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_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
23#include "xfs_mount.h"
24#include "xfs_inode.h"
25#include "xfs_trans.h"
26#include "xfs_inode_item.h"
27#include "xfs_error.h"
28#include "xfs_trace.h"
29#include "xfs_trans_priv.h"
30#include "xfs_log.h"
31
32
33kmem_zone_t *xfs_ili_zone; /* inode log item zone */
34
35static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
36{
37 return container_of(lip, struct xfs_inode_log_item, ili_item);
38}
39
40STATIC void
41xfs_inode_item_data_fork_size(
42 struct xfs_inode_log_item *iip,
43 int *nvecs,
44 int *nbytes)
45{
46 struct xfs_inode *ip = iip->ili_inode;
47
48 switch (ip->i_d.di_format) {
49 case XFS_DINODE_FMT_EXTENTS:
50 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
51 ip->i_d.di_nextents > 0 &&
52 ip->i_df.if_bytes > 0) {
53 /* worst case, doesn't subtract delalloc extents */
54 *nbytes += XFS_IFORK_DSIZE(ip);
55 *nvecs += 1;
56 }
57 break;
58 case XFS_DINODE_FMT_BTREE:
59 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
60 ip->i_df.if_broot_bytes > 0) {
61 *nbytes += ip->i_df.if_broot_bytes;
62 *nvecs += 1;
63 }
64 break;
65 case XFS_DINODE_FMT_LOCAL:
66 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
67 ip->i_df.if_bytes > 0) {
68 *nbytes += roundup(ip->i_df.if_bytes, 4);
69 *nvecs += 1;
70 }
71 break;
72
73 case XFS_DINODE_FMT_DEV:
74 case XFS_DINODE_FMT_UUID:
75 break;
76 default:
77 ASSERT(0);
78 break;
79 }
80}
81
82STATIC void
83xfs_inode_item_attr_fork_size(
84 struct xfs_inode_log_item *iip,
85 int *nvecs,
86 int *nbytes)
87{
88 struct xfs_inode *ip = iip->ili_inode;
89
90 switch (ip->i_d.di_aformat) {
91 case XFS_DINODE_FMT_EXTENTS:
92 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
93 ip->i_d.di_anextents > 0 &&
94 ip->i_afp->if_bytes > 0) {
95 /* worst case, doesn't subtract unused space */
96 *nbytes += XFS_IFORK_ASIZE(ip);
97 *nvecs += 1;
98 }
99 break;
100 case XFS_DINODE_FMT_BTREE:
101 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
102 ip->i_afp->if_broot_bytes > 0) {
103 *nbytes += ip->i_afp->if_broot_bytes;
104 *nvecs += 1;
105 }
106 break;
107 case XFS_DINODE_FMT_LOCAL:
108 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
109 ip->i_afp->if_bytes > 0) {
110 *nbytes += roundup(ip->i_afp->if_bytes, 4);
111 *nvecs += 1;
112 }
113 break;
114 default:
115 ASSERT(0);
116 break;
117 }
118}
119
120/*
121 * This returns the number of iovecs needed to log the given inode item.
122 *
123 * We need one iovec for the inode log format structure, one for the
124 * inode core, and possibly one for the inode data/extents/b-tree root
125 * and one for the inode attribute data/extents/b-tree root.
