Merge branch 'media_fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
[cris-mirror.git] / fs / xfs / xfs_inode.c
blobbe7cf625421f15a50a739af9dfc1166981a3fd06
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
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.
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.
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
18 #include <linux/log2.h>
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_types.h"
23 #include "xfs_bit.h"
24 #include "xfs_log.h"
25 #include "xfs_inum.h"
26 #include "xfs_trans.h"
27 #include "xfs_trans_priv.h"
28 #include "xfs_sb.h"
29 #include "xfs_ag.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_btree.h"
40 #include "xfs_btree_trace.h"
41 #include "xfs_alloc.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_bmap.h"
44 #include "xfs_error.h"
45 #include "xfs_utils.h"
46 #include "xfs_quota.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
51 kmem_zone_t *xfs_ifork_zone;
52 kmem_zone_t *xfs_inode_zone;
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
58 #define XFS_ITRUNC_MAX_EXTENTS 2
60 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
61 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
62 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
63 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
65 #ifdef DEBUG
67 * Make sure that the extents in the given memory buffer
68 * are valid.
70 STATIC void
71 xfs_validate_extents(
72 xfs_ifork_t *ifp,
73 int nrecs,
74 xfs_exntfmt_t fmt)
76 xfs_bmbt_irec_t irec;
77 xfs_bmbt_rec_host_t rec;
78 int i;
80 for (i = 0; i < nrecs; i++) {
81 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
82 rec.l0 = get_unaligned(&ep->l0);
83 rec.l1 = get_unaligned(&ep->l1);
84 xfs_bmbt_get_all(&rec, &irec);
85 if (fmt == XFS_EXTFMT_NOSTATE)
86 ASSERT(irec.br_state == XFS_EXT_NORM);
89 #else /* DEBUG */
90 #define xfs_validate_extents(ifp, nrecs, fmt)
91 #endif /* DEBUG */
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
97 #if defined(DEBUG)
98 void
99 xfs_inobp_check(
100 xfs_mount_t *mp,
101 xfs_buf_t *bp)
103 int i;
104 int j;
105 xfs_dinode_t *dip;
107 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
109 for (i = 0; i < j; i++) {
110 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
111 i * mp->m_sb.sb_inodesize);
112 if (!dip->di_next_unlinked) {
113 xfs_fs_cmn_err(CE_ALERT, mp,
114 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
115 bp);
116 ASSERT(dip->di_next_unlinked);
120 #endif
123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
127 STATIC int
128 xfs_imap_to_bp(
129 xfs_mount_t *mp,
130 xfs_trans_t *tp,
131 struct xfs_imap *imap,
132 xfs_buf_t **bpp,
133 uint buf_flags,
134 uint iget_flags)
136 int error;
137 int i;
138 int ni;
139 xfs_buf_t *bp;
141 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
142 (int)imap->im_len, buf_flags, &bp);
143 if (error) {
144 if (error != EAGAIN) {
145 cmn_err(CE_WARN,
146 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
147 "an error %d on %s. Returning error.",
148 error, mp->m_fsname);
149 } else {
150 ASSERT(buf_flags & XBF_TRYLOCK);
152 return error;
156 * Validate the magic number and version of every inode in the buffer
157 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
159 #ifdef DEBUG
160 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
161 #else /* usual case */
162 ni = 1;
163 #endif
165 for (i = 0; i < ni; i++) {
166 int di_ok;
167 xfs_dinode_t *dip;
169 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
170 (i << mp->m_sb.sb_inodelog));
171 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
172 XFS_DINODE_GOOD_VERSION(dip->di_version);
173 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
174 XFS_ERRTAG_ITOBP_INOTOBP,
175 XFS_RANDOM_ITOBP_INOTOBP))) {
176 if (iget_flags & XFS_IGET_UNTRUSTED) {
177 xfs_trans_brelse(tp, bp);
178 return XFS_ERROR(EINVAL);
180 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
181 XFS_ERRLEVEL_HIGH, mp, dip);
182 #ifdef DEBUG
183 cmn_err(CE_PANIC,
184 "Device %s - bad inode magic/vsn "
185 "daddr %lld #%d (magic=%x)",
186 XFS_BUFTARG_NAME(mp->m_ddev_targp),
187 (unsigned long long)imap->im_blkno, i,
188 be16_to_cpu(dip->di_magic));
189 #endif
190 xfs_trans_brelse(tp, bp);
191 return XFS_ERROR(EFSCORRUPTED);
195 xfs_inobp_check(mp, bp);
198 * Mark the buffer as an inode buffer now that it looks good
200 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
202 *bpp = bp;
203 return 0;
207 * This routine is called to map an inode number within a file
208 * system to the buffer containing the on-disk version of the
209 * inode. It returns a pointer to the buffer containing the
210 * on-disk inode in the bpp parameter, and in the dip parameter
211 * it returns a pointer to the on-disk inode within that buffer.
213 * If a non-zero error is returned, then the contents of bpp and
214 * dipp are undefined.
216 * Use xfs_imap() to determine the size and location of the
217 * buffer to read from disk.
220 xfs_inotobp(
221 xfs_mount_t *mp,
222 xfs_trans_t *tp,
223 xfs_ino_t ino,
224 xfs_dinode_t **dipp,
225 xfs_buf_t **bpp,
226 int *offset,
227 uint imap_flags)
229 struct xfs_imap imap;
230 xfs_buf_t *bp;
231 int error;
233 imap.im_blkno = 0;
234 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
235 if (error)
236 return error;
238 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
239 if (error)
240 return error;
242 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
243 *bpp = bp;
244 *offset = imap.im_boffset;
245 return 0;
250 * This routine is called to map an inode to the buffer containing
251 * the on-disk version of the inode. It returns a pointer to the
252 * buffer containing the on-disk inode in the bpp parameter, and in
253 * the dip parameter it returns a pointer to the on-disk inode within
254 * that buffer.
256 * If a non-zero error is returned, then the contents of bpp and
257 * dipp are undefined.
259 * The inode is expected to already been mapped to its buffer and read
260 * in once, thus we can use the mapping information stored in the inode
261 * rather than calling xfs_imap(). This allows us to avoid the overhead
262 * of looking at the inode btree for small block file systems
263 * (see xfs_imap()).
266 xfs_itobp(
267 xfs_mount_t *mp,
268 xfs_trans_t *tp,
269 xfs_inode_t *ip,
270 xfs_dinode_t **dipp,
271 xfs_buf_t **bpp,
272 uint buf_flags)
274 xfs_buf_t *bp;
275 int error;
277 ASSERT(ip->i_imap.im_blkno != 0);
279 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
280 if (error)
281 return error;
283 if (!bp) {
284 ASSERT(buf_flags & XBF_TRYLOCK);
285 ASSERT(tp == NULL);
286 *bpp = NULL;
287 return EAGAIN;
290 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
291 *bpp = bp;
292 return 0;
296 * Move inode type and inode format specific information from the
297 * on-disk inode to the in-core inode. For fifos, devs, and sockets
298 * this means set if_rdev to the proper value. For files, directories,
299 * and symlinks this means to bring in the in-line data or extent
300 * pointers. For a file in B-tree format, only the root is immediately
301 * brought in-core. The rest will be in-lined in if_extents when it
302 * is first referenced (see xfs_iread_extents()).
304 STATIC int
305 xfs_iformat(
306 xfs_inode_t *ip,
307 xfs_dinode_t *dip)
309 xfs_attr_shortform_t *atp;
310 int size;
311 int error;
312 xfs_fsize_t di_size;
313 ip->i_df.if_ext_max =
314 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
315 error = 0;
317 if (unlikely(be32_to_cpu(dip->di_nextents) +
318 be16_to_cpu(dip->di_anextents) >
319 be64_to_cpu(dip->di_nblocks))) {
320 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
321 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
322 (unsigned long long)ip->i_ino,
323 (int)(be32_to_cpu(dip->di_nextents) +
324 be16_to_cpu(dip->di_anextents)),
325 (unsigned long long)
326 be64_to_cpu(dip->di_nblocks));
327 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
328 ip->i_mount, dip);
329 return XFS_ERROR(EFSCORRUPTED);
332 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
333 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
334 "corrupt dinode %Lu, forkoff = 0x%x.",
335 (unsigned long long)ip->i_ino,
336 dip->di_forkoff);
337 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
338 ip->i_mount, dip);
339 return XFS_ERROR(EFSCORRUPTED);
342 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
343 !ip->i_mount->m_rtdev_targp)) {
344 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
345 "corrupt dinode %Lu, has realtime flag set.",
346 ip->i_ino);
347 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
348 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
349 return XFS_ERROR(EFSCORRUPTED);
352 switch (ip->i_d.di_mode & S_IFMT) {
353 case S_IFIFO:
354 case S_IFCHR:
355 case S_IFBLK:
356 case S_IFSOCK:
357 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
358 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
359 ip->i_mount, dip);
360 return XFS_ERROR(EFSCORRUPTED);
362 ip->i_d.di_size = 0;
363 ip->i_size = 0;
364 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
365 break;
367 case S_IFREG:
368 case S_IFLNK:
369 case S_IFDIR:
370 switch (dip->di_format) {
371 case XFS_DINODE_FMT_LOCAL:
373 * no local regular files yet
375 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
376 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
377 "corrupt inode %Lu "
378 "(local format for regular file).",
379 (unsigned long long) ip->i_ino);
380 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
381 XFS_ERRLEVEL_LOW,
382 ip->i_mount, dip);
383 return XFS_ERROR(EFSCORRUPTED);
386 di_size = be64_to_cpu(dip->di_size);
387 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
388 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
389 "corrupt inode %Lu "
390 "(bad size %Ld for local inode).",
391 (unsigned long long) ip->i_ino,
392 (long long) di_size);
393 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
394 XFS_ERRLEVEL_LOW,
395 ip->i_mount, dip);
396 return XFS_ERROR(EFSCORRUPTED);
399 size = (int)di_size;
400 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
401 break;
402 case XFS_DINODE_FMT_EXTENTS:
403 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
404 break;
405 case XFS_DINODE_FMT_BTREE:
406 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
407 break;
408 default:
409 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
410 ip->i_mount);
411 return XFS_ERROR(EFSCORRUPTED);
413 break;
415 default:
416 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
417 return XFS_ERROR(EFSCORRUPTED);
419 if (error) {
420 return error;
422 if (!XFS_DFORK_Q(dip))
423 return 0;
424 ASSERT(ip->i_afp == NULL);
425 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
426 ip->i_afp->if_ext_max =
427 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
428 switch (dip->di_aformat) {
429 case XFS_DINODE_FMT_LOCAL:
430 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
431 size = be16_to_cpu(atp->hdr.totsize);
433 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
434 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
435 "corrupt inode %Lu "
436 "(bad attr fork size %Ld).",
437 (unsigned long long) ip->i_ino,
438 (long long) size);
439 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
440 XFS_ERRLEVEL_LOW,
441 ip->i_mount, dip);
442 return XFS_ERROR(EFSCORRUPTED);
445 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
446 break;
447 case XFS_DINODE_FMT_EXTENTS:
448 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
449 break;
450 case XFS_DINODE_FMT_BTREE:
451 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
452 break;
453 default:
454 error = XFS_ERROR(EFSCORRUPTED);
455 break;
457 if (error) {
458 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
459 ip->i_afp = NULL;
460 xfs_idestroy_fork(ip, XFS_DATA_FORK);
462 return error;
466 * The file is in-lined in the on-disk inode.
467 * If it fits into if_inline_data, then copy
468 * it there, otherwise allocate a buffer for it
469 * and copy the data there. Either way, set
470 * if_data to point at the data.
471 * If we allocate a buffer for the data, make
472 * sure that its size is a multiple of 4 and
473 * record the real size in i_real_bytes.
475 STATIC int
476 xfs_iformat_local(
477 xfs_inode_t *ip,
478 xfs_dinode_t *dip,
479 int whichfork,
480 int size)
482 xfs_ifork_t *ifp;
483 int real_size;
486 * If the size is unreasonable, then something
487 * is wrong and we just bail out rather than crash in
488 * kmem_alloc() or memcpy() below.
490 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
491 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
492 "corrupt inode %Lu "
493 "(bad size %d for local fork, size = %d).",
494 (unsigned long long) ip->i_ino, size,
495 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
496 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
497 ip->i_mount, dip);
498 return XFS_ERROR(EFSCORRUPTED);
500 ifp = XFS_IFORK_PTR(ip, whichfork);
501 real_size = 0;
502 if (size == 0)
503 ifp->if_u1.if_data = NULL;
504 else if (size <= sizeof(ifp->if_u2.if_inline_data))
505 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
506 else {
507 real_size = roundup(size, 4);
508 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
510 ifp->if_bytes = size;
511 ifp->if_real_bytes = real_size;
512 if (size)
513 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
514 ifp->if_flags &= ~XFS_IFEXTENTS;
515 ifp->if_flags |= XFS_IFINLINE;
516 return 0;
520 * The file consists of a set of extents all
521 * of which fit into the on-disk inode.
522 * If there are few enough extents to fit into
523 * the if_inline_ext, then copy them there.
524 * Otherwise allocate a buffer for them and copy
525 * them into it. Either way, set if_extents
526 * to point at the extents.
528 STATIC int
529 xfs_iformat_extents(
530 xfs_inode_t *ip,
531 xfs_dinode_t *dip,
532 int whichfork)
534 xfs_bmbt_rec_t *dp;
535 xfs_ifork_t *ifp;
536 int nex;
537 int size;
538 int i;
540 ifp = XFS_IFORK_PTR(ip, whichfork);
541 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
542 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
545 * If the number of extents is unreasonable, then something
546 * is wrong and we just bail out rather than crash in
547 * kmem_alloc() or memcpy() below.
549 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
550 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
551 "corrupt inode %Lu ((a)extents = %d).",
552 (unsigned long long) ip->i_ino, nex);
553 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
554 ip->i_mount, dip);
555 return XFS_ERROR(EFSCORRUPTED);
558 ifp->if_real_bytes = 0;
559 if (nex == 0)
560 ifp->if_u1.if_extents = NULL;
561 else if (nex <= XFS_INLINE_EXTS)
562 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
563 else
564 xfs_iext_add(ifp, 0, nex);
566 ifp->if_bytes = size;
567 if (size) {
568 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
569 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
570 for (i = 0; i < nex; i++, dp++) {
571 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
572 ep->l0 = get_unaligned_be64(&dp->l0);
573 ep->l1 = get_unaligned_be64(&dp->l1);
575 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
576 if (whichfork != XFS_DATA_FORK ||
577 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
578 if (unlikely(xfs_check_nostate_extents(
579 ifp, 0, nex))) {
580 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
581 XFS_ERRLEVEL_LOW,
582 ip->i_mount);
583 return XFS_ERROR(EFSCORRUPTED);
586 ifp->if_flags |= XFS_IFEXTENTS;
587 return 0;
591 * The file has too many extents to fit into
592 * the inode, so they are in B-tree format.
593 * Allocate a buffer for the root of the B-tree
594 * and copy the root into it. The i_extents
595 * field will remain NULL until all of the
596 * extents are read in (when they are needed).
