Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_trans.h"
15#include "xfs_inode_item.h"
16#include "xfs_trace.h"
17#include "xfs_trans_priv.h"
18#include "xfs_buf_item.h"
19#include "xfs_log.h"
20#include "xfs_log_priv.h"
21#include "xfs_error.h"
22
23#include <linux/iversion.h>
24
25struct kmem_cache *xfs_ili_cache; /* inode log item */
26
27static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
28{
29 return container_of(lip, struct xfs_inode_log_item, ili_item);
30}
31
32/*
33 * The logged size of an inode fork is always the current size of the inode
34 * fork. This means that when an inode fork is relogged, the size of the logged
35 * region is determined by the current state, not the combination of the
36 * previously logged state + the current state. This is different relogging
37 * behaviour to most other log items which will retain the size of the
38 * previously logged changes when smaller regions are relogged.
39 *
40 * Hence operations that remove data from the inode fork (e.g. shortform
41 * dir/attr remove, extent form extent removal, etc), the size of the relogged
42 * inode gets -smaller- rather than stays the same size as the previously logged
43 * size and this can result in the committing transaction reducing the amount of
44 * space being consumed by the CIL.
45 */
46STATIC void
47xfs_inode_item_data_fork_size(
48 struct xfs_inode_log_item *iip,
49 int *nvecs,
50 int *nbytes)
51{
52 struct xfs_inode *ip = iip->ili_inode;
53
54 switch (ip->i_df.if_format) {
55 case XFS_DINODE_FMT_EXTENTS:
56 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
57 ip->i_df.if_nextents > 0 &&
58 ip->i_df.if_bytes > 0) {
59 /* worst case, doesn't subtract delalloc extents */
60 *nbytes += xfs_inode_data_fork_size(ip);
61 *nvecs += 1;
62 }
63 break;
64 case XFS_DINODE_FMT_BTREE:
65 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
66 ip->i_df.if_broot_bytes > 0) {
67 *nbytes += ip->i_df.if_broot_bytes;
68 *nvecs += 1;
69 }
70 break;
71 case XFS_DINODE_FMT_LOCAL:
72 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
73 ip->i_df.if_bytes > 0) {
74 *nbytes += xlog_calc_iovec_len(ip->i_df.if_bytes);
75 *nvecs += 1;
76 }
77 break;
78
79 case XFS_DINODE_FMT_DEV:
80 break;
81 default:
82 ASSERT(0);
83 break;
84 }
85}
86
87STATIC void
88xfs_inode_item_attr_fork_size(
89 struct xfs_inode_log_item *iip,
90 int *nvecs,
91 int *nbytes)
92{
93 struct xfs_inode *ip = iip->ili_inode;
94
95 switch (ip->i_af.if_format) {
96 case XFS_DINODE_FMT_EXTENTS:
97 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
98 ip->i_af.if_nextents > 0 &&
99 ip->i_af.if_bytes > 0) {
100 /* worst case, doesn't subtract unused space */
101 *nbytes += xfs_inode_attr_fork_size(ip);
102 *nvecs += 1;
103 }
104 break;
105 case XFS_DINODE_FMT_BTREE:
106 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
107 ip->i_af.if_broot_bytes > 0) {
108 *nbytes += ip->i_af.if_broot_bytes;
109 *nvecs += 1;
110 }
111 break;
112 case XFS_DINODE_FMT_LOCAL:
113 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
114 ip->i_af.if_bytes > 0) {
115 *nbytes += xlog_calc_iovec_len(ip->i_af.if_bytes);
116 *nvecs += 1;
117 }
118 break;
119 default:
120 ASSERT(0);
121 break;
122 }
123}
124
125/*
126 * This returns the number of iovecs needed to log the given inode item.
127 *
128 * We need one iovec for the inode log format structure, one for the
129 * inode core, and possibly one for the inode data/extents/b-tree root
130 * and one for the inode attribute data/extents/b-tree root.
131 */
132STATIC void
133xfs_inode_item_size(
134 struct xfs_log_item *lip,
135 int *nvecs,
136 int *nbytes)
137{
138 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
139 struct xfs_inode *ip = iip->ili_inode;
140
141 *nvecs += 2;
142 *nbytes += sizeof(struct xfs_inode_log_format) +
143 xfs_log_dinode_size(ip->i_mount);
144
145 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
146 if (xfs_inode_has_attr_fork(ip))
147 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
148}
149
150STATIC void
151xfs_inode_item_format_data_fork(
152 struct xfs_inode_log_item *iip,
153 struct xfs_inode_log_format *ilf,
154 struct xfs_log_vec *lv,
155 struct xfs_log_iovec **vecp)
156{
157 struct xfs_inode *ip = iip->ili_inode;
158 size_t data_bytes;
159
160 switch (ip->i_df.if_format) {
161 case XFS_DINODE_FMT_EXTENTS:
162 iip->ili_fields &=
163 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
164
165 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
166 ip->i_df.if_nextents > 0 &&
167 ip->i_df.if_bytes > 0) {
168 struct xfs_bmbt_rec *p;
169
170 ASSERT(xfs_iext_count(&ip->i_df) > 0);
171
172 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
173 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
174 xlog_finish_iovec(lv, *vecp, data_bytes);
175
176 ASSERT(data_bytes <= ip->i_df.if_bytes);
177
178 ilf->ilf_dsize = data_bytes;
179 ilf->ilf_size++;
180 } else {
181 iip->ili_fields &= ~XFS_ILOG_DEXT;
182 }
183 break;
184 case XFS_DINODE_FMT_BTREE:
185 iip->ili_fields &=
186 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
187
188 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
189 ip->i_df.if_broot_bytes > 0) {
190 ASSERT(ip->i_df.if_broot != NULL);
191 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
192 ip->i_df.if_broot,
193 ip->i_df.if_broot_bytes);
194 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
195 ilf->ilf_size++;
196 } else {
197 ASSERT(!(iip->ili_fields &
198 XFS_ILOG_DBROOT));
199 iip->ili_fields &= ~XFS_ILOG_DBROOT;
200 }
201 break;
202 case XFS_DINODE_FMT_LOCAL:
203 iip->ili_fields &=
204 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
205 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
206 ip->i_df.if_bytes > 0) {
207 ASSERT(ip->i_df.if_u1.if_data != NULL);
208 ASSERT(ip->i_disk_size > 0);
209 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
210 ip->i_df.if_u1.if_data,
211 ip->i_df.if_bytes);
212 ilf->ilf_dsize = (unsigned)ip->i_df.if_bytes;
213 ilf->ilf_size++;
214 } else {
215 iip->ili_fields &= ~XFS_ILOG_DDATA;
216 }
217 break;
218 case XFS_DINODE_FMT_DEV:
219 iip->ili_fields &=
220 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
221 if (iip->ili_fields & XFS_ILOG_DEV)
222 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
223 break;
224 default:
225 ASSERT(0);
226 break;
227 }
228}
229
230STATIC void
231xfs_inode_item_format_attr_fork(
232 struct xfs_inode_log_item *iip,
233 struct xfs_inode_log_format *ilf,
234 struct xfs_log_vec *lv,
235 struct xfs_log_iovec **vecp)
236{
237 struct xfs_inode *ip = iip->ili_inode;
238 size_t data_bytes;
239
240 switch (ip->i_af.if_format) {
241 case XFS_DINODE_FMT_EXTENTS:
242 iip->ili_fields &=
243 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
244
245 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
246 ip->i_af.if_nextents > 0 &&
247 ip->i_af.if_bytes > 0) {
248 struct xfs_bmbt_rec *p;
249
250 ASSERT(xfs_iext_count(&ip->i_af) ==
251 ip->i_af.if_nextents);
252
253 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
254 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
255 xlog_finish_iovec(lv, *vecp, data_bytes);
256
257 ilf->ilf_asize = data_bytes;
258 ilf->ilf_size++;
259 } else {
260 iip->ili_fields &= ~XFS_ILOG_AEXT;
261 }
262 break;
263 case XFS_DINODE_FMT_BTREE:
264 iip->ili_fields &=
265 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
266
267 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
268 ip->i_af.if_broot_bytes > 0) {
269 ASSERT(ip->i_af.if_broot != NULL);
270
271 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
272 ip->i_af.if_broot,
273 ip->i_af.if_broot_bytes);
274 ilf->ilf_asize = ip->i_af.if_broot_bytes;
275 ilf->ilf_size++;
276 } else {
277 iip->ili_fields &= ~XFS_ILOG_ABROOT;
278 }
279 break;
280 case XFS_DINODE_FMT_LOCAL:
281 iip->ili_fields &=
282 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
283
284 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
285 ip->i_af.if_bytes > 0) {
286 ASSERT(ip->i_af.if_u1.if_data != NULL);
287 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
288 ip->i_af.if_u1.if_data,
289 ip->i_af.if_bytes);
290 ilf->ilf_asize = (unsigned)ip->i_af.if_bytes;
291 ilf->ilf_size++;
292 } else {
293 iip->ili_fields &= ~XFS_ILOG_ADATA;
294 }
295 break;
296 default:
297 ASSERT(0);
298 break;
299 }
300}
301
302/*
303 * Convert an incore timestamp to a log timestamp. Note that the log format
304 * specifies host endian format!
