Linux 2.6.28-rc5
[cris-mirror.git] / fs / xfs / xfs_inode.c
bloba391b955df0176969727b3768d08e6162216e96c
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_imap.h"
27 #include "xfs_trans.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_sb.h"
30 #include "xfs_ag.h"
31 #include "xfs_dir2.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_btree.h"
44 #include "xfs_alloc.h"
45 #include "xfs_ialloc.h"
46 #include "xfs_bmap.h"
47 #include "xfs_rw.h"
48 #include "xfs_error.h"
49 #include "xfs_utils.h"
50 #include "xfs_dir2_trace.h"
51 #include "xfs_quota.h"
52 #include "xfs_acl.h"
53 #include "xfs_filestream.h"
54 #include "xfs_vnodeops.h"
56 kmem_zone_t *xfs_ifork_zone;
57 kmem_zone_t *xfs_inode_zone;
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
63 #define XFS_ITRUNC_MAX_EXTENTS 2
65 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
70 #ifdef DEBUG
72 * Make sure that the extents in the given memory buffer
73 * are valid.
75 STATIC void
76 xfs_validate_extents(
77 xfs_ifork_t *ifp,
78 int nrecs,
79 xfs_exntfmt_t fmt)
81 xfs_bmbt_irec_t irec;
82 xfs_bmbt_rec_host_t rec;
83 int i;
85 for (i = 0; i < nrecs; i++) {
86 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
87 rec.l0 = get_unaligned(&ep->l0);
88 rec.l1 = get_unaligned(&ep->l1);
89 xfs_bmbt_get_all(&rec, &irec);
90 if (fmt == XFS_EXTFMT_NOSTATE)
91 ASSERT(irec.br_state == XFS_EXT_NORM);
94 #else /* DEBUG */
95 #define xfs_validate_extents(ifp, nrecs, fmt)
96 #endif /* DEBUG */
99 * Check that none of the inode's in the buffer have a next
100 * unlinked field of 0.
102 #if defined(DEBUG)
103 void
104 xfs_inobp_check(
105 xfs_mount_t *mp,
106 xfs_buf_t *bp)
108 int i;
109 int j;
110 xfs_dinode_t *dip;
112 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
114 for (i = 0; i < j; i++) {
115 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
116 i * mp->m_sb.sb_inodesize);
117 if (!dip->di_next_unlinked) {
118 xfs_fs_cmn_err(CE_ALERT, mp,
119 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 bp);
121 ASSERT(dip->di_next_unlinked);
125 #endif
128 * Find the buffer associated with the given inode map
129 * We do basic validation checks on the buffer once it has been
130 * retrieved from disk.
132 STATIC int
133 xfs_imap_to_bp(
134 xfs_mount_t *mp,
135 xfs_trans_t *tp,
136 xfs_imap_t *imap,
137 xfs_buf_t **bpp,
138 uint buf_flags,
139 uint imap_flags)
141 int error;
142 int i;
143 int ni;
144 xfs_buf_t *bp;
146 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
147 (int)imap->im_len, buf_flags, &bp);
148 if (error) {
149 if (error != EAGAIN) {
150 cmn_err(CE_WARN,
151 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
152 "an error %d on %s. Returning error.",
153 error, mp->m_fsname);
154 } else {
155 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
157 return error;
161 * Validate the magic number and version of every inode in the buffer
162 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
164 #ifdef DEBUG
165 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
166 #else /* usual case */
167 ni = 1;
168 #endif
170 for (i = 0; i < ni; i++) {
171 int di_ok;
172 xfs_dinode_t *dip;
174 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
175 (i << mp->m_sb.sb_inodelog));
176 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
177 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
178 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
181 if (imap_flags & XFS_IMAP_BULKSTAT) {
182 xfs_trans_brelse(tp, bp);
183 return XFS_ERROR(EINVAL);
185 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
186 XFS_ERRLEVEL_HIGH, mp, dip);
187 #ifdef DEBUG
188 cmn_err(CE_PANIC,
189 "Device %s - bad inode magic/vsn "
190 "daddr %lld #%d (magic=%x)",
191 XFS_BUFTARG_NAME(mp->m_ddev_targp),
192 (unsigned long long)imap->im_blkno, i,
193 be16_to_cpu(dip->di_core.di_magic));
194 #endif
195 xfs_trans_brelse(tp, bp);
196 return XFS_ERROR(EFSCORRUPTED);
200 xfs_inobp_check(mp, bp);
203 * Mark the buffer as an inode buffer now that it looks good
205 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
207 *bpp = bp;
208 return 0;
212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
224 STATIC int
225 xfs_inotobp(
226 xfs_mount_t *mp,
227 xfs_trans_t *tp,
228 xfs_ino_t ino,
229 xfs_dinode_t **dipp,
230 xfs_buf_t **bpp,
231 int *offset)
233 xfs_imap_t imap;
234 xfs_buf_t *bp;
235 int error;
237 imap.im_blkno = 0;
238 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
239 if (error)
240 return error;
242 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, 0);
243 if (error)
244 return error;
246 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
247 *bpp = bp;
248 *offset = imap.im_boffset;
249 return 0;
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
258 * that buffer.