598 STATIC int
599 xfs_iformat_btree(
600 xfs_inode_t *ip,
601 xfs_dinode_t *dip,
602 int whichfork)
604 xfs_bmdr_block_t *dfp;
605 xfs_ifork_t *ifp;
606 /* REFERENCED */
607 int nrecs;
608 int size;
610 ifp = XFS_IFORK_PTR(ip, whichfork);
611 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
612 size = XFS_BMAP_BROOT_SPACE(dfp);
613 nrecs = be16_to_cpu(dfp->bb_numrecs);
616 * blow out if -- fork has less extents than can fit in
617 * fork (fork shouldn't be a btree format), root btree
618 * block has more records than can fit into the fork,
619 * or the number of extents is greater than the number of
620 * blocks.
622 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
623 || XFS_BMDR_SPACE_CALC(nrecs) >
624 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
625 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
626 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
627 "corrupt inode %Lu (btree).",
628 (unsigned long long) ip->i_ino);
629 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
630 ip->i_mount);
631 return XFS_ERROR(EFSCORRUPTED);
634 ifp->if_broot_bytes = size;
635 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
636 ASSERT(ifp->if_broot != NULL);
638 * Copy and convert from the on-disk structure
639 * to the in-memory structure.
641 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
642 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
643 ifp->if_broot, size);
644 ifp->if_flags &= ~XFS_IFEXTENTS;
645 ifp->if_flags |= XFS_IFBROOT;
647 return 0;
650 STATIC void
651 xfs_dinode_from_disk(
652 xfs_icdinode_t *to,
653 xfs_dinode_t *from)
655 to->di_magic = be16_to_cpu(from->di_magic);
656 to->di_mode = be16_to_cpu(from->di_mode);
657 to->di_version = from ->di_version;
658 to->di_format = from->di_format;
659 to->di_onlink = be16_to_cpu(from->di_onlink);
660 to->di_uid = be32_to_cpu(from->di_uid);
661 to->di_gid = be32_to_cpu(from->di_gid);
662 to->di_nlink = be32_to_cpu(from->di_nlink);
663 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
664 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
665 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
666 to->di_flushiter = be16_to_cpu(from->di_flushiter);
667 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
668 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
669 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
670 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
671 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
672 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
673 to->di_size = be64_to_cpu(from->di_size);
674 to->di_nblocks = be64_to_cpu(from->di_nblocks);
675 to->di_extsize = be32_to_cpu(from->di_extsize);
676 to->di_nextents = be32_to_cpu(from->di_nextents);
677 to->di_anextents = be16_to_cpu(from->di_anextents);
678 to->di_forkoff = from->di_forkoff;
679 to->di_aformat = from->di_aformat;
680 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
681 to->di_dmstate = be16_to_cpu(from->di_dmstate);
682 to->di_flags = be16_to_cpu(from->di_flags);
683 to->di_gen = be32_to_cpu(from->di_gen);
686 void
687 xfs_dinode_to_disk(
688 xfs_dinode_t *to,
689 xfs_icdinode_t *from)
691 to->di_magic = cpu_to_be16(from->di_magic);
692 to->di_mode = cpu_to_be16(from->di_mode);
693 to->di_version = from ->di_version;
694 to->di_format = from->di_format;
695 to->di_onlink = cpu_to_be16(from->di_onlink);
696 to->di_uid = cpu_to_be32(from->di_uid);
697 to->di_gid = cpu_to_be32(from->di_gid);
698 to->di_nlink = cpu_to_be32(from->di_nlink);
699 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
700 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
701 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
702 to->di_flushiter = cpu_to_be16(from->di_flushiter);
703 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
704 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
705 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
706 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
707 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
708 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
709 to->di_size = cpu_to_be64(from->di_size);
710 to->di_nblocks = cpu_to_be64(from->di_nblocks);
711 to->di_extsize = cpu_to_be32(from->di_extsize);
712 to->di_nextents = cpu_to_be32(from->di_nextents);
713 to->di_anextents = cpu_to_be16(from->di_anextents);
714 to->di_forkoff = from->di_forkoff;
715 to->di_aformat = from->di_aformat;
716 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
717 to->di_dmstate = cpu_to_be16(from->di_dmstate);
718 to->di_flags = cpu_to_be16(from->di_flags);
719 to->di_gen = cpu_to_be32(from->di_gen);
722 STATIC uint
723 _xfs_dic2xflags(
724 __uint16_t di_flags)
726 uint flags = 0;
728 if (di_flags & XFS_DIFLAG_ANY) {
729 if (di_flags & XFS_DIFLAG_REALTIME)
730 flags |= XFS_XFLAG_REALTIME;
731 if (di_flags & XFS_DIFLAG_PREALLOC)
732 flags |= XFS_XFLAG_PREALLOC;
733 if (di_flags & XFS_DIFLAG_IMMUTABLE)
734 flags |= XFS_XFLAG_IMMUTABLE;
735 if (di_flags & XFS_DIFLAG_APPEND)
736 flags |= XFS_XFLAG_APPEND;
737 if (di_flags & XFS_DIFLAG_SYNC)
738 flags |= XFS_XFLAG_SYNC;
739 if (di_flags & XFS_DIFLAG_NOATIME)
740 flags |= XFS_XFLAG_NOATIME;
741 if (di_flags & XFS_DIFLAG_NODUMP)
742 flags |= XFS_XFLAG_NODUMP;
743 if (di_flags & XFS_DIFLAG_RTINHERIT)
744 flags |= XFS_XFLAG_RTINHERIT;
745 if (di_flags & XFS_DIFLAG_PROJINHERIT)
746 flags |= XFS_XFLAG_PROJINHERIT;
747 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
748 flags |= XFS_XFLAG_NOSYMLINKS;
749 if (di_flags & XFS_DIFLAG_EXTSIZE)
750 flags |= XFS_XFLAG_EXTSIZE;
751 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
752 flags |= XFS_XFLAG_EXTSZINHERIT;
753 if (di_flags & XFS_DIFLAG_NODEFRAG)
754 flags |= XFS_XFLAG_NODEFRAG;
755 if (di_flags & XFS_DIFLAG_FILESTREAM)
756 flags |= XFS_XFLAG_FILESTREAM;
759 return flags;
762 uint
763 xfs_ip2xflags(
764 xfs_inode_t *ip)
766 xfs_icdinode_t *dic = &ip->i_d;
768 return _xfs_dic2xflags(dic->di_flags) |
769 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
772 uint
773 xfs_dic2xflags(
774 xfs_dinode_t *dip)
776 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
777 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
781 * Read the disk inode attributes into the in-core inode structure.
784 xfs_iread(
785 xfs_mount_t *mp,
786 xfs_trans_t *tp,
787 xfs_inode_t *ip,
788 uint iget_flags)
790 xfs_buf_t *bp;
791 xfs_dinode_t *dip;
792 int error;
795 * Fill in the location information in the in-core inode.
797 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
798 if (error)
799 return error;
802 * Get pointers to the on-disk inode and the buffer containing it.
804 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
805 XBF_LOCK, iget_flags);
806 if (error)
807 return error;
808 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
811 * If we got something that isn't an inode it means someone
812 * (nfs or dmi) has a stale handle.
814 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
815 #ifdef DEBUG
816 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
817 "dip->di_magic (0x%x) != "
818 "XFS_DINODE_MAGIC (0x%x)",
819 be16_to_cpu(dip->di_magic),
820 XFS_DINODE_MAGIC);
821 #endif /* DEBUG */
822 error = XFS_ERROR(EINVAL);
823 goto out_brelse;
827 * If the on-disk inode is already linked to a directory
828 * entry, copy all of the inode into the in-core inode.
829 * xfs_iformat() handles copying in the inode format
830 * specific information.
831 * Otherwise, just get the truly permanent information.
833 if (dip->di_mode) {
834 xfs_dinode_from_disk(&ip->i_d, dip);
835 error = xfs_iformat(ip, dip);
836 if (error) {
837 #ifdef DEBUG
838 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
839 "xfs_iformat() returned error %d",
840 error);
841 #endif /* DEBUG */
842 goto out_brelse;
844 } else {
845 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
846 ip->i_d.di_version = dip->di_version;
847 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
848 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
850 * Make sure to pull in the mode here as well in
851 * case the inode is released without being used.
852 * This ensures that xfs_inactive() will see that
853 * the inode is already free and not try to mess
854 * with the uninitialized part of it.
856 ip->i_d.di_mode = 0;
858 * Initialize the per-fork minima and maxima for a new
859 * inode here. xfs_iformat will do it for old inodes.
861 ip->i_df.if_ext_max =
862 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
866 * The inode format changed when we moved the link count and
867 * made it 32 bits long. If this is an old format inode,
868 * convert it in memory to look like a new one. If it gets
869 * flushed to disk we will convert back before flushing or
870 * logging it. We zero out the new projid field and the old link
871 * count field. We'll handle clearing the pad field (the remains
872 * of the old uuid field) when we actually convert the inode to
873 * the new format. We don't change the version number so that we
874 * can distinguish this from a real new format inode.
876 if (ip->i_d.di_version == 1) {
877 ip->i_d.di_nlink = ip->i_d.di_onlink;
878 ip->i_d.di_onlink = 0;
879 xfs_set_projid(ip, 0);
882 ip->i_delayed_blks = 0;
883 ip->i_size = ip->i_d.di_size;
886 * Mark the buffer containing the inode as something to keep
887 * around for a while. This helps to keep recently accessed
888 * meta-data in-core longer.
890 xfs_buf_set_ref(bp, XFS_INO_REF);
893 * Use xfs_trans_brelse() to release the buffer containing the
894 * on-disk inode, because it was acquired with xfs_trans_read_buf()
895 * in xfs_itobp() above. If tp is NULL, this is just a normal
896 * brelse(). If we're within a transaction, then xfs_trans_brelse()
897 * will only release the buffer if it is not dirty within the
898 * transaction. It will be OK to release the buffer in this case,
899 * because inodes on disk are never destroyed and we will be
900 * locking the new in-core inode before putting it in the hash
901 * table where other processes can find it. Thus we don't have
902 * to worry about the inode being changed just because we released
903 * the buffer.
905 out_brelse:
906 xfs_trans_brelse(tp, bp);
907 return error;
911 * Read in extents from a btree-format inode.
912 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
915 xfs_iread_extents(
916 xfs_trans_t *tp,
917 xfs_inode_t *ip,
918 int whichfork)
920 int error;
921 xfs_ifork_t *ifp;
922 xfs_extnum_t nextents;
924 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
925 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
926 ip->i_mount);
927 return XFS_ERROR(EFSCORRUPTED);
929 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
930 ifp = XFS_IFORK_PTR(ip, whichfork);
933 * We know that the size is valid (it's checked in iformat_btree)
935 ifp->if_lastex = NULLEXTNUM;
936 ifp->if_bytes = ifp->if_real_bytes = 0;
937 ifp->if_flags |= XFS_IFEXTENTS;
938 xfs_iext_add(ifp, 0, nextents);
939 error = xfs_bmap_read_extents(tp, ip, whichfork);
940 if (error) {
941 xfs_iext_destroy(ifp);
942 ifp->if_flags &= ~XFS_IFEXTENTS;
943 return error;
945 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
946 return 0;
950 * Allocate an inode on disk and return a copy of its in-core version.
951 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
952 * appropriately within the inode. The uid and gid for the inode are
953 * set according to the contents of the given cred structure.
955 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
956 * has a free inode available, call xfs_iget()
957 * to obtain the in-core version of the allocated inode. Finally,
958 * fill in the inode and log its initial contents. In this case,
959 * ialloc_context would be set to NULL and call_again set to false.
961 * If xfs_dialloc() does not have an available inode,
962 * it will replenish its supply by doing an allocation. Since we can
963 * only do one allocation within a transaction without deadlocks, we
964 * must commit the current transaction before returning the inode itself.
965 * In this case, therefore, we will set call_again to true and return.
966 * The caller should then commit the current transaction, start a new
967 * transaction, and call xfs_ialloc() again to actually get the inode.
969 * To ensure that some other process does not grab the inode that
970 * was allocated during the first call to xfs_ialloc(), this routine
971 * also returns the [locked] bp pointing to the head of the freelist
972 * as ialloc_context. The caller should hold this buffer across
973 * the commit and pass it back into this routine on the second call.
975 * If we are allocating quota inodes, we do not have a parent inode
976 * to attach to or associate with (i.e. pip == NULL) because they
977 * are not linked into the directory structure - they are attached
978 * directly to the superblock - and so have no parent.
981 xfs_ialloc(
982 xfs_trans_t *tp,
983 xfs_inode_t *pip,
984 mode_t mode,
985 xfs_nlink_t nlink,
986 xfs_dev_t rdev,
987 prid_t prid,
988 int okalloc,
989 xfs_buf_t **ialloc_context,
990 boolean_t *call_again,
991 xfs_inode_t **ipp)
993 xfs_ino_t ino;
994 xfs_inode_t *ip;
995 uint flags;
996 int error;
997 timespec_t tv;
998 int filestreams = 0;
1001 * Call the space management code to pick
1002 * the on-disk inode to be allocated.
1004 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1005 ialloc_context, call_again, &ino);
1006 if (error)
1007 return error;
1008 if (*call_again || ino == NULLFSINO) {
1009 *ipp = NULL;
1010 return 0;
1012 ASSERT(*ialloc_context == NULL);
1015 * Get the in-core inode with the lock held exclusively.
1016 * This is because we're setting fields here we need
1017 * to prevent others from looking at until we're done.
1019 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1020 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1021 if (error)
1022 return error;
1023 ASSERT(ip != NULL);
1025 ip->i_d.di_mode = (__uint16_t)mode;
1026 ip->i_d.di_onlink = 0;
1027 ip->i_d.di_nlink = nlink;
1028 ASSERT(ip->i_d.di_nlink == nlink);
1029 ip->i_d.di_uid = current_fsuid();
1030 ip->i_d.di_gid = current_fsgid();
1031 xfs_set_projid(ip, prid);
1032 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1035 * If the superblock version is up to where we support new format
1036 * inodes and this is currently an old format inode, then change
1037 * the inode version number now. This way we only do the conversion
1038 * here rather than here and in the flush/logging code.
1040 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1041 ip->i_d.di_version == 1) {
1042 ip->i_d.di_version = 2;
1044 * We've already zeroed the old link count, the projid field,
1045 * and the pad field.
1050 * Project ids won't be stored on disk if we are using a version 1 inode.
1052 if ((prid != 0) && (ip->i_d.di_version == 1))
1053 xfs_bump_ino_vers2(tp, ip);
1055 if (pip && XFS_INHERIT_GID(pip)) {
1056 ip->i_d.di_gid = pip->i_d.di_gid;
1057 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1058 ip->i_d.di_mode |= S_ISGID;
1063 * If the group ID of the new file does not match the effective group
1064 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1065 * (and only if the irix_sgid_inherit compatibility variable is set).
1067 if ((irix_sgid_inherit) &&
1068 (ip->i_d.di_mode & S_ISGID) &&
1069 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1070 ip->i_d.di_mode &= ~S_ISGID;
1073 ip->i_d.di_size = 0;
1074 ip->i_size = 0;
1075 ip->i_d.di_nextents = 0;
1076 ASSERT(ip->i_d.di_nblocks == 0);
1078 nanotime(&tv);
1079 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1080 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1081 ip->i_d.di_atime = ip->i_d.di_mtime;
1082 ip->i_d.di_ctime = ip->i_d.di_mtime;
1085 * di_gen will have been taken care of in xfs_iread.