305 */
306static inline xfs_log_timestamp_t
307xfs_inode_to_log_dinode_ts(
308 struct xfs_inode *ip,
309 const struct timespec64 tv)
310{
311 struct xfs_log_legacy_timestamp *lits;
312 xfs_log_timestamp_t its;
313
314 if (xfs_inode_has_bigtime(ip))
315 return xfs_inode_encode_bigtime(tv);
316
317 lits = (struct xfs_log_legacy_timestamp *)&its;
318 lits->t_sec = tv.tv_sec;
319 lits->t_nsec = tv.tv_nsec;
320
321 return its;
322}
323
324/*
325 * The legacy DMAPI fields are only present in the on-disk and in-log inodes,
326 * but not in the in-memory one. But we are guaranteed to have an inode buffer
327 * in memory when logging an inode, so we can just copy it from the on-disk
328 * inode to the in-log inode here so that recovery of file system with these
329 * fields set to non-zero values doesn't lose them. For all other cases we zero
330 * the fields.
331 */
332static void
333xfs_copy_dm_fields_to_log_dinode(
334 struct xfs_inode *ip,
335 struct xfs_log_dinode *to)
336{
337 struct xfs_dinode *dip;
338
339 dip = xfs_buf_offset(ip->i_itemp->ili_item.li_buf,
340 ip->i_imap.im_boffset);
341
342 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS)) {
343 to->di_dmevmask = be32_to_cpu(dip->di_dmevmask);
344 to->di_dmstate = be16_to_cpu(dip->di_dmstate);
345 } else {
346 to->di_dmevmask = 0;
347 to->di_dmstate = 0;
348 }
349}
350
351static inline void
352xfs_inode_to_log_dinode_iext_counters(
353 struct xfs_inode *ip,
354 struct xfs_log_dinode *to)
355{
356 if (xfs_inode_has_large_extent_counts(ip)) {
357 to->di_big_nextents = xfs_ifork_nextents(&ip->i_df);
358 to->di_big_anextents = xfs_ifork_nextents(&ip->i_af);
359 to->di_nrext64_pad = 0;
360 } else {
361 to->di_nextents = xfs_ifork_nextents(&ip->i_df);
362 to->di_anextents = xfs_ifork_nextents(&ip->i_af);
363 }
364}
365
366static void
367xfs_inode_to_log_dinode(
368 struct xfs_inode *ip,
369 struct xfs_log_dinode *to,
370 xfs_lsn_t lsn)
371{
372 struct inode *inode = VFS_I(ip);
373
374 to->di_magic = XFS_DINODE_MAGIC;
375 to->di_format = xfs_ifork_format(&ip->i_df);
376 to->di_uid = i_uid_read(inode);
377 to->di_gid = i_gid_read(inode);
378 to->di_projid_lo = ip->i_projid & 0xffff;
379 to->di_projid_hi = ip->i_projid >> 16;
380
381 memset(to->di_pad3, 0, sizeof(to->di_pad3));
382 to->di_atime = xfs_inode_to_log_dinode_ts(ip, inode->i_atime);
383 to->di_mtime = xfs_inode_to_log_dinode_ts(ip, inode->i_mtime);
384 to->di_ctime = xfs_inode_to_log_dinode_ts(ip, inode->i_ctime);
385 to->di_nlink = inode->i_nlink;
386 to->di_gen = inode->i_generation;
387 to->di_mode = inode->i_mode;
388
389 to->di_size = ip->i_disk_size;
390 to->di_nblocks = ip->i_nblocks;
391 to->di_extsize = ip->i_extsize;
392 to->di_forkoff = ip->i_forkoff;
393 to->di_aformat = xfs_ifork_format(&ip->i_af);
394 to->di_flags = ip->i_diflags;
395
396 xfs_copy_dm_fields_to_log_dinode(ip, to);
397
398 /* log a dummy value to ensure log structure is fully initialised */
399 to->di_next_unlinked = NULLAGINO;
400
401 if (xfs_has_v3inodes(ip->i_mount)) {
402 to->di_version = 3;
403 to->di_changecount = inode_peek_iversion(inode);
404 to->di_crtime = xfs_inode_to_log_dinode_ts(ip, ip->i_crtime);
405 to->di_flags2 = ip->i_diflags2;
406 to->di_cowextsize = ip->i_cowextsize;
407 to->di_ino = ip->i_ino;
408 to->di_lsn = lsn;
409 memset(to->di_pad2, 0, sizeof(to->di_pad2));
410 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
411 to->di_v3_pad = 0;
412 } else {
413 to->di_version = 2;
414 to->di_flushiter = ip->i_flushiter;
415 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
416 }
417
418 xfs_inode_to_log_dinode_iext_counters(ip, to);
419}
420
421/*
422 * Format the inode core. Current timestamp data is only in the VFS inode
423 * fields, so we need to grab them from there. Hence rather than just copying
424 * the XFS inode core structure, format the fields directly into the iovec.
425 */
426static void
427xfs_inode_item_format_core(
428 struct xfs_inode *ip,
429 struct xfs_log_vec *lv,
430 struct xfs_log_iovec **vecp)
431{
432 struct xfs_log_dinode *dic;
433
434 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
435 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
436 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
437}
438
439/*
440 * This is called to fill in the vector of log iovecs for the given inode
441 * log item. It fills the first item with an inode log format structure,
442 * the second with the on-disk inode structure, and a possible third and/or
443 * fourth with the inode data/extents/b-tree root and inode attributes
444 * data/extents/b-tree root.
445 *
446 * Note: Always use the 64 bit inode log format structure so we don't
447 * leave an uninitialised hole in the format item on 64 bit systems. Log
448 * recovery on 32 bit systems handles this just fine, so there's no reason
449 * for not using an initialising the properly padded structure all the time.
