xfs_dinode *dip;
2181 struct xfs_buf *ibp;
2182 xfs_agino_t old_value;
2183 int error;
2184
2185 ASSERT(xfs_verify_agino_or_null(mp, agno, next_agino));
2186
2187 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 0);
2188 if (error)
2189 return error;
2190
2191 /* Make sure the old pointer isn't garbage. */
2192 old_value = be32_to_cpu(dip->di_next_unlinked);
2193 if (!xfs_verify_agino_or_null(mp, agno, old_value)) {
2194 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
2195 sizeof(*dip), __this_address);
2196 error = -EFSCORRUPTED;
2197 goto out;
2198 }
2199
2200 /*
2201 * Since we're updating a linked list, we should never find that the
2202 * current pointer is the same as the new value, unless we're
2203 * terminating the list.
2204 */
2205 *old_next_agino = old_value;
2206 if (old_value == next_agino) {
2207 if (next_agino != NULLAGINO) {
2208 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
2209 dip, sizeof(*dip), __this_address);
2210 error = -EFSCORRUPTED;
2211 }
2212 goto out;
2213 }
2214
2215 /* Ok, update the new pointer. */
2216 xfs_iunlink_update_dinode(tp, agno, XFS_INO_TO_AGINO(mp, ip->i_ino),
2217 ibp, dip, &ip->i_imap, next_agino);
2218 return 0;
2219out:
2220 xfs_trans_brelse(tp, ibp);
2221 return error;
2222}
2223
2224/*
2225 * This is called when the inode's link count has gone to 0 or we are creating
2226 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2227 *
2228 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2229 * list when the inode is freed.
2230 */
2231STATIC int
2232xfs_iunlink(
2233 struct xfs_trans *tp,
2234 struct xfs_inode *ip)
2235{
2236 struct xfs_mount *mp = tp->t_mountp;
2237 struct xfs_agi *agi;
2238 struct xfs_buf *agibp;
2239 xfs_agino_t next_agino;
2240 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2241 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2242 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2243 int error;
2244
2245 ASSERT(VFS_I(ip)->i_nlink == 0);
2246 ASSERT(VFS_I(ip)->i_mode != 0);
2247 trace_xfs_iunlink(ip);
2248
2249 /* Get the agi buffer first. It ensures lock ordering on the list. */
2250 error = xfs_read_agi(mp, tp, agno, &agibp);
2251 if (error)
2252 return error;
2253 agi = agibp->b_addr;
2254
2255 /*
2256 * Get the index into the agi hash table for the list this inode will
2257 * go on. Make sure the pointer isn't garbage and that this inode
2258 * isn't already on the list.
2259 */
2260 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2261 if (next_agino == agino ||
2262 !xfs_verify_agino_or_null(mp, agno, next_agino)) {
2263 xfs_buf_mark_corrupt(agibp);
2264 return -EFSCORRUPTED;
2265 }
2266
2267 if (next_agino != NULLAGINO) {
2268 xfs_agino_t old_agino;
2269
2270 /*
2271 * There is already another inode in the bucket, so point this
2272 * inode to the current head of the list.
2273 */
2274 error = xfs_iunlink_update_inode(tp, ip, agno, next_agino,
2275 &old_agino);
2276 if (error)
2277 return error;
2278 ASSERT(old_agino == NULLAGINO);
2279
2280 /*
2281 * agino has been unlinked, add a backref from the next inode
2282 * back to agino.
2283 */
2284 error = xfs_iunlink_add_backref(agibp->b_pag, agino, next_agino);
2285 if (error)
2286 return error;
2287 }
2288
2289 /* Point the head of the list to point to this inode. */
2290 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index, agino);
2291}
2292
2293/* Return the imap, dinode pointer, and buffer for an inode. */
2294STATIC int
2295xfs_iunlink_map_ino(
2296 struct xfs_trans *tp,
2297 xfs_agnumber_t agno,
2298 xfs_agino_t agino,
2299 struct xfs_imap *imap,
2300 struct xfs_dinode **dipp,
2301 struct xfs_buf **bpp)
2302{
2303 struct xfs_mount *mp = tp->t_mountp;
2304 int error;
2305
2306 imap->im_blkno = 0;
2307 error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
2308 if (error) {
2309 xfs_warn(mp, "%s: xfs_imap returned error %d.",
2310 __func__, error);
2311 return error;
2312 }
2313
2314 error = xfs_imap_to_bp(mp, tp, imap, dipp, bpp, 0);
2315 if (error) {
2316 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2317 __func__, error);
2318 return error;
2319 }
2320
2321 return 0;
2322}
2323
2324/*
2325 * Walk the unlinked chain from @head_agino until we find the inode that
2326 * points to @target_agino. Return the inode number, map, dinode pointer,
2327 * and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
2328 *
2329 * @tp, @pag, @head_agino, and @target_agino are input parameters.
2330 * @agino, @imap, @dipp, and @bpp are all output parameters.
2331 *
2332 * Do not call this function if @target_agino is the head of the list.
2333 */
2334STATIC int
2335xfs_iunlink_map_prev(
2336 struct xfs_trans *tp,
2337 xfs_agnumber_t agno,
2338 xfs_agino_t head_agino,
2339 xfs_agino_t target_agino,
2340 xfs_agino_t *agino,
2341 struct xfs_imap *imap,
2342 struct xfs_dinode **dipp,
2343 struct xfs_buf **bpp,
2344 struct xfs_perag *pag)
2345{
2346 struct xfs_mount *mp = tp->t_mountp;
2347 xfs_agino_t next_agino;
2348 int error;
2349
2350 ASSERT(head_agino != target_agino);
2351 *bpp = NULL;
2352
2353 /* See if our backref cache can find it faster. */
2354 *agino = xfs_iunlink_lookup_backref(pag, target_agino);
2355 if (*agino != NULLAGINO) {
2356 error = xfs_iunlink_map_ino(tp, agno, *agino, imap, dipp, bpp);
2357 if (error)
2358 return error;
2359
2360 if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
2361 return 0;
2362
2363 /*
2364 * If we get here the cache contents were corrupt, so drop the
2365 * buffer and fall back to walking the bucket list.
