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/*
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_types.h"
21#include "xfs_log.h"
22#include "xfs_trans.h"
23#include "xfs_sb.h"
24#include "xfs_ag.h"
25#include "xfs_mount.h"
26#include "xfs_trans_priv.h"
27#include "xfs_bmap_btree.h"
28#include "xfs_dinode.h"
29#include "xfs_inode.h"
30#include "xfs_inode_item.h"
31#include "xfs_error.h"
32#include "xfs_trace.h"
33
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
42
43/*
44 * This returns the number of iovecs needed to log the given inode item.
45 *
46 * We need one iovec for the inode log format structure, one for the
47 * inode core, and possibly one for the inode data/extents/b-tree root
48 * and one for the inode attribute data/extents/b-tree root.
49 */
50STATIC uint
51xfs_inode_item_size(
52 struct xfs_log_item *lip)
53{
54 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
55 struct xfs_inode *ip = iip->ili_inode;
56 uint nvecs = 2;
57
58 switch (ip->i_d.di_format) {
59 case XFS_DINODE_FMT_EXTENTS:
60 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
61 ip->i_d.di_nextents > 0 &&
62 ip->i_df.if_bytes > 0)
63 nvecs++;
64 break;
65
66 case XFS_DINODE_FMT_BTREE:
67 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
68 ip->i_df.if_broot_bytes > 0)
69 nvecs++;
70 break;
71
72 case XFS_DINODE_FMT_LOCAL:
73 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
74 ip->i_df.if_bytes > 0)
75 nvecs++;
76 break;
77
78 case XFS_DINODE_FMT_DEV:
79 case XFS_DINODE_FMT_UUID:
80 break;
81
82 default:
83 ASSERT(0);
84 break;
85 }
86
87 if (!XFS_IFORK_Q(ip))
88 return nvecs;
89
90
91 /*
92 * Log any necessary attribute data.
93 */
94 switch (ip->i_d.di_aformat) {
95 case XFS_DINODE_FMT_EXTENTS:
96 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
97 ip->i_d.di_anextents > 0 &&
98 ip->i_afp->if_bytes > 0)
99 nvecs++;
100 break;
101
102 case XFS_DINODE_FMT_BTREE:
103 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
104 ip->i_afp->if_broot_bytes > 0)
105 nvecs++;
106 break;
107
108 case XFS_DINODE_FMT_LOCAL:
109 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
110 ip->i_afp->if_bytes > 0)
111 nvecs++;
112 break;
113
114 default:
115 ASSERT(0);
116 break;
117 }
118
119 return nvecs;
120}
121
122/*
123 * xfs_inode_item_format_extents - convert in-core extents to on-disk form
124 *
125 * For either the data or attr fork in extent format, we need to endian convert
126 * the in-core extent as we place them into the on-disk inode. In this case, we
127 * need to do this conversion before we write the extents into the log. Because
128 * we don't have the disk inode to write into here, we allocate a buffer and
129 * format the extents into it via xfs_iextents_copy(). We free the buffer in
130 * the unlock routine after the copy for the log has been made.
131 *
132 * In the case of the data fork, the in-core and on-disk fork sizes can be
133 * different due to delayed allocation extents. We only log on-disk extents
134 * here, so always use the physical fork size to determine the size of the
135 * buffer we need to allocate.
136 */
137STATIC void
138xfs_inode_item_format_extents(
139 struct xfs_inode *ip,
140 struct xfs_log_iovec *vecp,
141 int whichfork,
142 int type)
143{
144 xfs_bmbt_rec_t *ext_buffer;
145
146 ext_buffer = kmem_alloc(XFS_IFORK_SIZE(ip, whichfork), KM_SLEEP);
147 if (whichfork == XFS_DATA_FORK)
148 ip->i_itemp->ili_extents_buf = ext_buffer;
149 else
150 ip->i_itemp->ili_aextents_buf = ext_buffer;
151
152 vecp->i_addr = ext_buffer;
153 vecp->i_len = xfs_iextents_copy(ip, ext_buffer, whichfork);
154 vecp->i_type = type;
155}
156
157/*
158 * This is called to fill in the vector of log iovecs for the
159 * given inode log item. It fills the first item with an inode
160 * log format structure, the second with the on-disk inode structure,
161 * and a possible third and/or fourth with the inode data/extents/b-tree
162 * root and inode attributes data/extents/b-tree root.
