Loading...
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_buf_item.h"
31#include "xfs_log.h"
32
33#include <linux/iversion.h>
34
35kmem_zone_t *xfs_ili_zone; /* inode log item zone */
36
37static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
38{
39 return container_of(lip, struct xfs_inode_log_item, ili_item);
40}
41
42STATIC void
43xfs_inode_item_data_fork_size(
44 struct xfs_inode_log_item *iip,
45 int *nvecs,
46 int *nbytes)
47{
48 struct xfs_inode *ip = iip->ili_inode;
49
50 switch (ip->i_d.di_format) {
51 case XFS_DINODE_FMT_EXTENTS:
52 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
53 ip->i_d.di_nextents > 0 &&
54 ip->i_df.if_bytes > 0) {
55 /* worst case, doesn't subtract delalloc extents */
56 *nbytes += XFS_IFORK_DSIZE(ip);
57 *nvecs += 1;
58 }
59 break;
60 case XFS_DINODE_FMT_BTREE:
61 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
62 ip->i_df.if_broot_bytes > 0) {
63 *nbytes += ip->i_df.if_broot_bytes;
64 *nvecs += 1;
65 }
66 break;
67 case XFS_DINODE_FMT_LOCAL:
68 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
69 ip->i_df.if_bytes > 0) {
70 *nbytes += roundup(ip->i_df.if_bytes, 4);
71 *nvecs += 1;
72 }
73 break;
74
75 case XFS_DINODE_FMT_DEV:
76 break;
77 default:
78 ASSERT(0);
79 break;
80 }
81}
82
83STATIC void
84xfs_inode_item_attr_fork_size(
85 struct xfs_inode_log_item *iip,
86 int *nvecs,
87 int *nbytes)
88{
89 struct xfs_inode *ip = iip->ili_inode;
90
91 switch (ip->i_d.di_aformat) {
92 case XFS_DINODE_FMT_EXTENTS:
93 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
94 ip->i_d.di_anextents > 0 &&
95 ip->i_afp->if_bytes > 0) {
96 /* worst case, doesn't subtract unused space */
97 *nbytes += XFS_IFORK_ASIZE(ip);
98 *nvecs += 1;
99 }
100 break;
101 case XFS_DINODE_FMT_BTREE:
102 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
103 ip->i_afp->if_broot_bytes > 0) {
104 *nbytes += ip->i_afp->if_broot_bytes;
105 *nvecs += 1;
106 }
107 break;
108 case XFS_DINODE_FMT_LOCAL:
109 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
110 ip->i_afp->if_bytes > 0) {
111 *nbytes += roundup(ip->i_afp->if_bytes, 4);
112 *nvecs += 1;
113 }
114 break;
115 default:
116 ASSERT(0);
117 break;
118 }
119}
120
121/*
122 * This returns the number of iovecs needed to log the given inode item.
123 *
124 * We need one iovec for the inode log format structure, one for the
125 * inode core, and possibly one for the inode data/extents/b-tree root
126 * and one for the inode attribute data/extents/b-tree root.
127 */
128STATIC void
129xfs_inode_item_size(
130 struct xfs_log_item *lip,
131 int *nvecs,
132 int *nbytes)
133{
134 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
135 struct xfs_inode *ip = iip->ili_inode;
136
137 *nvecs += 2;
138 *nbytes += sizeof(struct xfs_inode_log_format) +
139 xfs_log_dinode_size(ip->i_d.di_version);
140
141 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
142 if (XFS_IFORK_Q(ip))
143 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
144}
145
146STATIC void
147xfs_inode_item_format_data_fork(
148 struct xfs_inode_log_item *iip,
149 struct xfs_inode_log_format *ilf,
150 struct xfs_log_vec *lv,
151 struct xfs_log_iovec **vecp)
152{
153 struct xfs_inode *ip = iip->ili_inode;
154 size_t data_bytes;
155
156 switch (ip->i_d.di_format) {
157 case XFS_DINODE_FMT_EXTENTS:
158 iip->ili_fields &=
159 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
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(xfs_iext_count(&ip->i_df) > 0);
167
168 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
169 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
170 xlog_finish_iovec(lv, *vecp, data_bytes);
171
172 ASSERT(data_bytes <= ip->i_df.if_bytes);
173
174 ilf->ilf_dsize = data_bytes;
175 ilf->ilf_size++;
176 } else {
177 iip->ili_fields &= ~XFS_ILOG_DEXT;
178 }
179 break;
180 case XFS_DINODE_FMT_BTREE:
181 iip->ili_fields &=
182 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
183
184 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
185 ip->i_df.if_broot_bytes > 0) {
186 ASSERT(ip->i_df.if_broot != NULL);
187 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
188 ip->i_df.if_broot,
189 ip->i_df.if_broot_bytes);
190 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
191 ilf->ilf_size++;
192 } else {
193 ASSERT(!(iip->ili_fields &
194 XFS_ILOG_DBROOT));
195 iip->ili_fields &= ~XFS_ILOG_DBROOT;
196 }
197 break;
198 case XFS_DINODE_FMT_LOCAL:
199 iip->ili_fields &=
200 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
201 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
202 ip->i_df.if_bytes > 0) {
203 /*
204 * Round i_bytes up to a word boundary.
205 * The underlying memory is guaranteed to
206 * to be there by xfs_idata_realloc().
207 */
208 data_bytes = roundup(ip->i_df.if_bytes, 4);
209 ASSERT(ip->i_df.if_real_bytes == 0 ||
210 ip->i_df.if_real_bytes >= data_bytes);
211 ASSERT(ip->i_df.if_u1.if_data != NULL);
212 ASSERT(ip->i_d.di_size > 0);
213 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
214 ip->i_df.if_u1.if_data, data_bytes);
215 ilf->ilf_dsize = (unsigned)data_bytes;
216 ilf->ilf_size++;
217 } else {
218 iip->ili_fields &= ~XFS_ILOG_DDATA;
219 }
220 break;
221 case XFS_DINODE_FMT_DEV:
222 iip->ili_fields &=
223 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
224 if (iip->ili_fields & XFS_ILOG_DEV)
225 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
226 break;
227 default:
228 ASSERT(0);
229 break;
230 }
231}
232
233STATIC void
234xfs_inode_item_format_attr_fork(
235 struct xfs_inode_log_item *iip,
236 struct xfs_inode_log_format *ilf,
237 struct xfs_log_vec *lv,
238 struct xfs_log_iovec **vecp)
239{
240 struct xfs_inode *ip = iip->ili_inode;
241 size_t data_bytes;
242
243 switch (ip->i_d.di_aformat) {
244 case XFS_DINODE_FMT_EXTENTS:
245 iip->ili_fields &=
246 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
247
248 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
249 ip->i_d.di_anextents > 0 &&
250 ip->i_afp->if_bytes > 0) {
251 struct xfs_bmbt_rec *p;
252
253 ASSERT(xfs_iext_count(ip->i_afp) ==
254 ip->i_d.di_anextents);
255
256 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
257 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
258 xlog_finish_iovec(lv, *vecp, data_bytes);
259
260 ilf->ilf_asize = data_bytes;
261 ilf->ilf_size++;
262 } else {
263 iip->ili_fields &= ~XFS_ILOG_AEXT;
264 }
265 break;
266 case XFS_DINODE_FMT_BTREE:
267 iip->ili_fields &=
268 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
269
270 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
271 ip->i_afp->if_broot_bytes > 0) {
272 ASSERT(ip->i_afp->if_broot != NULL);
273
274 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
275 ip->i_afp->if_broot,
276 ip->i_afp->if_broot_bytes);
277 ilf->ilf_asize = ip->i_afp->if_broot_bytes;
278 ilf->ilf_size++;
279 } else {
280 iip->ili_fields &= ~XFS_ILOG_ABROOT;
281 }
282 break;
283 case XFS_DINODE_FMT_LOCAL:
284 iip->ili_fields &=
285 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
286
287 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
288 ip->i_afp->if_bytes > 0) {
289 /*
290 * Round i_bytes up to a word boundary.