450 */
451STATIC void
452xfs_inode_item_format(
453 struct xfs_log_item *lip,
454 struct xfs_log_vec *lv)
455{
456 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
457 struct xfs_inode *ip = iip->ili_inode;
458 struct xfs_log_iovec *vecp = NULL;
459 struct xfs_inode_log_format *ilf;
460
461 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
462 ilf->ilf_type = XFS_LI_INODE;
463 ilf->ilf_ino = ip->i_ino;
464 ilf->ilf_blkno = ip->i_imap.im_blkno;
465 ilf->ilf_len = ip->i_imap.im_len;
466 ilf->ilf_boffset = ip->i_imap.im_boffset;
467 ilf->ilf_fields = XFS_ILOG_CORE;
468 ilf->ilf_size = 2; /* format + core */
469
470 /*
471 * make sure we don't leak uninitialised data into the log in the case
472 * when we don't log every field in the inode.
473 */
474 ilf->ilf_dsize = 0;
475 ilf->ilf_asize = 0;
476 ilf->ilf_pad = 0;
477 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
478
479 xlog_finish_iovec(lv, vecp, sizeof(*ilf));
480
481 xfs_inode_item_format_core(ip, lv, &vecp);
482 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
483 if (xfs_inode_has_attr_fork(ip)) {
484 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
485 } else {
486 iip->ili_fields &=
487 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
488 }
489
490 /* update the format with the exact fields we actually logged */
491 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
492}
493
494/*
495 * This is called to pin the inode associated with the inode log
496 * item in memory so it cannot be written out.
497 */
498STATIC void
499xfs_inode_item_pin(
500 struct xfs_log_item *lip)
501{
502 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
503
504 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
505 ASSERT(lip->li_buf);
506
507 trace_xfs_inode_pin(ip, _RET_IP_);
508 atomic_inc(&ip->i_pincount);
509}
510
511
512/*
513 * This is called to unpin the inode associated with the inode log
514 * item which was previously pinned with a call to xfs_inode_item_pin().
515 *
516 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
517 *
518 * Note that unpin can race with inode cluster buffer freeing marking the buffer
519 * stale. In that case, flush completions are run from the buffer unpin call,
520 * which may happen before the inode is unpinned. If we lose the race, there
521 * will be no buffer attached to the log item, but the inode will be marked
522 * XFS_ISTALE.
523 */
524STATIC void
525xfs_inode_item_unpin(
526 struct xfs_log_item *lip,
527 int remove)
528{
529 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
530
531 trace_xfs_inode_unpin(ip, _RET_IP_);
532 ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
533 ASSERT(atomic_read(&ip->i_pincount) > 0);
534 if (atomic_dec_and_test(&ip->i_pincount))
535 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
536}
537
538STATIC uint
539xfs_inode_item_push(
540 struct xfs_log_item *lip,
541 struct list_head *buffer_list)
542 __releases(&lip->li_ailp->ail_lock)
543 __acquires(&lip->li_ailp->ail_lock)
544{
545 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
546 struct xfs_inode *ip = iip->ili_inode;
547 struct xfs_buf *bp = lip->li_buf;
548 uint rval = XFS_ITEM_SUCCESS;
549 int error;
550
551 if (!bp || (ip->i_flags & XFS_ISTALE)) {
552 /*
553 * Inode item/buffer is being aborted due to cluster
554 * buffer deletion. Trigger a log force to have that operation
555 * completed and items removed from the AIL before the next push
556 * attempt.
557 */
558 return XFS_ITEM_PINNED;
559 }
560
561 if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp))
562 return XFS_ITEM_PINNED;
563
564 if (xfs_iflags_test(ip, XFS_IFLUSHING))
565 return XFS_ITEM_FLUSHING;
566
567 if (!xfs_buf_trylock(bp))
568 return XFS_ITEM_LOCKED;
569
570 spin_unlock(&lip->li_ailp->ail_lock);
571
572 /*
573 * We need to hold a reference for flushing the cluster buffer as it may
574 * fail the buffer without IO submission. In which case, we better get a
575 * reference for that completion because otherwise we don't get a
576 * reference for IO until we queue the buffer for delwri submission.
577 */
578 xfs_buf_hold(bp);
579 error = xfs_iflush_cluster(bp);
580 if (!error) {
581 if (!xfs_buf_delwri_queue(bp, buffer_list))
582 rval = XFS_ITEM_FLUSHING;
583 xfs_buf_relse(bp);
584 } else {
585 /*
586 * Release the buffer if we were unable to flush anything. On
587 * any other error, the buffer has already been released.
588 */
589 if (error == -EAGAIN)
590 xfs_buf_relse(bp);
591 rval = XFS_ITEM_LOCKED;
592 }
593
594 spin_lock(&lip->li_ailp->ail_lock);
595 return rval;
596}
597
598/*
599 * Unlock the inode associated with the inode log item.
600 */
601STATIC void
602xfs_inode_item_release(
603 struct xfs_log_item *lip)
604{
605 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
606 struct xfs_inode *ip = iip->ili_inode;
607 unsigned short lock_flags;
608
609 ASSERT(ip->i_itemp != NULL);
610 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
611
612 lock_flags = iip->ili_lock_flags;
613 iip->ili_lock_flags = 0;
614 if (lock_flags)
615 xfs_iunlock(ip, lock_flags);
616}
617
618/*
619 * This is called to find out where the oldest active copy of the inode log
620 * item in the on disk log resides now that the last log write of it completed
621 * at the given lsn. Since we always re-log all dirty data in an inode, the
622 * latest copy in the on disk log is the only one that matters. Therefore,
623 * simply return the given lsn.
624 *
625 * If the inode has been marked stale because the cluster is being freed, we
626 * don't want to (re-)insert this inode into the AIL. There is a race condition
627 * where the cluster buffer may be unpinned before the inode is inserted into
628 * the AIL during transaction committed processing. If the buffer is unpinned
629 * before the inode item has been committed and inserted, then it is possible
630 * for the buffer to be written and IO completes before the inode is inserted
631 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
632 * AIL which will never get removed. It will, however, get reclaimed which
633 * triggers an assert in xfs_inode_free() complaining about freein an inode
634 * still in the AIL.
635 *
636 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
637 * transaction committed code knows that it does not need to do any further
638 * processing on the item.
639 */
640STATIC xfs_lsn_t
641xfs_inode_item_committed(
642 struct xfs_log_item *lip,
643 xfs_lsn_t lsn)
644{
645 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
646 struct xfs_inode *ip = iip->ili_inode;
647
648 if (xfs_iflags_test(ip, XFS_ISTALE)) {
649 xfs_inode_item_unpin(lip, 0);
650 return -1;
651 }
652 return lsn;
653}
654
655STATIC void
656xfs_inode_item_committing(
657 struct xfs_log_item *lip,
658 xfs_csn_t seq)
659{
660 INODE_ITEM(lip)->ili_commit_seq = seq;
661 return xfs_inode_item_release(lip);
662}
663
664static const struct xfs_item_ops xfs_inode_item_ops = {
665 .iop_size = xfs_inode_item_size,
666 .iop_format = xfs_inode_item_format,
667 .iop_pin = xfs_inode_item_pin,
668 .iop_unpin = xfs_inode_item_unpin,
669 .iop_release = xfs_inode_item_release,
670 .iop_committed = xfs_inode_item_committed,
671 .iop_push = xfs_inode_item_push,
672 .iop_committing = xfs_inode_item_committing,
673};
674
675
676/*
677 * Initialize the inode log item for a newly allocated (in-core) inode.
678 */
679void
680xfs_inode_item_init(
681 struct xfs_inode *ip,
682 struct xfs_mount *mp)
683{
684 struct xfs_inode_log_item *iip;
685
686 ASSERT(ip->i_itemp == NULL);
687 iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_cache,
688 GFP_KERNEL | __GFP_NOFAIL);
689
690 iip->ili_inode = ip;
691 spin_lock_init(&iip->ili_lock);
692 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
693 &xfs_inode_item_ops);
694}
695
696/*
697 * Free the inode log item and any memory hanging off of it.