1087 ip->i_d.di_extsize = 0;
1088 ip->i_d.di_dmevmask = 0;
1089 ip->i_d.di_dmstate = 0;
1090 ip->i_d.di_flags = 0;
1091 flags = XFS_ILOG_CORE;
1092 switch (mode & S_IFMT) {
1093 case S_IFIFO:
1094 case S_IFCHR:
1095 case S_IFBLK:
1096 case S_IFSOCK:
1097 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1098 ip->i_df.if_u2.if_rdev = rdev;
1099 ip->i_df.if_flags = 0;
1100 flags |= XFS_ILOG_DEV;
1101 break;
1102 case S_IFREG:
1104 * we can't set up filestreams until after the VFS inode
1105 * is set up properly.
1107 if (pip && xfs_inode_is_filestream(pip))
1108 filestreams = 1;
1109 /* fall through */
1110 case S_IFDIR:
1111 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1112 uint di_flags = 0;
1114 if ((mode & S_IFMT) == S_IFDIR) {
1115 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1116 di_flags |= XFS_DIFLAG_RTINHERIT;
1117 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1118 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1119 ip->i_d.di_extsize = pip->i_d.di_extsize;
1121 } else if ((mode & S_IFMT) == S_IFREG) {
1122 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1123 di_flags |= XFS_DIFLAG_REALTIME;
1124 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1125 di_flags |= XFS_DIFLAG_EXTSIZE;
1126 ip->i_d.di_extsize = pip->i_d.di_extsize;
1129 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1130 xfs_inherit_noatime)
1131 di_flags |= XFS_DIFLAG_NOATIME;
1132 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1133 xfs_inherit_nodump)
1134 di_flags |= XFS_DIFLAG_NODUMP;
1135 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1136 xfs_inherit_sync)
1137 di_flags |= XFS_DIFLAG_SYNC;
1138 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1139 xfs_inherit_nosymlinks)
1140 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1141 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1142 di_flags |= XFS_DIFLAG_PROJINHERIT;
1143 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1144 xfs_inherit_nodefrag)
1145 di_flags |= XFS_DIFLAG_NODEFRAG;
1146 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1147 di_flags |= XFS_DIFLAG_FILESTREAM;
1148 ip->i_d.di_flags |= di_flags;
1150 /* FALLTHROUGH */
1151 case S_IFLNK:
1152 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1153 ip->i_df.if_flags = XFS_IFEXTENTS;
1154 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1155 ip->i_df.if_u1.if_extents = NULL;
1156 break;
1157 default:
1158 ASSERT(0);
1161 * Attribute fork settings for new inode.
1163 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1164 ip->i_d.di_anextents = 0;
1167 * Log the new values stuffed into the inode.
1169 xfs_trans_log_inode(tp, ip, flags);
1171 /* now that we have an i_mode we can setup inode ops and unlock */
1172 xfs_setup_inode(ip);
1174 /* now we have set up the vfs inode we can associate the filestream */
1175 if (filestreams) {
1176 error = xfs_filestream_associate(pip, ip);
1177 if (error < 0)
1178 return -error;
1179 if (!error)
1180 xfs_iflags_set(ip, XFS_IFILESTREAM);
1183 *ipp = ip;
1184 return 0;
1188 * Check to make sure that there are no blocks allocated to the
1189 * file beyond the size of the file. We don't check this for
1190 * files with fixed size extents or real time extents, but we
1191 * at least do it for regular files.
1193 #ifdef DEBUG
1194 void
1195 xfs_isize_check(
1196 xfs_mount_t *mp,
1197 xfs_inode_t *ip,
1198 xfs_fsize_t isize)
1200 xfs_fileoff_t map_first;
1201 int nimaps;
1202 xfs_bmbt_irec_t imaps[2];
1204 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1205 return;
1207 if (XFS_IS_REALTIME_INODE(ip))
1208 return;
1210 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1211 return;
1213 nimaps = 2;
1214 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1216 * The filesystem could be shutting down, so bmapi may return
1217 * an error.
1219 if (xfs_bmapi(NULL, ip, map_first,
1220 (XFS_B_TO_FSB(mp,
1221 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1222 map_first),
1223 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1224 NULL))
1225 return;
1226 ASSERT(nimaps == 1);
1227 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1229 #endif /* DEBUG */
1232 * Calculate the last possible buffered byte in a file. This must
1233 * include data that was buffered beyond the EOF by the write code.
1234 * This also needs to deal with overflowing the xfs_fsize_t type
1235 * which can happen for sizes near the limit.
1237 * We also need to take into account any blocks beyond the EOF. It
1238 * may be the case that they were buffered by a write which failed.
1239 * In that case the pages will still be in memory, but the inode size
1240 * will never have been updated.
1242 STATIC xfs_fsize_t
1243 xfs_file_last_byte(
1244 xfs_inode_t *ip)
1246 xfs_mount_t *mp;
1247 xfs_fsize_t last_byte;
1248 xfs_fileoff_t last_block;
1249 xfs_fileoff_t size_last_block;
1250 int error;
1252 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1254 mp = ip->i_mount;
1256 * Only check for blocks beyond the EOF if the extents have
1257 * been read in. This eliminates the need for the inode lock,
1258 * and it also saves us from looking when it really isn't
1259 * necessary.
1261 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1262 xfs_ilock(ip, XFS_ILOCK_SHARED);
1263 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1264 XFS_DATA_FORK);
1265 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1266 if (error) {
1267 last_block = 0;
1269 } else {
1270 last_block = 0;
1272 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1273 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1275 last_byte = XFS_FSB_TO_B(mp, last_block);
1276 if (last_byte < 0) {
1277 return XFS_MAXIOFFSET(mp);
1279 last_byte += (1 << mp->m_writeio_log);
1280 if (last_byte < 0) {
1281 return XFS_MAXIOFFSET(mp);
1283 return last_byte;
1287 * Start the truncation of the file to new_size. The new size
1288 * must be smaller than the current size. This routine will
1289 * clear the buffer and page caches of file data in the removed
1290 * range, and xfs_itruncate_finish() will remove the underlying
1291 * disk blocks.
1293 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1294 * must NOT have the inode lock held at all. This is because we're
1295 * calling into the buffer/page cache code and we can't hold the
1296 * inode lock when we do so.
1298 * We need to wait for any direct I/Os in flight to complete before we
1299 * proceed with the truncate. This is needed to prevent the extents
1300 * being read or written by the direct I/Os from being removed while the
1301 * I/O is in flight as there is no other method of synchronising
1302 * direct I/O with the truncate operation. Also, because we hold
1303 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1304 * started until the truncate completes and drops the lock. Essentially,
1305 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1306 * ordering between direct I/Os and the truncate operation.
1308 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1309 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1310 * in the case that the caller is locking things out of order and
1311 * may not be able to call xfs_itruncate_finish() with the inode lock
1312 * held without dropping the I/O lock. If the caller must drop the
1313 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1314 * must be called again with all the same restrictions as the initial
1315 * call.
1318 xfs_itruncate_start(
1319 xfs_inode_t *ip,
1320 uint flags,
1321 xfs_fsize_t new_size)
1323 xfs_fsize_t last_byte;
1324 xfs_off_t toss_start;
1325 xfs_mount_t *mp;
1326 int error = 0;
1328 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1329 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1330 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1331 (flags == XFS_ITRUNC_MAYBE));
1333 mp = ip->i_mount;
1335 /* wait for the completion of any pending DIOs */
1336 if (new_size == 0 || new_size < ip->i_size)
1337 xfs_ioend_wait(ip);
1340 * Call toss_pages or flushinval_pages to get rid of pages
1341 * overlapping the region being removed. We have to use
1342 * the less efficient flushinval_pages in the case that the
1343 * caller may not be able to finish the truncate without
1344 * dropping the inode's I/O lock. Make sure
1345 * to catch any pages brought in by buffers overlapping
1346 * the EOF by searching out beyond the isize by our
1347 * block size. We round new_size up to a block boundary
1348 * so that we don't toss things on the same block as
1349 * new_size but before it.
1351 * Before calling toss_page or flushinval_pages, make sure to
1352 * call remapf() over the same region if the file is mapped.
1353 * This frees up mapped file references to the pages in the
1354 * given range and for the flushinval_pages case it ensures
1355 * that we get the latest mapped changes flushed out.
1357 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1358 toss_start = XFS_FSB_TO_B(mp, toss_start);
1359 if (toss_start < 0) {
1361 * The place to start tossing is beyond our maximum
1362 * file size, so there is no way that the data extended
1363 * out there.
1365 return 0;
1367 last_byte = xfs_file_last_byte(ip);
1368 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
1369 if (last_byte > toss_start) {
1370 if (flags & XFS_ITRUNC_DEFINITE) {
1371 xfs_tosspages(ip, toss_start,
1372 -1, FI_REMAPF_LOCKED);
1373 } else {
1374 error = xfs_flushinval_pages(ip, toss_start,
1375 -1, FI_REMAPF_LOCKED);
1379 #ifdef DEBUG
1380 if (new_size == 0) {
1381 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1383 #endif
1384 return error;
1388 * Shrink the file to the given new_size. The new size must be smaller than
1389 * the current size. This will free up the underlying blocks in the removed
1390 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1392 * The transaction passed to this routine must have made a permanent log
1393 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1394 * given transaction and start new ones, so make sure everything involved in
1395 * the transaction is tidy before calling here. Some transaction will be
1396 * returned to the caller to be committed. The incoming transaction must
1397 * already include the inode, and both inode locks must be held exclusively.
1398 * The inode must also be "held" within the transaction. On return the inode
1399 * will be "held" within the returned transaction. This routine does NOT
1400 * require any disk space to be reserved for it within the transaction.
1402 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1403 * indicates the fork which is to be truncated. For the attribute fork we only
1404 * support truncation to size 0.
1406 * We use the sync parameter to indicate whether or not the first transaction
1407 * we perform might have to be synchronous. For the attr fork, it needs to be
1408 * so if the unlink of the inode is not yet known to be permanent in the log.
1409 * This keeps us from freeing and reusing the blocks of the attribute fork
1410 * before the unlink of the inode becomes permanent.
1412 * For the data fork, we normally have to run synchronously if we're being
1413 * called out of the inactive path or we're being called out of the create path
1414 * where we're truncating an existing file. Either way, the truncate needs to
1415 * be sync so blocks don't reappear in the file with altered data in case of a
1416 * crash. wsync filesystems can run the first case async because anything that
1417 * shrinks the inode has to run sync so by the time we're called here from
1418 * inactive, the inode size is permanently set to 0.
1420 * Calls from the truncate path always need to be sync unless we're in a wsync
1421 * filesystem and the file has already been unlinked.
1423 * The caller is responsible for correctly setting the sync parameter. It gets
1424 * too hard for us to guess here which path we're being called out of just
1425 * based on inode state.
1427 * If we get an error, we must return with the inode locked and linked into the
1428 * current transaction. This keeps things simple for the higher level code,
1429 * because it always knows that the inode is locked and held in the transaction
1430 * that returns to it whether errors occur or not. We don't mark the inode
1431 * dirty on error so that transactions can be easily aborted if possible.
1434 xfs_itruncate_finish(
1435 xfs_trans_t **tp,
1436 xfs_inode_t *ip,
1437 xfs_fsize_t new_size,
1438 int fork,
1439 int sync)
1441 xfs_fsblock_t first_block;
1442 xfs_fileoff_t first_unmap_block;
1443 xfs_fileoff_t last_block;
1444 xfs_filblks_t unmap_len=0;
1445 xfs_mount_t *mp;
1446 xfs_trans_t *ntp;
1447 int done;
1448 int committed;
1449 xfs_bmap_free_t free_list;
1450 int error;
1452 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1453 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1454 ASSERT(*tp != NULL);
1455 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1456 ASSERT(ip->i_transp == *tp);
1457 ASSERT(ip->i_itemp != NULL);
1458 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1461 ntp = *tp;
1462 mp = (ntp)->t_mountp;
1463 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1466 * We only support truncating the entire attribute fork.
1468 if (fork == XFS_ATTR_FORK) {
1469 new_size = 0LL;
1471 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1472 trace_xfs_itruncate_finish_start(ip, new_size);
1475 * The first thing we do is set the size to new_size permanently
1476 * on disk. This way we don't have to worry about anyone ever
1477 * being able to look at the data being freed even in the face
1478 * of a crash. What we're getting around here is the case where
1479 * we free a block, it is allocated to another file, it is written
1480 * to, and then we crash. If the new data gets written to the
1481 * file but the log buffers containing the free and reallocation
1482 * don't, then we'd end up with garbage in the blocks being freed.
1483 * As long as we make the new_size permanent before actually
1484 * freeing any blocks it doesn't matter if they get writtten to.
1486 * The callers must signal into us whether or not the size
1487 * setting here must be synchronous. There are a few cases
1488 * where it doesn't have to be synchronous. Those cases
1489 * occur if the file is unlinked and we know the unlink is
1490 * permanent or if the blocks being truncated are guaranteed
1491 * to be beyond the inode eof (regardless of the link count)
1492 * and the eof value is permanent. Both of these cases occur
1493 * only on wsync-mounted filesystems. In those cases, we're
1494 * guaranteed that no user will ever see the data in the blocks
1495 * that are being truncated so the truncate can run async.
1496 * In the free beyond eof case, the file may wind up with
1497 * more blocks allocated to it than it needs if we crash
1498 * and that won't get fixed until the next time the file
1499 * is re-opened and closed but that's ok as that shouldn't
1500 * be too many blocks.
1502 * However, we can't just make all wsync xactions run async
1503 * because there's one call out of the create path that needs
1504 * to run sync where it's truncating an existing file to size
1505 * 0 whose size is > 0.
1507 * It's probably possible to come up with a test in this
1508 * routine that would correctly distinguish all the above
1509 * cases from the values of the function parameters and the
1510 * inode state but for sanity's sake, I've decided to let the
1511 * layers above just tell us. It's simpler to correctly figure
1512 * out in the layer above exactly under what conditions we
1513 * can run async and I think it's easier for others read and
1514 * follow the logic in case something has to be changed.
1515 * cscope is your friend -- rcc.
1517 * The attribute fork is much simpler.
1519 * For the attribute fork we allow the caller to tell us whether
1520 * the unlink of the inode that led to this call is yet permanent
1521 * in the on disk log. If it is not and we will be freeing extents
1522 * in this inode then we make the first transaction synchronous
1523 * to make sure that the unlink is permanent by the time we free
1524 * the blocks.
1526 if (fork == XFS_DATA_FORK) {
1527 if (ip->i_d.di_nextents > 0) {
1529 * If we are not changing the file size then do
1530 * not update the on-disk file size - we may be
1531 * called from xfs_inactive_free_eofblocks(). If we
1532 * update the on-disk file size and then the system
1533 * crashes before the contents of the file are
1534 * flushed to disk then the files may be full of
1535 * holes (ie NULL files bug).