291 * The underlying memory is guaranteed to
292 * to be there by xfs_idata_realloc().
293 */
294 data_bytes = roundup(ip->i_afp->if_bytes, 4);
295 ASSERT(ip->i_afp->if_real_bytes == 0 ||
296 ip->i_afp->if_real_bytes >= data_bytes);
297 ASSERT(ip->i_afp->if_u1.if_data != NULL);
298 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
299 ip->i_afp->if_u1.if_data,
300 data_bytes);
301 ilf->ilf_asize = (unsigned)data_bytes;
302 ilf->ilf_size++;
303 } else {
304 iip->ili_fields &= ~XFS_ILOG_ADATA;
305 }
306 break;
307 default:
308 ASSERT(0);
309 break;
310 }
311}
312
313static void
314xfs_inode_to_log_dinode(
315 struct xfs_inode *ip,
316 struct xfs_log_dinode *to,
317 xfs_lsn_t lsn)
318{
319 struct xfs_icdinode *from = &ip->i_d;
320 struct inode *inode = VFS_I(ip);
321
322 to->di_magic = XFS_DINODE_MAGIC;
323
324 to->di_version = from->di_version;
325 to->di_format = from->di_format;
326 to->di_uid = from->di_uid;
327 to->di_gid = from->di_gid;
328 to->di_projid_lo = from->di_projid_lo;
329 to->di_projid_hi = from->di_projid_hi;
330
331 memset(to->di_pad, 0, sizeof(to->di_pad));
332 memset(to->di_pad3, 0, sizeof(to->di_pad3));
333 to->di_atime.t_sec = inode->i_atime.tv_sec;
334 to->di_atime.t_nsec = inode->i_atime.tv_nsec;
335 to->di_mtime.t_sec = inode->i_mtime.tv_sec;
336 to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
337 to->di_ctime.t_sec = inode->i_ctime.tv_sec;
338 to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
339 to->di_nlink = inode->i_nlink;
340 to->di_gen = inode->i_generation;
341 to->di_mode = inode->i_mode;
342
343 to->di_size = from->di_size;
344 to->di_nblocks = from->di_nblocks;
345 to->di_extsize = from->di_extsize;
346 to->di_nextents = from->di_nextents;
347 to->di_anextents = from->di_anextents;
348 to->di_forkoff = from->di_forkoff;
349 to->di_aformat = from->di_aformat;
350 to->di_dmevmask = from->di_dmevmask;
351 to->di_dmstate = from->di_dmstate;
352 to->di_flags = from->di_flags;
353
354 /* log a dummy value to ensure log structure is fully initialised */
355 to->di_next_unlinked = NULLAGINO;
356
357 if (from->di_version == 3) {
358 to->di_changecount = inode_peek_iversion(inode);
359 to->di_crtime.t_sec = from->di_crtime.t_sec;
360 to->di_crtime.t_nsec = from->di_crtime.t_nsec;
361 to->di_flags2 = from->di_flags2;
362 to->di_cowextsize = from->di_cowextsize;
363 to->di_ino = ip->i_ino;
364 to->di_lsn = lsn;
365 memset(to->di_pad2, 0, sizeof(to->di_pad2));
366 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
367 to->di_flushiter = 0;
368 } else {
369 to->di_flushiter = from->di_flushiter;
370 }
371}
372
373/*
374 * Format the inode core. Current timestamp data is only in the VFS inode
375 * fields, so we need to grab them from there. Hence rather than just copying
376 * the XFS inode core structure, format the fields directly into the iovec.
377 */
378static void
379xfs_inode_item_format_core(
380 struct xfs_inode *ip,
381 struct xfs_log_vec *lv,
382 struct xfs_log_iovec **vecp)
383{
384 struct xfs_log_dinode *dic;
385
386 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
387 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
388 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_d.di_version));
389}
390
391/*
392 * This is called to fill in the vector of log iovecs for the given inode
393 * log item. It fills the first item with an inode log format structure,
394 * the second with the on-disk inode structure, and a possible third and/or
395 * fourth with the inode data/extents/b-tree root and inode attributes
396 * data/extents/b-tree root.
397 *
398 * Note: Always use the 64 bit inode log format structure so we don't
399 * leave an uninitialised hole in the format item on 64 bit systems. Log
400 * recovery on 32 bit systems handles this just fine, so there's no reason
401 * for not using an initialising the properly padded structure all the time.
402 */
403STATIC void
404xfs_inode_item_format(
405 struct xfs_log_item *lip,
406 struct xfs_log_vec *lv)
407{
408 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
409 struct xfs_inode *ip = iip->ili_inode;
410 struct xfs_log_iovec *vecp = NULL;
411 struct xfs_inode_log_format *ilf;
412
413 ASSERT(ip->i_d.di_version > 1);
414
415 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
416 ilf->ilf_type = XFS_LI_INODE;
417 ilf->ilf_ino = ip->i_ino;
418 ilf->ilf_blkno = ip->i_imap.im_blkno;
419 ilf->ilf_len = ip->i_imap.im_len;
420 ilf->ilf_boffset = ip->i_imap.im_boffset;
421 ilf->ilf_fields = XFS_ILOG_CORE;
422 ilf->ilf_size = 2; /* format + core */
423
424 /*
425 * make sure we don't leak uninitialised data into the log in the case
426 * when we don't log every field in the inode.