2366 */
2367 xfs_trans_brelse(tp, *bpp);
2368 *bpp = NULL;
2369 WARN_ON_ONCE(1);
2370 }
2371
2372 trace_xfs_iunlink_map_prev_fallback(mp, agno);
2373
2374 /* Otherwise, walk the entire bucket until we find it. */
2375 next_agino = head_agino;
2376 while (next_agino != target_agino) {
2377 xfs_agino_t unlinked_agino;
2378
2379 if (*bpp)
2380 xfs_trans_brelse(tp, *bpp);
2381
2382 *agino = next_agino;
2383 error = xfs_iunlink_map_ino(tp, agno, next_agino, imap, dipp,
2384 bpp);
2385 if (error)
2386 return error;
2387
2388 unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
2389 /*
2390 * Make sure this pointer is valid and isn't an obvious
2391 * infinite loop.
2392 */
2393 if (!xfs_verify_agino(mp, agno, unlinked_agino) ||
2394 next_agino == unlinked_agino) {
2395 XFS_CORRUPTION_ERROR(__func__,
2396 XFS_ERRLEVEL_LOW, mp,
2397 *dipp, sizeof(**dipp));
2398 error = -EFSCORRUPTED;
2399 return error;
2400 }
2401 next_agino = unlinked_agino;
2402 }
2403
2404 return 0;
2405}
2406
2407/*
2408 * Pull the on-disk inode from the AGI unlinked list.
2409 */
2410STATIC int
2411xfs_iunlink_remove(
2412 struct xfs_trans *tp,
2413 struct xfs_inode *ip)
2414{
2415 struct xfs_mount *mp = tp->t_mountp;
2416 struct xfs_agi *agi;
2417 struct xfs_buf *agibp;
2418 struct xfs_buf *last_ibp;
2419 struct xfs_dinode *last_dip = NULL;
2420 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
2421 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2422 xfs_agino_t next_agino;
2423 xfs_agino_t head_agino;
2424 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2425 int error;
2426
2427 trace_xfs_iunlink_remove(ip);
2428
2429 /* Get the agi buffer first. It ensures lock ordering on the list. */
2430 error = xfs_read_agi(mp, tp, agno, &agibp);
2431 if (error)
2432 return error;
2433 agi = agibp->b_addr;
2434
2435 /*
2436 * Get the index into the agi hash table for the list this inode will
2437 * go on. Make sure the head pointer isn't garbage.
2438 */
2439 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2440 if (!xfs_verify_agino(mp, agno, head_agino)) {
2441 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2442 agi, sizeof(*agi));
2443 return -EFSCORRUPTED;
2444 }
2445
2446 /*
2447 * Set our inode's next_unlinked pointer to NULL and then return
2448 * the old pointer value so that we can update whatever was previous
2449 * to us in the list to point to whatever was next in the list.
2450 */
2451 error = xfs_iunlink_update_inode(tp, ip, agno, NULLAGINO, &next_agino);
2452 if (error)
2453 return error;
2454
2455 /*
2456 * If there was a backref pointing from the next inode back to this
2457 * one, remove it because we've removed this inode from the list.
2458 *
2459 * Later, if this inode was in the middle of the list we'll update
2460 * this inode's backref to point from the next inode.
2461 */
2462 if (next_agino != NULLAGINO) {
2463 error = xfs_iunlink_change_backref(agibp->b_pag, next_agino,
2464 NULLAGINO);
2465 if (error)
2466 return error;
2467 }
2468
2469 if (head_agino != agino) {
2470 struct xfs_imap imap;
2471 xfs_agino_t prev_agino;
2472
2473 /* We need to search the list for the inode being freed. */
2474 error = xfs_iunlink_map_prev(tp, agno, head_agino, agino,
2475 &prev_agino, &imap, &last_dip, &last_ibp,
2476 agibp->b_pag);
2477 if (error)
2478 return error;
2479
2480 /* Point the previous inode on the list to the next inode. */
2481 xfs_iunlink_update_dinode(tp, agno, prev_agino, last_ibp,
2482 last_dip, &imap, next_agino);
2483
2484 /*
2485 * Now we deal with the backref for this inode. If this inode
2486 * pointed at a real inode, change the backref that pointed to
2487 * us to point to our old next. If this inode was the end of
2488 * the list, delete the backref that pointed to us. Note that
2489 * change_backref takes care of deleting the backref if
2490 * next_agino is NULLAGINO.
2491 */
2492 return xfs_iunlink_change_backref(agibp->b_pag, agino,
2493 next_agino);
2494 }
2495
2496 /* Point the head of the list to the next unlinked inode. */
2497 return xfs_iunlink_update_bucket(tp, agno, agibp, bucket_index,
2498 next_agino);
2499}
2500
2501/*
2502 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2503 * mark it stale. We should only find clean inodes in this lookup that aren't
2504 * already stale.
2505 */
2506static void
2507xfs_ifree_mark_inode_stale(
2508 struct xfs_buf *bp,
2509 struct xfs_inode *free_ip,
2510 xfs_ino_t inum)
2511{
2512 struct xfs_mount *mp = bp->b_mount;
2513 struct xfs_perag *pag = bp->b_pag;
2514 struct xfs_inode_log_item *iip;
2515 struct xfs_inode *ip;
2516
2517retry:
2518 rcu_read_lock();
2519 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2520
2521 /* Inode not in memory, nothing to do */
2522 if (!ip) {
2523 rcu_read_unlock();
2524 return;
2525 }
2526
2527 /*
2528 * because this is an RCU protected lookup, we could find a recently
2529 * freed or even reallocated inode during the lookup. We need to check
2530 * under the i_flags_lock for a valid inode here. Skip it if it is not
2531 * valid, the wrong inode or stale.
2532 */
2533 spin_lock(&ip->i_flags_lock);
2534 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE)) {
2535 spin_unlock(&ip->i_flags_lock);
2536 rcu_read_unlock();
2537 return;
2538 }
2539
2540 /*
2541 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2542 * other inodes that we did not find in the list attached to the buffer
2543 * and are not already marked stale. If we can't lock it, back off and
2544 * retry.
2545 */
2546 if (ip != free_ip) {
2547 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2548 spin_unlock(&ip->i_flags_lock);
2549 rcu_read_unlock();
2550 delay(1);
2551 goto retry;
2552 }
2553 }
2554 ip->i_flags |= XFS_ISTALE;
2555 spin_unlock(&ip->i_flags_lock);
2556 rcu_read_unlock();
2557
2558 /*
2559 * If we can't get the flush lock, the inode is already attached. All
2560 * we needed to do here is mark the inode stale so buffer IO completion
2561 * will remove it from the AIL.