163 */
164STATIC void
165xfs_inode_item_format(
166 struct xfs_log_item *lip,
167 struct xfs_log_iovec *vecp)
168{
169 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
170 struct xfs_inode *ip = iip->ili_inode;
171 uint nvecs;
172 size_t data_bytes;
173 xfs_mount_t *mp;
174
175 vecp->i_addr = &iip->ili_format;
176 vecp->i_len = sizeof(xfs_inode_log_format_t);
177 vecp->i_type = XLOG_REG_TYPE_IFORMAT;
178 vecp++;
179 nvecs = 1;
180
181 vecp->i_addr = &ip->i_d;
182 vecp->i_len = sizeof(struct xfs_icdinode);
183 vecp->i_type = XLOG_REG_TYPE_ICORE;
184 vecp++;
185 nvecs++;
186
187 /*
188 * If this is really an old format inode, then we need to
189 * log it as such. This means that we have to copy the link
190 * count from the new field to the old. We don't have to worry
191 * about the new fields, because nothing trusts them as long as
192 * the old inode version number is there. If the superblock already
193 * has a new version number, then we don't bother converting back.
194 */
195 mp = ip->i_mount;
196 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
197 if (ip->i_d.di_version == 1) {
198 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
199 /*
200 * Convert it back.
201 */
202 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
203 ip->i_d.di_onlink = ip->i_d.di_nlink;
204 } else {
205 /*
206 * The superblock version has already been bumped,
207 * so just make the conversion to the new inode
208 * format permanent.
209 */
210 ip->i_d.di_version = 2;
211 ip->i_d.di_onlink = 0;
212 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
213 }
214 }
215
216 switch (ip->i_d.di_format) {
217 case XFS_DINODE_FMT_EXTENTS:
218 iip->ili_fields &=
219 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
220 XFS_ILOG_DEV | XFS_ILOG_UUID);
221
222 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
223 ip->i_d.di_nextents > 0 &&
224 ip->i_df.if_bytes > 0) {
225 ASSERT(ip->i_df.if_u1.if_extents != NULL);
226 ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
227 ASSERT(iip->ili_extents_buf == NULL);
228
229#ifdef XFS_NATIVE_HOST
230 if (ip->i_d.di_nextents == ip->i_df.if_bytes /
231 (uint)sizeof(xfs_bmbt_rec_t)) {
232 /*
233 * There are no delayed allocation
234 * extents, so just point to the
235 * real extents array.
236 */
237 vecp->i_addr = ip->i_df.if_u1.if_extents;
238 vecp->i_len = ip->i_df.if_bytes;
239 vecp->i_type = XLOG_REG_TYPE_IEXT;
240 } else
241#endif
242 {
243 xfs_inode_item_format_extents(ip, vecp,
244 XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
245 }
246 ASSERT(vecp->i_len <= ip->i_df.if_bytes);
247 iip->ili_format.ilf_dsize = vecp->i_len;
248 vecp++;
249 nvecs++;
250 } else {
251 iip->ili_fields &= ~XFS_ILOG_DEXT;
252 }
253 break;
254
255 case XFS_DINODE_FMT_BTREE:
256 iip->ili_fields &=
257 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
258 XFS_ILOG_DEV | XFS_ILOG_UUID);
259
260 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
261 ip->i_df.if_broot_bytes > 0) {
262 ASSERT(ip->i_df.if_broot != NULL);
263 vecp->i_addr = ip->i_df.if_broot;
264 vecp->i_len = ip->i_df.if_broot_bytes;
265 vecp->i_type = XLOG_REG_TYPE_IBROOT;
266 vecp++;
267 nvecs++;
268 iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
269 } else {
270 ASSERT(!(iip->ili_fields &
271 XFS_ILOG_DBROOT));
272#ifdef XFS_TRANS_DEBUG
273 if (iip->ili_root_size > 0) {
274 ASSERT(iip->ili_root_size ==
275 ip->i_df.if_broot_bytes);
276 ASSERT(memcmp(iip->ili_orig_root,
277 ip->i_df.if_broot,
278 iip->ili_root_size) == 0);
279 } else {
280 ASSERT(ip->i_df.if_broot_bytes == 0);
281 }
282#endif
283 iip->ili_fields &= ~XFS_ILOG_DBROOT;
284 }
285 break;
286
287 case XFS_DINODE_FMT_LOCAL:
288 iip->ili_fields &=
289 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
290 XFS_ILOG_DEV | XFS_ILOG_UUID);
291 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
292 ip->i_df.if_bytes > 0) {
293 ASSERT(ip->i_df.if_u1.if_data != NULL);
294 ASSERT(ip->i_d.di_size > 0);
295
296 vecp->i_addr = ip->i_df.if_u1.if_data;
297 /*
298 * Round i_bytes up to a word boundary.