1537 if (ip->i_size != new_size) {
1538 ip->i_d.di_size = new_size;
1539 ip->i_size = new_size;
1540 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1543 } else if (sync) {
1544 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1545 if (ip->i_d.di_anextents > 0)
1546 xfs_trans_set_sync(ntp);
1548 ASSERT(fork == XFS_DATA_FORK ||
1549 (fork == XFS_ATTR_FORK &&
1550 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1551 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1554 * Since it is possible for space to become allocated beyond
1555 * the end of the file (in a crash where the space is allocated
1556 * but the inode size is not yet updated), simply remove any
1557 * blocks which show up between the new EOF and the maximum
1558 * possible file size. If the first block to be removed is
1559 * beyond the maximum file size (ie it is the same as last_block),
1560 * then there is nothing to do.
1562 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1563 ASSERT(first_unmap_block <= last_block);
1564 done = 0;
1565 if (last_block == first_unmap_block) {
1566 done = 1;
1567 } else {
1568 unmap_len = last_block - first_unmap_block + 1;
1570 while (!done) {
1572 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1573 * will tell us whether it freed the entire range or
1574 * not. If this is a synchronous mount (wsync),
1575 * then we can tell bunmapi to keep all the
1576 * transactions asynchronous since the unlink
1577 * transaction that made this inode inactive has
1578 * already hit the disk. There's no danger of
1579 * the freed blocks being reused, there being a
1580 * crash, and the reused blocks suddenly reappearing
1581 * in this file with garbage in them once recovery
1582 * runs.
1584 xfs_bmap_init(&free_list, &first_block);
1585 error = xfs_bunmapi(ntp, ip,
1586 first_unmap_block, unmap_len,
1587 xfs_bmapi_aflag(fork),
1588 XFS_ITRUNC_MAX_EXTENTS,
1589 &first_block, &free_list,
1590 &done);
1591 if (error) {
1593 * If the bunmapi call encounters an error,
1594 * return to the caller where the transaction
1595 * can be properly aborted. We just need to
1596 * make sure we're not holding any resources
1597 * that we were not when we came in.
1599 xfs_bmap_cancel(&free_list);
1600 return error;
1604 * Duplicate the transaction that has the permanent
1605 * reservation and commit the old transaction.
1607 error = xfs_bmap_finish(tp, &free_list, &committed);
1608 ntp = *tp;
1609 if (committed)
1610 xfs_trans_ijoin(ntp, ip);
1612 if (error) {
1614 * If the bmap finish call encounters an error, return
1615 * to the caller where the transaction can be properly
1616 * aborted. We just need to make sure we're not
1617 * holding any resources that we were not when we came
1618 * in.
1620 * Aborting from this point might lose some blocks in
1621 * the file system, but oh well.
1623 xfs_bmap_cancel(&free_list);
1624 return error;
1627 if (committed) {
1629 * Mark the inode dirty so it will be logged and
1630 * moved forward in the log as part of every commit.
1632 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1635 ntp = xfs_trans_dup(ntp);
1636 error = xfs_trans_commit(*tp, 0);
1637 *tp = ntp;
1639 xfs_trans_ijoin(ntp, ip);
1641 if (error)
1642 return error;
1644 * transaction commit worked ok so we can drop the extra ticket
1645 * reference that we gained in xfs_trans_dup()
1647 xfs_log_ticket_put(ntp->t_ticket);
1648 error = xfs_trans_reserve(ntp, 0,
1649 XFS_ITRUNCATE_LOG_RES(mp), 0,
1650 XFS_TRANS_PERM_LOG_RES,
1651 XFS_ITRUNCATE_LOG_COUNT);
1652 if (error)
1653 return error;
1656 * Only update the size in the case of the data fork, but
1657 * always re-log the inode so that our permanent transaction
1658 * can keep on rolling it forward in the log.
1660 if (fork == XFS_DATA_FORK) {
1661 xfs_isize_check(mp, ip, new_size);
1663 * If we are not changing the file size then do
1664 * not update the on-disk file size - we may be
1665 * called from xfs_inactive_free_eofblocks(). If we
1666 * update the on-disk file size and then the system
1667 * crashes before the contents of the file are
1668 * flushed to disk then the files may be full of
1669 * holes (ie NULL files bug).
1671 if (ip->i_size != new_size) {
1672 ip->i_d.di_size = new_size;
1673 ip->i_size = new_size;
1676 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1677 ASSERT((new_size != 0) ||
1678 (fork == XFS_ATTR_FORK) ||
1679 (ip->i_delayed_blks == 0));
1680 ASSERT((new_size != 0) ||
1681 (fork == XFS_ATTR_FORK) ||
1682 (ip->i_d.di_nextents == 0));
1683 trace_xfs_itruncate_finish_end(ip, new_size);
1684 return 0;
1688 * This is called when the inode's link count goes to 0.
1689 * We place the on-disk inode on a list in the AGI. It
1690 * will be pulled from this list when the inode is freed.
1693 xfs_iunlink(
1694 xfs_trans_t *tp,
1695 xfs_inode_t *ip)
1697 xfs_mount_t *mp;
1698 xfs_agi_t *agi;
1699 xfs_dinode_t *dip;
1700 xfs_buf_t *agibp;
1701 xfs_buf_t *ibp;
1702 xfs_agino_t agino;
1703 short bucket_index;
1704 int offset;
1705 int error;
1707 ASSERT(ip->i_d.di_nlink == 0);
1708 ASSERT(ip->i_d.di_mode != 0);
1709 ASSERT(ip->i_transp == tp);
1711 mp = tp->t_mountp;
1714 * Get the agi buffer first. It ensures lock ordering
1715 * on the list.
1717 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1718 if (error)
1719 return error;
1720 agi = XFS_BUF_TO_AGI(agibp);
1723 * Get the index into the agi hash table for the
1724 * list this inode will go on.
1726 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1727 ASSERT(agino != 0);
1728 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1729 ASSERT(agi->agi_unlinked[bucket_index]);
1730 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1732 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1734 * There is already another inode in the bucket we need
1735 * to add ourselves to. Add us at the front of the list.
1736 * Here we put the head pointer into our next pointer,
1737 * and then we fall through to point the head at us.
1739 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1740 if (error)
1741 return error;
1743 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1744 /* both on-disk, don't endian flip twice */
1745 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1746 offset = ip->i_imap.im_boffset +
1747 offsetof(xfs_dinode_t, di_next_unlinked);
1748 xfs_trans_inode_buf(tp, ibp);
1749 xfs_trans_log_buf(tp, ibp, offset,
1750 (offset + sizeof(xfs_agino_t) - 1));
1751 xfs_inobp_check(mp, ibp);
1755 * Point the bucket head pointer at the inode being inserted.
1757 ASSERT(agino != 0);
1758 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1759 offset = offsetof(xfs_agi_t, agi_unlinked) +
1760 (sizeof(xfs_agino_t) * bucket_index);
1761 xfs_trans_log_buf(tp, agibp, offset,
1762 (offset + sizeof(xfs_agino_t) - 1));
1763 return 0;
1767 * Pull the on-disk inode from the AGI unlinked list.
1769 STATIC int
1770 xfs_iunlink_remove(
1771 xfs_trans_t *tp,
1772 xfs_inode_t *ip)
1774 xfs_ino_t next_ino;
1775 xfs_mount_t *mp;
1776 xfs_agi_t *agi;
1777 xfs_dinode_t *dip;
1778 xfs_buf_t *agibp;
1779 xfs_buf_t *ibp;
1780 xfs_agnumber_t agno;
1781 xfs_agino_t agino;
1782 xfs_agino_t next_agino;
1783 xfs_buf_t *last_ibp;
1784 xfs_dinode_t *last_dip = NULL;
1785 short bucket_index;
1786 int offset, last_offset = 0;
1787 int error;
1789 mp = tp->t_mountp;
1790 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1793 * Get the agi buffer first. It ensures lock ordering
1794 * on the list.
1796 error = xfs_read_agi(mp, tp, agno, &agibp);
1797 if (error)
1798 return error;
1800 agi = XFS_BUF_TO_AGI(agibp);
1803 * Get the index into the agi hash table for the
1804 * list this inode will go on.
1806 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1807 ASSERT(agino != 0);
1808 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1809 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1810 ASSERT(agi->agi_unlinked[bucket_index]);
1812 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1814 * We're at the head of the list. Get the inode's
1815 * on-disk buffer to see if there is anyone after us
1816 * on the list. Only modify our next pointer if it
1817 * is not already NULLAGINO. This saves us the overhead
1818 * of dealing with the buffer when there is no need to
1819 * change it.
1821 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1822 if (error) {
1823 cmn_err(CE_WARN,
1824 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1825 error, mp->m_fsname);
1826 return error;
1828 next_agino = be32_to_cpu(dip->di_next_unlinked);
1829 ASSERT(next_agino != 0);
1830 if (next_agino != NULLAGINO) {
1831 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1832 offset = ip->i_imap.im_boffset +
1833 offsetof(xfs_dinode_t, di_next_unlinked);
1834 xfs_trans_inode_buf(tp, ibp);
1835 xfs_trans_log_buf(tp, ibp, offset,
1836 (offset + sizeof(xfs_agino_t) - 1));
1837 xfs_inobp_check(mp, ibp);
1838 } else {
1839 xfs_trans_brelse(tp, ibp);
1842 * Point the bucket head pointer at the next inode.
1844 ASSERT(next_agino != 0);
1845 ASSERT(next_agino != agino);
1846 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1847 offset = offsetof(xfs_agi_t, agi_unlinked) +
1848 (sizeof(xfs_agino_t) * bucket_index);
1849 xfs_trans_log_buf(tp, agibp, offset,
1850 (offset + sizeof(xfs_agino_t) - 1));
1851 } else {
1853 * We need to search the list for the inode being freed.
1855 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1856 last_ibp = NULL;
1857 while (next_agino != agino) {
1859 * If the last inode wasn't the one pointing to
1860 * us, then release its buffer since we're not
1861 * going to do anything with it.
1863 if (last_ibp != NULL) {
1864 xfs_trans_brelse(tp, last_ibp);
1866 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1867 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1868 &last_ibp, &last_offset, 0);
1869 if (error) {
1870 cmn_err(CE_WARN,
1871 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1872 error, mp->m_fsname);
1873 return error;
1875 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1876 ASSERT(next_agino != NULLAGINO);
1877 ASSERT(next_agino != 0);
1880 * Now last_ibp points to the buffer previous to us on
1881 * the unlinked list. Pull us from the list.
1883 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1884 if (error) {
1885 cmn_err(CE_WARN,
1886 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1887 error, mp->m_fsname);
1888 return error;
1890 next_agino = be32_to_cpu(dip->di_next_unlinked);
1891 ASSERT(next_agino != 0);
1892 ASSERT(next_agino != agino);
1893 if (next_agino != NULLAGINO) {
1894 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1895 offset = ip->i_imap.im_boffset +
1896 offsetof(xfs_dinode_t, di_next_unlinked);
1897 xfs_trans_inode_buf(tp, ibp);
1898 xfs_trans_log_buf(tp, ibp, offset,
1899 (offset + sizeof(xfs_agino_t) - 1));
1900 xfs_inobp_check(mp, ibp);
1901 } else {
1902 xfs_trans_brelse(tp, ibp);
1905 * Point the previous inode on the list to the next inode.
1907 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1908 ASSERT(next_agino != 0);
1909 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1910 xfs_trans_inode_buf(tp, last_ibp);
1911 xfs_trans_log_buf(tp, last_ibp, offset,
1912 (offset + sizeof(xfs_agino_t) - 1));
1913 xfs_inobp_check(mp, last_ibp);
1915 return 0;
1919 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1920 * inodes that are in memory - they all must be marked stale and attached to
1921 * the cluster buffer.
1923 STATIC void
1924 xfs_ifree_cluster(
1925 xfs_inode_t *free_ip,
1926 xfs_trans_t *tp,
1927 xfs_ino_t inum)
1929 xfs_mount_t *mp = free_ip->i_mount;
1930 int blks_per_cluster;
1931 int nbufs;
1932 int ninodes;
1933 int i, j;
1934 xfs_daddr_t blkno;
1935 xfs_buf_t *bp;
1936 xfs_inode_t *ip;
1937 xfs_inode_log_item_t *iip;
1938 xfs_log_item_t *lip;
1939 struct xfs_perag *pag;
1941 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1942 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1943 blks_per_cluster = 1;
1944 ninodes = mp->m_sb.sb_inopblock;
1945 nbufs = XFS_IALLOC_BLOCKS(mp);
1946 } else {
1947 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1948 mp->m_sb.sb_blocksize;
1949 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1950 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1953 for (j = 0; j < nbufs; j++, inum += ninodes) {
1954 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1955 XFS_INO_TO_AGBNO(mp, inum));
1958 * We obtain and lock the backing buffer first in the process
1959 * here, as we have to ensure that any dirty inode that we
1960 * can't get the flush lock on is attached to the buffer.
1961 * If we scan the in-memory inodes first, then buffer IO can
1962 * complete before we get a lock on it, and hence we may fail
1963 * to mark all the active inodes on the buffer stale.
1965 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1966 mp->m_bsize * blks_per_cluster,
1967 XBF_LOCK);
1970 * Walk the inodes already attached to the buffer and mark them
1971 * stale. These will all have the flush locks held, so an
1972 * in-memory inode walk can't lock them. By marking them all
1973 * stale first, we will not attempt to lock them in the loop
1974 * below as the XFS_ISTALE flag will be set.
1976 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1977 while (lip) {
1978 if (lip->li_type == XFS_LI_INODE) {
1979 iip = (xfs_inode_log_item_t *)lip;
1980 ASSERT(iip->ili_logged == 1);
1981 lip->li_cb = xfs_istale_done;
1982 xfs_trans_ail_copy_lsn(mp->m_ail,
1983 &iip->ili_flush_lsn,
1984 &iip->ili_item.li_lsn);
1985 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1987 lip = lip->li_bio_list;
1992 * For each inode in memory attempt to add it to the inode
1993 * buffer and set it up for being staled on buffer IO
1994 * completion. This is safe as we've locked out tail pushing
1995 * and flushing by locking the buffer.
1997 * We have already marked every inode that was part of a
1998 * transaction stale above, which means there is no point in
1999 * even trying to lock them.
2001 for (i = 0; i < ninodes; i++) {
2002 retry:
2003 rcu_read_lock();
2004 ip = radix_tree_lookup(&pag->pag_ici_root,
2005 XFS_INO_TO_AGINO(mp, (inum + i)));
2007 /* Inode not in memory, nothing to do */
2008 if (!ip) {
2009 rcu_read_unlock();
2010 continue;
2014 * because this is an RCU protected lookup, we could
2015 * find a recently freed or even reallocated inode
2016 * during the lookup. We need to check under the
2017 * i_flags_lock for a valid inode here. Skip it if it
2018 * is not valid, the wrong inode or stale.
2020 spin_lock(&ip->i_flags_lock);
2021 if (ip->i_ino != inum + i ||
2022 __xfs_iflags_test(ip, XFS_ISTALE)) {
2023 spin_unlock(&ip->i_flags_lock);
2024 rcu_read_unlock();
2025 continue;
2027 spin_unlock(&ip->i_flags_lock);
2030 * Don't try to lock/unlock the current inode, but we
2031 * _cannot_ skip the other inodes that we did not find
2032 * in the list attached to the buffer and are not
2033 * already marked stale. If we can't lock it, back off
2034 * and retry.