698 */
699void
700xfs_inode_item_destroy(
701 struct xfs_inode *ip)
702{
703 struct xfs_inode_log_item *iip = ip->i_itemp;
704
705 ASSERT(iip->ili_item.li_buf == NULL);
706
707 ip->i_itemp = NULL;
708 kmem_free(iip->ili_item.li_lv_shadow);
709 kmem_cache_free(xfs_ili_cache, iip);
710}
711
712
713/*
714 * We only want to pull the item from the AIL if it is actually there
715 * and its location in the log has not changed since we started the
716 * flush. Thus, we only bother if the inode's lsn has not changed.
717 */
718static void
719xfs_iflush_ail_updates(
720 struct xfs_ail *ailp,
721 struct list_head *list)
722{
723 struct xfs_log_item *lip;
724 xfs_lsn_t tail_lsn = 0;
725
726 /* this is an opencoded batch version of xfs_trans_ail_delete */
727 spin_lock(&ailp->ail_lock);
728 list_for_each_entry(lip, list, li_bio_list) {
729 xfs_lsn_t lsn;
730
731 clear_bit(XFS_LI_FAILED, &lip->li_flags);
732 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
733 continue;
734
735 /*
736 * dgc: Not sure how this happens, but it happens very
737 * occassionaly via generic/388. xfs_iflush_abort() also
738 * silently handles this same "under writeback but not in AIL at
739 * shutdown" condition via xfs_trans_ail_delete().
740 */
741 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
742 ASSERT(xlog_is_shutdown(lip->li_log));
743 continue;
744 }
745
746 lsn = xfs_ail_delete_one(ailp, lip);
747 if (!tail_lsn && lsn)
748 tail_lsn = lsn;
749 }
750 xfs_ail_update_finish(ailp, tail_lsn);
751}
752
753/*
754 * Walk the list of inodes that have completed their IOs. If they are clean
755 * remove them from the list and dissociate them from the buffer. Buffers that
756 * are still dirty remain linked to the buffer and on the list. Caller must
757 * handle them appropriately.
758 */
759static void
760xfs_iflush_finish(
761 struct xfs_buf *bp,
762 struct list_head *list)
763{
764 struct xfs_log_item *lip, *n;
765
766 list_for_each_entry_safe(lip, n, list, li_bio_list) {
767 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
768 bool drop_buffer = false;
769
770 spin_lock(&iip->ili_lock);
771
772 /*
773 * Remove the reference to the cluster buffer if the inode is
774 * clean in memory and drop the buffer reference once we've
775 * dropped the locks we hold.
776 */
777 ASSERT(iip->ili_item.li_buf == bp);
778 if (!iip->ili_fields) {
779 iip->ili_item.li_buf = NULL;
780 list_del_init(&lip->li_bio_list);
781 drop_buffer = true;
782 }
783 iip->ili_last_fields = 0;
784 iip->ili_flush_lsn = 0;
785 spin_unlock(&iip->ili_lock);
786 xfs_iflags_clear(iip->ili_inode, XFS_IFLUSHING);
787 if (drop_buffer)
788 xfs_buf_rele(bp);
789 }
790}
791
792/*
793 * Inode buffer IO completion routine. It is responsible for removing inodes
794 * attached to the buffer from the AIL if they have not been re-logged and
795 * completing the inode flush.
796 */
797void
798xfs_buf_inode_iodone(
799 struct xfs_buf *bp)
800{
801 struct xfs_log_item *lip, *n;
802 LIST_HEAD(flushed_inodes);
803 LIST_HEAD(ail_updates);
804
805 /*
806 * Pull the attached inodes from the buffer one at a time and take the
807 * appropriate action on them.
808 */
809 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
810 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
811
812 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
813 xfs_iflush_abort(iip->ili_inode);
814 continue;
815 }
816 if (!iip->ili_last_fields)
817 continue;
818
819 /* Do an unlocked check for needing the AIL lock. */
820 if (iip->ili_flush_lsn == lip->li_lsn ||
821 test_bit(XFS_LI_FAILED, &lip->li_flags))
822 list_move_tail(&lip->li_bio_list, &ail_updates);
823 else
824 list_move_tail(&lip->li_bio_list, &flushed_inodes);
825 }
826
827 if (!list_empty(&ail_updates)) {
828 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
829 list_splice_tail(&ail_updates, &flushed_inodes);
830 }
831
832 xfs_iflush_finish(bp, &flushed_inodes);
833 if (!list_empty(&flushed_inodes))
834 list_splice_tail(&flushed_inodes, &bp->b_li_list);
835}
836
837void
838xfs_buf_inode_io_fail(
839 struct xfs_buf *bp)
840{
841 struct xfs_log_item *lip;
842
843 list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
844 set_bit(XFS_LI_FAILED, &lip->li_flags);
845}
846
847/*
848 * Clear the inode logging fields so no more flushes are attempted. If we are
849 * on a buffer list, it is now safe to remove it because the buffer is
850 * guaranteed to be locked. The caller will drop the reference to the buffer
851 * the log item held.
852 */
853static void
854xfs_iflush_abort_clean(
855 struct xfs_inode_log_item *iip)
856{
857 iip->ili_last_fields = 0;
858 iip->ili_fields = 0;
859 iip->ili_fsync_fields = 0;
860 iip->ili_flush_lsn = 0;
861 iip->ili_item.li_buf = NULL;
862 list_del_init(&iip->ili_item.li_bio_list);
863}
864
865/*
866 * Abort flushing the inode from a context holding the cluster buffer locked.
867 *
868 * This is the normal runtime method of aborting writeback of an inode that is
869 * attached to a cluster buffer. It occurs when the inode and the backing
870 * cluster buffer have been freed (i.e. inode is XFS_ISTALE), or when cluster
871 * flushing or buffer IO completion encounters a log shutdown situation.
872 *
873 * If we need to abort inode writeback and we don't already hold the buffer
874 * locked, call xfs_iflush_shutdown_abort() instead as this should only ever be
875 * necessary in a shutdown situation.
876 */
877void
878xfs_iflush_abort(
879 struct xfs_inode *ip)
880{
881 struct xfs_inode_log_item *iip = ip->i_itemp;
882 struct xfs_buf *bp;
883
884 if (!iip) {
885 /* clean inode, nothing to do */
886 xfs_iflags_clear(ip, XFS_IFLUSHING);
887 return;
888 }
889
890 /*
891 * Remove the inode item from the AIL before we clear its internal
892 * state. Whilst the inode is in the AIL, it should have a valid buffer
893 * pointer for push operations to access - it is only safe to remove the
894 * inode from the buffer once it has been removed from the AIL.
895 *
896 * We also clear the failed bit before removing the item from the AIL
897 * as xfs_trans_ail_delete()->xfs_clear_li_failed() will release buffer
898 * references the inode item owns and needs to hold until we've fully
899 * aborted the inode log item and detached it from the buffer.
900 */
901 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
902 xfs_trans_ail_delete(&iip->ili_item, 0);
903
904 /*
905 * Grab the inode buffer so can we release the reference the inode log
906 * item holds on it.
907 */
908 spin_lock(&iip->ili_lock);
909 bp = iip->ili_item.li_buf;
910 xfs_iflush_abort_clean(iip);
911 spin_unlock(&iip->ili_lock);
912
913 xfs_iflags_clear(ip, XFS_IFLUSHING);
914 if (bp)
915 xfs_buf_rele(bp);
916}
917
918/*
919 * Abort an inode flush in the case of a shutdown filesystem. This can be called
920 * from anywhere with just an inode reference and does not require holding the
921 * inode cluster buffer locked. If the inode is attached to a cluster buffer,
922 * it will grab and lock it safely, then abort the inode flush.