427 */
428 ilf->ilf_dsize = 0;
429 ilf->ilf_asize = 0;
430 ilf->ilf_pad = 0;
431 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
432
433 xlog_finish_iovec(lv, vecp, sizeof(*ilf));
434
435 xfs_inode_item_format_core(ip, lv, &vecp);
436 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
437 if (XFS_IFORK_Q(ip)) {
438 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
439 } else {
440 iip->ili_fields &=
441 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
442 }
443
444 /* update the format with the exact fields we actually logged */
445 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
446}
447
448/*
449 * This is called to pin the inode associated with the inode log
450 * item in memory so it cannot be written out.
451 */
452STATIC void
453xfs_inode_item_pin(
454 struct xfs_log_item *lip)
455{
456 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
457
458 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
459
460 trace_xfs_inode_pin(ip, _RET_IP_);
461 atomic_inc(&ip->i_pincount);
462}
463
464
465/*
466 * This is called to unpin the inode associated with the inode log
467 * item which was previously pinned with a call to xfs_inode_item_pin().
468 *
469 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
470 */
471STATIC void
472xfs_inode_item_unpin(
473 struct xfs_log_item *lip,
474 int remove)
475{
476 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
477
478 trace_xfs_inode_unpin(ip, _RET_IP_);
479 ASSERT(atomic_read(&ip->i_pincount) > 0);
480 if (atomic_dec_and_test(&ip->i_pincount))
481 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
482}
483
484/*
485 * Callback used to mark a buffer with XFS_LI_FAILED when items in the buffer
486 * have been failed during writeback
487 *
488 * This informs the AIL that the inode is already flush locked on the next push,
489 * and acquires a hold on the buffer to ensure that it isn't reclaimed before
490 * dirty data makes it to disk.
491 */
492STATIC void
493xfs_inode_item_error(
494 struct xfs_log_item *lip,
495 struct xfs_buf *bp)
496{
497 ASSERT(xfs_isiflocked(INODE_ITEM(lip)->ili_inode));
498 xfs_set_li_failed(lip, bp);
499}
500
501STATIC uint
502xfs_inode_item_push(
503 struct xfs_log_item *lip,
504 struct list_head *buffer_list)
505 __releases(&lip->li_ailp->ail_lock)
506 __acquires(&lip->li_ailp->ail_lock)
507{
508 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
509 struct xfs_inode *ip = iip->ili_inode;
510 struct xfs_buf *bp = lip->li_buf;
511 uint rval = XFS_ITEM_SUCCESS;
512 int error;
513
514 if (xfs_ipincount(ip) > 0)
515 return XFS_ITEM_PINNED;
516
517 /*
518 * The buffer containing this item failed to be written back
519 * previously. Resubmit the buffer for IO.
520 */
521 if (lip->li_flags & XFS_LI_FAILED) {
522 if (!xfs_buf_trylock(bp))
523 return XFS_ITEM_LOCKED;
524
525 if (!xfs_buf_resubmit_failed_buffers(bp, buffer_list))
526 rval = XFS_ITEM_FLUSHING;
527
528 xfs_buf_unlock(bp);
529 return rval;
530 }
531
532 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
533 return XFS_ITEM_LOCKED;
534
535 /*
536 * Re-check the pincount now that we stabilized the value by
537 * taking the ilock.
538 */
539 if (xfs_ipincount(ip) > 0) {
540 rval = XFS_ITEM_PINNED;
541 goto out_unlock;
542 }
543
544 /*
545 * Stale inode items should force out the iclog.
546 */
547 if (ip->i_flags & XFS_ISTALE) {
548 rval = XFS_ITEM_PINNED;
549 goto out_unlock;
550 }
551
552 /*
553 * Someone else is already flushing the inode. Nothing we can do
554 * here but wait for the flush to finish and remove the item from
555 * the AIL.
556 */
557 if (!xfs_iflock_nowait(ip)) {
558 rval = XFS_ITEM_FLUSHING;
559 goto out_unlock;
560 }
561
562 ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
563 ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
564
565 spin_unlock(&lip->li_ailp->ail_lock);
566
567 error = xfs_iflush(ip, &bp);
568 if (!error) {
569 if (!xfs_buf_delwri_queue(bp, buffer_list))
570 rval = XFS_ITEM_FLUSHING;
571 xfs_buf_relse(bp);
572 }
573
574 spin_lock(&lip->li_ailp->ail_lock);
575out_unlock:
576 xfs_iunlock(ip, XFS_ILOCK_SHARED);
577 return rval;
578}
579
580/*
581 * Unlock the inode associated with the inode log item.
582 */
583STATIC void
584xfs_inode_item_unlock(
585 struct xfs_log_item *lip)
586{
587 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
588 struct xfs_inode *ip = iip->ili_inode;
589 unsigned short lock_flags;
590
591 ASSERT(ip->i_itemp != NULL);
592 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
593
594 lock_flags = iip->ili_lock_flags;
595 iip->ili_lock_flags = 0;
596 if (lock_flags)
597 xfs_iunlock(ip, lock_flags);
598}
599
600/*
601 * This is called to find out where the oldest active copy of the inode log
602 * item in the on disk log resides now that the last log write of it completed
603 * at the given lsn. Since we always re-log all dirty data in an inode, the
604 * latest copy in the on disk log is the only one that matters. Therefore,
605 * simply return the given lsn.
606 *
607 * If the inode has been marked stale because the cluster is being freed, we
608 * don't want to (re-)insert this inode into the AIL. There is a race condition
609 * where the cluster buffer may be unpinned before the inode is inserted into
610 * the AIL during transaction committed processing. If the buffer is unpinned
611 * before the inode item has been committed and inserted, then it is possible
612 * for the buffer to be written and IO completes before the inode is inserted
613 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
614 * AIL which will never get removed. It will, however, get reclaimed which
615 * triggers an assert in xfs_inode_free() complaining about freein an inode
616 * still in the AIL.
617 *
618 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
619 * transaction committed code knows that it does not need to do any further
620 * processing on the item.
621 */
622STATIC xfs_lsn_t
623xfs_inode_item_committed(
624 struct xfs_log_item *lip,
625 xfs_lsn_t lsn)
626{
627 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
628 struct xfs_inode *ip = iip->ili_inode;
629
630 if (xfs_iflags_test(ip, XFS_ISTALE)) {
631 xfs_inode_item_unpin(lip, 0);
632 return -1;
633 }
634 return lsn;
635}
636
637STATIC void
638xfs_inode_item_committing(
639 struct xfs_log_item *lip,
640 xfs_lsn_t lsn)
641{
642 INODE_ITEM(lip)->ili_last_lsn = lsn;
643}
644
645/*
646 * This is the ops vector shared by all buf log items.