2562 */
2563 iip = ip->i_itemp;
2564 if (!xfs_iflock_nowait(ip)) {
2565 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2566 ASSERT(iip->ili_last_fields);
2567 goto out_iunlock;
2568 }
2569
2570 /*
2571 * Inodes not attached to the buffer can be released immediately.
2572 * Everything else has to go through xfs_iflush_abort() on journal
2573 * commit as the flock synchronises removal of the inode from the
2574 * cluster buffer against inode reclaim.
2575 */
2576 if (!iip || list_empty(&iip->ili_item.li_bio_list)) {
2577 xfs_ifunlock(ip);
2578 goto out_iunlock;
2579 }
2580
2581 /* we have a dirty inode in memory that has not yet been flushed. */
2582 spin_lock(&iip->ili_lock);
2583 iip->ili_last_fields = iip->ili_fields;
2584 iip->ili_fields = 0;
2585 iip->ili_fsync_fields = 0;
2586 spin_unlock(&iip->ili_lock);
2587 ASSERT(iip->ili_last_fields);
2588
2589out_iunlock:
2590 if (ip != free_ip)
2591 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2592}
2593
2594/*
2595 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2596 * inodes that are in memory - they all must be marked stale and attached to
2597 * the cluster buffer.
2598 */
2599STATIC int
2600xfs_ifree_cluster(
2601 struct xfs_inode *free_ip,
2602 struct xfs_trans *tp,
2603 struct xfs_icluster *xic)
2604{
2605 struct xfs_mount *mp = free_ip->i_mount;
2606 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2607 struct xfs_buf *bp;
2608 xfs_daddr_t blkno;
2609 xfs_ino_t inum = xic->first_ino;
2610 int nbufs;
2611 int i, j;
2612 int ioffset;
2613 int error;
2614
2615 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2616
2617 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2618 /*
2619 * The allocation bitmap tells us which inodes of the chunk were
2620 * physically allocated. Skip the cluster if an inode falls into
2621 * a sparse region.
2622 */
2623 ioffset = inum - xic->first_ino;
2624 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2625 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2626 continue;
2627 }
2628
2629 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2630 XFS_INO_TO_AGBNO(mp, inum));
2631
2632 /*
2633 * We obtain and lock the backing buffer first in the process
2634 * here, as we have to ensure that any dirty inode that we
2635 * can't get the flush lock on is attached to the buffer.
2636 * If we scan the in-memory inodes first, then buffer IO can
2637 * complete before we get a lock on it, and hence we may fail
2638 * to mark all the active inodes on the buffer stale.
2639 */
2640 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2641 mp->m_bsize * igeo->blocks_per_cluster,
2642 XBF_UNMAPPED, &bp);
2643 if (error)
2644 return error;
2645
2646 /*
2647 * This buffer may not have been correctly initialised as we
2648 * didn't read it from disk. That's not important because we are
2649 * only using to mark the buffer as stale in the log, and to
2650 * attach stale cached inodes on it. That means it will never be
2651 * dispatched for IO. If it is, we want to know about it, and we
2652 * want it to fail. We can acheive this by adding a write
2653 * verifier to the buffer.
2654 */
2655 bp->b_ops = &xfs_inode_buf_ops;
2656
2657 /*
2658 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2659 * too. This requires lookups, and will skip inodes that we've
2660 * already marked XFS_ISTALE.
2661 */
2662 for (i = 0; i < igeo->inodes_per_cluster; i++)
2663 xfs_ifree_mark_inode_stale(bp, free_ip, inum + i);
2664
2665 xfs_trans_stale_inode_buf(tp, bp);
2666 xfs_trans_binval(tp, bp);
2667 }
2668 return 0;
2669}
2670
2671/*
2672 * This is called to return an inode to the inode free list.
2673 * The inode should already be truncated to 0 length and have
2674 * no pages associated with it. This routine also assumes that
2675 * the inode is already a part of the transaction.
2676 *
2677 * The on-disk copy of the inode will have been added to the list
2678 * of unlinked inodes in the AGI. We need to remove the inode from
2679 * that list atomically with respect to freeing it here.
2680 */
2681int
2682xfs_ifree(
2683 struct xfs_trans *tp,
2684 struct xfs_inode *ip)
2685{
2686 int error;
2687 struct xfs_icluster xic = { 0 };
2688 struct xfs_inode_log_item *iip = ip->i_itemp;
2689
2690 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2691 ASSERT(VFS_I(ip)->i_nlink == 0);
2692 ASSERT(ip->i_df.if_nextents == 0);
2693 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2694 ASSERT(ip->i_d.di_nblocks == 0);
2695
2696 /*
2697 * Pull the on-disk inode from the AGI unlinked list.
2698 */
2699 error = xfs_iunlink_remove(tp, ip);
2700 if (error)
2701 return error;
2702
2703 error = xfs_difree(tp, ip->i_ino, &xic);
2704 if (error)
2705 return error;
2706
2707 /*
2708 * Free any local-format data sitting around before we reset the
2709 * data fork to extents format. Note that the attr fork data has
2710 * already been freed by xfs_attr_inactive.
2711 */
2712 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2713 kmem_free(ip->i_df.if_u1.if_data);
2714 ip->i_df.if_u1.if_data = NULL;
2715 ip->i_df.if_bytes = 0;
2716 }
2717
2718 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2719 ip->i_d.di_flags = 0;
2720 ip->i_d.di_flags2 = 0;
2721 ip->i_d.di_dmevmask = 0;
2722 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2723 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2724
2725 /* Don't attempt to replay owner changes for a deleted inode */
2726 spin_lock(&iip->ili_lock);
2727 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2728 spin_unlock(&iip->ili_lock);
2729
2730 /*
2731 * Bump the generation count so no one will be confused
2732 * by reincarnations of this inode.
2733 */
2734 VFS_I(ip)->i_generation++;
2735 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2736
2737 if (xic.deleted)
2738 error = xfs_ifree_cluster(ip, tp, &xic);
2739
2740 return error;
2741}
2742
2743/*
2744 * This is called to unpin an inode. The caller must have the inode locked
2745 * in at least shared mode so that the buffer cannot be subsequently pinned
2746 * once someone is waiting for it to be unpinned.