299 * The underlying memory is guaranteed to
300 * to be there by xfs_idata_realloc().
301 */
302 data_bytes = roundup(ip->i_df.if_bytes, 4);
303 ASSERT((ip->i_df.if_real_bytes == 0) ||
304 (ip->i_df.if_real_bytes == data_bytes));
305 vecp->i_len = (int)data_bytes;
306 vecp->i_type = XLOG_REG_TYPE_ILOCAL;
307 vecp++;
308 nvecs++;
309 iip->ili_format.ilf_dsize = (unsigned)data_bytes;
310 } else {
311 iip->ili_fields &= ~XFS_ILOG_DDATA;
312 }
313 break;
314
315 case XFS_DINODE_FMT_DEV:
316 iip->ili_fields &=
317 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
318 XFS_ILOG_DEXT | XFS_ILOG_UUID);
319 if (iip->ili_fields & XFS_ILOG_DEV) {
320 iip->ili_format.ilf_u.ilfu_rdev =
321 ip->i_df.if_u2.if_rdev;
322 }
323 break;
324
325 case XFS_DINODE_FMT_UUID:
326 iip->ili_fields &=
327 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
328 XFS_ILOG_DEXT | XFS_ILOG_DEV);
329 if (iip->ili_fields & XFS_ILOG_UUID) {
330 iip->ili_format.ilf_u.ilfu_uuid =
331 ip->i_df.if_u2.if_uuid;
332 }
333 break;
334
335 default:
336 ASSERT(0);
337 break;
338 }
339
340 /*
341 * If there are no attributes associated with the file, then we're done.
342 */
343 if (!XFS_IFORK_Q(ip)) {
344 iip->ili_fields &=
345 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
346 goto out;
347 }
348
349 switch (ip->i_d.di_aformat) {
350 case XFS_DINODE_FMT_EXTENTS:
351 iip->ili_fields &=
352 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
353
354 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
355 ip->i_d.di_anextents > 0 &&
356 ip->i_afp->if_bytes > 0) {
357 ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
358 ip->i_d.di_anextents);
359 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
360#ifdef XFS_NATIVE_HOST
361 /*
362 * There are not delayed allocation extents
363 * for attributes, so just point at the array.
364 */
365 vecp->i_addr = ip->i_afp->if_u1.if_extents;
366 vecp->i_len = ip->i_afp->if_bytes;
367 vecp->i_type = XLOG_REG_TYPE_IATTR_EXT;
368#else
369 ASSERT(iip->ili_aextents_buf == NULL);
370 xfs_inode_item_format_extents(ip, vecp,
371 XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
372#endif
373 iip->ili_format.ilf_asize = vecp->i_len;
374 vecp++;
375 nvecs++;
376 } else {
377 iip->ili_fields &= ~XFS_ILOG_AEXT;
378 }
379 break;
380
381 case XFS_DINODE_FMT_BTREE:
382 iip->ili_fields &=
383 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
384
385 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
386 ip->i_afp->if_broot_bytes > 0) {
387 ASSERT(ip->i_afp->if_broot != NULL);
388
389 vecp->i_addr = ip->i_afp->if_broot;
390 vecp->i_len = ip->i_afp->if_broot_bytes;
391 vecp->i_type = XLOG_REG_TYPE_IATTR_BROOT;
392 vecp++;
393 nvecs++;
394 iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
395 } else {
396 iip->ili_fields &= ~XFS_ILOG_ABROOT;
397 }
398 break;
399
400 case XFS_DINODE_FMT_LOCAL:
401 iip->ili_fields &=
402 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
403
404 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
405 ip->i_afp->if_bytes > 0) {
406 ASSERT(ip->i_afp->if_u1.if_data != NULL);
407
408 vecp->i_addr = ip->i_afp->if_u1.if_data;
409 /*
410 * Round i_bytes up to a word boundary.