647 */
648static const struct xfs_item_ops xfs_inode_item_ops = {
649 .iop_size = xfs_inode_item_size,
650 .iop_format = xfs_inode_item_format,
651 .iop_pin = xfs_inode_item_pin,
652 .iop_unpin = xfs_inode_item_unpin,
653 .iop_unlock = xfs_inode_item_unlock,
654 .iop_committed = xfs_inode_item_committed,
655 .iop_push = xfs_inode_item_push,
656 .iop_committing = xfs_inode_item_committing,
657 .iop_error = xfs_inode_item_error
658};
659
660
661/*
662 * Initialize the inode log item for a newly allocated (in-core) inode.
663 */
664void
665xfs_inode_item_init(
666 struct xfs_inode *ip,
667 struct xfs_mount *mp)
668{
669 struct xfs_inode_log_item *iip;
670
671 ASSERT(ip->i_itemp == NULL);
672 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
673
674 iip->ili_inode = ip;
675 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
676 &xfs_inode_item_ops);
677}
678
679/*
680 * Free the inode log item and any memory hanging off of it.
681 */
682void
683xfs_inode_item_destroy(
684 xfs_inode_t *ip)
685{
686 kmem_free(ip->i_itemp->ili_item.li_lv_shadow);
687 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
688}
689
690
691/*
692 * This is the inode flushing I/O completion routine. It is called
693 * from interrupt level when the buffer containing the inode is
694 * flushed to disk. It is responsible for removing the inode item
695 * from the AIL if it has not been re-logged, and unlocking the inode's
696 * flush lock.
697 *
698 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
699 * list for other inodes that will run this function. We remove them from the
700 * buffer list so we can process all the inode IO completions in one AIL lock
701 * traversal.
702 */
703void
704xfs_iflush_done(
705 struct xfs_buf *bp,
706 struct xfs_log_item *lip)
707{
708 struct xfs_inode_log_item *iip;
709 struct xfs_log_item *blip, *n;
710 struct xfs_ail *ailp = lip->li_ailp;
711 int need_ail = 0;
712 LIST_HEAD(tmp);
713
714 /*
715 * Scan the buffer IO completions for other inodes being completed and
716 * attach them to the current inode log item.
717 */
718
719 list_add_tail(&lip->li_bio_list, &tmp);
720
721 list_for_each_entry_safe(blip, n, &bp->b_li_list, li_bio_list) {
722 if (lip->li_cb != xfs_iflush_done)
723 continue;
724
725 list_move_tail(&blip->li_bio_list, &tmp);
726 /*
727 * while we have the item, do the unlocked check for needing
728 * the AIL lock.
729 */
730 iip = INODE_ITEM(blip);
731 if ((iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn) ||
732 (blip->li_flags & XFS_LI_FAILED))
733 need_ail++;
734 }
735
736 /* make sure we capture the state of the initial inode. */
737 iip = INODE_ITEM(lip);
738 if ((iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn) ||
739 lip->li_flags & XFS_LI_FAILED)
740 need_ail++;
741
742 /*
743 * We only want to pull the item from the AIL if it is
744 * actually there and its location in the log has not
745 * changed since we started the flush. Thus, we only bother
746 * if the ili_logged flag is set and the inode's lsn has not
747 * changed. First we check the lsn outside
748 * the lock since it's cheaper, and then we recheck while
749 * holding the lock before removing the inode from the AIL.
750 */
751 if (need_ail) {
752 bool mlip_changed = false;
753
754 /* this is an opencoded batch version of xfs_trans_ail_delete */
755 spin_lock(&ailp->ail_lock);
756 list_for_each_entry(blip, &tmp, li_bio_list) {
757 if (INODE_ITEM(blip)->ili_logged &&
758 blip->li_lsn == INODE_ITEM(blip)->ili_flush_lsn)
759 mlip_changed |= xfs_ail_delete_one(ailp, blip);
760 else {
761 xfs_clear_li_failed(blip);
762 }
763 }
764
765 if (mlip_changed) {
766 if (!XFS_FORCED_SHUTDOWN(ailp->ail_mount))
767 xlog_assign_tail_lsn_locked(ailp->ail_mount);
768 if (list_empty(&ailp->ail_head))
769 wake_up_all(&ailp->ail_empty);
770 }
771 spin_unlock(&ailp->ail_lock);
772
773 if (mlip_changed)
774 xfs_log_space_wake(ailp->ail_mount);
775 }
776
777 /*
778 * clean up and unlock the flush lock now we are done. We can clear the
779 * ili_last_fields bits now that we know that the data corresponding to
780 * them is safely on disk.
781 */
782 list_for_each_entry_safe(blip, n, &tmp, li_bio_list) {
783 list_del_init(&blip->li_bio_list);
784 iip = INODE_ITEM(blip);
785 iip->ili_logged = 0;
786 iip->ili_last_fields = 0;
787 xfs_ifunlock(iip->ili_inode);
788 }
789 list_del(&tmp);
790}
791
792/*
793 * This is the inode flushing abort routine. It is called from xfs_iflush when
794 * the filesystem is shutting down to clean up the inode state. It is
795 * responsible for removing the inode item from the AIL if it has not been
796 * re-logged, and unlocking the inode's flush lock.
797 */
798void
799xfs_iflush_abort(
800 xfs_inode_t *ip,
801 bool stale)
802{
803 xfs_inode_log_item_t *iip = ip->i_itemp;
804
805 if (iip) {
806 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
807 xfs_trans_ail_remove(&iip->ili_item,
808 stale ? SHUTDOWN_LOG_IO_ERROR :
809 SHUTDOWN_CORRUPT_INCORE);
810 }
811 iip->ili_logged = 0;
812 /*
813 * Clear the ili_last_fields bits now that we know that the
814 * data corresponding to them is safely on disk.
815 */
816 iip->ili_last_fields = 0;
817 /*
818 * Clear the inode logging fields so no more flushes are
819 * attempted.
820 */
821 iip->ili_fields = 0;
822 iip->ili_fsync_fields = 0;
823 }
824 /*
825 * Release the inode's flush lock since we're done with it.