2747 */
2748static void
2749xfs_iunpin(
2750 struct xfs_inode *ip)
2751{
2752 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2753
2754 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2755
2756 /* Give the log a push to start the unpinning I/O */
2757 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL);
2758
2759}
2760
2761static void
2762__xfs_iunpin_wait(
2763 struct xfs_inode *ip)
2764{
2765 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2766 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2767
2768 xfs_iunpin(ip);
2769
2770 do {
2771 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2772 if (xfs_ipincount(ip))
2773 io_schedule();
2774 } while (xfs_ipincount(ip));
2775 finish_wait(wq, &wait.wq_entry);
2776}
2777
2778void
2779xfs_iunpin_wait(
2780 struct xfs_inode *ip)
2781{
2782 if (xfs_ipincount(ip))
2783 __xfs_iunpin_wait(ip);
2784}
2785
2786/*
2787 * Removing an inode from the namespace involves removing the directory entry
2788 * and dropping the link count on the inode. Removing the directory entry can
2789 * result in locking an AGF (directory blocks were freed) and removing a link
2790 * count can result in placing the inode on an unlinked list which results in
2791 * locking an AGI.
2792 *
2793 * The big problem here is that we have an ordering constraint on AGF and AGI
2794 * locking - inode allocation locks the AGI, then can allocate a new extent for
2795 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2796 * removes the inode from the unlinked list, requiring that we lock the AGI
2797 * first, and then freeing the inode can result in an inode chunk being freed
2798 * and hence freeing disk space requiring that we lock an AGF.
2799 *
2800 * Hence the ordering that is imposed by other parts of the code is AGI before
2801 * AGF. This means we cannot remove the directory entry before we drop the inode
2802 * reference count and put it on the unlinked list as this results in a lock
2803 * order of AGF then AGI, and this can deadlock against inode allocation and
2804 * freeing. Therefore we must drop the link counts before we remove the
2805 * directory entry.
2806 *
2807 * This is still safe from a transactional point of view - it is not until we
2808 * get to xfs_defer_finish() that we have the possibility of multiple
2809 * transactions in this operation. Hence as long as we remove the directory
2810 * entry and drop the link count in the first transaction of the remove
2811 * operation, there are no transactional constraints on the ordering here.
2812 */
2813int
2814xfs_remove(
2815 xfs_inode_t *dp,
2816 struct xfs_name *name,
2817 xfs_inode_t *ip)
2818{
2819 xfs_mount_t *mp = dp->i_mount;
2820 xfs_trans_t *tp = NULL;
2821 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2822 int error = 0;
2823 uint resblks;
2824
2825 trace_xfs_remove(dp, name);
2826
2827 if (XFS_FORCED_SHUTDOWN(mp))
2828 return -EIO;
2829
2830 error = xfs_qm_dqattach(dp);
2831 if (error)
2832 goto std_return;
2833
2834 error = xfs_qm_dqattach(ip);
2835 if (error)
2836 goto std_return;
2837
2838 /*
2839 * We try to get the real space reservation first,
2840 * allowing for directory btree deletion(s) implying
2841 * possible bmap insert(s). If we can't get the space
2842 * reservation then we use 0 instead, and avoid the bmap
2843 * btree insert(s) in the directory code by, if the bmap
2844 * insert tries to happen, instead trimming the LAST
2845 * block from the directory.
2846 */
2847 resblks = XFS_REMOVE_SPACE_RES(mp);
2848 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
2849 if (error == -ENOSPC) {
2850 resblks = 0;
2851 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
2852 &tp);
2853 }
2854 if (error) {
2855 ASSERT(error != -ENOSPC);
2856 goto std_return;
2857 }
2858
2859 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
2860
2861 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2862 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2863
2864 /*
2865 * If we're removing a directory perform some additional validation.
2866 */
2867 if (is_dir) {
2868 ASSERT(VFS_I(ip)->i_nlink >= 2);
2869 if (VFS_I(ip)->i_nlink != 2) {
2870 error = -ENOTEMPTY;
2871 goto out_trans_cancel;
2872 }
2873 if (!xfs_dir_isempty(ip)) {
2874 error = -ENOTEMPTY;
2875 goto out_trans_cancel;
2876 }
2877
2878 /* Drop the link from ip's "..". */
2879 error = xfs_droplink(tp, dp);
2880 if (error)
2881 goto out_trans_cancel;
2882
2883 /* Drop the "." link from ip to self. */
2884 error = xfs_droplink(tp, ip);
2885 if (error)
2886 goto out_trans_cancel;
2887 } else {
2888 /*
2889 * When removing a non-directory we need to log the parent
2890 * inode here. For a directory this is done implicitly
2891 * by the xfs_droplink call for the ".." entry.
2892 */
2893 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2894 }
2895 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2896
2897 /* Drop the link from dp to ip. */
2898 error = xfs_droplink(tp, ip);
2899 if (error)
2900 goto out_trans_cancel;
2901
2902 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2903 if (error) {
2904 ASSERT(error != -ENOENT);
2905 goto out_trans_cancel;
2906 }
2907
2908 /*
2909 * If this is a synchronous mount, make sure that the
2910 * remove transaction goes to disk before returning to
2911 * the user.
2912 */
2913 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2914 xfs_trans_set_sync(tp);
2915
2916 error = xfs_trans_commit(tp);
2917 if (error)
2918 goto std_return;
2919
2920 if (is_dir && xfs_inode_is_filestream(ip))
2921 xfs_filestream_deassociate(ip);
2922
2923 return 0;
2924
2925 out_trans_cancel:
2926 xfs_trans_cancel(tp);
2927 std_return:
2928 return error;
2929}
2930
2931/*
2932 * Enter all inodes for a rename transaction into a sorted array.
2933 */
2934#define __XFS_SORT_INODES 5
2935STATIC void
2936xfs_sort_for_rename(
2937 struct xfs_inode *dp1, /* in: old (source) directory inode */
2938 struct xfs_inode *dp2, /* in: new (target) directory inode */
2939 struct xfs_inode *ip1, /* in: inode of old entry */
2940 struct xfs_inode *ip2, /* in: inode of new entry */
2941 struct xfs_inode *wip, /* in: whiteout inode */
2942 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2943 int *num_inodes) /* in/out: inodes in array */
2944{
2945 int i, j;
2946
2947 ASSERT(*num_inodes == __XFS_SORT_INODES);
2948 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2949
2950 /*
2951 * i_tab contains a list of pointers to inodes. We initialize
2952 * the table here & we'll sort it. We will then use it to
2953 * order the acquisition of the inode locks.
2954 *
2955 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2956 */
2957 i = 0;
2958 i_tab[i++] = dp1;
2959 i_tab[i++] = dp2;
2960 i_tab[i++] = ip1;
2961 if (ip2)
2962 i_tab[i++] = ip2;
2963 if (wip)
2964 i_tab[i++] = wip;
2965 *num_inodes = i;
2966
2967 /*
2968 * Sort the elements via bubble sort. (Remember, there are at
2969 * most 5 elements to sort, so this is adequate.)