126 */
127STATIC void
128xfs_inode_item_size(
129 struct xfs_log_item *lip,
130 int *nvecs,
131 int *nbytes)
132{
133 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
134 struct xfs_inode *ip = iip->ili_inode;
135
136 *nvecs += 2;
137 *nbytes += sizeof(struct xfs_inode_log_format) +
138 xfs_log_dinode_size(ip->i_d.di_version);
139
140 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);
141 if (XFS_IFORK_Q(ip))
142 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);
143}
144
145STATIC void
146xfs_inode_item_format_data_fork(
147 struct xfs_inode_log_item *iip,
148 struct xfs_inode_log_format *ilf,
149 struct xfs_log_vec *lv,
150 struct xfs_log_iovec **vecp)
151{
152 struct xfs_inode *ip = iip->ili_inode;
153 size_t data_bytes;
154
155 switch (ip->i_d.di_format) {
156 case XFS_DINODE_FMT_EXTENTS:
157 iip->ili_fields &=
158 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
159 XFS_ILOG_DEV | XFS_ILOG_UUID);
160
161 if ((iip->ili_fields & XFS_ILOG_DEXT) &&
162 ip->i_d.di_nextents > 0 &&
163 ip->i_df.if_bytes > 0) {
164 struct xfs_bmbt_rec *p;
165
166 ASSERT(ip->i_df.if_u1.if_extents != NULL);
167 ASSERT(ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) > 0);
168
169 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);
170 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);
171 xlog_finish_iovec(lv, *vecp, data_bytes);
172
173 ASSERT(data_bytes <= ip->i_df.if_bytes);
174
175 ilf->ilf_dsize = data_bytes;
176 ilf->ilf_size++;
177 } else {
178 iip->ili_fields &= ~XFS_ILOG_DEXT;
179 }
180 break;
181 case XFS_DINODE_FMT_BTREE:
182 iip->ili_fields &=
183 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT |
184 XFS_ILOG_DEV | XFS_ILOG_UUID);
185
186 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
187 ip->i_df.if_broot_bytes > 0) {
188 ASSERT(ip->i_df.if_broot != NULL);
189 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
190 ip->i_df.if_broot,
191 ip->i_df.if_broot_bytes);
192 ilf->ilf_dsize = ip->i_df.if_broot_bytes;
193 ilf->ilf_size++;
194 } else {
195 ASSERT(!(iip->ili_fields &
196 XFS_ILOG_DBROOT));
197 iip->ili_fields &= ~XFS_ILOG_DBROOT;
198 }
199 break;
200 case XFS_DINODE_FMT_LOCAL:
201 iip->ili_fields &=
202 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT |
203 XFS_ILOG_DEV | XFS_ILOG_UUID);
204 if ((iip->ili_fields & XFS_ILOG_DDATA) &&
205 ip->i_df.if_bytes > 0) {
206 /*
207 * Round i_bytes up to a word boundary.
208 * The underlying memory is guaranteed to
209 * to be there by xfs_idata_realloc().
210 */
211 data_bytes = roundup(ip->i_df.if_bytes, 4);
212 ASSERT(ip->i_df.if_real_bytes == 0 ||
213 ip->i_df.if_real_bytes == data_bytes);
214 ASSERT(ip->i_df.if_u1.if_data != NULL);
215 ASSERT(ip->i_d.di_size > 0);
216 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
217 ip->i_df.if_u1.if_data, data_bytes);
218 ilf->ilf_dsize = (unsigned)data_bytes;
219 ilf->ilf_size++;
220 } else {
221 iip->ili_fields &= ~XFS_ILOG_DDATA;
222 }
223 break;
224 case XFS_DINODE_FMT_DEV:
225 iip->ili_fields &=
226 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
227 XFS_ILOG_DEXT | XFS_ILOG_UUID);
228 if (iip->ili_fields & XFS_ILOG_DEV)
229 ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
230 break;
231 case XFS_DINODE_FMT_UUID:
232 iip->ili_fields &=
233 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
234 XFS_ILOG_DEXT | XFS_ILOG_DEV);
235 if (iip->ili_fields & XFS_ILOG_UUID)
236 ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
237 break;
238 default:
239 ASSERT(0);
240 break;
241 }
242}
243
244STATIC void
245xfs_inode_item_format_attr_fork(
246 struct xfs_inode_log_item *iip,
247 struct xfs_inode_log_format *ilf,
248 struct xfs_log_vec *lv,
249 struct xfs_log_iovec **vecp)
250{
251 struct xfs_inode *ip = iip->ili_inode;
252 size_t data_bytes;
253
254 switch (ip->i_d.di_aformat) {
255 case XFS_DINODE_FMT_EXTENTS:
256 iip->ili_fields &=
257 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);
258
259 if ((iip->ili_fields & XFS_ILOG_AEXT) &&
260 ip->i_d.di_anextents > 0 &&
261 ip->i_afp->if_bytes > 0) {
262 struct xfs_bmbt_rec *p;
263
264 ASSERT(ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) ==
265 ip->i_d.di_anextents);
266 ASSERT(ip->i_afp->if_u1.if_extents != NULL);
267
268 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);
269 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);
270 xlog_finish_iovec(lv, *vecp, data_bytes);
271
272 ilf->ilf_asize = data_bytes;
273 ilf->ilf_size++;
274 } else {
275 iip->ili_fields &= ~XFS_ILOG_AEXT;
276 }
277 break;
278 case XFS_DINODE_FMT_BTREE:
279 iip->ili_fields &=
280 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);
281
282 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&
283 ip->i_afp->if_broot_bytes > 0) {
284 ASSERT(ip->i_afp->if_broot != NULL);
285
286 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
287 ip->i_afp->if_broot,
288 ip->i_afp->if_broot_bytes);
289 ilf->ilf_asize = ip->i_afp->if_broot_bytes;
290 ilf->ilf_size++;
291 } else {
292 iip->ili_fields &= ~XFS_ILOG_ABROOT;
293 }
294 break;
295 case XFS_DINODE_FMT_LOCAL:
296 iip->ili_fields &=
297 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);
298
299 if ((iip->ili_fields & XFS_ILOG_ADATA) &&
300 ip->i_afp->if_bytes > 0) {
301 /*
302 * Round i_bytes up to a word boundary.