2036 if (ip != free_ip &&
2037 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2038 rcu_read_unlock();
2039 delay(1);
2040 goto retry;
2042 rcu_read_unlock();
2044 xfs_iflock(ip);
2045 xfs_iflags_set(ip, XFS_ISTALE);
2048 * we don't need to attach clean inodes or those only
2049 * with unlogged changes (which we throw away, anyway).
2051 iip = ip->i_itemp;
2052 if (!iip || xfs_inode_clean(ip)) {
2053 ASSERT(ip != free_ip);
2054 ip->i_update_core = 0;
2055 xfs_ifunlock(ip);
2056 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2057 continue;
2060 iip->ili_last_fields = iip->ili_format.ilf_fields;
2061 iip->ili_format.ilf_fields = 0;
2062 iip->ili_logged = 1;
2063 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2064 &iip->ili_item.li_lsn);
2066 xfs_buf_attach_iodone(bp, xfs_istale_done,
2067 &iip->ili_item);
2069 if (ip != free_ip)
2070 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2073 xfs_trans_stale_inode_buf(tp, bp);
2074 xfs_trans_binval(tp, bp);
2077 xfs_perag_put(pag);
2081 * This is called to return an inode to the inode free list.
2082 * The inode should already be truncated to 0 length and have
2083 * no pages associated with it. This routine also assumes that
2084 * the inode is already a part of the transaction.
2086 * The on-disk copy of the inode will have been added to the list
2087 * of unlinked inodes in the AGI. We need to remove the inode from
2088 * that list atomically with respect to freeing it here.
2091 xfs_ifree(
2092 xfs_trans_t *tp,
2093 xfs_inode_t *ip,
2094 xfs_bmap_free_t *flist)
2096 int error;
2097 int delete;
2098 xfs_ino_t first_ino;
2099 xfs_dinode_t *dip;
2100 xfs_buf_t *ibp;
2102 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2103 ASSERT(ip->i_transp == tp);
2104 ASSERT(ip->i_d.di_nlink == 0);
2105 ASSERT(ip->i_d.di_nextents == 0);
2106 ASSERT(ip->i_d.di_anextents == 0);
2107 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2108 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2109 ASSERT(ip->i_d.di_nblocks == 0);
2112 * Pull the on-disk inode from the AGI unlinked list.
2114 error = xfs_iunlink_remove(tp, ip);
2115 if (error != 0) {
2116 return error;
2119 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2120 if (error != 0) {
2121 return error;
2123 ip->i_d.di_mode = 0; /* mark incore inode as free */
2124 ip->i_d.di_flags = 0;
2125 ip->i_d.di_dmevmask = 0;
2126 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2127 ip->i_df.if_ext_max =
2128 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2129 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2130 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2132 * Bump the generation count so no one will be confused
2133 * by reincarnations of this inode.
2135 ip->i_d.di_gen++;
2137 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2139 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
2140 if (error)
2141 return error;
2144 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2145 * from picking up this inode when it is reclaimed (its incore state
2146 * initialzed but not flushed to disk yet). The in-core di_mode is
2147 * already cleared and a corresponding transaction logged.
2148 * The hack here just synchronizes the in-core to on-disk
2149 * di_mode value in advance before the actual inode sync to disk.
2150 * This is OK because the inode is already unlinked and would never
2151 * change its di_mode again for this inode generation.
2152 * This is a temporary hack that would require a proper fix
2153 * in the future.
2155 dip->di_mode = 0;
2157 if (delete) {
2158 xfs_ifree_cluster(ip, tp, first_ino);
2161 return 0;
2165 * Reallocate the space for if_broot based on the number of records
2166 * being added or deleted as indicated in rec_diff. Move the records
2167 * and pointers in if_broot to fit the new size. When shrinking this
2168 * will eliminate holes between the records and pointers created by
2169 * the caller. When growing this will create holes to be filled in
2170 * by the caller.
2172 * The caller must not request to add more records than would fit in
2173 * the on-disk inode root. If the if_broot is currently NULL, then
2174 * if we adding records one will be allocated. The caller must also
2175 * not request that the number of records go below zero, although
2176 * it can go to zero.
2178 * ip -- the inode whose if_broot area is changing
2179 * ext_diff -- the change in the number of records, positive or negative,
2180 * requested for the if_broot array.
2182 void
2183 xfs_iroot_realloc(
2184 xfs_inode_t *ip,
2185 int rec_diff,
2186 int whichfork)
2188 struct xfs_mount *mp = ip->i_mount;
2189 int cur_max;
2190 xfs_ifork_t *ifp;
2191 struct xfs_btree_block *new_broot;
2192 int new_max;
2193 size_t new_size;
2194 char *np;
2195 char *op;
2198 * Handle the degenerate case quietly.
2200 if (rec_diff == 0) {
2201 return;
2204 ifp = XFS_IFORK_PTR(ip, whichfork);
2205 if (rec_diff > 0) {
2207 * If there wasn't any memory allocated before, just
2208 * allocate it now and get out.
2210 if (ifp->if_broot_bytes == 0) {
2211 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2212 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2213 ifp->if_broot_bytes = (int)new_size;
2214 return;
2218 * If there is already an existing if_broot, then we need
2219 * to realloc() it and shift the pointers to their new
2220 * location. The records don't change location because
2221 * they are kept butted up against the btree block header.
2223 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2224 new_max = cur_max + rec_diff;
2225 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2226 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
2227 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2228 KM_SLEEP | KM_NOFS);
2229 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2230 ifp->if_broot_bytes);
2231 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2232 (int)new_size);
2233 ifp->if_broot_bytes = (int)new_size;
2234 ASSERT(ifp->if_broot_bytes <=
2235 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2236 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2237 return;
2241 * rec_diff is less than 0. In this case, we are shrinking the
2242 * if_broot buffer. It must already exist. If we go to zero
2243 * records, just get rid of the root and clear the status bit.
2245 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2246 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
2247 new_max = cur_max + rec_diff;
2248 ASSERT(new_max >= 0);
2249 if (new_max > 0)
2250 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2251 else
2252 new_size = 0;
2253 if (new_size > 0) {
2254 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
2256 * First copy over the btree block header.
2258 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
2259 } else {
2260 new_broot = NULL;
2261 ifp->if_flags &= ~XFS_IFBROOT;
2265 * Only copy the records and pointers if there are any.
2267 if (new_max > 0) {
2269 * First copy the records.
2271 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2272 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
2273 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2276 * Then copy the pointers.
2278 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2279 ifp->if_broot_bytes);
2280 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2281 (int)new_size);
2282 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2284 kmem_free(ifp->if_broot);
2285 ifp->if_broot = new_broot;
2286 ifp->if_broot_bytes = (int)new_size;
2287 ASSERT(ifp->if_broot_bytes <=
2288 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2289 return;
2294 * This is called when the amount of space needed for if_data
2295 * is increased or decreased. The change in size is indicated by
2296 * the number of bytes that need to be added or deleted in the
2297 * byte_diff parameter.
2299 * If the amount of space needed has decreased below the size of the
2300 * inline buffer, then switch to using the inline buffer. Otherwise,
2301 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2302 * to what is needed.
2304 * ip -- the inode whose if_data area is changing
2305 * byte_diff -- the change in the number of bytes, positive or negative,
2306 * requested for the if_data array.
2308 void
2309 xfs_idata_realloc(
2310 xfs_inode_t *ip,
2311 int byte_diff,
2312 int whichfork)
2314 xfs_ifork_t *ifp;
2315 int new_size;
2316 int real_size;
2318 if (byte_diff == 0) {
2319 return;
2322 ifp = XFS_IFORK_PTR(ip, whichfork);
2323 new_size = (int)ifp->if_bytes + byte_diff;
2324 ASSERT(new_size >= 0);
2326 if (new_size == 0) {
2327 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2328 kmem_free(ifp->if_u1.if_data);
2330 ifp->if_u1.if_data = NULL;
2331 real_size = 0;
2332 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2334 * If the valid extents/data can fit in if_inline_ext/data,
2335 * copy them from the malloc'd vector and free it.
2337 if (ifp->if_u1.if_data == NULL) {
2338 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2339 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2340 ASSERT(ifp->if_real_bytes != 0);
2341 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2342 new_size);
2343 kmem_free(ifp->if_u1.if_data);
2344 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2346 real_size = 0;
2347 } else {
2349 * Stuck with malloc/realloc.
2350 * For inline data, the underlying buffer must be
2351 * a multiple of 4 bytes in size so that it can be
2352 * logged and stay on word boundaries. We enforce
2353 * that here.
2355 real_size = roundup(new_size, 4);
2356 if (ifp->if_u1.if_data == NULL) {
2357 ASSERT(ifp->if_real_bytes == 0);
2358 ifp->if_u1.if_data = kmem_alloc(real_size,
2359 KM_SLEEP | KM_NOFS);
2360 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2362 * Only do the realloc if the underlying size
2363 * is really changing.
2365 if (ifp->if_real_bytes != real_size) {
2366 ifp->if_u1.if_data =
2367 kmem_realloc(ifp->if_u1.if_data,
2368 real_size,
2369 ifp->if_real_bytes,
2370 KM_SLEEP | KM_NOFS);
2372 } else {
2373 ASSERT(ifp->if_real_bytes == 0);
2374 ifp->if_u1.if_data = kmem_alloc(real_size,
2375 KM_SLEEP | KM_NOFS);
2376 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2377 ifp->if_bytes);
2380 ifp->if_real_bytes = real_size;
2381 ifp->if_bytes = new_size;
2382 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2385 void
2386 xfs_idestroy_fork(
2387 xfs_inode_t *ip,
2388 int whichfork)
2390 xfs_ifork_t *ifp;
2392 ifp = XFS_IFORK_PTR(ip, whichfork);
2393 if (ifp->if_broot != NULL) {
2394 kmem_free(ifp->if_broot);
2395 ifp->if_broot = NULL;
2399 * If the format is local, then we can't have an extents
2400 * array so just look for an inline data array. If we're
2401 * not local then we may or may not have an extents list,
2402 * so check and free it up if we do.
2404 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2405 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2406 (ifp->if_u1.if_data != NULL)) {
2407 ASSERT(ifp->if_real_bytes != 0);
2408 kmem_free(ifp->if_u1.if_data);
2409 ifp->if_u1.if_data = NULL;
2410 ifp->if_real_bytes = 0;
2412 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2413 ((ifp->if_flags & XFS_IFEXTIREC) ||
2414 ((ifp->if_u1.if_extents != NULL) &&
2415 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2416 ASSERT(ifp->if_real_bytes != 0);
2417 xfs_iext_destroy(ifp);
2419 ASSERT(ifp->if_u1.if_extents == NULL ||
2420 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2421 ASSERT(ifp->if_real_bytes == 0);
2422 if (whichfork == XFS_ATTR_FORK) {
2423 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2424 ip->i_afp = NULL;
2429 * This is called to unpin an inode. The caller must have the inode locked
2430 * in at least shared mode so that the buffer cannot be subsequently pinned
2431 * once someone is waiting for it to be unpinned.
2433 static void
2434 xfs_iunpin_nowait(
2435 struct xfs_inode *ip)
2437 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2439 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2441 /* Give the log a push to start the unpinning I/O */
2442 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2446 void
2447 xfs_iunpin_wait(
2448 struct xfs_inode *ip)
2450 if (xfs_ipincount(ip)) {
2451 xfs_iunpin_nowait(ip);
2452 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2457 * xfs_iextents_copy()
2459 * This is called to copy the REAL extents (as opposed to the delayed
2460 * allocation extents) from the inode into the given buffer. It
2461 * returns the number of bytes copied into the buffer.
2463 * If there are no delayed allocation extents, then we can just
2464 * memcpy() the extents into the buffer. Otherwise, we need to
2465 * examine each extent in turn and skip those which are delayed.
2468 xfs_iextents_copy(
2469 xfs_inode_t *ip,
2470 xfs_bmbt_rec_t *dp,
2471 int whichfork)
2473 int copied;
2474 int i;
2475 xfs_ifork_t *ifp;
2476 int nrecs;
2477 xfs_fsblock_t start_block;
2479 ifp = XFS_IFORK_PTR(ip, whichfork);
2480 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2481 ASSERT(ifp->if_bytes > 0);
2483 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2484 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2485 ASSERT(nrecs > 0);
2488 * There are some delayed allocation extents in the
2489 * inode, so copy the extents one at a time and skip
2490 * the delayed ones. There must be at least one
2491 * non-delayed extent.
2493 copied = 0;
2494 for (i = 0; i < nrecs; i++) {
2495 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2496 start_block = xfs_bmbt_get_startblock(ep);
2497 if (isnullstartblock(start_block)) {
2499 * It's a delayed allocation extent, so skip it.
2501 continue;
2504 /* Translate to on disk format */
2505 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2506 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2507 dp++;
2508 copied++;
2510 ASSERT(copied != 0);
2511 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2513 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2517 * Each of the following cases stores data into the same region
2518 * of the on-disk inode, so only one of them can be valid at
2519 * any given time. While it is possible to have conflicting formats
2520 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2521 * in EXTENTS format, this can only happen when the fork has
2522 * changed formats after being modified but before being flushed.
2523 * In these cases, the format always takes precedence, because the
2524 * format indicates the current state of the fork.
2526 /*ARGSUSED*/
2527 STATIC void
2528 xfs_iflush_fork(
2529 xfs_inode_t *ip,
2530 xfs_dinode_t *dip,
2531 xfs_inode_log_item_t *iip,
2532 int whichfork,
2533 xfs_buf_t *bp)
2535 char *cp;
2536 xfs_ifork_t *ifp;
2537 xfs_mount_t *mp;
2538 #ifdef XFS_TRANS_DEBUG
2539 int first;
2540 #endif
2541 static const short brootflag[2] =
2542 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2543 static const short dataflag[2] =
2544 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2545 static const short extflag[2] =
2546 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2548 if (!iip)
2549 return;
2550 ifp = XFS_IFORK_PTR(ip, whichfork);
2552 * This can happen if we gave up in iformat in an error path,
2553 * for the attribute fork.