826 */
827 xfs_ifunlock(ip);
828}
829
830void
831xfs_istale_done(
832 struct xfs_buf *bp,
833 struct xfs_log_item *lip)
834{
835 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
836}
837
838/*
839 * convert an xfs_inode_log_format struct from the old 32 bit version
840 * (which can have different field alignments) to the native 64 bit version
841 */
842int
843xfs_inode_item_format_convert(
844 struct xfs_log_iovec *buf,
845 struct xfs_inode_log_format *in_f)
846{
847 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;
848
849 if (buf->i_len != sizeof(*in_f32))
850 return -EFSCORRUPTED;
851
852 in_f->ilf_type = in_f32->ilf_type;
853 in_f->ilf_size = in_f32->ilf_size;
854 in_f->ilf_fields = in_f32->ilf_fields;
855 in_f->ilf_asize = in_f32->ilf_asize;
856 in_f->ilf_dsize = in_f32->ilf_dsize;
857 in_f->ilf_ino = in_f32->ilf_ino;
858 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
859 in_f->ilf_blkno = in_f32->ilf_blkno;
860 in_f->ilf_len = in_f32->ilf_len;
861 in_f->ilf_boffset = in_f32->ilf_boffset;
862 return 0;
863}
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_error.h"
21
22#include <linux/iversion.h>
23
24kmem_zone_t *xfs_ili_zone; /* inode log item zone */
25
26static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
27{
28 return container_of(lip, struct xfs_inode_log_item, ili_item);
29}
30
31STATIC void
32xfs_inode_item_data_fork_size(
33 struct xfs_inode_log_item *iip,
34 int *nvecs,
35 int *nbytes)
36{
37 struct xfs_inode *ip = iip->ili_inode;
38
39 switch (ip->i_df.if_format) {
40 case XFS_DINODE_FMT_EXTENTS:
41 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
42 ip->i_df.if_nextents > 0 &&
43 ip->i_df.if_bytes > 0) {
44 /* worst case, doesn't subtract delalloc extents */
45 *nbytes += XFS_IFORK_DSIZE(ip);
46 *nvecs += 1;
47 }
48 break;
49 case XFS_DINODE_FMT_BTREE:
50 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
51 ip->i_df.if_broot_bytes > 0) {
52 *nbytes += ip->i_df.if_broot_bytes;
53 *nvecs += 1;
54 }
55 break;
56 case XFS_DINODE_FMT_LOCAL:
57 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
58 ip->i_df.if_bytes > 0) {
59 *nbytes += roundup(ip->i_df.if_bytes, 4);
60 *nvecs += 1;
61 }
62 break;
63
64 case XFS_DINODE_FMT_DEV:
65 break;
66 default:
67 ASSERT(0);
68 break;
69 }
70}
71
72STATIC void
73xfs_inode_item_attr_fork_size(
74 struct xfs_inode_log_item *iip,
75 int *nvecs,
76 int *nbytes)
77{
78 struct xfs_inode *ip = iip->ili_inode;
79
80 switch (ip->i_afp->if_format) {
81 case XFS_DINODE_FMT_EXTENTS:
82 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
83 ip->i_afp->if_nextents > 0 &&
84 ip->i_afp->if_bytes > 0) {
85 /* worst case, doesn't subtract unused space */
86 *nbytes += XFS_IFORK_ASIZE(ip);
87 *nvecs += 1;
88 }
89 break;
90 case XFS_DINODE_FMT_BTREE:
91 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
92 ip->i_afp->if_broot_bytes > 0) {
93 *nbytes += ip->i_afp->if_broot_bytes;
94 *nvecs += 1;
95 }
96 break;
97 case XFS_DINODE_FMT_LOCAL:
98 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
99 ip->i_afp->if_bytes > 0) {
100 *nbytes += roundup(ip->i_afp->if_bytes, 4);
101 *nvecs += 1;
102 }
103 break;
104 default:
105 ASSERT(0);
106 break;
107 }
108}
109
110/*
111 * This returns the number of iovecs needed to log the given inode item.
112 *
113 * We need one iovec for the inode log format structure, one for the
114 * inode core, and possibly one for the inode data/extents/b-tree root
115 * and one for the inode attribute data/extents/b-tree root.
116 */
117STATIC void
118xfs_inode_item_size(
119 struct xfs_log_item *lip,
120 int *nvecs,
121 int *nbytes)
122{
123 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
124 struct xfs_inode *ip = iip->ili_inode;
125
126 *nvecs += 2;
127 *nbytes += sizeof(struct xfs_inode_log_format) +
128 xfs_log_dinode_size(ip->i_mount);
129
130 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
131 if (XFS_IFORK_Q(ip))
132 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
133}
134
135STATIC void
136xfs_inode_item_format_data_fork(
137 struct xfs_inode_log_item *iip,
138 struct xfs_inode_log_format *ilf,
139 struct xfs_log_vec *lv,
140 struct xfs_log_iovec **vecp)
141{
142 struct xfs_inode *ip = iip->ili_inode;
143 size_t data_bytes;
144
145 switch (ip->i_df.if_format) {
146 case XFS_DINODE_FMT_EXTENTS:
147 iip->ili_fields &=
148 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
149
150 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
151 ip->i_df.if_nextents > 0 &&
152 ip->i_df.if_bytes > 0) {
153 struct xfs_bmbt_rec *p;
154
155 ASSERT(xfs_iext_count(&ip->i_df) > 0);
156
157 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
158 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
159 xlog_finish_iovec(lv, *vecp, data_bytes);
160
161 ASSERT(data_bytes <= ip->i_df.if_bytes);
162
163 ilf->ilf_dsize = data_bytes;
164 ilf->ilf_size++;
165 } else {
166 iip->ili_fields &= ~XFS_ILOG_DEXT;
167 }
168 break;
169 case XFS_DINODE_FMT_BTREE:
170 iip->ili_fields &=
171 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);
172
173 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
174 ip->i_df.if_broot_bytes > 0) {
175 ASSERT(ip->i_df.if_broot != NULL);
176 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
177 ip->i_df.if_broot,
178 ip->i_df.if_broot_bytes);
179 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
180 ilf->ilf_size++;
181 } else {
182 ASSERT(!(iip->ili_fields &
183 XFS_ILOG_DBROOT));
184 iip->ili_fields &= ~XFS_ILOG_DBROOT;
185 }
186 break;
187 case XFS_DINODE_FMT_LOCAL:
188 iip->ili_fields &=
189 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);
190 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
191 ip->i_df.if_bytes > 0) {
192 /*
193 * Round i_bytes up to a word boundary.
194 * The underlying memory is guaranteed
195 * to be there by xfs_idata_realloc().