411 * The underlying memory is guaranteed to
412 * to be there by xfs_idata_realloc().
413 */
414 data_bytes = roundup(ip->i_afp->if_bytes, 4);
415 ASSERT((ip->i_afp->if_real_bytes == 0) ||
416 (ip->i_afp->if_real_bytes == data_bytes));
417 vecp->i_len = (int)data_bytes;
418 vecp->i_type = XLOG_REG_TYPE_IATTR_LOCAL;
419 vecp++;
420 nvecs++;
421 iip->ili_format.ilf_asize = (unsigned)data_bytes;
422 } else {
423 iip->ili_fields &= ~XFS_ILOG_ADATA;
424 }
425 break;
426
427 default:
428 ASSERT(0);
429 break;
430 }
431
432out:
433 /*
434 * Now update the log format that goes out to disk from the in-core
435 * values. We always write the inode core to make the arithmetic
436 * games in recovery easier, which isn't a big deal as just about any
437 * transaction would dirty it anyway.
438 */
439 iip->ili_format.ilf_fields = XFS_ILOG_CORE |
440 (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
441 iip->ili_format.ilf_size = nvecs;
442}
443
444
445/*
446 * This is called to pin the inode associated with the inode log
447 * item in memory so it cannot be written out.
448 */
449STATIC void
450xfs_inode_item_pin(
451 struct xfs_log_item *lip)
452{
453 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
454
455 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
456
457 trace_xfs_inode_pin(ip, _RET_IP_);
458 atomic_inc(&ip->i_pincount);
459}
460
461
462/*
463 * This is called to unpin the inode associated with the inode log
464 * item which was previously pinned with a call to xfs_inode_item_pin().
465 *
466 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
467 */
468STATIC void
469xfs_inode_item_unpin(
470 struct xfs_log_item *lip,
471 int remove)
472{
473 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
474
475 trace_xfs_inode_unpin(ip, _RET_IP_);
476 ASSERT(atomic_read(&ip->i_pincount) > 0);
477 if (atomic_dec_and_test(&ip->i_pincount))
478 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
479}
480
481STATIC uint
482xfs_inode_item_push(
483 struct xfs_log_item *lip,
484 struct list_head *buffer_list)
485{
486 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
487 struct xfs_inode *ip = iip->ili_inode;
488 struct xfs_buf *bp = NULL;
489 uint rval = XFS_ITEM_SUCCESS;
490 int error;
491
492 if (xfs_ipincount(ip) > 0)
493 return XFS_ITEM_PINNED;
494
495 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
496 return XFS_ITEM_LOCKED;
497
498 /*
499 * Re-check the pincount now that we stabilized the value by
500 * taking the ilock.
501 */
502 if (xfs_ipincount(ip) > 0) {
503 rval = XFS_ITEM_PINNED;
504 goto out_unlock;
505 }
506
507 /*
508 * Stale inode items should force out the iclog.
509 */
510 if (ip->i_flags & XFS_ISTALE) {
511 rval = XFS_ITEM_PINNED;
512 goto out_unlock;
513 }
514
515 /*
516 * Someone else is already flushing the inode. Nothing we can do
517 * here but wait for the flush to finish and remove the item from
518 * the AIL.