2555 if (!ifp) {
2556 ASSERT(whichfork == XFS_ATTR_FORK);
2557 return;
2559 cp = XFS_DFORK_PTR(dip, whichfork);
2560 mp = ip->i_mount;
2561 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2562 case XFS_DINODE_FMT_LOCAL:
2563 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2564 (ifp->if_bytes > 0)) {
2565 ASSERT(ifp->if_u1.if_data != NULL);
2566 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2567 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2569 break;
2571 case XFS_DINODE_FMT_EXTENTS:
2572 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2573 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2574 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2575 (ifp->if_bytes == 0));
2576 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2577 (ifp->if_bytes > 0));
2578 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2579 (ifp->if_bytes > 0)) {
2580 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2581 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2582 whichfork);
2584 break;
2586 case XFS_DINODE_FMT_BTREE:
2587 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2588 (ifp->if_broot_bytes > 0)) {
2589 ASSERT(ifp->if_broot != NULL);
2590 ASSERT(ifp->if_broot_bytes <=
2591 (XFS_IFORK_SIZE(ip, whichfork) +
2592 XFS_BROOT_SIZE_ADJ));
2593 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2594 (xfs_bmdr_block_t *)cp,
2595 XFS_DFORK_SIZE(dip, mp, whichfork));
2597 break;
2599 case XFS_DINODE_FMT_DEV:
2600 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2601 ASSERT(whichfork == XFS_DATA_FORK);
2602 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2604 break;
2606 case XFS_DINODE_FMT_UUID:
2607 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2608 ASSERT(whichfork == XFS_DATA_FORK);
2609 memcpy(XFS_DFORK_DPTR(dip),
2610 &ip->i_df.if_u2.if_uuid,
2611 sizeof(uuid_t));
2613 break;
2615 default:
2616 ASSERT(0);
2617 break;
2621 STATIC int
2622 xfs_iflush_cluster(
2623 xfs_inode_t *ip,
2624 xfs_buf_t *bp)
2626 xfs_mount_t *mp = ip->i_mount;
2627 struct xfs_perag *pag;
2628 unsigned long first_index, mask;
2629 unsigned long inodes_per_cluster;
2630 int ilist_size;
2631 xfs_inode_t **ilist;
2632 xfs_inode_t *iq;
2633 int nr_found;
2634 int clcount = 0;
2635 int bufwasdelwri;
2636 int i;
2638 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2640 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2641 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2642 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2643 if (!ilist)
2644 goto out_put;
2646 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2647 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2648 rcu_read_lock();
2649 /* really need a gang lookup range call here */
2650 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2651 first_index, inodes_per_cluster);
2652 if (nr_found == 0)
2653 goto out_free;
2655 for (i = 0; i < nr_found; i++) {
2656 iq = ilist[i];
2657 if (iq == ip)
2658 continue;
2661 * because this is an RCU protected lookup, we could find a
2662 * recently freed or even reallocated inode during the lookup.
2663 * We need to check under the i_flags_lock for a valid inode
2664 * here. Skip it if it is not valid or the wrong inode.
2666 spin_lock(&ip->i_flags_lock);
2667 if (!ip->i_ino ||
2668 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2669 spin_unlock(&ip->i_flags_lock);
2670 continue;
2672 spin_unlock(&ip->i_flags_lock);
2675 * Do an un-protected check to see if the inode is dirty and
2676 * is a candidate for flushing. These checks will be repeated
2677 * later after the appropriate locks are acquired.
2679 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2680 continue;
2683 * Try to get locks. If any are unavailable or it is pinned,
2684 * then this inode cannot be flushed and is skipped.
2687 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2688 continue;
2689 if (!xfs_iflock_nowait(iq)) {
2690 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2691 continue;
2693 if (xfs_ipincount(iq)) {
2694 xfs_ifunlock(iq);
2695 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2696 continue;
2700 * arriving here means that this inode can be flushed. First
2701 * re-check that it's dirty before flushing.
2703 if (!xfs_inode_clean(iq)) {
2704 int error;
2705 error = xfs_iflush_int(iq, bp);
2706 if (error) {
2707 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2708 goto cluster_corrupt_out;
2710 clcount++;
2711 } else {
2712 xfs_ifunlock(iq);
2714 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2717 if (clcount) {
2718 XFS_STATS_INC(xs_icluster_flushcnt);
2719 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2722 out_free:
2723 rcu_read_unlock();
2724 kmem_free(ilist);
2725 out_put:
2726 xfs_perag_put(pag);
2727 return 0;
2730 cluster_corrupt_out:
2732 * Corruption detected in the clustering loop. Invalidate the
2733 * inode buffer and shut down the filesystem.
2735 rcu_read_unlock();
2737 * Clean up the buffer. If it was B_DELWRI, just release it --
2738 * brelse can handle it with no problems. If not, shut down the
2739 * filesystem before releasing the buffer.
2741 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2742 if (bufwasdelwri)
2743 xfs_buf_relse(bp);
2745 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2747 if (!bufwasdelwri) {
2749 * Just like incore_relse: if we have b_iodone functions,
2750 * mark the buffer as an error and call them. Otherwise
2751 * mark it as stale and brelse.
2753 if (XFS_BUF_IODONE_FUNC(bp)) {
2754 XFS_BUF_UNDONE(bp);
2755 XFS_BUF_STALE(bp);
2756 XFS_BUF_ERROR(bp,EIO);
2757 xfs_buf_ioend(bp, 0);
2758 } else {
2759 XFS_BUF_STALE(bp);
2760 xfs_buf_relse(bp);
2765 * Unlocks the flush lock
2767 xfs_iflush_abort(iq);
2768 kmem_free(ilist);
2769 xfs_perag_put(pag);
2770 return XFS_ERROR(EFSCORRUPTED);
2774 * xfs_iflush() will write a modified inode's changes out to the
2775 * inode's on disk home. The caller must have the inode lock held
2776 * in at least shared mode and the inode flush completion must be
2777 * active as well. The inode lock will still be held upon return from
2778 * the call and the caller is free to unlock it.
2779 * The inode flush will be completed when the inode reaches the disk.
2780 * The flags indicate how the inode's buffer should be written out.
2783 xfs_iflush(
2784 xfs_inode_t *ip,
2785 uint flags)
2787 xfs_inode_log_item_t *iip;
2788 xfs_buf_t *bp;
2789 xfs_dinode_t *dip;
2790 xfs_mount_t *mp;
2791 int error;
2793 XFS_STATS_INC(xs_iflush_count);
2795 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2796 ASSERT(!completion_done(&ip->i_flush));
2797 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2798 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2800 iip = ip->i_itemp;
2801 mp = ip->i_mount;
2804 * We can't flush the inode until it is unpinned, so wait for it if we
2805 * are allowed to block. We know noone new can pin it, because we are
2806 * holding the inode lock shared and you need to hold it exclusively to
2807 * pin the inode.
2809 * If we are not allowed to block, force the log out asynchronously so
2810 * that when we come back the inode will be unpinned. If other inodes
2811 * in the same cluster are dirty, they will probably write the inode
2812 * out for us if they occur after the log force completes.
2814 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2815 xfs_iunpin_nowait(ip);
2816 xfs_ifunlock(ip);
2817 return EAGAIN;
2819 xfs_iunpin_wait(ip);
2822 * For stale inodes we cannot rely on the backing buffer remaining
2823 * stale in cache for the remaining life of the stale inode and so
2824 * xfs_itobp() below may give us a buffer that no longer contains
2825 * inodes below. We have to check this after ensuring the inode is
2826 * unpinned so that it is safe to reclaim the stale inode after the
2827 * flush call.
2829 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2830 xfs_ifunlock(ip);
2831 return 0;
2835 * This may have been unpinned because the filesystem is shutting
2836 * down forcibly. If that's the case we must not write this inode
2837 * to disk, because the log record didn't make it to disk!
2839 if (XFS_FORCED_SHUTDOWN(mp)) {
2840 ip->i_update_core = 0;
2841 if (iip)
2842 iip->ili_format.ilf_fields = 0;
2843 xfs_ifunlock(ip);
2844 return XFS_ERROR(EIO);
2848 * Get the buffer containing the on-disk inode.
2850 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2851 (flags & SYNC_WAIT) ? XBF_LOCK : XBF_TRYLOCK);
2852 if (error || !bp) {
2853 xfs_ifunlock(ip);
2854 return error;
2858 * First flush out the inode that xfs_iflush was called with.
2860 error = xfs_iflush_int(ip, bp);
2861 if (error)
2862 goto corrupt_out;
2865 * If the buffer is pinned then push on the log now so we won't
2866 * get stuck waiting in the write for too long.
2868 if (XFS_BUF_ISPINNED(bp))
2869 xfs_log_force(mp, 0);
2872 * inode clustering:
2873 * see if other inodes can be gathered into this write
2875 error = xfs_iflush_cluster(ip, bp);
2876 if (error)
2877 goto cluster_corrupt_out;
2879 if (flags & SYNC_WAIT)
2880 error = xfs_bwrite(mp, bp);
2881 else
2882 xfs_bdwrite(mp, bp);
2883 return error;
2885 corrupt_out:
2886 xfs_buf_relse(bp);
2887 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2888 cluster_corrupt_out:
2890 * Unlocks the flush lock
2892 xfs_iflush_abort(ip);
2893 return XFS_ERROR(EFSCORRUPTED);
2897 STATIC int
2898 xfs_iflush_int(
2899 xfs_inode_t *ip,
2900 xfs_buf_t *bp)
2902 xfs_inode_log_item_t *iip;
2903 xfs_dinode_t *dip;
2904 xfs_mount_t *mp;
2905 #ifdef XFS_TRANS_DEBUG
2906 int first;
2907 #endif
2909 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2910 ASSERT(!completion_done(&ip->i_flush));
2911 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2912 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2914 iip = ip->i_itemp;
2915 mp = ip->i_mount;
2917 /* set *dip = inode's place in the buffer */
2918 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2921 * Clear i_update_core before copying out the data.
2922 * This is for coordination with our timestamp updates
2923 * that don't hold the inode lock. They will always
2924 * update the timestamps BEFORE setting i_update_core,
2925 * so if we clear i_update_core after they set it we
2926 * are guaranteed to see their updates to the timestamps.
2927 * I believe that this depends on strongly ordered memory
2928 * semantics, but we have that. We use the SYNCHRONIZE
2929 * macro to make sure that the compiler does not reorder
2930 * the i_update_core access below the data copy below.
2932 ip->i_update_core = 0;
2933 SYNCHRONIZE();
2936 * Make sure to get the latest timestamps from the Linux inode.
2938 xfs_synchronize_times(ip);
2940 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
2941 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2942 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2943 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2944 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2945 goto corrupt_out;
2947 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2948 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2949 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2950 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2951 ip->i_ino, ip, ip->i_d.di_magic);
2952 goto corrupt_out;
2954 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2955 if (XFS_TEST_ERROR(
2956 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2957 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2958 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2959 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2960 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
2961 ip->i_ino, ip);
2962 goto corrupt_out;
2964 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2965 if (XFS_TEST_ERROR(
2966 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2967 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2968 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2969 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2970 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2971 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
2972 ip->i_ino, ip);
2973 goto corrupt_out;
2976 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2977 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2978 XFS_RANDOM_IFLUSH_5)) {
2979 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2980 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
2981 ip->i_ino,
2982 ip->i_d.di_nextents + ip->i_d.di_anextents,
2983 ip->i_d.di_nblocks,
2984 ip);
2985 goto corrupt_out;
2987 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2988 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2989 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2990 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2991 ip->i_ino, ip->i_d.di_forkoff, ip);
2992 goto corrupt_out;
2995 * bump the flush iteration count, used to detect flushes which
2996 * postdate a log record during recovery.
2999 ip->i_d.di_flushiter++;
3002 * Copy the dirty parts of the inode into the on-disk
3003 * inode. We always copy out the core of the inode,
3004 * because if the inode is dirty at all the core must
3005 * be.
3007 xfs_dinode_to_disk(dip, &ip->i_d);
3009 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3010 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3011 ip->i_d.di_flushiter = 0;
3014 * If this is really an old format inode and the superblock version
3015 * has not been updated to support only new format inodes, then
3016 * convert back to the old inode format. If the superblock version
3017 * has been updated, then make the conversion permanent.
3019 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3020 if (ip->i_d.di_version == 1) {
3021 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3023 * Convert it back.
3025 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3026 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3027 } else {
3029 * The superblock version has already been bumped,
3030 * so just make the conversion to the new inode
3031 * format permanent.
3033 ip->i_d.di_version = 2;
3034 dip->di_version = 2;
3035 ip->i_d.di_onlink = 0;
3036 dip->di_onlink = 0;
3037 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3038 memset(&(dip->di_pad[0]), 0,
3039 sizeof(dip->di_pad));
3040 ASSERT(xfs_get_projid(ip) == 0);
3044 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3045 if (XFS_IFORK_Q(ip))
3046 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3047 xfs_inobp_check(mp, bp);
3050 * We've recorded everything logged in the inode, so we'd
3051 * like to clear the ilf_fields bits so we don't log and
3052 * flush things unnecessarily. However, we can't stop
3053 * logging all this information until the data we've copied
3054 * into the disk buffer is written to disk. If we did we might
3055 * overwrite the copy of the inode in the log with all the
3056 * data after re-logging only part of it, and in the face of
3057 * a crash we wouldn't have all the data we need to recover.
3059 * What we do is move the bits to the ili_last_fields field.
3060 * When logging the inode, these bits are moved back to the
3061 * ilf_fields field. In the xfs_iflush_done() routine we
3062 * clear ili_last_fields, since we know that the information
3063 * those bits represent is permanently on disk. As long as
3064 * the flush completes before the inode is logged again, then
3065 * both ilf_fields and ili_last_fields will be cleared.
3067 * We can play with the ilf_fields bits here, because the inode
3068 * lock must be held exclusively in order to set bits there
3069 * and the flush lock protects the ili_last_fields bits.
3070 * Set ili_logged so the flush done
3071 * routine can tell whether or not to look in the AIL.
3072 * Also, store the current LSN of the inode so that we can tell
3073 * whether the item has moved in the AIL from xfs_iflush_done().
3074 * In order to read the lsn we need the AIL lock, because
3075 * it is a 64 bit value that cannot be read atomically.
3077 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3078 iip->ili_last_fields = iip->ili_format.ilf_fields;
3079 iip->ili_format.ilf_fields = 0;
3080 iip->ili_logged = 1;
3082 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3083 &iip->ili_item.li_lsn);
3086 * Attach the function xfs_iflush_done to the inode's
3087 * buffer. This will remove the inode from the AIL
3088 * and unlock the inode's flush lock when the inode is
3089 * completely written to disk.
3091 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3093 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3094 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3095 } else {
3097 * We're flushing an inode which is not in the AIL and has
3098 * not been logged but has i_update_core set. For this
3099 * case we can use a B_DELWRI flush and immediately drop
3100 * the inode flush lock because we can avoid the whole
3101 * AIL state thing. It's OK to drop the flush lock now,
3102 * because we've already locked the buffer and to do anything
3103 * you really need both.
3105 if (iip != NULL) {
3106 ASSERT(iip->ili_logged == 0);
3107 ASSERT(iip->ili_last_fields == 0);
3108 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3110 xfs_ifunlock(ip);
3113 return 0;
3115 corrupt_out:
3116 return XFS_ERROR(EFSCORRUPTED);
3120 * Return a pointer to the extent record at file index idx.