923 */
924void
925xfs_iflush_shutdown_abort(
926 struct xfs_inode *ip)
927{
928 struct xfs_inode_log_item *iip = ip->i_itemp;
929 struct xfs_buf *bp;
930
931 if (!iip) {
932 /* clean inode, nothing to do */
933 xfs_iflags_clear(ip, XFS_IFLUSHING);
934 return;
935 }
936
937 spin_lock(&iip->ili_lock);
938 bp = iip->ili_item.li_buf;
939 if (!bp) {
940 spin_unlock(&iip->ili_lock);
941 xfs_iflush_abort(ip);
942 return;
943 }
944
945 /*
946 * We have to take a reference to the buffer so that it doesn't get
947 * freed when we drop the ili_lock and then wait to lock the buffer.
948 * We'll clean up the extra reference after we pick up the ili_lock
949 * again.
950 */
951 xfs_buf_hold(bp);
952 spin_unlock(&iip->ili_lock);
953 xfs_buf_lock(bp);
954
955 spin_lock(&iip->ili_lock);
956 if (!iip->ili_item.li_buf) {
957 /*
958 * Raced with another removal, hold the only reference
959 * to bp now. Inode should not be in the AIL now, so just clean
960 * up and return;
961 */
962 ASSERT(list_empty(&iip->ili_item.li_bio_list));
963 ASSERT(!test_bit(XFS_LI_IN_AIL, &iip->ili_item.li_flags));
964 xfs_iflush_abort_clean(iip);
965 spin_unlock(&iip->ili_lock);
966 xfs_iflags_clear(ip, XFS_IFLUSHING);
967 xfs_buf_relse(bp);
968 return;
969 }
970
971 /*
972 * Got two references to bp. The first will get dropped by
973 * xfs_iflush_abort() when the item is removed from the buffer list, but
974 * we can't drop our reference until _abort() returns because we have to
975 * unlock the buffer as well. Hence we abort and then unlock and release
976 * our reference to the buffer.
977 */
978 ASSERT(iip->ili_item.li_buf == bp);
979 spin_unlock(&iip->ili_lock);
980 xfs_iflush_abort(ip);
981 xfs_buf_relse(bp);
982}
983
984
985/*
986 * convert an xfs_inode_log_format struct from the old 32 bit version
987 * (which can have different field alignments) to the native 64 bit version
988 */
989int
990xfs_inode_item_format_convert(
991 struct xfs_log_iovec *buf,
992 struct xfs_inode_log_format *in_f)
993{
994 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;
995
996 if (buf->i_len != sizeof(*in_f32)) {
997 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
998 return -EFSCORRUPTED;
999 }
1000
1001 in_f->ilf_type = in_f32->ilf_type;
1002 in_f->ilf_size = in_f32->ilf_size;
1003 in_f->ilf_fields = in_f32->ilf_fields;
1004 in_f->ilf_asize = in_f32->ilf_asize;
1005 in_f->ilf_dsize = in_f32->ilf_dsize;
1006 in_f->ilf_ino = in_f32->ilf_ino;
1007 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
1008 in_f->ilf_blkno = in_f32->ilf_blkno;
1009 in_f->ilf_len = in_f32->ilf_len;
1010 in_f->ilf_boffset = in_f32->ilf_boffset;
1011 return 0;
1012}
1/*
2 * Copyright (c) 2000-2002,2005 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 "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
23#include "xfs_mount.h"
24#include "xfs_inode.h"
25#include "xfs_trans.h"
26#include "xfs_inode_item.h"
27#include "xfs_error.h"
28#include "xfs_trace.h"
29#include "xfs_trans_priv.h"
30#include "xfs_log.h"
31
32
33kmem_zone_t *xfs_ili_zone; /* inode log item zone */
34
35static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
36{
37 return container_of(lip, struct xfs_inode_log_item, ili_item);
38}
39
40STATIC void
41xfs_inode_item_data_fork_size(
42 struct xfs_inode_log_item *iip,
43 int *nvecs,
44 int *nbytes)
45{
46 struct xfs_inode *ip = iip->ili_inode;
47
48 switch (ip->i_d.di_format) {
49 case XFS_DINODE_FMT_EXTENTS:
50 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
51 ip->i_d.di_nextents > 0 &&
52 ip->i_df.if_bytes > 0) {
53 /* worst case, doesn't subtract delalloc extents */
54 *nbytes += XFS_IFORK_DSIZE(ip);
55 *nvecs += 1;
56 }
57 break;
58 case XFS_DINODE_FMT_BTREE:
59 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
60 ip->i_df.if_broot_bytes > 0) {
61 *nbytes += ip->i_df.if_broot_bytes;
62 *nvecs += 1;
63 }
64 break;
65 case XFS_DINODE_FMT_LOCAL:
66 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
67 ip->i_df.if_bytes > 0) {
68 *nbytes += roundup(ip->i_df.if_bytes, 4);
69 *nvecs += 1;
70 }
71 break;
72
73 case XFS_DINODE_FMT_DEV:
74 case XFS_DINODE_FMT_UUID:
75 break;
76 default:
77 ASSERT(0);
78 break;
79 }
80}
81
82STATIC void
83xfs_inode_item_attr_fork_size(
84 struct xfs_inode_log_item *iip,
85 int *nvecs,
86 int *nbytes)
87{
88 struct xfs_inode *ip = iip->ili_inode;
89
90 switch (ip->i_d.di_aformat) {
91 case XFS_DINODE_FMT_EXTENTS:
92 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
93 ip->i_d.di_anextents > 0 &&
94 ip->i_afp->if_bytes > 0) {
95 /* worst case, doesn't subtract unused space */
96 *nbytes += XFS_IFORK_ASIZE(ip);
97 *nvecs += 1;
98 }
99 break;
100 case XFS_DINODE_FMT_BTREE:
101 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
102 ip->i_afp->if_broot_bytes > 0) {
103 *nbytes += ip->i_afp->if_broot_bytes;
104 *nvecs += 1;
105 }
106 break;
107 case XFS_DINODE_FMT_LOCAL:
108 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
109 ip->i_afp->if_bytes > 0) {
110 *nbytes += roundup(ip->i_afp->if_bytes, 4);
111 *nvecs += 1;
112 }
113 break;
114 default:
115 ASSERT(0);
116 break;
117 }
118}
119
120/*
121 * This returns the number of iovecs needed to log the given inode item.
122 *
123 * We need one iovec for the inode log format structure, one for the
124 * inode core, and possibly one for the inode data/extents/b-tree root
125 * and one for the inode attribute data/extents/b-tree root.