519 */
520 if (!xfs_iflock_nowait(ip)) {
521 rval = XFS_ITEM_FLUSHING;
522 goto out_unlock;
523 }
524
525 ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
526 ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
527
528 spin_unlock(&lip->li_ailp->xa_lock);
529
530 error = xfs_iflush(ip, &bp);
531 if (!error) {
532 if (!xfs_buf_delwri_queue(bp, buffer_list))
533 rval = XFS_ITEM_FLUSHING;
534 xfs_buf_relse(bp);
535 }
536
537 spin_lock(&lip->li_ailp->xa_lock);
538out_unlock:
539 xfs_iunlock(ip, XFS_ILOCK_SHARED);
540 return rval;
541}
542
543/*
544 * Unlock the inode associated with the inode log item.
545 * Clear the fields of the inode and inode log item that
546 * are specific to the current transaction. If the
547 * hold flags is set, do not unlock the inode.
548 */
549STATIC void
550xfs_inode_item_unlock(
551 struct xfs_log_item *lip)
552{
553 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
554 struct xfs_inode *ip = iip->ili_inode;
555 unsigned short lock_flags;
556
557 ASSERT(ip->i_itemp != NULL);
558 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
559
560 /*
561 * If the inode needed a separate buffer with which to log
562 * its extents, then free it now.
563 */
564 if (iip->ili_extents_buf != NULL) {
565 ASSERT(ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS);
566 ASSERT(ip->i_d.di_nextents > 0);
567 ASSERT(iip->ili_fields & XFS_ILOG_DEXT);
568 ASSERT(ip->i_df.if_bytes > 0);
569 kmem_free(iip->ili_extents_buf);
570 iip->ili_extents_buf = NULL;
571 }
572 if (iip->ili_aextents_buf != NULL) {
573 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS);
574 ASSERT(ip->i_d.di_anextents > 0);
575 ASSERT(iip->ili_fields & XFS_ILOG_AEXT);
576 ASSERT(ip->i_afp->if_bytes > 0);
577 kmem_free(iip->ili_aextents_buf);
578 iip->ili_aextents_buf = NULL;
579 }
580
581 lock_flags = iip->ili_lock_flags;
582 iip->ili_lock_flags = 0;
583 if (lock_flags)
584 xfs_iunlock(ip, lock_flags);
585}
586
587/*
588 * This is called to find out where the oldest active copy of the inode log
589 * item in the on disk log resides now that the last log write of it completed
590 * at the given lsn. Since we always re-log all dirty data in an inode, the
591 * latest copy in the on disk log is the only one that matters. Therefore,
592 * simply return the given lsn.
593 *
594 * If the inode has been marked stale because the cluster is being freed, we
595 * don't want to (re-)insert this inode into the AIL. There is a race condition
596 * where the cluster buffer may be unpinned before the inode is inserted into
597 * the AIL during transaction committed processing. If the buffer is unpinned
598 * before the inode item has been committed and inserted, then it is possible
599 * for the buffer to be written and IO completes before the inode is inserted
600 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
601 * AIL which will never get removed. It will, however, get reclaimed which
602 * triggers an assert in xfs_inode_free() complaining about freein an inode
603 * still in the AIL.
604 *
605 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
606 * transaction committed code knows that it does not need to do any further
607 * processing on the item.
608 */
609STATIC xfs_lsn_t
610xfs_inode_item_committed(
611 struct xfs_log_item *lip,
612 xfs_lsn_t lsn)
613{
614 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
615 struct xfs_inode *ip = iip->ili_inode;
616
617 if (xfs_iflags_test(ip, XFS_ISTALE)) {
618 xfs_inode_item_unpin(lip, 0);
619 return -1;
620 }
621 return lsn;
622}
623
624/*
625 * XXX rcc - this one really has to do something. Probably needs
626 * to stamp in a new field in the incore inode.
627 */
628STATIC void
629xfs_inode_item_committing(
630 struct xfs_log_item *lip,
631 xfs_lsn_t lsn)
632{
633 INODE_ITEM(lip)->ili_last_lsn = lsn;
634}
635
636/*
637 * This is the ops vector shared by all buf log items.