3122 xfs_bmbt_rec_host_t *
3123 xfs_iext_get_ext(
3124 xfs_ifork_t *ifp, /* inode fork pointer */
3125 xfs_extnum_t idx) /* index of target extent */
3127 ASSERT(idx >= 0);
3128 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3129 return ifp->if_u1.if_ext_irec->er_extbuf;
3130 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3131 xfs_ext_irec_t *erp; /* irec pointer */
3132 int erp_idx = 0; /* irec index */
3133 xfs_extnum_t page_idx = idx; /* ext index in target list */
3135 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3136 return &erp->er_extbuf[page_idx];
3137 } else if (ifp->if_bytes) {
3138 return &ifp->if_u1.if_extents[idx];
3139 } else {
3140 return NULL;
3145 * Insert new item(s) into the extent records for incore inode
3146 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3148 void
3149 xfs_iext_insert(
3150 xfs_inode_t *ip, /* incore inode pointer */
3151 xfs_extnum_t idx, /* starting index of new items */
3152 xfs_extnum_t count, /* number of inserted items */
3153 xfs_bmbt_irec_t *new, /* items to insert */
3154 int state) /* type of extent conversion */
3156 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3157 xfs_extnum_t i; /* extent record index */
3159 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3161 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3162 xfs_iext_add(ifp, idx, count);
3163 for (i = idx; i < idx + count; i++, new++)
3164 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3168 * This is called when the amount of space required for incore file
3169 * extents needs to be increased. The ext_diff parameter stores the
3170 * number of new extents being added and the idx parameter contains
3171 * the extent index where the new extents will be added. If the new
3172 * extents are being appended, then we just need to (re)allocate and
3173 * initialize the space. Otherwise, if the new extents are being
3174 * inserted into the middle of the existing entries, a bit more work
3175 * is required to make room for the new extents to be inserted. The
3176 * caller is responsible for filling in the new extent entries upon
3177 * return.
3179 void
3180 xfs_iext_add(
3181 xfs_ifork_t *ifp, /* inode fork pointer */
3182 xfs_extnum_t idx, /* index to begin adding exts */
3183 int ext_diff) /* number of extents to add */
3185 int byte_diff; /* new bytes being added */
3186 int new_size; /* size of extents after adding */
3187 xfs_extnum_t nextents; /* number of extents in file */
3189 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3190 ASSERT((idx >= 0) && (idx <= nextents));
3191 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3192 new_size = ifp->if_bytes + byte_diff;
3194 * If the new number of extents (nextents + ext_diff)
3195 * fits inside the inode, then continue to use the inline
3196 * extent buffer.
3198 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3199 if (idx < nextents) {
3200 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3201 &ifp->if_u2.if_inline_ext[idx],
3202 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3203 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3205 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3206 ifp->if_real_bytes = 0;
3207 ifp->if_lastex = nextents + ext_diff;
3210 * Otherwise use a linear (direct) extent list.
3211 * If the extents are currently inside the inode,
3212 * xfs_iext_realloc_direct will switch us from
3213 * inline to direct extent allocation mode.
3215 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3216 xfs_iext_realloc_direct(ifp, new_size);
3217 if (idx < nextents) {
3218 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3219 &ifp->if_u1.if_extents[idx],
3220 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3221 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3224 /* Indirection array */
3225 else {
3226 xfs_ext_irec_t *erp;
3227 int erp_idx = 0;
3228 int page_idx = idx;
3230 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3231 if (ifp->if_flags & XFS_IFEXTIREC) {
3232 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3233 } else {
3234 xfs_iext_irec_init(ifp);
3235 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3236 erp = ifp->if_u1.if_ext_irec;
3238 /* Extents fit in target extent page */
3239 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3240 if (page_idx < erp->er_extcount) {
3241 memmove(&erp->er_extbuf[page_idx + ext_diff],
3242 &erp->er_extbuf[page_idx],
3243 (erp->er_extcount - page_idx) *
3244 sizeof(xfs_bmbt_rec_t));
3245 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3247 erp->er_extcount += ext_diff;
3248 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3250 /* Insert a new extent page */
3251 else if (erp) {
3252 xfs_iext_add_indirect_multi(ifp,
3253 erp_idx, page_idx, ext_diff);
3256 * If extent(s) are being appended to the last page in
3257 * the indirection array and the new extent(s) don't fit
3258 * in the page, then erp is NULL and erp_idx is set to
3259 * the next index needed in the indirection array.
3261 else {
3262 int count = ext_diff;
3264 while (count) {
3265 erp = xfs_iext_irec_new(ifp, erp_idx);
3266 erp->er_extcount = count;
3267 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3268 if (count) {
3269 erp_idx++;
3274 ifp->if_bytes = new_size;
3278 * This is called when incore extents are being added to the indirection
3279 * array and the new extents do not fit in the target extent list. The
3280 * erp_idx parameter contains the irec index for the target extent list
3281 * in the indirection array, and the idx parameter contains the extent
3282 * index within the list. The number of extents being added is stored
3283 * in the count parameter.
3285 * |-------| |-------|
3286 * | | | | idx - number of extents before idx
3287 * | idx | | count |
3288 * | | | | count - number of extents being inserted at idx
3289 * |-------| |-------|
3290 * | count | | nex2 | nex2 - number of extents after idx + count
3291 * |-------| |-------|
3293 void
3294 xfs_iext_add_indirect_multi(
3295 xfs_ifork_t *ifp, /* inode fork pointer */
3296 int erp_idx, /* target extent irec index */
3297 xfs_extnum_t idx, /* index within target list */
3298 int count) /* new extents being added */
3300 int byte_diff; /* new bytes being added */
3301 xfs_ext_irec_t *erp; /* pointer to irec entry */
3302 xfs_extnum_t ext_diff; /* number of extents to add */
3303 xfs_extnum_t ext_cnt; /* new extents still needed */
3304 xfs_extnum_t nex2; /* extents after idx + count */
3305 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3306 int nlists; /* number of irec's (lists) */
3308 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3309 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3310 nex2 = erp->er_extcount - idx;
3311 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3314 * Save second part of target extent list
3315 * (all extents past */
3316 if (nex2) {
3317 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3318 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3319 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3320 erp->er_extcount -= nex2;
3321 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3322 memset(&erp->er_extbuf[idx], 0, byte_diff);
3326 * Add the new extents to the end of the target
3327 * list, then allocate new irec record(s) and
3328 * extent buffer(s) as needed to store the rest
3329 * of the new extents.
3331 ext_cnt = count;
3332 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3333 if (ext_diff) {
3334 erp->er_extcount += ext_diff;
3335 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3336 ext_cnt -= ext_diff;
3338 while (ext_cnt) {
3339 erp_idx++;
3340 erp = xfs_iext_irec_new(ifp, erp_idx);
3341 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3342 erp->er_extcount = ext_diff;
3343 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3344 ext_cnt -= ext_diff;
3347 /* Add nex2 extents back to indirection array */
3348 if (nex2) {
3349 xfs_extnum_t ext_avail;
3350 int i;
3352 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3353 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3354 i = 0;
3356 * If nex2 extents fit in the current page, append
3357 * nex2_ep after the new extents.
3359 if (nex2 <= ext_avail) {
3360 i = erp->er_extcount;
3363 * Otherwise, check if space is available in the
3364 * next page.
3366 else if ((erp_idx < nlists - 1) &&
3367 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3368 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3369 erp_idx++;
3370 erp++;
3371 /* Create a hole for nex2 extents */
3372 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3373 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3376 * Final choice, create a new extent page for
3377 * nex2 extents.
3379 else {
3380 erp_idx++;
3381 erp = xfs_iext_irec_new(ifp, erp_idx);
3383 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3384 kmem_free(nex2_ep);
3385 erp->er_extcount += nex2;
3386 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3391 * This is called when the amount of space required for incore file
3392 * extents needs to be decreased. The ext_diff parameter stores the
3393 * number of extents to be removed and the idx parameter contains
3394 * the extent index where the extents will be removed from.
3396 * If the amount of space needed has decreased below the linear
3397 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3398 * extent array. Otherwise, use kmem_realloc() to adjust the
3399 * size to what is needed.
3401 void
3402 xfs_iext_remove(
3403 xfs_inode_t *ip, /* incore inode pointer */
3404 xfs_extnum_t idx, /* index to begin removing exts */
3405 int ext_diff, /* number of extents to remove */
3406 int state) /* type of extent conversion */
3408 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3409 xfs_extnum_t nextents; /* number of extents in file */
3410 int new_size; /* size of extents after removal */
3412 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3414 ASSERT(ext_diff > 0);
3415 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3416 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3418 if (new_size == 0) {
3419 xfs_iext_destroy(ifp);
3420 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3421 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3422 } else if (ifp->if_real_bytes) {
3423 xfs_iext_remove_direct(ifp, idx, ext_diff);
3424 } else {
3425 xfs_iext_remove_inline(ifp, idx, ext_diff);
3427 ifp->if_bytes = new_size;
3431 * This removes ext_diff extents from the inline buffer, beginning
3432 * at extent index idx.
3434 void
3435 xfs_iext_remove_inline(
3436 xfs_ifork_t *ifp, /* inode fork pointer */
3437 xfs_extnum_t idx, /* index to begin removing exts */
3438 int ext_diff) /* number of extents to remove */
3440 int nextents; /* number of extents in file */
3442 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3443 ASSERT(idx < XFS_INLINE_EXTS);
3444 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3445 ASSERT(((nextents - ext_diff) > 0) &&
3446 (nextents - ext_diff) < XFS_INLINE_EXTS);
3448 if (idx + ext_diff < nextents) {
3449 memmove(&ifp->if_u2.if_inline_ext[idx],
3450 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3451 (nextents - (idx + ext_diff)) *
3452 sizeof(xfs_bmbt_rec_t));
3453 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3454 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3455 } else {
3456 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3457 ext_diff * sizeof(xfs_bmbt_rec_t));
3462 * This removes ext_diff extents from a linear (direct) extent list,
3463 * beginning at extent index idx. If the extents are being removed
3464 * from the end of the list (ie. truncate) then we just need to re-
3465 * allocate the list to remove the extra space. Otherwise, if the
3466 * extents are being removed from the middle of the existing extent
3467 * entries, then we first need to move the extent records beginning
3468 * at idx + ext_diff up in the list to overwrite the records being
3469 * removed, then remove the extra space via kmem_realloc.
3471 void
3472 xfs_iext_remove_direct(
3473 xfs_ifork_t *ifp, /* inode fork pointer */
3474 xfs_extnum_t idx, /* index to begin removing exts */
3475 int ext_diff) /* number of extents to remove */
3477 xfs_extnum_t nextents; /* number of extents in file */
3478 int new_size; /* size of extents after removal */
3480 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3481 new_size = ifp->if_bytes -
3482 (ext_diff * sizeof(xfs_bmbt_rec_t));
3483 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3485 if (new_size == 0) {
3486 xfs_iext_destroy(ifp);
3487 return;
3489 /* Move extents up in the list (if needed) */
3490 if (idx + ext_diff < nextents) {
3491 memmove(&ifp->if_u1.if_extents[idx],
3492 &ifp->if_u1.if_extents[idx + ext_diff],
3493 (nextents - (idx + ext_diff)) *
3494 sizeof(xfs_bmbt_rec_t));
3496 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3497 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3499 * Reallocate the direct extent list. If the extents
3500 * will fit inside the inode then xfs_iext_realloc_direct
3501 * will switch from direct to inline extent allocation
3502 * mode for us.
3504 xfs_iext_realloc_direct(ifp, new_size);
3505 ifp->if_bytes = new_size;
3509 * This is called when incore extents are being removed from the
3510 * indirection array and the extents being removed span multiple extent
3511 * buffers. The idx parameter contains the file extent index where we
3512 * want to begin removing extents, and the count parameter contains
3513 * how many extents need to be removed.
3515 * |-------| |-------|
3516 * | nex1 | | | nex1 - number of extents before idx
3517 * |-------| | count |
3518 * | | | | count - number of extents being removed at idx
3519 * | count | |-------|
3520 * | | | nex2 | nex2 - number of extents after idx + count
3521 * |-------| |-------|
3523 void
3524 xfs_iext_remove_indirect(
3525 xfs_ifork_t *ifp, /* inode fork pointer */
3526 xfs_extnum_t idx, /* index to begin removing extents */
3527 int count) /* number of extents to remove */
3529 xfs_ext_irec_t *erp; /* indirection array pointer */
3530 int erp_idx = 0; /* indirection array index */
3531 xfs_extnum_t ext_cnt; /* extents left to remove */
3532 xfs_extnum_t ext_diff; /* extents to remove in current list */
3533 xfs_extnum_t nex1; /* number of extents before idx */
3534 xfs_extnum_t nex2; /* extents after idx + count */
3535 int page_idx = idx; /* index in target extent list */
3537 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3538 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3539 ASSERT(erp != NULL);
3540 nex1 = page_idx;
3541 ext_cnt = count;
3542 while (ext_cnt) {
3543 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3544 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3546 * Check for deletion of entire list;
3547 * xfs_iext_irec_remove() updates extent offsets.
3549 if (ext_diff == erp->er_extcount) {
3550 xfs_iext_irec_remove(ifp, erp_idx);
3551 ext_cnt -= ext_diff;
3552 nex1 = 0;
3553 if (ext_cnt) {
3554 ASSERT(erp_idx < ifp->if_real_bytes /
3555 XFS_IEXT_BUFSZ);
3556 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3557 nex1 = 0;
3558 continue;
3559 } else {
3560 break;
3563 /* Move extents up (if needed) */
3564 if (nex2) {
3565 memmove(&erp->er_extbuf[nex1],
3566 &erp->er_extbuf[nex1 + ext_diff],
3567 nex2 * sizeof(xfs_bmbt_rec_t));
3569 /* Zero out rest of page */
3570 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3571 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3572 /* Update remaining counters */
3573 erp->er_extcount -= ext_diff;
3574 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3575 ext_cnt -= ext_diff;
3576 nex1 = 0;
3577 erp_idx++;
3578 erp++;
3580 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3581 xfs_iext_irec_compact(ifp);
3585 * Create, destroy, or resize a linear (direct) block of extents.
3587 void
3588 xfs_iext_realloc_direct(
3589 xfs_ifork_t *ifp, /* inode fork pointer */
3590 int new_size) /* new size of extents */
3592 int rnew_size; /* real new size of extents */
3594 rnew_size = new_size;
3596 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3597 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3598 (new_size != ifp->if_real_bytes)));
3600 /* Free extent records */
3601 if (new_size == 0) {
3602 xfs_iext_destroy(ifp);
3604 /* Resize direct extent list and zero any new bytes */
3605 else if (ifp->if_real_bytes) {
3606 /* Check if extents will fit inside the inode */
3607 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3608 xfs_iext_direct_to_inline(ifp, new_size /
3609 (uint)sizeof(xfs_bmbt_rec_t));
3610 ifp->if_bytes = new_size;
3611 return;
3613 if (!is_power_of_2(new_size)){
3614 rnew_size = roundup_pow_of_two(new_size);
3616 if (rnew_size != ifp->if_real_bytes) {
3617 ifp->if_u1.if_extents =
3618 kmem_realloc(ifp->if_u1.if_extents,
3619 rnew_size,
3620 ifp->if_real_bytes, KM_NOFS);
3622 if (rnew_size > ifp->if_real_bytes) {
3623 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3624 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3625 rnew_size - ifp->if_real_bytes);
3629 * Switch from the inline extent buffer to a direct
3630 * extent list. Be sure to include the inline extent
3631 * bytes in new_size.
3633 else {
3634 new_size += ifp->if_bytes;
3635 if (!is_power_of_2(new_size)) {
3636 rnew_size = roundup_pow_of_two(new_size);
3638 xfs_iext_inline_to_direct(ifp, rnew_size);
3640 ifp->if_real_bytes = rnew_size;
3641 ifp->if_bytes = new_size;
3645 * Switch from linear (direct) extent records to inline buffer.