2970 */
2971 for (i = 0; i < *num_inodes; i++) {
2972 for (j = 1; j < *num_inodes; j++) {
2973 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2974 struct xfs_inode *temp = i_tab[j];
2975 i_tab[j] = i_tab[j-1];
2976 i_tab[j-1] = temp;
2977 }
2978 }
2979 }
2980}
2981
2982static int
2983xfs_finish_rename(
2984 struct xfs_trans *tp)
2985{
2986 /*
2987 * If this is a synchronous mount, make sure that the rename transaction
2988 * goes to disk before returning to the user.
2989 */
2990 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2991 xfs_trans_set_sync(tp);
2992
2993 return xfs_trans_commit(tp);
2994}
2995
2996/*
2997 * xfs_cross_rename()
2998 *
2999 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall
3000 */
3001STATIC int
3002xfs_cross_rename(
3003 struct xfs_trans *tp,
3004 struct xfs_inode *dp1,
3005 struct xfs_name *name1,
3006 struct xfs_inode *ip1,
3007 struct xfs_inode *dp2,
3008 struct xfs_name *name2,
3009 struct xfs_inode *ip2,
3010 int spaceres)
3011{
3012 int error = 0;
3013 int ip1_flags = 0;
3014 int ip2_flags = 0;
3015 int dp2_flags = 0;
3016
3017 /* Swap inode number for dirent in first parent */
3018 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
3019 if (error)
3020 goto out_trans_abort;
3021
3022 /* Swap inode number for dirent in second parent */
3023 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
3024 if (error)
3025 goto out_trans_abort;
3026
3027 /*
3028 * If we're renaming one or more directories across different parents,
3029 * update the respective ".." entries (and link counts) to match the new
3030 * parents.
3031 */
3032 if (dp1 != dp2) {
3033 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3034
3035 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
3036 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
3037 dp1->i_ino, spaceres);
3038 if (error)
3039 goto out_trans_abort;
3040
3041 /* transfer ip2 ".." reference to dp1 */
3042 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
3043 error = xfs_droplink(tp, dp2);
3044 if (error)
3045 goto out_trans_abort;
3046 xfs_bumplink(tp, dp1);
3047 }
3048
3049 /*
3050 * Although ip1 isn't changed here, userspace needs
3051 * to be warned about the change, so that applications
3052 * relying on it (like backup ones), will properly
3053 * notify the change
3054 */
3055 ip1_flags |= XFS_ICHGTIME_CHG;
3056 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3057 }
3058
3059 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
3060 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
3061 dp2->i_ino, spaceres);
3062 if (error)
3063 goto out_trans_abort;
3064
3065 /* transfer ip1 ".." reference to dp2 */
3066 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
3067 error = xfs_droplink(tp, dp1);
3068 if (error)
3069 goto out_trans_abort;
3070 xfs_bumplink(tp, dp2);
3071 }
3072
3073 /*
3074 * Although ip2 isn't changed here, userspace needs
3075 * to be warned about the change, so that applications
3076 * relying on it (like backup ones), will properly
3077 * notify the change
3078 */
3079 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
3080 ip2_flags |= XFS_ICHGTIME_CHG;
3081 }
3082 }
3083
3084 if (ip1_flags) {
3085 xfs_trans_ichgtime(tp, ip1, ip1_flags);
3086 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
3087 }
3088 if (ip2_flags) {
3089 xfs_trans_ichgtime(tp, ip2, ip2_flags);
3090 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
3091 }
3092 if (dp2_flags) {
3093 xfs_trans_ichgtime(tp, dp2, dp2_flags);
3094 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
3095 }
3096 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3097 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
3098 return xfs_finish_rename(tp);
3099
3100out_trans_abort:
3101 xfs_trans_cancel(tp);
3102 return error;
3103}
3104
3105/*
3106 * xfs_rename_alloc_whiteout()
3107 *
3108 * Return a referenced, unlinked, unlocked inode that can be used as a
3109 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
3110 * crash between allocating the inode and linking it into the rename transaction
3111 * recovery will free the inode and we won't leak it.
3112 */
3113static int
3114xfs_rename_alloc_whiteout(
3115 struct xfs_inode *dp,
3116 struct xfs_inode **wip)
3117{
3118 struct xfs_inode *tmpfile;
3119 int error;
3120
3121 error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile);
3122 if (error)
3123 return error;
3124
3125 /*
3126 * Prepare the tmpfile inode as if it were created through the VFS.
3127 * Complete the inode setup and flag it as linkable. nlink is already
3128 * zero, so we can skip the drop_nlink.
3129 */
3130 xfs_setup_iops(tmpfile);
3131 xfs_finish_inode_setup(tmpfile);
3132 VFS_I(tmpfile)->i_state |= I_LINKABLE;
3133
3134 *wip = tmpfile;
3135 return 0;
3136}
3137
3138/*
3139 * xfs_rename
3140 */
3141int
3142xfs_rename(
3143 struct xfs_inode *src_dp,
3144 struct xfs_name *src_name,
3145 struct xfs_inode *src_ip,
3146 struct xfs_inode *target_dp,
3147 struct xfs_name *target_name,
3148 struct xfs_inode *target_ip,
3149 unsigned int flags)
3150{
3151 struct xfs_mount *mp = src_dp->i_mount;
3152 struct xfs_trans *tp;
3153 struct xfs_inode *wip = NULL; /* whiteout inode */
3154 struct xfs_inode *inodes[__XFS_SORT_INODES];
3155 struct xfs_buf *agibp;
3156 int num_inodes = __XFS_SORT_INODES;
3157 bool new_parent = (src_dp != target_dp);
3158 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
3159 int spaceres;
3160 int error;
3161
3162 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
3163
3164 if ((flags & RENAME_EXCHANGE) && !target_ip)
3165 return -EINVAL;
3166
3167 /*
3168 * If we are doing a whiteout operation, allocate the whiteout inode
3169 * we will be placing at the target and ensure the type is set
3170 * appropriately.