196 */
197 data_bytes = roundup(ip->i_df.if_bytes, 4);
198 ASSERT(ip->i_df.if_u1.if_data != NULL);
199 ASSERT(ip->i_d.di_size > 0);
200 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
201 ip->i_df.if_u1.if_data, data_bytes);
202 ilf->ilf_dsize = (unsigned)data_bytes;
203 ilf->ilf_size++;
204 } else {
205 iip->ili_fields &= ~XFS_ILOG_DDATA;
206 }
207 break;
208 case XFS_DINODE_FMT_DEV:
209 iip->ili_fields &=
210 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);
211 if (iip->ili_fields & XFS_ILOG_DEV)
212 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);
213 break;
214 default:
215 ASSERT(0);
216 break;
217 }
218}
219
220STATIC void
221xfs_inode_item_format_attr_fork(
222 struct xfs_inode_log_item *iip,
223 struct xfs_inode_log_format *ilf,
224 struct xfs_log_vec *lv,
225 struct xfs_log_iovec **vecp)
226{
227 struct xfs_inode *ip = iip->ili_inode;
228 size_t data_bytes;
229
230 switch (ip->i_afp->if_format) {
231 case XFS_DINODE_FMT_EXTENTS:
232 iip->ili_fields &=
233 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
234
235 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
236 ip->i_afp->if_nextents > 0 &&
237 ip->i_afp->if_bytes > 0) {
238 struct xfs_bmbt_rec *p;
239
240 ASSERT(xfs_iext_count(ip->i_afp) ==
241 ip->i_afp->if_nextents);
242
243 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
244 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
245 xlog_finish_iovec(lv, *vecp, data_bytes);
246
247 ilf->ilf_asize = data_bytes;
248 ilf->ilf_size++;
249 } else {
250 iip->ili_fields &= ~XFS_ILOG_AEXT;
251 }
252 break;
253 case XFS_DINODE_FMT_BTREE:
254 iip->ili_fields &=
255 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
256
257 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
258 ip->i_afp->if_broot_bytes > 0) {
259 ASSERT(ip->i_afp->if_broot != NULL);
260
261 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
262 ip->i_afp->if_broot,
263 ip->i_afp->if_broot_bytes);
264 ilf->ilf_asize = ip->i_afp->if_broot_bytes;
265 ilf->ilf_size++;
266 } else {
267 iip->ili_fields &= ~XFS_ILOG_ABROOT;
268 }
269 break;
270 case XFS_DINODE_FMT_LOCAL:
271 iip->ili_fields &=
272 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
273
274 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
275 ip->i_afp->if_bytes > 0) {
276 /*
277 * Round i_bytes up to a word boundary.
278 * The underlying memory is guaranteed
279 * to be there by xfs_idata_realloc().
280 */
281 data_bytes = roundup(ip->i_afp->if_bytes, 4);
282 ASSERT(ip->i_afp->if_u1.if_data != NULL);
283 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
284 ip->i_afp->if_u1.if_data,
285 data_bytes);
286 ilf->ilf_asize = (unsigned)data_bytes;
287 ilf->ilf_size++;
288 } else {
289 iip->ili_fields &= ~XFS_ILOG_ADATA;
290 }
291 break;
292 default:
293 ASSERT(0);
294 break;
295 }
296}
297
298static void
299xfs_inode_to_log_dinode(
300 struct xfs_inode *ip,
301 struct xfs_log_dinode *to,
302 xfs_lsn_t lsn)
303{
304 struct xfs_icdinode *from = &ip->i_d;
305 struct inode *inode = VFS_I(ip);
306
307 to->di_magic = XFS_DINODE_MAGIC;
308 to->di_format = xfs_ifork_format(&ip->i_df);
309 to->di_uid = i_uid_read(inode);
310 to->di_gid = i_gid_read(inode);
311 to->di_projid_lo = from->di_projid & 0xffff;
312 to->di_projid_hi = from->di_projid >> 16;
313
314 memset(to->di_pad, 0, sizeof(to->di_pad));
315 memset(to->di_pad3, 0, sizeof(to->di_pad3));
316 to->di_atime.t_sec = inode->i_atime.tv_sec;
317 to->di_atime.t_nsec = inode->i_atime.tv_nsec;
318 to->di_mtime.t_sec = inode->i_mtime.tv_sec;
319 to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
320 to->di_ctime.t_sec = inode->i_ctime.tv_sec;
321 to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
322 to->di_nlink = inode->i_nlink;
323 to->di_gen = inode->i_generation;
324 to->di_mode = inode->i_mode;
325
326 to->di_size = from->di_size;
327 to->di_nblocks = from->di_nblocks;
328 to->di_extsize = from->di_extsize;
329 to->di_nextents = xfs_ifork_nextents(&ip->i_df);
330 to->di_anextents = xfs_ifork_nextents(ip->i_afp);
331 to->di_forkoff = from->di_forkoff;
332 to->di_aformat = xfs_ifork_format(ip->i_afp);
333 to->di_dmevmask = from->di_dmevmask;
334 to->di_dmstate = from->di_dmstate;
335 to->di_flags = from->di_flags;
336
337 /* log a dummy value to ensure log structure is fully initialised */
338 to->di_next_unlinked = NULLAGINO;
339
340 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
341 to->di_version = 3;
342 to->di_changecount = inode_peek_iversion(inode);
343 to->di_crtime.t_sec = from->di_crtime.tv_sec;
344 to->di_crtime.t_nsec = from->di_crtime.tv_nsec;
345 to->di_flags2 = from->di_flags2;
346 to->di_cowextsize = from->di_cowextsize;
347 to->di_ino = ip->i_ino;
348 to->di_lsn = lsn;
349 memset(to->di_pad2, 0, sizeof(to->di_pad2));
350 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
351 to->di_flushiter = 0;
352 } else {
353 to->di_version = 2;
354 to->di_flushiter = from->di_flushiter;
355 }
356}
357
358/*
359 * Format the inode core. Current timestamp data is only in the VFS inode
360 * fields, so we need to grab them from there. Hence rather than just copying
361 * the XFS inode core structure, format the fields directly into the iovec.
362 */
363static void
364xfs_inode_item_format_core(
365 struct xfs_inode *ip,
366 struct xfs_log_vec *lv,
367 struct xfs_log_iovec **vecp)
368{
369 struct xfs_log_dinode *dic;
370
371 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
372 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
373 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));
374}
375
376/*
377 * This is called to fill in the vector of log iovecs for the given inode
378 * log item. It fills the first item with an inode log format structure,
379 * the second with the on-disk inode structure, and a possible third and/or
380 * fourth with the inode data/extents/b-tree root and inode attributes
381 * data/extents/b-tree root.
382 *
383 * Note: Always use the 64 bit inode log format structure so we don't
384 * leave an uninitialised hole in the format item on 64 bit systems. Log
385 * recovery on 32 bit systems handles this just fine, so there's no reason
386 * for not using an initialising the properly padded structure all the time.
387 */
388STATIC void
389xfs_inode_item_format(
390 struct xfs_log_item *lip,
391 struct xfs_log_vec *lv)
392{
393 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
394 struct xfs_inode *ip = iip->ili_inode;
395 struct xfs_log_iovec *vecp = NULL;
396 struct xfs_inode_log_format *ilf;
397
398 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
399 ilf->ilf_type = XFS_LI_INODE;
400 ilf->ilf_ino = ip->i_ino;
401 ilf->ilf_blkno = ip->i_imap.im_blkno;
402 ilf->ilf_len = ip->i_imap.im_len;
403 ilf->ilf_boffset = ip->i_imap.im_boffset;
404 ilf->ilf_fields = XFS_ILOG_CORE;
405 ilf->ilf_size = 2; /* format + core */
406
407 /*
408 * make sure we don't leak uninitialised data into the log in the case
409 * when we don't log every field in the inode.