303 * The underlying memory is guaranteed to
304 * to be there by xfs_idata_realloc().
305 */
306 data_bytes = roundup(ip->i_afp->if_bytes, 4);
307 ASSERT(ip->i_afp->if_real_bytes == 0 ||
308 ip->i_afp->if_real_bytes == data_bytes);
309 ASSERT(ip->i_afp->if_u1.if_data != NULL);
310 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
311 ip->i_afp->if_u1.if_data,
312 data_bytes);
313 ilf->ilf_asize = (unsigned)data_bytes;
314 ilf->ilf_size++;
315 } else {
316 iip->ili_fields &= ~XFS_ILOG_ADATA;
317 }
318 break;
319 default:
320 ASSERT(0);
321 break;
322 }
323}
324
325static void
326xfs_inode_to_log_dinode(
327 struct xfs_inode *ip,
328 struct xfs_log_dinode *to,
329 xfs_lsn_t lsn)
330{
331 struct xfs_icdinode *from = &ip->i_d;
332 struct inode *inode = VFS_I(ip);
333
334 to->di_magic = XFS_DINODE_MAGIC;
335
336 to->di_version = from->di_version;
337 to->di_format = from->di_format;
338 to->di_uid = from->di_uid;
339 to->di_gid = from->di_gid;
340 to->di_projid_lo = from->di_projid_lo;
341 to->di_projid_hi = from->di_projid_hi;
342
343 memset(to->di_pad, 0, sizeof(to->di_pad));
344 memset(to->di_pad3, 0, sizeof(to->di_pad3));
345 to->di_atime.t_sec = inode->i_atime.tv_sec;
346 to->di_atime.t_nsec = inode->i_atime.tv_nsec;
347 to->di_mtime.t_sec = inode->i_mtime.tv_sec;
348 to->di_mtime.t_nsec = inode->i_mtime.tv_nsec;
349 to->di_ctime.t_sec = inode->i_ctime.tv_sec;
350 to->di_ctime.t_nsec = inode->i_ctime.tv_nsec;
351 to->di_nlink = inode->i_nlink;
352 to->di_gen = inode->i_generation;
353 to->di_mode = inode->i_mode;
354
355 to->di_size = from->di_size;
356 to->di_nblocks = from->di_nblocks;
357 to->di_extsize = from->di_extsize;
358 to->di_nextents = from->di_nextents;
359 to->di_anextents = from->di_anextents;
360 to->di_forkoff = from->di_forkoff;
361 to->di_aformat = from->di_aformat;
362 to->di_dmevmask = from->di_dmevmask;
363 to->di_dmstate = from->di_dmstate;
364 to->di_flags = from->di_flags;
365
366 if (from->di_version == 3) {
367 to->di_changecount = inode->i_version;
368 to->di_crtime.t_sec = from->di_crtime.t_sec;
369 to->di_crtime.t_nsec = from->di_crtime.t_nsec;
370 to->di_flags2 = from->di_flags2;
371
372 to->di_ino = ip->i_ino;
373 to->di_lsn = lsn;
374 memset(to->di_pad2, 0, sizeof(to->di_pad2));
375 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
376 to->di_flushiter = 0;
377 } else {
378 to->di_flushiter = from->di_flushiter;
379 }
380}
381
382/*
383 * Format the inode core. Current timestamp data is only in the VFS inode
384 * fields, so we need to grab them from there. Hence rather than just copying
385 * the XFS inode core structure, format the fields directly into the iovec.