126 */
127STATIC void
128xfs_inode_item_size(
129 struct xfs_log_item *lip,
130 int *nvecs,
131 int *nbytes)
132{
133 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
134 struct xfs_inode *ip = iip->ili_inode;
135
136 *nvecs += 2;
137 *nbytes += sizeof(struct xfs_inode_log_format) +
138 xfs_log_dinode_size(ip->i_d.di_version);
139
140 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
141 if (XFS_IFORK_Q(ip))
142 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
143}
144
145STATIC void
146xfs_inode_item_format_data_fork(
147 struct xfs_inode_log_item *iip,
148 struct xfs_inode_log_format *ilf,
149 struct xfs_log_vec *lv,
150 struct xfs_log_iovec **vecp)
151{
152 struct xfs_inode *ip = iip->ili_inode;
153 size_t data_bytes;
154
155 switch (ip->i_d.di_format) {
156 case XFS_DINODE_FMT_EXTENTS:
157 iip->ili_fields &=
158 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
159 XFS_ILOG_DEV | XFS_ILOG_UUID);
160
161 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
162 ip->i_d.di_nextents > 0 &&
163 ip->i_df.if_bytes > 0) {
164 struct xfs_bmbt_rec *p;
165
166 ASSERT(ip->i_df.if_u1.if_extents != NULL);
167 ASSERT(xfs_iext_count(&ip->i_df) > 0);
168
169 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
170 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
171 xlog_finish_iovec(lv, *vecp, data_bytes);
172
173 ASSERT(data_bytes <= ip->i_df.if_bytes);
174
175 ilf->ilf_dsize = data_bytes;
176 ilf->ilf_size++;
177 } else {
178 iip->ili_fields &= ~XFS_ILOG_DEXT;
179 }
180 break;
181 case XFS_DINODE_FMT_BTREE:
182 iip->ili_fields &=
183 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
184 XFS_ILOG_DEV | XFS_ILOG_UUID);
185
186 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
187 ip->i_df.if_broot_bytes > 0) {
188 ASSERT(ip->i_df.if_broot != NULL);
189 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
190 ip->i_df.if_broot,
191 ip->i_df.if_broot_bytes);
192 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
193 ilf->ilf_size++;
194 } else {
195 ASSERT(!(iip->ili_fields &
196 XFS_ILOG_DBROOT));
197 iip->ili_fields &= ~XFS_ILOG_DBROOT;
198 }
199 break;
200 case XFS_DINODE_FMT_LOCAL:
201 iip->ili_fields &=
202 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
203 XFS_ILOG_DEV | XFS_ILOG_UUID);
204 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
205 ip->i_df.if_bytes > 0) {
206 /*
207 * Round i_bytes up to a word boundary.
208 * The underlying memory is guaranteed to
209 * to be there by xfs_idata_realloc().
210 */
211 data_bytes = roundup(ip->i_df.if_bytes, 4);
212 ASSERT(ip->i_df.if_real_bytes == 0 ||
213 ip->i_df.if_real_bytes >= data_bytes);
214 ASSERT(ip->i_df.if_u1.if_data != NULL);
215 ASSERT(ip->i_d.di_size > 0);
216 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
217 ip->i_df.if_u1.if_data, data_bytes);
218 ilf->ilf_dsize = (unsigned)data_bytes;
219 ilf->ilf_size++;
220 } else {
221 iip->ili_fields &= ~XFS_ILOG_DDATA;
222 }
223 break;
224 case XFS_DINODE_FMT_DEV:
225 iip->ili_fields &=
226 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
227 XFS_ILOG_DEXT | XFS_ILOG_UUID);
228 if (iip->ili_fields & XFS_ILOG_DEV)
229 ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
230 break;
231 case XFS_DINODE_FMT_UUID:
232 iip->ili_fields &=
233 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
234 XFS_ILOG_DEXT | XFS_ILOG_DEV);
235 if (iip->ili_fields & XFS_ILOG_UUID)
236 ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
237 break;
238 default:
239 ASSERT(0);
240 break;
241 }
242}
243
244STATIC void
245xfs_inode_item_format_attr_fork(
246 struct xfs_inode_log_item *iip,
247 struct xfs_inode_log_format *ilf,
248 struct xfs_log_vec *lv,
249 struct xfs_log_iovec **vecp)
250{
251 struct xfs_inode *ip = iip->ili_inode;
252 size_t data_bytes;
253
254 switch (ip->i_d.di_aformat) {
255 case XFS_DINODE_FMT_EXTENTS:
256 iip->ili_fields &=
257 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
258
259 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
260 ip->i_d.di_anextents > 0 &&
261 ip->i_afp->if_bytes > 0) {
262 struct xfs_bmbt_rec *p;
263
264 ASSERT(xfs_iext_count(ip->i_afp) ==
265 ip->i_d.di_anextents);
266 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
267
268 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
269 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
270 xlog_finish_iovec(lv, *vecp, data_bytes);
271
272 ilf->ilf_asize = data_bytes;
273 ilf->ilf_size++;
274 } else {
275 iip->ili_fields &= ~XFS_ILOG_AEXT;
276 }
277 break;
278 case XFS_DINODE_FMT_BTREE:
279 iip->ili_fields &=
280 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
281
282 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
283 ip->i_afp->if_broot_bytes > 0) {
284 ASSERT(ip->i_afp->if_broot != NULL);
285
286 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
287 ip->i_afp->if_broot,
288 ip->i_afp->if_broot_bytes);
289 ilf->ilf_asize = ip->i_afp->if_broot_bytes;
290 ilf->ilf_size++;
291 } else {
292 iip->ili_fields &= ~XFS_ILOG_ABROOT;
293 }
294 break;
295 case XFS_DINODE_FMT_LOCAL:
296 iip->ili_fields &=
297 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
298
299 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
300 ip->i_afp->if_bytes > 0) {
301 /*
302 * Round i_bytes up to a word boundary.
303 * The underlying memory is guaranteed to
304 * to be there by xfs_idata_realloc().
305 */
306 data_bytes = roundup(ip->i_afp->if_bytes, 4);
307 ASSERT(ip->i_afp->if_real_bytes == 0 ||
308 ip->i_afp->if_real_bytes >= data_bytes);
309 ASSERT(ip->i_afp->if_u1.if_data != NULL);
310 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
311 ip->i_afp->if_u1.if_data,
312 data_bytes);
313 ilf->ilf_asize = (unsigned)data_bytes;
314 ilf->ilf_size++;
315 } else {
316 iip->ili_fields &= ~XFS_ILOG_ADATA;
317 }
318 break;
319 default:
320 ASSERT(0);
321 break;
322 }
323}
324
325static void
326xfs_inode_to_log_dinode(
327 struct xfs_inode *ip,
328 struct xfs_log_dinode *to,
329 xfs_lsn_t lsn)
330{
331 struct xfs_icdinode *from = &ip->i_d;
332 struct inode *inode = VFS_I(ip);
333
334 to->di_magic = XFS_DINODE_MAGIC;
335
336 to->di_version = from->di_version;
337 to->di_format = from->di_format;
338 to->di_uid = from->di_uid;
339 to->di_gid = from->di_gid;
340 to->di_projid_lo = from->di_projid_lo;
341 to->di_projid_hi = from->di_projid_hi;
342
343 memset(to->di_pad, 0, sizeof(to->di_pad));
344 memset(to->di_pad3, 0, sizeof(to->di_pad3));
345 to->di_atime.t_sec = inode->i_atime.tv_sec;
346 to->di_atime.t_nsec = inode->i_atime.tv_nsec;
347 to->di_mtime.t_sec = inode->i_mtime.tv_sec;
348 to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
349 to->di_ctime.t_sec = inode->i_ctime.tv_sec;
350 to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
351 to->di_nlink = inode->i_nlink;
352 to->di_gen = inode->i_generation;
353 to->di_mode = inode->i_mode;
354
355 to->di_size = from->di_size;
356 to->di_nblocks = from->di_nblocks;
357 to->di_extsize = from->di_extsize;
358 to->di_nextents = from->di_nextents;
359 to->di_anextents = from->di_anextents;
360 to->di_forkoff = from->di_forkoff;
361 to->di_aformat = from->di_aformat;
362 to->di_dmevmask = from->di_dmevmask;
363 to->di_dmstate = from->di_dmstate;
364 to->di_flags = from->di_flags;
365
366 if (from->di_version == 3) {
367 to->di_changecount = inode->i_version;
368 to->di_crtime.t_sec = from->di_crtime.t_sec;
369 to->di_crtime.t_nsec = from->di_crtime.t_nsec;
370 to->di_flags2 = from->di_flags2;
371 to->di_cowextsize = from->di_cowextsize;
372 to->di_ino = ip->i_ino;
373 to->di_lsn = lsn;
374 memset(to->di_pad2, 0, sizeof(to->di_pad2));
375 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
376 to->di_flushiter = 0;
377 } else {
378 to->di_flushiter = from->di_flushiter;
379 }
380}
381
382/*
383 * Format the inode core. Current timestamp data is only in the VFS inode
384 * fields, so we need to grab them from there. Hence rather than just copying
385 * the XFS inode core structure, format the fields directly into the iovec.