638 */
639static const struct xfs_item_ops xfs_inode_item_ops = {
640 .iop_size = xfs_inode_item_size,
641 .iop_format = xfs_inode_item_format,
642 .iop_pin = xfs_inode_item_pin,
643 .iop_unpin = xfs_inode_item_unpin,
644 .iop_unlock = xfs_inode_item_unlock,
645 .iop_committed = xfs_inode_item_committed,
646 .iop_push = xfs_inode_item_push,
647 .iop_committing = xfs_inode_item_committing
648};
649
650
651/*
652 * Initialize the inode log item for a newly allocated (in-core) inode.
653 */
654void
655xfs_inode_item_init(
656 struct xfs_inode *ip,
657 struct xfs_mount *mp)
658{
659 struct xfs_inode_log_item *iip;
660
661 ASSERT(ip->i_itemp == NULL);
662 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
663
664 iip->ili_inode = ip;
665 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
666 &xfs_inode_item_ops);
667 iip->ili_format.ilf_type = XFS_LI_INODE;
668 iip->ili_format.ilf_ino = ip->i_ino;
669 iip->ili_format.ilf_blkno = ip->i_imap.im_blkno;
670 iip->ili_format.ilf_len = ip->i_imap.im_len;
671 iip->ili_format.ilf_boffset = ip->i_imap.im_boffset;
672}
673
674/*
675 * Free the inode log item and any memory hanging off of it.
676 */
677void
678xfs_inode_item_destroy(
679 xfs_inode_t *ip)
680{
681#ifdef XFS_TRANS_DEBUG
682 if (ip->i_itemp->ili_root_size != 0) {
683 kmem_free(ip->i_itemp->ili_orig_root);
684 }
685#endif
686 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
687}
688
689
690/*
691 * This is the inode flushing I/O completion routine. It is called
692 * from interrupt level when the buffer containing the inode is
693 * flushed to disk. It is responsible for removing the inode item
694 * from the AIL if it has not been re-logged, and unlocking the inode's
695 * flush lock.
696 *
697 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
698 * list for other inodes that will run this function. We remove them from the
699 * buffer list so we can process all the inode IO completions in one AIL lock
700 * traversal.
701 */
702void
703xfs_iflush_done(
704 struct xfs_buf *bp,
705 struct xfs_log_item *lip)
706{
707 struct xfs_inode_log_item *iip;
708 struct xfs_log_item *blip;
709 struct xfs_log_item *next;
710 struct xfs_log_item *prev;
711 struct xfs_ail *ailp = lip->li_ailp;
712 int need_ail = 0;
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 blip = bp->b_fspriv;
719 prev = NULL;
720 while (blip != NULL) {
721 if (lip->li_cb != xfs_iflush_done) {
722 prev = blip;
723 blip = blip->li_bio_list;
724 continue;
725 }
726
727 /* remove from list */
728 next = blip->li_bio_list;
729 if (!prev) {
730 bp->b_fspriv = next;
731 } else {
732 prev->li_bio_list = next;
733 }
734
735 /* add to current list */
736 blip->li_bio_list = lip->li_bio_list;
737 lip->li_bio_list = blip;
738
739 /*
740 * while we have the item, do the unlocked check for needing
741 * the AIL lock.
742 */
743 iip = INODE_ITEM(blip);
744 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
745 need_ail++;
746
747 blip = next;
748 }
749
750 /* make sure we capture the state of the initial inode. */
751 iip = INODE_ITEM(lip);
752 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
753 need_ail++;
754
755 /*
756 * We only want to pull the item from the AIL if it is
757 * actually there and its location in the log has not
758 * changed since we started the flush. Thus, we only bother
759 * if the ili_logged flag is set and the inode's lsn has not
760 * changed. First we check the lsn outside
761 * the lock since it's cheaper, and then we recheck while
762 * holding the lock before removing the inode from the AIL.
763 */
764 if (need_ail) {
765 struct xfs_log_item *log_items[need_ail];
766 int i = 0;
767 spin_lock(&ailp->xa_lock);
768 for (blip = lip; blip; blip = blip->li_bio_list) {
769 iip = INODE_ITEM(blip);
770 if (iip->ili_logged &&
771 blip->li_lsn == iip->ili_flush_lsn) {
772 log_items[i++] = blip;
773 }
774 ASSERT(i <= need_ail);
775 }
776 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
777 xfs_trans_ail_delete_bulk(ailp, log_items, i,
778 SHUTDOWN_CORRUPT_INCORE);
779 }
780
781
782 /*
783 * clean up and unlock the flush lock now we are done. We can clear the
784 * ili_last_fields bits now that we know that the data corresponding to
785 * them is safely on disk.