386 */
387static void
388xfs_inode_item_format_core(
389 struct xfs_inode *ip,
390 struct xfs_log_vec *lv,
391 struct xfs_log_iovec **vecp)
392{
393 struct xfs_log_dinode *dic;
394
395 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);
396 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);
397 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_d.di_version));
398}
399
400/*
401 * This is called to fill in the vector of log iovecs for the given inode
402 * log item. It fills the first item with an inode log format structure,
403 * the second with the on-disk inode structure, and a possible third and/or
404 * fourth with the inode data/extents/b-tree root and inode attributes
405 * data/extents/b-tree root.
406 */
407STATIC void
408xfs_inode_item_format(
409 struct xfs_log_item *lip,
410 struct xfs_log_vec *lv)
411{
412 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
413 struct xfs_inode *ip = iip->ili_inode;
414 struct xfs_inode_log_format *ilf;
415 struct xfs_log_iovec *vecp = NULL;
416
417 ASSERT(ip->i_d.di_version > 1);
418
419 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);
420 ilf->ilf_type = XFS_LI_INODE;
421 ilf->ilf_ino = ip->i_ino;
422 ilf->ilf_blkno = ip->i_imap.im_blkno;
423 ilf->ilf_len = ip->i_imap.im_len;
424 ilf->ilf_boffset = ip->i_imap.im_boffset;
425 ilf->ilf_fields = XFS_ILOG_CORE;
426 ilf->ilf_size = 2; /* format + core */
427 xlog_finish_iovec(lv, vecp, sizeof(struct xfs_inode_log_format));
428
429 xfs_inode_item_format_core(ip, lv, &vecp);
430 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
431 if (XFS_IFORK_Q(ip)) {
432 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
433 } else {
434 iip->ili_fields &=
435 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
436 }
437
438 /* update the format with the exact fields we actually logged */
439 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
440}
441
442/*
443 * This is called to pin the inode associated with the inode log
444 * item in memory so it cannot be written out.
445 */
446STATIC void
447xfs_inode_item_pin(
448 struct xfs_log_item *lip)
449{
450 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
451
452 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
453
454 trace_xfs_inode_pin(ip, _RET_IP_);
455 atomic_inc(&ip->i_pincount);
456}
457
458
459/*
460 * This is called to unpin the inode associated with the inode log
461 * item which was previously pinned with a call to xfs_inode_item_pin().
462 *
463 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.
464 */
465STATIC void
466xfs_inode_item_unpin(
467 struct xfs_log_item *lip,
468 int remove)
469{
470 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;
471
472 trace_xfs_inode_unpin(ip, _RET_IP_);
473 ASSERT(atomic_read(&ip->i_pincount) > 0);
474 if (atomic_dec_and_test(&ip->i_pincount))
475 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
476}
477
478STATIC uint
479xfs_inode_item_push(
480 struct xfs_log_item *lip,
481 struct list_head *buffer_list)
482{
483 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
484 struct xfs_inode *ip = iip->ili_inode;
485 struct xfs_buf *bp = NULL;
486 uint rval = XFS_ITEM_SUCCESS;
487 int error;
488
489 if (xfs_ipincount(ip) > 0)
490 return XFS_ITEM_PINNED;
491
492 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
493 return XFS_ITEM_LOCKED;
494
495 /*
496 * Re-check the pincount now that we stabilized the value by
497 * taking the ilock.
498 */
499 if (xfs_ipincount(ip) > 0) {
500 rval = XFS_ITEM_PINNED;
501 goto out_unlock;
502 }
503
504 /*
505 * Stale inode items should force out the iclog.