3171 */
3172 if (flags & RENAME_WHITEOUT) {
3173 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
3174 error = xfs_rename_alloc_whiteout(target_dp, &wip);
3175 if (error)
3176 return error;
3177
3178 /* setup target dirent info as whiteout */
3179 src_name->type = XFS_DIR3_FT_CHRDEV;
3180 }
3181
3182 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3183 inodes, &num_inodes);
3184
3185 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3186 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3187 if (error == -ENOSPC) {
3188 spaceres = 0;
3189 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3190 &tp);
3191 }
3192 if (error)
3193 goto out_release_wip;
3194
3195 /*
3196 * Attach the dquots to the inodes
3197 */
3198 error = xfs_qm_vop_rename_dqattach(inodes);
3199 if (error)
3200 goto out_trans_cancel;
3201
3202 /*
3203 * Lock all the participating inodes. Depending upon whether
3204 * the target_name exists in the target directory, and
3205 * whether the target directory is the same as the source
3206 * directory, we can lock from 2 to 4 inodes.
3207 */
3208 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3209
3210 /*
3211 * Join all the inodes to the transaction. From this point on,
3212 * we can rely on either trans_commit or trans_cancel to unlock
3213 * them.
3214 */
3215 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3216 if (new_parent)
3217 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3218 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3219 if (target_ip)
3220 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3221 if (wip)
3222 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3223
3224 /*
3225 * If we are using project inheritance, we only allow renames
3226 * into our tree when the project IDs are the same; else the
3227 * tree quota mechanism would be circumvented.
3228 */
3229 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
3230 target_dp->i_d.di_projid != src_ip->i_d.di_projid)) {
3231 error = -EXDEV;
3232 goto out_trans_cancel;
3233 }
3234
3235 /* RENAME_EXCHANGE is unique from here on. */
3236 if (flags & RENAME_EXCHANGE)
3237 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3238 target_dp, target_name, target_ip,
3239 spaceres);
3240
3241 /*
3242 * Check for expected errors before we dirty the transaction
3243 * so we can return an error without a transaction abort.
3244 */
3245 if (target_ip == NULL) {
3246 /*
3247 * If there's no space reservation, check the entry will
3248 * fit before actually inserting it.
3249 */
3250 if (!spaceres) {
3251 error = xfs_dir_canenter(tp, target_dp, target_name);
3252 if (error)
3253 goto out_trans_cancel;
3254 }
3255 } else {
3256 /*
3257 * If target exists and it's a directory, check that whether
3258 * it can be destroyed.
3259 */
3260 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3261 (!xfs_dir_isempty(target_ip) ||
3262 (VFS_I(target_ip)->i_nlink > 2))) {
3263 error = -EEXIST;
3264 goto out_trans_cancel;
3265 }
3266 }
3267
3268 /*
3269 * Directory entry creation below may acquire the AGF. Remove
3270 * the whiteout from the unlinked list first to preserve correct
3271 * AGI/AGF locking order. This dirties the transaction so failures
3272 * after this point will abort and log recovery will clean up the
3273 * mess.
3274 *
3275 * For whiteouts, we need to bump the link count on the whiteout
3276 * inode. After this point, we have a real link, clear the tmpfile
3277 * state flag from the inode so it doesn't accidentally get misused
3278 * in future.
3279 */
3280 if (wip) {
3281 ASSERT(VFS_I(wip)->i_nlink == 0);
3282 error = xfs_iunlink_remove(tp, wip);
3283 if (error)
3284 goto out_trans_cancel;
3285
3286 xfs_bumplink(tp, wip);
3287 VFS_I(wip)->i_state &= ~I_LINKABLE;
3288 }
3289
3290 /*
3291 * Set up the target.
3292 */
3293 if (target_ip == NULL) {
3294 /*
3295 * If target does not exist and the rename crosses
3296 * directories, adjust the target directory link count
3297 * to account for the ".." reference from the new entry.
3298 */
3299 error = xfs_dir_createname(tp, target_dp, target_name,
3300 src_ip->i_ino, spaceres);
3301 if (error)
3302 goto out_trans_cancel;
3303
3304 xfs_trans_ichgtime(tp, target_dp,
3305 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3306
3307 if (new_parent && src_is_directory) {
3308 xfs_bumplink(tp, target_dp);
3309 }
3310 } else { /* target_ip != NULL */
3311 /*
3312 * Link the source inode under the target name.
3313 * If the source inode is a directory and we are moving
3314 * it across directories, its ".." entry will be
3315 * inconsistent until we replace that down below.
3316 *
3317 * In case there is already an entry with the same
3318 * name at the destination directory, remove it first.
3319 */
3320
3321 /*
3322 * Check whether the replace operation will need to allocate
3323 * blocks. This happens when the shortform directory lacks
3324 * space and we have to convert it to a block format directory.
3325 * When more blocks are necessary, we must lock the AGI first
3326 * to preserve locking order (AGI -> AGF).
3327 */
3328 if (xfs_dir2_sf_replace_needblock(target_dp, src_ip->i_ino)) {
3329 error = xfs_read_agi(mp, tp,
3330 XFS_INO_TO_AGNO(mp, target_ip->i_ino),
3331 &agibp);
3332 if (error)
3333 goto out_trans_cancel;
3334 }
3335
3336 error = xfs_dir_replace(tp, target_dp, target_name,
3337 src_ip->i_ino, spaceres);
3338 if (error)
3339 goto out_trans_cancel;
3340
3341 xfs_trans_ichgtime(tp, target_dp,
3342 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3343
3344 /*
3345 * Decrement the link count on the target since the target
3346 * dir no longer points to it.
3347 */
3348 error = xfs_droplink(tp, target_ip);
3349 if (error)
3350 goto out_trans_cancel;
3351
3352 if (src_is_directory) {
3353 /*
3354 * Drop the link from the old "." entry.
3355 */
3356 error = xfs_droplink(tp, target_ip);
3357 if (error)
3358 goto out_trans_cancel;
3359 }
3360 } /* target_ip != NULL */
3361
3362 /*
3363 * Remove the source.
3364 */
3365 if (new_parent && src_is_directory) {
3366 /*
3367 * Rewrite the ".." entry to point to the new
3368 * directory.
3369 */
3370 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3371 target_dp->i_ino, spaceres);
3372 ASSERT(error != -EEXIST);
3373 if (error)
3374 goto out_trans_cancel;
3375 }
3376
3377 /*
3378 * We always want to hit the ctime on the source inode.
3379 *
3380 * This isn't strictly required by the standards since the source
3381 * inode isn't really being changed, but old unix file systems did
3382 * it and some incremental backup programs won't work without it.
3383 */
3384 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3385 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3386
3387 /*
3388 * Adjust the link count on src_dp. This is necessary when
3389 * renaming a directory, either within one parent when
3390 * the target existed, or across two parent directories.