386 */
387static void
388xfs_inode_item_format_core(
389 struct xfs_inode *ip,
390 struct xfs_log_vec *lv,
391 struct xfs_log_iovec **vecp)
392{
393 struct xfs_log_dinode *dic;
394
395 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
396 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
397 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_d.di_version));
398}
399
400/*
401 * This is called to fill in the vector of log iovecs for the given inode
402 * log item. It fills the first item with an inode log format structure,
403 * the second with the on-disk inode structure, and a possible third and/or
404 * fourth with the inode data/extents/b-tree root and inode attributes
405 * data/extents/b-tree root.
406 */
407STATIC void
408xfs_inode_item_format(
409 struct xfs_log_item *lip,
410 struct xfs_log_vec *lv)
411{
412 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
413 struct xfs_inode *ip = iip->ili_inode;
414 struct xfs_inode_log_format *ilf;
415 struct xfs_log_iovec *vecp = NULL;
416
417 ASSERT(ip->i_d.di_version > 1);
418
419 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
420 ilf->ilf_type = XFS_LI_INODE;
421 ilf->ilf_ino = ip->i_ino;
422 ilf->ilf_blkno = ip->i_imap.im_blkno;
423 ilf->ilf_len = ip->i_imap.im_len;
424 ilf->ilf_boffset = ip->i_imap.im_boffset;
425 ilf->ilf_fields = XFS_ILOG_CORE;
426 ilf->ilf_size = 2; /* format + core */
427 xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
428
429 xfs_inode_item_format_core(ip, lv, &vecp);
430 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
431 if (XFS_IFORK_Q(ip)) {
432 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
433 } else {
434 iip->ili_fields &=
435 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
436 }
437
438 /* update the format with the exact fields we actually logged */
439 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
440}
441
442/*
443 * This is called to pin the inode associated with the inode log
444 * item in memory so it cannot be written out.
445 */
446STATIC void
447xfs_inode_item_pin(
448 struct xfs_log_item *lip)
449{
450 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
451
452 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
453
454 trace_xfs_inode_pin(ip, _RET_IP_);
455 atomic_inc(&ip->i_pincount);
456}
457
458
459/*
460 * This is called to unpin the inode associated with the inode log
461 * item which was previously pinned with a call to xfs_inode_item_pin().
462 *
463 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
464 */
465STATIC void
466xfs_inode_item_unpin(
467 struct xfs_log_item *lip,
468 int remove)
469{
470 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
471
472 trace_xfs_inode_unpin(ip, _RET_IP_);
473 ASSERT(atomic_read(&ip->i_pincount) > 0);
474 if (atomic_dec_and_test(&ip->i_pincount))
475 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
476}
477
478STATIC uint
479xfs_inode_item_push(
480 struct xfs_log_item *lip,
481 struct list_head *buffer_list)
482 __releases(&lip->li_ailp->xa_lock)
483 __acquires(&lip->li_ailp->xa_lock)
484{
485 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
486 struct xfs_inode *ip = iip->ili_inode;
487 struct xfs_buf *bp = NULL;
488 uint rval = XFS_ITEM_SUCCESS;
489 int error;
490
491 if (xfs_ipincount(ip) > 0)
492 return XFS_ITEM_PINNED;
493
494 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
495 return XFS_ITEM_LOCKED;
496
497 /*
498 * Re-check the pincount now that we stabilized the value by
499 * taking the ilock.
500 */
501 if (xfs_ipincount(ip) > 0) {
502 rval = XFS_ITEM_PINNED;
503 goto out_unlock;
504 }
505
506 /*
507 * Stale inode items should force out the iclog.
508 */
509 if (ip->i_flags & XFS_ISTALE) {
510 rval = XFS_ITEM_PINNED;
511 goto out_unlock;
512 }
513
514 /*
515 * Someone else is already flushing the inode. Nothing we can do
516 * here but wait for the flush to finish and remove the item from
517 * the AIL.
518 */
519 if (!xfs_iflock_nowait(ip)) {
520 rval = XFS_ITEM_FLUSHING;
521 goto out_unlock;
522 }
523
524 ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
525 ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
526
527 spin_unlock(&lip->li_ailp->xa_lock);
528
529 error = xfs_iflush(ip, &bp);
530 if (!error) {
531 if (!xfs_buf_delwri_queue(bp, buffer_list))
532 rval = XFS_ITEM_FLUSHING;
533 xfs_buf_relse(bp);
534 }
535
536 spin_lock(&lip->li_ailp->xa_lock);
537out_unlock:
538 xfs_iunlock(ip, XFS_ILOCK_SHARED);
539 return rval;
540}
541
542/*
543 * Unlock the inode associated with the inode log item.
544 * Clear the fields of the inode and inode log item that
545 * are specific to the current transaction. If the
546 * hold flags is set, do not unlock the inode.
547 */
548STATIC void
549xfs_inode_item_unlock(
550 struct xfs_log_item *lip)
551{
552 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
553 struct xfs_inode *ip = iip->ili_inode;
554 unsigned short lock_flags;
555
556 ASSERT(ip->i_itemp != NULL);
557 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
558
559 lock_flags = iip->ili_lock_flags;
560 iip->ili_lock_flags = 0;
561 if (lock_flags)
562 xfs_iunlock(ip, lock_flags);
563}
564
565/*
566 * This is called to find out where the oldest active copy of the inode log
567 * item in the on disk log resides now that the last log write of it completed
568 * at the given lsn. Since we always re-log all dirty data in an inode, the
569 * latest copy in the on disk log is the only one that matters. Therefore,
570 * simply return the given lsn.
571 *
572 * If the inode has been marked stale because the cluster is being freed, we
573 * don't want to (re-)insert this inode into the AIL. There is a race condition
574 * where the cluster buffer may be unpinned before the inode is inserted into
575 * the AIL during transaction committed processing. If the buffer is unpinned
576 * before the inode item has been committed and inserted, then it is possible
577 * for the buffer to be written and IO completes before the inode is inserted
578 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
579 * AIL which will never get removed. It will, however, get reclaimed which
580 * triggers an assert in xfs_inode_free() complaining about freein an inode
581 * still in the AIL.
582 *
583 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
584 * transaction committed code knows that it does not need to do any further
585 * processing on the item.
586 */
587STATIC xfs_lsn_t
588xfs_inode_item_committed(
589 struct xfs_log_item *lip,
590 xfs_lsn_t lsn)
591{
592 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
593 struct xfs_inode *ip = iip->ili_inode;
594
595 if (xfs_iflags_test(ip, XFS_ISTALE)) {
596 xfs_inode_item_unpin(lip, 0);
597 return -1;
598 }
599 return lsn;
600}
601
602/*
603 * XXX rcc - this one really has to do something. Probably needs
604 * to stamp in a new field in the incore inode.
605 */
606STATIC void
607xfs_inode_item_committing(
608 struct xfs_log_item *lip,
609 xfs_lsn_t lsn)
610{
611 INODE_ITEM(lip)->ili_last_lsn = lsn;
612}
613
614/*
615 * This is the ops vector shared by all buf log items.