3647 void
3648 xfs_iext_direct_to_inline(
3649 xfs_ifork_t *ifp, /* inode fork pointer */
3650 xfs_extnum_t nextents) /* number of extents in file */
3652 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3653 ASSERT(nextents <= XFS_INLINE_EXTS);
3655 * The inline buffer was zeroed when we switched
3656 * from inline to direct extent allocation mode,
3657 * so we don't need to clear it here.
3659 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3660 nextents * sizeof(xfs_bmbt_rec_t));
3661 kmem_free(ifp->if_u1.if_extents);
3662 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3663 ifp->if_real_bytes = 0;
3667 * Switch from inline buffer to linear (direct) extent records.
3668 * new_size should already be rounded up to the next power of 2
3669 * by the caller (when appropriate), so use new_size as it is.
3670 * However, since new_size may be rounded up, we can't update
3671 * if_bytes here. It is the caller's responsibility to update
3672 * if_bytes upon return.
3674 void
3675 xfs_iext_inline_to_direct(
3676 xfs_ifork_t *ifp, /* inode fork pointer */
3677 int new_size) /* number of extents in file */
3679 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3680 memset(ifp->if_u1.if_extents, 0, new_size);
3681 if (ifp->if_bytes) {
3682 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3683 ifp->if_bytes);
3684 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3685 sizeof(xfs_bmbt_rec_t));
3687 ifp->if_real_bytes = new_size;
3691 * Resize an extent indirection array to new_size bytes.
3693 STATIC void
3694 xfs_iext_realloc_indirect(
3695 xfs_ifork_t *ifp, /* inode fork pointer */
3696 int new_size) /* new indirection array size */
3698 int nlists; /* number of irec's (ex lists) */
3699 int size; /* current indirection array size */
3701 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3702 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3703 size = nlists * sizeof(xfs_ext_irec_t);
3704 ASSERT(ifp->if_real_bytes);
3705 ASSERT((new_size >= 0) && (new_size != size));
3706 if (new_size == 0) {
3707 xfs_iext_destroy(ifp);
3708 } else {
3709 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3710 kmem_realloc(ifp->if_u1.if_ext_irec,
3711 new_size, size, KM_NOFS);
3716 * Switch from indirection array to linear (direct) extent allocations.
3718 STATIC void
3719 xfs_iext_indirect_to_direct(
3720 xfs_ifork_t *ifp) /* inode fork pointer */
3722 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3723 xfs_extnum_t nextents; /* number of extents in file */
3724 int size; /* size of file extents */
3726 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3727 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3728 ASSERT(nextents <= XFS_LINEAR_EXTS);
3729 size = nextents * sizeof(xfs_bmbt_rec_t);
3731 xfs_iext_irec_compact_pages(ifp);
3732 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3734 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3735 kmem_free(ifp->if_u1.if_ext_irec);
3736 ifp->if_flags &= ~XFS_IFEXTIREC;
3737 ifp->if_u1.if_extents = ep;
3738 ifp->if_bytes = size;
3739 if (nextents < XFS_LINEAR_EXTS) {
3740 xfs_iext_realloc_direct(ifp, size);
3745 * Free incore file extents.
3747 void
3748 xfs_iext_destroy(
3749 xfs_ifork_t *ifp) /* inode fork pointer */
3751 if (ifp->if_flags & XFS_IFEXTIREC) {
3752 int erp_idx;
3753 int nlists;
3755 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3756 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3757 xfs_iext_irec_remove(ifp, erp_idx);
3759 ifp->if_flags &= ~XFS_IFEXTIREC;
3760 } else if (ifp->if_real_bytes) {
3761 kmem_free(ifp->if_u1.if_extents);
3762 } else if (ifp->if_bytes) {
3763 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3764 sizeof(xfs_bmbt_rec_t));
3766 ifp->if_u1.if_extents = NULL;
3767 ifp->if_real_bytes = 0;
3768 ifp->if_bytes = 0;
3772 * Return a pointer to the extent record for file system block bno.
3774 xfs_bmbt_rec_host_t * /* pointer to found extent record */
3775 xfs_iext_bno_to_ext(
3776 xfs_ifork_t *ifp, /* inode fork pointer */
3777 xfs_fileoff_t bno, /* block number to search for */
3778 xfs_extnum_t *idxp) /* index of target extent */
3780 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3781 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3782 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3783 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3784 int high; /* upper boundary in search */
3785 xfs_extnum_t idx = 0; /* index of target extent */
3786 int low; /* lower boundary in search */
3787 xfs_extnum_t nextents; /* number of file extents */
3788 xfs_fileoff_t startoff = 0; /* start offset of extent */
3790 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3791 if (nextents == 0) {
3792 *idxp = 0;
3793 return NULL;
3795 low = 0;
3796 if (ifp->if_flags & XFS_IFEXTIREC) {
3797 /* Find target extent list */
3798 int erp_idx = 0;
3799 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3800 base = erp->er_extbuf;
3801 high = erp->er_extcount - 1;
3802 } else {
3803 base = ifp->if_u1.if_extents;
3804 high = nextents - 1;
3806 /* Binary search extent records */
3807 while (low <= high) {
3808 idx = (low + high) >> 1;
3809 ep = base + idx;
3810 startoff = xfs_bmbt_get_startoff(ep);
3811 blockcount = xfs_bmbt_get_blockcount(ep);
3812 if (bno < startoff) {
3813 high = idx - 1;
3814 } else if (bno >= startoff + blockcount) {
3815 low = idx + 1;
3816 } else {
3817 /* Convert back to file-based extent index */
3818 if (ifp->if_flags & XFS_IFEXTIREC) {
3819 idx += erp->er_extoff;
3821 *idxp = idx;
3822 return ep;
3825 /* Convert back to file-based extent index */
3826 if (ifp->if_flags & XFS_IFEXTIREC) {
3827 idx += erp->er_extoff;
3829 if (bno >= startoff + blockcount) {
3830 if (++idx == nextents) {
3831 ep = NULL;
3832 } else {
3833 ep = xfs_iext_get_ext(ifp, idx);
3836 *idxp = idx;
3837 return ep;
3841 * Return a pointer to the indirection array entry containing the
3842 * extent record for filesystem block bno. Store the index of the
3843 * target irec in *erp_idxp.
3845 xfs_ext_irec_t * /* pointer to found extent record */
3846 xfs_iext_bno_to_irec(
3847 xfs_ifork_t *ifp, /* inode fork pointer */
3848 xfs_fileoff_t bno, /* block number to search for */
3849 int *erp_idxp) /* irec index of target ext list */
3851 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3852 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3853 int erp_idx; /* indirection array index */
3854 int nlists; /* number of extent irec's (lists) */
3855 int high; /* binary search upper limit */
3856 int low; /* binary search lower limit */
3858 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3859 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3860 erp_idx = 0;
3861 low = 0;
3862 high = nlists - 1;
3863 while (low <= high) {
3864 erp_idx = (low + high) >> 1;
3865 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3866 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3867 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3868 high = erp_idx - 1;
3869 } else if (erp_next && bno >=
3870 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3871 low = erp_idx + 1;
3872 } else {
3873 break;
3876 *erp_idxp = erp_idx;
3877 return erp;
3881 * Return a pointer to the indirection array entry containing the
3882 * extent record at file extent index *idxp. Store the index of the
3883 * target irec in *erp_idxp and store the page index of the target
3884 * extent record in *idxp.
3886 xfs_ext_irec_t *
3887 xfs_iext_idx_to_irec(
3888 xfs_ifork_t *ifp, /* inode fork pointer */
3889 xfs_extnum_t *idxp, /* extent index (file -> page) */
3890 int *erp_idxp, /* pointer to target irec */
3891 int realloc) /* new bytes were just added */
3893 xfs_ext_irec_t *prev; /* pointer to previous irec */
3894 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3895 int erp_idx; /* indirection array index */
3896 int nlists; /* number of irec's (ex lists) */
3897 int high; /* binary search upper limit */
3898 int low; /* binary search lower limit */
3899 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3901 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3902 ASSERT(page_idx >= 0 && page_idx <=
3903 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
3904 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3905 erp_idx = 0;
3906 low = 0;
3907 high = nlists - 1;
3909 /* Binary search extent irec's */
3910 while (low <= high) {
3911 erp_idx = (low + high) >> 1;
3912 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3913 prev = erp_idx > 0 ? erp - 1 : NULL;
3914 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3915 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3916 high = erp_idx - 1;
3917 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3918 (page_idx == erp->er_extoff + erp->er_extcount &&
3919 !realloc)) {
3920 low = erp_idx + 1;
3921 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3922 erp->er_extcount == XFS_LINEAR_EXTS) {
3923 ASSERT(realloc);
3924 page_idx = 0;
3925 erp_idx++;
3926 erp = erp_idx < nlists ? erp + 1 : NULL;
3927 break;
3928 } else {
3929 page_idx -= erp->er_extoff;
3930 break;
3933 *idxp = page_idx;
3934 *erp_idxp = erp_idx;
3935 return(erp);
3939 * Allocate and initialize an indirection array once the space needed
3940 * for incore extents increases above XFS_IEXT_BUFSZ.
3942 void
3943 xfs_iext_irec_init(
3944 xfs_ifork_t *ifp) /* inode fork pointer */
3946 xfs_ext_irec_t *erp; /* indirection array pointer */
3947 xfs_extnum_t nextents; /* number of extents in file */
3949 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3950 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3951 ASSERT(nextents <= XFS_LINEAR_EXTS);
3953 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3955 if (nextents == 0) {
3956 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3957 } else if (!ifp->if_real_bytes) {
3958 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3959 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3960 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3962 erp->er_extbuf = ifp->if_u1.if_extents;
3963 erp->er_extcount = nextents;
3964 erp->er_extoff = 0;
3966 ifp->if_flags |= XFS_IFEXTIREC;
3967 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3968 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3969 ifp->if_u1.if_ext_irec = erp;
3971 return;
3975 * Allocate and initialize a new entry in the indirection array.
3977 xfs_ext_irec_t *
3978 xfs_iext_irec_new(
3979 xfs_ifork_t *ifp, /* inode fork pointer */
3980 int erp_idx) /* index for new irec */
3982 xfs_ext_irec_t *erp; /* indirection array pointer */
3983 int i; /* loop counter */
3984 int nlists; /* number of irec's (ex lists) */
3986 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3987 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3989 /* Resize indirection array */
3990 xfs_iext_realloc_indirect(ifp, ++nlists *
3991 sizeof(xfs_ext_irec_t));
3993 * Move records down in the array so the
3994 * new page can use erp_idx.
3996 erp = ifp->if_u1.if_ext_irec;
3997 for (i = nlists - 1; i > erp_idx; i--) {
3998 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4000 ASSERT(i == erp_idx);
4002 /* Initialize new extent record */
4003 erp = ifp->if_u1.if_ext_irec;
4004 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4005 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4006 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4007 erp[erp_idx].er_extcount = 0;
4008 erp[erp_idx].er_extoff = erp_idx > 0 ?
4009 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4010 return (&erp[erp_idx]);
4014 * Remove a record from the indirection array.
4016 void
4017 xfs_iext_irec_remove(
4018 xfs_ifork_t *ifp, /* inode fork pointer */
4019 int erp_idx) /* irec index to remove */
4021 xfs_ext_irec_t *erp; /* indirection array pointer */
4022 int i; /* loop counter */
4023 int nlists; /* number of irec's (ex lists) */
4025 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4026 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4027 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4028 if (erp->er_extbuf) {
4029 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4030 -erp->er_extcount);
4031 kmem_free(erp->er_extbuf);
4033 /* Compact extent records */
4034 erp = ifp->if_u1.if_ext_irec;
4035 for (i = erp_idx; i < nlists - 1; i++) {
4036 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4039 * Manually free the last extent record from the indirection
4040 * array. A call to xfs_iext_realloc_indirect() with a size
4041 * of zero would result in a call to xfs_iext_destroy() which
4042 * would in turn call this function again, creating a nasty
4043 * infinite loop.
4045 if (--nlists) {
4046 xfs_iext_realloc_indirect(ifp,
4047 nlists * sizeof(xfs_ext_irec_t));
4048 } else {
4049 kmem_free(ifp->if_u1.if_ext_irec);
4051 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4055 * This is called to clean up large amounts of unused memory allocated
4056 * by the indirection array. Before compacting anything though, verify
4057 * that the indirection array is still needed and switch back to the
4058 * linear extent list (or even the inline buffer) if possible. The
4059 * compaction policy is as follows:
4061 * Full Compaction: Extents fit into a single page (or inline buffer)
4062 * Partial Compaction: Extents occupy less than 50% of allocated space
4063 * No Compaction: Extents occupy at least 50% of allocated space
4065 void
4066 xfs_iext_irec_compact(
4067 xfs_ifork_t *ifp) /* inode fork pointer */
4069 xfs_extnum_t nextents; /* number of extents in file */
4070 int nlists; /* number of irec's (ex lists) */
4072 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4073 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4074 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4076 if (nextents == 0) {
4077 xfs_iext_destroy(ifp);
4078 } else if (nextents <= XFS_INLINE_EXTS) {
4079 xfs_iext_indirect_to_direct(ifp);
4080 xfs_iext_direct_to_inline(ifp, nextents);
4081 } else if (nextents <= XFS_LINEAR_EXTS) {
4082 xfs_iext_indirect_to_direct(ifp);
4083 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4084 xfs_iext_irec_compact_pages(ifp);
4089 * Combine extents from neighboring extent pages.
4091 void
4092 xfs_iext_irec_compact_pages(
4093 xfs_ifork_t *ifp) /* inode fork pointer */
4095 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4096 int erp_idx = 0; /* indirection array index */
4097 int nlists; /* number of irec's (ex lists) */
4099 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4100 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4101 while (erp_idx < nlists - 1) {
4102 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4103 erp_next = erp + 1;
4104 if (erp_next->er_extcount <=
4105 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4106 memcpy(&erp->er_extbuf[erp->er_extcount],
4107 erp_next->er_extbuf, erp_next->er_extcount *
4108 sizeof(xfs_bmbt_rec_t));
4109 erp->er_extcount += erp_next->er_extcount;
4111 * Free page before removing extent record
4112 * so er_extoffs don't get modified in
4113 * xfs_iext_irec_remove.
4115 kmem_free(erp_next->er_extbuf);
4116 erp_next->er_extbuf = NULL;
4117 xfs_iext_irec_remove(ifp, erp_idx + 1);
4118 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4119 } else {
4120 erp_idx++;
4126 * This is called to update the er_extoff field in the indirection
4127 * array when extents have been added or removed from one of the
4128 * extent lists. erp_idx contains the irec index to begin updating
4129 * at and ext_diff contains the number of extents that were added
4130 * or removed.
4132 void
4133 xfs_iext_irec_update_extoffs(
4134 xfs_ifork_t *ifp, /* inode fork pointer */
4135 int erp_idx, /* irec index to update */
4136 int ext_diff) /* number of new extents */
4138 int i; /* loop counter */
4139 int nlists; /* number of irec's (ex lists */
4141 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4142 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4143 for (i = erp_idx; i < nlists; i++) {
4144 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;