786 */
787 for (blip = lip; blip; blip = next) {
788 next = blip->li_bio_list;
789 blip->li_bio_list = NULL;
790
791 iip = INODE_ITEM(blip);
792 iip->ili_logged = 0;
793 iip->ili_last_fields = 0;
794 xfs_ifunlock(iip->ili_inode);
795 }
796}
797
798/*
799 * This is the inode flushing abort routine. It is called from xfs_iflush when
800 * the filesystem is shutting down to clean up the inode state. It is
801 * responsible for removing the inode item from the AIL if it has not been
802 * re-logged, and unlocking the inode's flush lock.
803 */
804void
805xfs_iflush_abort(
806 xfs_inode_t *ip,
807 bool stale)
808{
809 xfs_inode_log_item_t *iip = ip->i_itemp;
810
811 if (iip) {
812 struct xfs_ail *ailp = iip->ili_item.li_ailp;
813 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
814 spin_lock(&ailp->xa_lock);
815 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
816 /* xfs_trans_ail_delete() drops the AIL lock. */
817 xfs_trans_ail_delete(ailp, &iip->ili_item,
818 stale ?
819 SHUTDOWN_LOG_IO_ERROR :
820 SHUTDOWN_CORRUPT_INCORE);
821 } else
822 spin_unlock(&ailp->xa_lock);
823 }
824 iip->ili_logged = 0;
825 /*
826 * Clear the ili_last_fields bits now that we know that the
827 * data corresponding to them is safely on disk.
828 */
829 iip->ili_last_fields = 0;
830 /*
831 * Clear the inode logging fields so no more flushes are
832 * attempted.
833 */
834 iip->ili_fields = 0;
835 }
836 /*
837 * Release the inode's flush lock since we're done with it.
838 */
839 xfs_ifunlock(ip);
840}
841
842void
843xfs_istale_done(
844 struct xfs_buf *bp,
845 struct xfs_log_item *lip)
846{
847 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
848}
849
850/*
851 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
852 * (which can have different field alignments) to the native version
853 */
854int
855xfs_inode_item_format_convert(
856 xfs_log_iovec_t *buf,
857 xfs_inode_log_format_t *in_f)
858{
859 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
860 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
861
862 in_f->ilf_type = in_f32->ilf_type;
863 in_f->ilf_size = in_f32->ilf_size;
864 in_f->ilf_fields = in_f32->ilf_fields;
865 in_f->ilf_asize = in_f32->ilf_asize;
866 in_f->ilf_dsize = in_f32->ilf_dsize;
867 in_f->ilf_ino = in_f32->ilf_ino;
868 /* copy biggest field of ilf_u */
869 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
870 in_f32->ilf_u.ilfu_uuid.__u_bits,
871 sizeof(uuid_t));
872 in_f->ilf_blkno = in_f32->ilf_blkno;
873 in_f->ilf_len = in_f32->ilf_len;
874 in_f->ilf_boffset = in_f32->ilf_boffset;
875 return 0;
876 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
877 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
878
879 in_f->ilf_type = in_f64->ilf_type;
880 in_f->ilf_size = in_f64->ilf_size;
881 in_f->ilf_fields = in_f64->ilf_fields;
882 in_f->ilf_asize = in_f64->ilf_asize;
883 in_f->ilf_dsize = in_f64->ilf_dsize;
884 in_f->ilf_ino = in_f64->ilf_ino;
885 /* copy biggest field of ilf_u */
886 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
887 in_f64->ilf_u.ilfu_uuid.__u_bits,
888 sizeof(uuid_t));
889 in_f->ilf_blkno = in_f64->ilf_blkno;
890 in_f->ilf_len = in_f64->ilf_len;
891 in_f->ilf_boffset = in_f64->ilf_boffset;
892 return 0;
893 }
894 return EFSCORRUPTED;
895}