506 */
507 if (ip->i_flags & XFS_ISTALE) {
508 rval = XFS_ITEM_PINNED;
509 goto out_unlock;
510 }
511
512 /*
513 * Someone else is already flushing the inode. Nothing we can do
514 * here but wait for the flush to finish and remove the item from
515 * the AIL.
516 */
517 if (!xfs_iflock_nowait(ip)) {
518 rval = XFS_ITEM_FLUSHING;
519 goto out_unlock;
520 }
521
522 ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
523 ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
524
525 spin_unlock(&lip->li_ailp->xa_lock);
526
527 error = xfs_iflush(ip, &bp);
528 if (!error) {
529 if (!xfs_buf_delwri_queue(bp, buffer_list))
530 rval = XFS_ITEM_FLUSHING;
531 xfs_buf_relse(bp);
532 }
533
534 spin_lock(&lip->li_ailp->xa_lock);
535out_unlock:
536 xfs_iunlock(ip, XFS_ILOCK_SHARED);
537 return rval;
538}
539
540/*
541 * Unlock the inode associated with the inode log item.
542 * Clear the fields of the inode and inode log item that
543 * are specific to the current transaction. If the
544 * hold flags is set, do not unlock the inode.
545 */
546STATIC void
547xfs_inode_item_unlock(
548 struct xfs_log_item *lip)
549{
550 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
551 struct xfs_inode *ip = iip->ili_inode;
552 unsigned short lock_flags;
553
554 ASSERT(ip->i_itemp != NULL);
555 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
556
557 lock_flags = iip->ili_lock_flags;
558 iip->ili_lock_flags = 0;
559 if (lock_flags)
560 xfs_iunlock(ip, lock_flags);
561}
562
563/*
564 * This is called to find out where the oldest active copy of the inode log
565 * item in the on disk log resides now that the last log write of it completed
566 * at the given lsn. Since we always re-log all dirty data in an inode, the
567 * latest copy in the on disk log is the only one that matters. Therefore,
568 * simply return the given lsn.
569 *
570 * If the inode has been marked stale because the cluster is being freed, we
571 * don't want to (re-)insert this inode into the AIL. There is a race condition
572 * where the cluster buffer may be unpinned before the inode is inserted into
573 * the AIL during transaction committed processing. If the buffer is unpinned
574 * before the inode item has been committed and inserted, then it is possible
575 * for the buffer to be written and IO completes before the inode is inserted
576 * into the AIL. In that case, we'd be inserting a clean, stale inode into the
577 * AIL which will never get removed. It will, however, get reclaimed which
578 * triggers an assert in xfs_inode_free() complaining about freein an inode
579 * still in the AIL.
580 *
581 * To avoid this, just unpin the inode directly and return a LSN of -1 so the
582 * transaction committed code knows that it does not need to do any further
583 * processing on the item.
584 */
585STATIC xfs_lsn_t
586xfs_inode_item_committed(
587 struct xfs_log_item *lip,
588 xfs_lsn_t lsn)
589{
590 struct xfs_inode_log_item *iip = INODE_ITEM(lip);
591 struct xfs_inode *ip = iip->ili_inode;
592
593 if (xfs_iflags_test(ip, XFS_ISTALE)) {
594 xfs_inode_item_unpin(lip, 0);
595 return -1;
596 }
597 return lsn;
598}
599
600/*
601 * XXX rcc - this one really has to do something. Probably needs
602 * to stamp in a new field in the incore inode.
603 */
604STATIC void
605xfs_inode_item_committing(
606 struct xfs_log_item *lip,
607 xfs_lsn_t lsn)
608{
609 INODE_ITEM(lip)->ili_last_lsn = lsn;
610}
611
612/*
613 * This is the ops vector shared by all buf log items.
614 */
615static const struct xfs_item_ops xfs_inode_item_ops = {
616 .iop_size = xfs_inode_item_size,
617 .iop_format = xfs_inode_item_format,
618 .iop_pin = xfs_inode_item_pin,
619 .iop_unpin = xfs_inode_item_unpin,
620 .iop_unlock = xfs_inode_item_unlock,
621 .iop_committed = xfs_inode_item_committed,
622 .iop_push = xfs_inode_item_push,
623 .iop_committing = xfs_inode_item_committing
624};
625
626
627/*
628 * Initialize the inode log item for a newly allocated (in-core) inode.