616 */
617static const struct xfs_item_ops xfs_inode_item_ops = {
618 .iop_size = xfs_inode_item_size,
619 .iop_format = xfs_inode_item_format,
620 .iop_pin = xfs_inode_item_pin,
621 .iop_unpin = xfs_inode_item_unpin,
622 .iop_unlock = xfs_inode_item_unlock,
623 .iop_committed = xfs_inode_item_committed,
624 .iop_push = xfs_inode_item_push,
625 .iop_committing = xfs_inode_item_committing
626};
627
628
629/*
630 * Initialize the inode log item for a newly allocated (in-core) inode.
631 */
632void
633xfs_inode_item_init(
634 struct xfs_inode *ip,
635 struct xfs_mount *mp)
636{
637 struct xfs_inode_log_item *iip;
638
639 ASSERT(ip->i_itemp == NULL);
640 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
641
642 iip->ili_inode = ip;
643 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
644 &xfs_inode_item_ops);
645}
646
647/*
648 * Free the inode log item and any memory hanging off of it.
649 */
650void
651xfs_inode_item_destroy(
652 xfs_inode_t *ip)
653{
654 kmem_free(ip->i_itemp->ili_item.li_lv_shadow);
655 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
656}
657
658
659/*
660 * This is the inode flushing I/O completion routine. It is called
661 * from interrupt level when the buffer containing the inode is
662 * flushed to disk. It is responsible for removing the inode item
663 * from the AIL if it has not been re-logged, and unlocking the inode's
664 * flush lock.
665 *
666 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
667 * list for other inodes that will run this function. We remove them from the
668 * buffer list so we can process all the inode IO completions in one AIL lock
669 * traversal.
670 */
671void
672xfs_iflush_done(
673 struct xfs_buf *bp,
674 struct xfs_log_item *lip)
675{
676 struct xfs_inode_log_item *iip;
677 struct xfs_log_item *blip;
678 struct xfs_log_item *next;
679 struct xfs_log_item *prev;
680 struct xfs_ail *ailp = lip->li_ailp;
681 int need_ail = 0;
682
683 /*
684 * Scan the buffer IO completions for other inodes being completed and
685 * attach them to the current inode log item.
686 */
687 blip = bp->b_fspriv;
688 prev = NULL;
689 while (blip != NULL) {
690 if (blip->li_cb != xfs_iflush_done) {
691 prev = blip;
692 blip = blip->li_bio_list;
693 continue;
694 }
695
696 /* remove from list */
697 next = blip->li_bio_list;
698 if (!prev) {
699 bp->b_fspriv = next;
700 } else {
701 prev->li_bio_list = next;
702 }
703
704 /* add to current list */
705 blip->li_bio_list = lip->li_bio_list;
706 lip->li_bio_list = blip;
707
708 /*
709 * while we have the item, do the unlocked check for needing
710 * the AIL lock.
711 */
712 iip = INODE_ITEM(blip);
713 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
714 need_ail++;
715
716 blip = next;
717 }
718
719 /* make sure we capture the state of the initial inode. */
720 iip = INODE_ITEM(lip);
721 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
722 need_ail++;
723
724 /*
725 * We only want to pull the item from the AIL if it is
726 * actually there and its location in the log has not
727 * changed since we started the flush. Thus, we only bother
728 * if the ili_logged flag is set and the inode's lsn has not
729 * changed. First we check the lsn outside
730 * the lock since it's cheaper, and then we recheck while
731 * holding the lock before removing the inode from the AIL.
732 */
733 if (need_ail) {
734 struct xfs_log_item *log_items[need_ail];
735 int i = 0;
736 spin_lock(&ailp->xa_lock);
737 for (blip = lip; blip; blip = blip->li_bio_list) {
738 iip = INODE_ITEM(blip);
739 if (iip->ili_logged &&
740 blip->li_lsn == iip->ili_flush_lsn) {
741 log_items[i++] = blip;
742 }
743 ASSERT(i <= need_ail);
744 }
745 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
746 xfs_trans_ail_delete_bulk(ailp, log_items, i,
747 SHUTDOWN_CORRUPT_INCORE);
748 }
749
750
751 /*
752 * clean up and unlock the flush lock now we are done. We can clear the
753 * ili_last_fields bits now that we know that the data corresponding to
754 * them is safely on disk.
755 */
756 for (blip = lip; blip; blip = next) {
757 next = blip->li_bio_list;
758 blip->li_bio_list = NULL;
759
760 iip = INODE_ITEM(blip);
761 iip->ili_logged = 0;
762 iip->ili_last_fields = 0;
763 xfs_ifunlock(iip->ili_inode);
764 }
765}
766
767/*
768 * This is the inode flushing abort routine. It is called from xfs_iflush when
769 * the filesystem is shutting down to clean up the inode state. It is
770 * responsible for removing the inode item from the AIL if it has not been
771 * re-logged, and unlocking the inode's flush lock.
772 */
773void
774xfs_iflush_abort(
775 xfs_inode_t *ip,
776 bool stale)
777{
778 xfs_inode_log_item_t *iip = ip->i_itemp;
779
780 if (iip) {
781 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
782 xfs_trans_ail_remove(&iip->ili_item,
783 stale ? SHUTDOWN_LOG_IO_ERROR :
784 SHUTDOWN_CORRUPT_INCORE);
785 }
786 iip->ili_logged = 0;
787 /*
788 * Clear the ili_last_fields bits now that we know that the
789 * data corresponding to them is safely on disk.
790 */
791 iip->ili_last_fields = 0;
792 /*
793 * Clear the inode logging fields so no more flushes are
794 * attempted.
795 */
796 iip->ili_fields = 0;
797 iip->ili_fsync_fields = 0;
798 }
799 /*
800 * Release the inode's flush lock since we're done with it.
801 */
802 xfs_ifunlock(ip);
803}
804
805void
806xfs_istale_done(
807 struct xfs_buf *bp,
808 struct xfs_log_item *lip)
809{
810 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
811}
812
813/*
814 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
815 * (which can have different field alignments) to the native version
816 */
817int
818xfs_inode_item_format_convert(
819 xfs_log_iovec_t *buf,
820 xfs_inode_log_format_t *in_f)
821{
822 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
823 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
824
825 in_f->ilf_type = in_f32->ilf_type;
826 in_f->ilf_size = in_f32->ilf_size;
827 in_f->ilf_fields = in_f32->ilf_fields;
828 in_f->ilf_asize = in_f32->ilf_asize;
829 in_f->ilf_dsize = in_f32->ilf_dsize;
830 in_f->ilf_ino = in_f32->ilf_ino;
831 /* copy biggest field of ilf_u */
832 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
833 in_f32->ilf_u.ilfu_uuid.__u_bits,
834 sizeof(uuid_t));
835 in_f->ilf_blkno = in_f32->ilf_blkno;
836 in_f->ilf_len = in_f32->ilf_len;
837 in_f->ilf_boffset = in_f32->ilf_boffset;
838 return 0;
839 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
840 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
841
842 in_f->ilf_type = in_f64->ilf_type;
843 in_f->ilf_size = in_f64->ilf_size;
844 in_f->ilf_fields = in_f64->ilf_fields;
845 in_f->ilf_asize = in_f64->ilf_asize;
846 in_f->ilf_dsize = in_f64->ilf_dsize;
847 in_f->ilf_ino = in_f64->ilf_ino;
848 /* copy biggest field of ilf_u */
849 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
850 in_f64->ilf_u.ilfu_uuid.__u_bits,
851 sizeof(uuid_t));
852 in_f->ilf_blkno = in_f64->ilf_blkno;
853 in_f->ilf_len = in_f64->ilf_len;
854 in_f->ilf_boffset = in_f64->ilf_boffset;
855 return 0;
856 }
857 return -EFSCORRUPTED;
858}