410 */
411 ilf->ilf_dsize = 0;
412 ilf->ilf_asize = 0;
413 ilf->ilf_pad = 0;
414 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));
415
416 xlog_finish_iovec(lv, vecp, sizeof(*ilf));
417
418 xfs_inode_item_format_core(ip, lv, &vecp);
419 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
420 if (XFS_IFORK_Q(ip)) {
421 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
422 } else {
423 iip->ili_fields &=
424 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
425 }
426
427 /* update the format with the exact fields we actually logged */
428 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
429}
430
431/*
432 * This is called to pin the inode associated with the inode log
433 * item in memory so it cannot be written out.
434 */
435STATIC void
436xfs_inode_item_pin(
437 struct xfs_log_item *lip)
438{
439 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
440
441 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
442 ASSERT(lip->li_buf);
443
444 trace_xfs_inode_pin(ip, _RET_IP_);
445 atomic_inc(&ip->i_pincount);
446}
447
448
449/*
450 * This is called to unpin the inode associated with the inode log
451 * item which was previously pinned with a call to xfs_inode_item_pin().
452 *
453 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
454 *
455 * Note that unpin can race with inode cluster buffer freeing marking the buffer
456 * stale. In that case, flush completions are run from the buffer unpin call,
457 * which may happen before the inode is unpinned. If we lose the race, there
458 * will be no buffer attached to the log item, but the inode will be marked
459 * XFS_ISTALE.
460 */
461STATIC void
462xfs_inode_item_unpin(
463 struct xfs_log_item *lip,
464 int remove)
465{
466 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
467
468 trace_xfs_inode_unpin(ip, _RET_IP_);
469 ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));
470 ASSERT(atomic_read(&ip->i_pincount) > 0);
471 if (atomic_dec_and_test(&ip->i_pincount))
472 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
473}
474
475STATIC uint
476xfs_inode_item_push(
477 struct xfs_log_item *lip,
478 struct list_head *buffer_list)
479 __releases(&lip->li_ailp->ail_lock)
480 __acquires(&lip->li_ailp->ail_lock)
481{
482 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
483 struct xfs_inode *ip = iip->ili_inode;
484 struct xfs_buf *bp = lip->li_buf;
485 uint rval = XFS_ITEM_SUCCESS;
486 int error;
487
488 ASSERT(iip->ili_item.li_buf);
489
490 if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp) ||
491 (ip->i_flags & XFS_ISTALE))
492 return XFS_ITEM_PINNED;
493
494 /* If the inode is already flush locked, we're already flushing. */
495 if (xfs_isiflocked(ip))
496 return XFS_ITEM_FLUSHING;
497
498 if (!xfs_buf_trylock(bp))
499 return XFS_ITEM_LOCKED;
500
501 spin_unlock(&lip->li_ailp->ail_lock);
502
503 /*
504 * We need to hold a reference for flushing the cluster buffer as it may
505 * fail the buffer without IO submission. In which case, we better get a
506 * reference for that completion because otherwise we don't get a
507 * reference for IO until we queue the buffer for delwri submission.
508 */
509 xfs_buf_hold(bp);
510 error = xfs_iflush_cluster(bp);
511 if (!error) {
512 if (!xfs_buf_delwri_queue(bp, buffer_list))
513 rval = XFS_ITEM_FLUSHING;
514 xfs_buf_relse(bp);
515 } else {
516 /*
517 * Release the buffer if we were unable to flush anything. On
518 * any other error, the buffer has already been released.
519 */
520 if (error == -EAGAIN)
521 xfs_buf_relse(bp);
522 rval = XFS_ITEM_LOCKED;
523 }
524
525 spin_lock(&lip->li_ailp->ail_lock);
526 return rval;
527}
528
529/*
530 * Unlock the inode associated with the inode log item.
531 */
532STATIC void
533xfs_inode_item_release(
534 struct xfs_log_item *lip)
535{
536 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
537 struct xfs_inode *ip = iip->ili_inode;
538 unsigned short lock_flags;
539
540 ASSERT(ip->i_itemp != NULL);
541 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
542
543 lock_flags = iip->ili_lock_flags;
544 iip->ili_lock_flags = 0;
545 if (lock_flags)
546 xfs_iunlock(ip, lock_flags);
547}
548
549/*
550 * This is called to find out where the oldest active copy of the inode log
551 * item in the on disk log resides now that the last log write of it completed
552 * at the given lsn. Since we always re-log all dirty data in an inode, the
553 * latest copy in the on disk log is the only one that matters. Therefore,
554 * simply return the given lsn.
555 *
556 * If the inode has been marked stale because the cluster is being freed, we
557 * don't want to (re-)insert this inode into the AIL. There is a race condition
558 * where the cluster buffer may be unpinned before the inode is inserted into
559 * the AIL during transaction committed processing. If the buffer is unpinned
560 * before the inode item has been committed and inserted, then it is possible
561 * for the buffer to be written and IO completes before the inode is inserted
562 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
563 * AIL which will never get removed. It will, however, get reclaimed which
564 * triggers an assert in xfs_inode_free() complaining about freein an inode
565 * still in the AIL.
566 *
567 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
568 * transaction committed code knows that it does not need to do any further
569 * processing on the item.
570 */
571STATIC xfs_lsn_t
572xfs_inode_item_committed(
573 struct xfs_log_item *lip,
574 xfs_lsn_t lsn)
575{
576 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
577 struct xfs_inode *ip = iip->ili_inode;
578
579 if (xfs_iflags_test(ip, XFS_ISTALE)) {
580 xfs_inode_item_unpin(lip, 0);
581 return -1;
582 }
583 return lsn;
584}
585
586STATIC void
587xfs_inode_item_committing(
588 struct xfs_log_item *lip,
589 xfs_lsn_t commit_lsn)
590{
591 INODE_ITEM(lip)->ili_last_lsn = commit_lsn;
592 return xfs_inode_item_release(lip);
593}
594
595static const struct xfs_item_ops xfs_inode_item_ops = {
596 .iop_size = xfs_inode_item_size,
597 .iop_format = xfs_inode_item_format,
598 .iop_pin = xfs_inode_item_pin,
599 .iop_unpin = xfs_inode_item_unpin,
600 .iop_release = xfs_inode_item_release,
601 .iop_committed = xfs_inode_item_committed,
602 .iop_push = xfs_inode_item_push,
603 .iop_committing = xfs_inode_item_committing,
604};
605
606
607/*
608 * Initialize the inode log item for a newly allocated (in-core) inode.
609 */
610void
611xfs_inode_item_init(
612 struct xfs_inode *ip,
613 struct xfs_mount *mp)
614{
615 struct xfs_inode_log_item *iip;
616
617 ASSERT(ip->i_itemp == NULL);
618 iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_zone,
619 GFP_KERNEL | __GFP_NOFAIL);
620
621 iip->ili_inode = ip;
622 spin_lock_init(&iip->ili_lock);
623 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
624 &xfs_inode_item_ops);
625}
626
627/*
628 * Free the inode log item and any memory hanging off of it.