3391 */
3392 if (src_is_directory && (new_parent || target_ip != NULL)) {
3393
3394 /*
3395 * Decrement link count on src_directory since the
3396 * entry that's moved no longer points to it.
3397 */
3398 error = xfs_droplink(tp, src_dp);
3399 if (error)
3400 goto out_trans_cancel;
3401 }
3402
3403 /*
3404 * For whiteouts, we only need to update the source dirent with the
3405 * inode number of the whiteout inode rather than removing it
3406 * altogether.
3407 */
3408 if (wip) {
3409 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3410 spaceres);
3411 } else
3412 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3413 spaceres);
3414 if (error)
3415 goto out_trans_cancel;
3416
3417 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3418 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3419 if (new_parent)
3420 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3421
3422 error = xfs_finish_rename(tp);
3423 if (wip)
3424 xfs_irele(wip);
3425 return error;
3426
3427out_trans_cancel:
3428 xfs_trans_cancel(tp);
3429out_release_wip:
3430 if (wip)
3431 xfs_irele(wip);
3432 return error;
3433}
3434
3435static int
3436xfs_iflush(
3437 struct xfs_inode *ip,
3438 struct xfs_buf *bp)
3439{
3440 struct xfs_inode_log_item *iip = ip->i_itemp;
3441 struct xfs_dinode *dip;
3442 struct xfs_mount *mp = ip->i_mount;
3443 int error;
3444
3445 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3446 ASSERT(xfs_isiflocked(ip));
3447 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3448 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3449 ASSERT(iip->ili_item.li_buf == bp);
3450
3451 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3452
3453 /*
3454 * We don't flush the inode if any of the following checks fail, but we
3455 * do still update the log item and attach to the backing buffer as if
3456 * the flush happened. This is a formality to facilitate predictable
3457 * error handling as the caller will shutdown and fail the buffer.
3458 */
3459 error = -EFSCORRUPTED;
3460 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3461 mp, XFS_ERRTAG_IFLUSH_1)) {
3462 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3463 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
3464 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3465 goto flush_out;
3466 }
3467 if (S_ISREG(VFS_I(ip)->i_mode)) {
3468 if (XFS_TEST_ERROR(
3469 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3470 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3471 mp, XFS_ERRTAG_IFLUSH_3)) {
3472 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3473 "%s: Bad regular inode %Lu, ptr "PTR_FMT,
3474 __func__, ip->i_ino, ip);
3475 goto flush_out;
3476 }
3477 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3478 if (XFS_TEST_ERROR(
3479 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3480 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3481 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3482 mp, XFS_ERRTAG_IFLUSH_4)) {
3483 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3484 "%s: Bad directory inode %Lu, ptr "PTR_FMT,
3485 __func__, ip->i_ino, ip);
3486 goto flush_out;
3487 }
3488 }
3489 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
3490 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3491 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3492 "%s: detected corrupt incore inode %Lu, "
3493 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
3494 __func__, ip->i_ino,
3495 ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
3496 ip->i_d.di_nblocks, ip);
3497 goto flush_out;
3498 }
3499 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3500 mp, XFS_ERRTAG_IFLUSH_6)) {
3501 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3502 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
3503 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3504 goto flush_out;
3505 }
3506
3507 /*
3508 * Inode item log recovery for v2 inodes are dependent on the
3509 * di_flushiter count for correct sequencing. We bump the flush
3510 * iteration count so we can detect flushes which postdate a log record
3511 * during recovery. This is redundant as we now log every change and
3512 * hence this can't happen but we need to still do it to ensure
3513 * backwards compatibility with old kernels that predate logging all
3514 * inode changes.
3515 */
3516 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
3517 ip->i_d.di_flushiter++;
3518
3519 /*
3520 * If there are inline format data / attr forks attached to this inode,
3521 * make sure they are not corrupt.
3522 */
3523 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3524 xfs_ifork_verify_local_data(ip))
3525 goto flush_out;
3526 if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
3527 xfs_ifork_verify_local_attr(ip))
3528 goto flush_out;
3529
3530 /*
3531 * Copy the dirty parts of the inode into the on-disk inode. We always
3532 * copy out the core of the inode, because if the inode is dirty at all
3533 * the core must be.
3534 */
3535 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3536
3537 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3538 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3539 ip->i_d.di_flushiter = 0;
3540
3541 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3542 if (XFS_IFORK_Q(ip))
3543 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3544
3545 /*
3546 * We've recorded everything logged in the inode, so we'd like to clear
3547 * the ili_fields bits so we don't log and flush things unnecessarily.
3548 * However, we can't stop logging all this information until the data
3549 * we've copied into the disk buffer is written to disk. If we did we
3550 * might overwrite the copy of the inode in the log with all the data
3551 * after re-logging only part of it, and in the face of a crash we
3552 * wouldn't have all the data we need to recover.
3553 *
3554 * What we do is move the bits to the ili_last_fields field. When
3555 * logging the inode, these bits are moved back to the ili_fields field.
3556 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3557 * know that the information those bits represent is permanently on
3558 * disk. As long as the flush completes before the inode is logged
3559 * again, then both ili_fields and ili_last_fields will be cleared.
3560 */
3561 error = 0;
3562flush_out:
3563 spin_lock(&iip->ili_lock);
3564 iip->ili_last_fields = iip->ili_fields;
3565 iip->ili_fields = 0;
3566 iip->ili_fsync_fields = 0;
3567 spin_unlock(&iip->ili_lock);
3568
3569 /*
3570 * Store the current LSN of the inode so that we can tell whether the
3571 * item has moved in the AIL from xfs_iflush_done().
3572 */
3573 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3574 &iip->ili_item.li_lsn);
3575
3576 /* generate the checksum. */
3577 xfs_dinode_calc_crc(mp, dip);
3578 return error;
3579}
3580
3581/*
3582 * Non-blocking flush of dirty inode metadata into the backing buffer.
3583 *
3584 * The caller must have a reference to the inode and hold the cluster buffer
3585 * locked. The function will walk across all the inodes on the cluster buffer it
3586 * can find and lock without blocking, and flush them to the cluster buffer.
3587 *
3588 * On successful flushing of at least one inode, the caller must write out the
3589 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3590 * the caller needs to release the buffer. On failure, the filesystem will be
3591 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3592 * will be returned.