629 */
630void
631xfs_inode_item_init(
632 struct xfs_inode *ip,
633 struct xfs_mount *mp)
634{
635 struct xfs_inode_log_item *iip;
636
637 ASSERT(ip->i_itemp == NULL);
638 iip = ip->i_itemp = kmem_zone_zalloc(xfs_ili_zone, KM_SLEEP);
639
640 iip->ili_inode = ip;
641 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,
642 &xfs_inode_item_ops);
643}
644
645/*
646 * Free the inode log item and any memory hanging off of it.
647 */
648void
649xfs_inode_item_destroy(
650 xfs_inode_t *ip)
651{
652 kmem_zone_free(xfs_ili_zone, ip->i_itemp);
653}
654
655
656/*
657 * This is the inode flushing I/O completion routine. It is called
658 * from interrupt level when the buffer containing the inode is
659 * flushed to disk. It is responsible for removing the inode item
660 * from the AIL if it has not been re-logged, and unlocking the inode's
661 * flush lock.
662 *
663 * To reduce AIL lock traffic as much as possible, we scan the buffer log item
664 * list for other inodes that will run this function. We remove them from the
665 * buffer list so we can process all the inode IO completions in one AIL lock
666 * traversal.
667 */
668void
669xfs_iflush_done(
670 struct xfs_buf *bp,
671 struct xfs_log_item *lip)
672{
673 struct xfs_inode_log_item *iip;
674 struct xfs_log_item *blip;
675 struct xfs_log_item *next;
676 struct xfs_log_item *prev;
677 struct xfs_ail *ailp = lip->li_ailp;
678 int need_ail = 0;
679
680 /*
681 * Scan the buffer IO completions for other inodes being completed and
682 * attach them to the current inode log item.
683 */
684 blip = bp->b_fspriv;
685 prev = NULL;
686 while (blip != NULL) {
687 if (blip->li_cb != xfs_iflush_done) {
688 prev = blip;
689 blip = blip->li_bio_list;
690 continue;
691 }
692
693 /* remove from list */
694 next = blip->li_bio_list;
695 if (!prev) {
696 bp->b_fspriv = next;
697 } else {
698 prev->li_bio_list = next;
699 }
700
701 /* add to current list */
702 blip->li_bio_list = lip->li_bio_list;
703 lip->li_bio_list = blip;
704
705 /*
706 * while we have the item, do the unlocked check for needing
707 * the AIL lock.
708 */
709 iip = INODE_ITEM(blip);
710 if (iip->ili_logged && blip->li_lsn == iip->ili_flush_lsn)
711 need_ail++;
712
713 blip = next;
714 }
715
716 /* make sure we capture the state of the initial inode. */
717 iip = INODE_ITEM(lip);
718 if (iip->ili_logged && lip->li_lsn == iip->ili_flush_lsn)
719 need_ail++;
720
721 /*
722 * We only want to pull the item from the AIL if it is
723 * actually there and its location in the log has not
724 * changed since we started the flush. Thus, we only bother
725 * if the ili_logged flag is set and the inode's lsn has not
726 * changed. First we check the lsn outside
727 * the lock since it's cheaper, and then we recheck while
728 * holding the lock before removing the inode from the AIL.
729 */
730 if (need_ail) {
731 struct xfs_log_item *log_items[need_ail];
732 int i = 0;
733 spin_lock(&ailp->xa_lock);
734 for (blip = lip; blip; blip = blip->li_bio_list) {
735 iip = INODE_ITEM(blip);
736 if (iip->ili_logged &&
737 blip->li_lsn == iip->ili_flush_lsn) {
738 log_items[i++] = blip;
739 }
740 ASSERT(i <= need_ail);
741 }
742 /* xfs_trans_ail_delete_bulk() drops the AIL lock. */
743 xfs_trans_ail_delete_bulk(ailp, log_items, i,
744 SHUTDOWN_CORRUPT_INCORE);
745 }
746
747
748 /*
749 * clean up and unlock the flush lock now we are done. We can clear the
750 * ili_last_fields bits now that we know that the data corresponding to
751 * them is safely on disk.