629 */
630void
631xfs_inode_item_destroy(
632 struct xfs_inode *ip)
633{
634 struct xfs_inode_log_item *iip = ip->i_itemp;
635
636 ASSERT(iip->ili_item.li_buf == NULL);
637
638 ip->i_itemp = NULL;
639 kmem_free(iip->ili_item.li_lv_shadow);
640 kmem_cache_free(xfs_ili_zone, iip);
641}
642
643
644/*
645 * We only want to pull the item from the AIL if it is actually there
646 * and its location in the log has not changed since we started the
647 * flush. Thus, we only bother if the inode's lsn has not changed.
648 */
649static void
650xfs_iflush_ail_updates(
651 struct xfs_ail *ailp,
652 struct list_head *list)
653{
654 struct xfs_log_item *lip;
655 xfs_lsn_t tail_lsn = 0;
656
657 /* this is an opencoded batch version of xfs_trans_ail_delete */
658 spin_lock(&ailp->ail_lock);
659 list_for_each_entry(lip, list, li_bio_list) {
660 xfs_lsn_t lsn;
661
662 clear_bit(XFS_LI_FAILED, &lip->li_flags);
663 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)
664 continue;
665
666 lsn = xfs_ail_delete_one(ailp, lip);
667 if (!tail_lsn && lsn)
668 tail_lsn = lsn;
669 }
670 xfs_ail_update_finish(ailp, tail_lsn);
671}
672
673/*
674 * Walk the list of inodes that have completed their IOs. If they are clean
675 * remove them from the list and dissociate them from the buffer. Buffers that
676 * are still dirty remain linked to the buffer and on the list. Caller must
677 * handle them appropriately.
678 */
679static void
680xfs_iflush_finish(
681 struct xfs_buf *bp,
682 struct list_head *list)
683{
684 struct xfs_log_item *lip, *n;
685
686 list_for_each_entry_safe(lip, n, list, li_bio_list) {
687 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
688 bool drop_buffer = false;
689
690 spin_lock(&iip->ili_lock);
691
692 /*
693 * Remove the reference to the cluster buffer if the inode is
694 * clean in memory and drop the buffer reference once we've
695 * dropped the locks we hold.
696 */
697 ASSERT(iip->ili_item.li_buf == bp);
698 if (!iip->ili_fields) {
699 iip->ili_item.li_buf = NULL;
700 list_del_init(&lip->li_bio_list);
701 drop_buffer = true;
702 }
703 iip->ili_last_fields = 0;
704 iip->ili_flush_lsn = 0;
705 spin_unlock(&iip->ili_lock);
706 xfs_ifunlock(iip->ili_inode);
707 if (drop_buffer)
708 xfs_buf_rele(bp);
709 }
710}
711
712/*
713 * Inode buffer IO completion routine. It is responsible for removing inodes
714 * attached to the buffer from the AIL if they have not been re-logged, as well
715 * as completing the flush and unlocking the inode.
716 */
717void
718xfs_iflush_done(
719 struct xfs_buf *bp)
720{
721 struct xfs_log_item *lip, *n;
722 LIST_HEAD(flushed_inodes);
723 LIST_HEAD(ail_updates);
724
725 /*
726 * Pull the attached inodes from the buffer one at a time and take the
727 * appropriate action on them.
728 */
729 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
730 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
731
732 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {
733 xfs_iflush_abort(iip->ili_inode);
734 continue;
735 }
736 if (!iip->ili_last_fields)
737 continue;
738
739 /* Do an unlocked check for needing the AIL lock. */
740 if (iip->ili_flush_lsn == lip->li_lsn ||
741 test_bit(XFS_LI_FAILED, &lip->li_flags))
742 list_move_tail(&lip->li_bio_list, &ail_updates);
743 else
744 list_move_tail(&lip->li_bio_list, &flushed_inodes);
745 }
746
747 if (!list_empty(&ail_updates)) {
748 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);
749 list_splice_tail(&ail_updates, &flushed_inodes);
750 }
751
752 xfs_iflush_finish(bp, &flushed_inodes);
753 if (!list_empty(&flushed_inodes))
754 list_splice_tail(&flushed_inodes, &bp->b_li_list);
755}
756
757/*
758 * This is the inode flushing abort routine. It is called from xfs_iflush when
759 * the filesystem is shutting down to clean up the inode state. It is
760 * responsible for removing the inode item from the AIL if it has not been
761 * re-logged, and unlocking the inode's flush lock.
762 */
763void
764xfs_iflush_abort(
765 struct xfs_inode *ip)
766{
767 struct xfs_inode_log_item *iip = ip->i_itemp;
768 struct xfs_buf *bp = NULL;
769
770 if (iip) {
771 /*
772 * Clear the failed bit before removing the item from the AIL so
773 * xfs_trans_ail_delete() doesn't try to clear and release the
774 * buffer attached to the log item before we are done with it.
775 */
776 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);
777 xfs_trans_ail_delete(&iip->ili_item, 0);
778
779 /*
780 * Clear the inode logging fields so no more flushes are
781 * attempted.
782 */
783 spin_lock(&iip->ili_lock);
784 iip->ili_last_fields = 0;
785 iip->ili_fields = 0;
786 iip->ili_fsync_fields = 0;
787 iip->ili_flush_lsn = 0;
788 bp = iip->ili_item.li_buf;
789 iip->ili_item.li_buf = NULL;
790 list_del_init(&iip->ili_item.li_bio_list);
791 spin_unlock(&iip->ili_lock);
792 }
793 xfs_ifunlock(ip);
794 if (bp)
795 xfs_buf_rele(bp);
796}
797
798/*
799 * convert an xfs_inode_log_format struct from the old 32 bit version
800 * (which can have different field alignments) to the native 64 bit version
801 */
802int
803xfs_inode_item_format_convert(
804 struct xfs_log_iovec *buf,
805 struct xfs_inode_log_format *in_f)
806{
807 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;
808
809 if (buf->i_len != sizeof(*in_f32)) {
810 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
811 return -EFSCORRUPTED;
812 }
813
814 in_f->ilf_type = in_f32->ilf_type;
815 in_f->ilf_size = in_f32->ilf_size;
816 in_f->ilf_fields = in_f32->ilf_fields;
817 in_f->ilf_asize = in_f32->ilf_asize;
818 in_f->ilf_dsize = in_f32->ilf_dsize;
819 in_f->ilf_ino = in_f32->ilf_ino;
820 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));
821 in_f->ilf_blkno = in_f32->ilf_blkno;
822 in_f->ilf_len = in_f32->ilf_len;
823 in_f->ilf_boffset = in_f32->ilf_boffset;
824 return 0;
825}