3593 */
3594int
3595xfs_iflush_cluster(
3596 struct xfs_buf *bp)
3597{
3598 struct xfs_mount *mp = bp->b_mount;
3599 struct xfs_log_item *lip, *n;
3600 struct xfs_inode *ip;
3601 struct xfs_inode_log_item *iip;
3602 int clcount = 0;
3603 int error = 0;
3604
3605 /*
3606 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3607 * can remove itself from the list.
3608 */
3609 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3610 iip = (struct xfs_inode_log_item *)lip;
3611 ip = iip->ili_inode;
3612
3613 /*
3614 * Quick and dirty check to avoid locks if possible.
3615 */
3616 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLOCK))
3617 continue;
3618 if (xfs_ipincount(ip))
3619 continue;
3620
3621 /*
3622 * The inode is still attached to the buffer, which means it is
3623 * dirty but reclaim might try to grab it. Check carefully for
3624 * that, and grab the ilock while still holding the i_flags_lock
3625 * to guarantee reclaim will not be able to reclaim this inode
3626 * once we drop the i_flags_lock.
3627 */
3628 spin_lock(&ip->i_flags_lock);
3629 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3630 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLOCK)) {
3631 spin_unlock(&ip->i_flags_lock);
3632 continue;
3633 }
3634
3635 /*
3636 * ILOCK will pin the inode against reclaim and prevent
3637 * concurrent transactions modifying the inode while we are
3638 * flushing the inode.
3639 */
3640 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3641 spin_unlock(&ip->i_flags_lock);
3642 continue;
3643 }
3644 spin_unlock(&ip->i_flags_lock);
3645
3646 /*
3647 * Skip inodes that are already flush locked as they have
3648 * already been written to the buffer.
3649 */
3650 if (!xfs_iflock_nowait(ip)) {
3651 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3652 continue;
3653 }
3654
3655 /*
3656 * Abort flushing this inode if we are shut down because the
3657 * inode may not currently be in the AIL. This can occur when
3658 * log I/O failure unpins the inode without inserting into the
3659 * AIL, leaving a dirty/unpinned inode attached to the buffer
3660 * that otherwise looks like it should be flushed.
3661 */
3662 if (XFS_FORCED_SHUTDOWN(mp)) {
3663 xfs_iunpin_wait(ip);
3664 /* xfs_iflush_abort() drops the flush lock */
3665 xfs_iflush_abort(ip);
3666 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3667 error = -EIO;
3668 continue;
3669 }
3670
3671 /* don't block waiting on a log force to unpin dirty inodes */
3672 if (xfs_ipincount(ip)) {
3673 xfs_ifunlock(ip);
3674 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3675 continue;
3676 }
3677
3678 if (!xfs_inode_clean(ip))
3679 error = xfs_iflush(ip, bp);
3680 else
3681 xfs_ifunlock(ip);
3682 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3683 if (error)
3684 break;
3685 clcount++;
3686 }
3687
3688 if (error) {
3689 bp->b_flags |= XBF_ASYNC;
3690 xfs_buf_ioend_fail(bp);
3691 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3692 return error;
3693 }
3694
3695 if (!clcount)
3696 return -EAGAIN;
3697
3698 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3699 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3700 return 0;
3701
3702}
3703
3704/* Release an inode. */
3705void
3706xfs_irele(
3707 struct xfs_inode *ip)
3708{
3709 trace_xfs_irele(ip, _RET_IP_);
3710 iput(VFS_I(ip));
3711}
3712
3713/*
3714 * Ensure all commited transactions touching the inode are written to the log.
3715 */
3716int
3717xfs_log_force_inode(
3718 struct xfs_inode *ip)
3719{
3720 xfs_lsn_t lsn = 0;
3721
3722 xfs_ilock(ip, XFS_ILOCK_SHARED);
3723 if (xfs_ipincount(ip))
3724 lsn = ip->i_itemp->ili_last_lsn;
3725 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3726
3727 if (!lsn)
3728 return 0;
3729 return xfs_log_force_lsn(ip->i_mount, lsn, XFS_LOG_SYNC, NULL);
3730}
3731
3732/*
3733 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3734 * abide vfs locking order (lowest pointer value goes first) and breaking the
3735 * layout leases before proceeding. The loop is needed because we cannot call
3736 * the blocking break_layout() with the iolocks held, and therefore have to
3737 * back out both locks.
3738 */
3739static int
3740xfs_iolock_two_inodes_and_break_layout(
3741 struct inode *src,
3742 struct inode *dest)
3743{
3744 int error;
3745
3746 if (src > dest)
3747 swap(src, dest);
3748
3749retry:
3750 /* Wait to break both inodes' layouts before we start locking. */
3751 error = break_layout(src, true);
3752 if (error)
3753 return error;
3754 if (src != dest) {
3755 error = break_layout(dest, true);
3756 if (error)
3757 return error;
3758 }
3759
3760 /* Lock one inode and make sure nobody got in and leased it. */
3761 inode_lock(src);
3762 error = break_layout(src, false);
3763 if (error) {
3764 inode_unlock(src);
3765 if (error == -EWOULDBLOCK)
3766 goto retry;
3767 return error;
3768 }
3769
3770 if (src == dest)
3771 return 0;
3772
3773 /* Lock the other inode and make sure nobody got in and leased it. */
3774 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3775 error = break_layout(dest, false);
3776 if (error) {
3777 inode_unlock(src);
3778 inode_unlock(dest);
3779 if (error == -EWOULDBLOCK)
3780 goto retry;
3781 return error;
3782 }
3783
3784 return 0;
3785}
3786
3787/*
3788 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3789 * mmap activity.
3790 */
3791int
3792xfs_ilock2_io_mmap(
3793 struct xfs_inode *ip1,
3794 struct xfs_inode *ip2)
3795{
3796 int ret;
3797
3798 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3799 if (ret)
3800 return ret;
3801 if (ip1 == ip2)
3802 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3803 else
3804 xfs_lock_two_inodes(ip1, XFS_MMAPLOCK_EXCL,
3805 ip2, XFS_MMAPLOCK_EXCL);
3806 return 0;
3807}
3808
3809/* Unlock both inodes to allow IO and mmap activity. */
3810void
3811xfs_iunlock2_io_mmap(
3812 struct xfs_inode *ip1,
3813 struct xfs_inode *ip2)
3814{
3815 bool same_inode = (ip1 == ip2);
3816
3817 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3818 if (!same_inode)
3819 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3820 inode_unlock(VFS_I(ip2));
3821 if (!same_inode)
3822 inode_unlock(VFS_I(ip1));
3823}
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}