752 */
753 for (blip = lip; blip; blip = next) {
754 next = blip->li_bio_list;
755 blip->li_bio_list = NULL;
756
757 iip = INODE_ITEM(blip);
758 iip->ili_logged = 0;
759 iip->ili_last_fields = 0;
760 xfs_ifunlock(iip->ili_inode);
761 }
762}
763
764/*
765 * This is the inode flushing abort routine. It is called from xfs_iflush when
766 * the filesystem is shutting down to clean up the inode state. It is
767 * responsible for removing the inode item from the AIL if it has not been
768 * re-logged, and unlocking the inode's flush lock.
769 */
770void
771xfs_iflush_abort(
772 xfs_inode_t *ip,
773 bool stale)
774{
775 xfs_inode_log_item_t *iip = ip->i_itemp;
776
777 if (iip) {
778 if (iip->ili_item.li_flags & XFS_LI_IN_AIL) {
779 xfs_trans_ail_remove(&iip->ili_item,
780 stale ? SHUTDOWN_LOG_IO_ERROR :
781 SHUTDOWN_CORRUPT_INCORE);
782 }
783 iip->ili_logged = 0;
784 /*
785 * Clear the ili_last_fields bits now that we know that the
786 * data corresponding to them is safely on disk.
787 */
788 iip->ili_last_fields = 0;
789 /*
790 * Clear the inode logging fields so no more flushes are
791 * attempted.
792 */
793 iip->ili_fields = 0;
794 iip->ili_fsync_fields = 0;
795 }
796 /*
797 * Release the inode's flush lock since we're done with it.
798 */
799 xfs_ifunlock(ip);
800}
801
802void
803xfs_istale_done(
804 struct xfs_buf *bp,
805 struct xfs_log_item *lip)
806{
807 xfs_iflush_abort(INODE_ITEM(lip)->ili_inode, true);
808}
809
810/*
811 * convert an xfs_inode_log_format struct from either 32 or 64 bit versions
812 * (which can have different field alignments) to the native version
813 */
814int
815xfs_inode_item_format_convert(
816 xfs_log_iovec_t *buf,
817 xfs_inode_log_format_t *in_f)
818{
819 if (buf->i_len == sizeof(xfs_inode_log_format_32_t)) {
820 xfs_inode_log_format_32_t *in_f32 = buf->i_addr;
821
822 in_f->ilf_type = in_f32->ilf_type;
823 in_f->ilf_size = in_f32->ilf_size;
824 in_f->ilf_fields = in_f32->ilf_fields;
825 in_f->ilf_asize = in_f32->ilf_asize;
826 in_f->ilf_dsize = in_f32->ilf_dsize;
827 in_f->ilf_ino = in_f32->ilf_ino;
828 /* copy biggest field of ilf_u */
829 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
830 in_f32->ilf_u.ilfu_uuid.__u_bits,
831 sizeof(uuid_t));
832 in_f->ilf_blkno = in_f32->ilf_blkno;
833 in_f->ilf_len = in_f32->ilf_len;
834 in_f->ilf_boffset = in_f32->ilf_boffset;
835 return 0;
836 } else if (buf->i_len == sizeof(xfs_inode_log_format_64_t)){
837 xfs_inode_log_format_64_t *in_f64 = buf->i_addr;
838
839 in_f->ilf_type = in_f64->ilf_type;
840 in_f->ilf_size = in_f64->ilf_size;
841 in_f->ilf_fields = in_f64->ilf_fields;
842 in_f->ilf_asize = in_f64->ilf_asize;
843 in_f->ilf_dsize = in_f64->ilf_dsize;
844 in_f->ilf_ino = in_f64->ilf_ino;
845 /* copy biggest field of ilf_u */
846 memcpy(in_f->ilf_u.ilfu_uuid.__u_bits,
847 in_f64->ilf_u.ilfu_uuid.__u_bits,
848 sizeof(uuid_t));
849 in_f->ilf_blkno = in_f64->ilf_blkno;
850 in_f->ilf_len = in_f64->ilf_len;
851 in_f->ilf_boffset = in_f64->ilf_boffset;
852 return 0;
853 }
854 return -EFSCORRUPTED;
855}