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
263 * If the inode is new and has not yet been initialized, use xfs_imap()
264 * to determine the size and location of the buffer to read from disk.
265 * If the inode has already been mapped to its buffer and read in once,
266 * then use the mapping information stored in the inode rather than
267 * calling xfs_imap(). This allows us to avoid the overhead of looking
268 * at the inode btree for small block file systems (see xfs_dilocate()).
269 * We can tell whether the inode has been mapped in before by comparing
270 * its disk block address to 0. Only uninitialized inodes will have
271 * 0 for the disk block address.
274 xfs_itobp(
275 xfs_mount_t *mp,
276 xfs_trans_t *tp,
277 xfs_inode_t *ip,
278 xfs_dinode_t **dipp,
279 xfs_buf_t **bpp,
280 xfs_daddr_t bno,
281 uint imap_flags,
282 uint buf_flags)
284 xfs_imap_t imap;
285 xfs_buf_t *bp;
286 int error;
288 if (ip->i_blkno == (xfs_daddr_t)0) {
289 imap.im_blkno = bno;
290 error = xfs_imap(mp, tp, ip->i_ino, &imap,
291 XFS_IMAP_LOOKUP | imap_flags);
292 if (error)
293 return error;
296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
299 ip->i_blkno = imap.im_blkno;
300 ip->i_len = imap.im_len;
301 ip->i_boffset = imap.im_boffset;
302 } else {
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
307 imap.im_blkno = ip->i_blkno;
308 imap.im_len = ip->i_len;
309 imap.im_boffset = ip->i_boffset;
311 ASSERT(bno == 0 || bno == imap.im_blkno);
313 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
314 if (error)
315 return error;
317 if (!bp) {
318 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
319 ASSERT(tp == NULL);
320 *bpp = NULL;
321 return EAGAIN;
324 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
325 *bpp = bp;
326 return 0;
330 * Move inode type and inode format specific information from the
331 * on-disk inode to the in-core inode. For fifos, devs, and sockets
332 * this means set if_rdev to the proper value. For files, directories,
333 * and symlinks this means to bring in the in-line data or extent
334 * pointers. For a file in B-tree format, only the root is immediately
335 * brought in-core. The rest will be in-lined in if_extents when it
336 * is first referenced (see xfs_iread_extents()).
338 STATIC int
339 xfs_iformat(
340 xfs_inode_t *ip,
341 xfs_dinode_t *dip)
343 xfs_attr_shortform_t *atp;
344 int size;
345 int error;
346 xfs_fsize_t di_size;
347 ip->i_df.if_ext_max =
348 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
349 error = 0;
351 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
352 be16_to_cpu(dip->di_core.di_anextents) >
353 be64_to_cpu(dip->di_core.di_nblocks))) {
354 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
355 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
356 (unsigned long long)ip->i_ino,
357 (int)(be32_to_cpu(dip->di_core.di_nextents) +
358 be16_to_cpu(dip->di_core.di_anextents)),
359 (unsigned long long)
360 be64_to_cpu(dip->di_core.di_nblocks));
361 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
362 ip->i_mount, dip);
363 return XFS_ERROR(EFSCORRUPTED);
366 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
367 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
368 "corrupt dinode %Lu, forkoff = 0x%x.",
369 (unsigned long long)ip->i_ino,
370 dip->di_core.di_forkoff);
371 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
372 ip->i_mount, dip);
373 return XFS_ERROR(EFSCORRUPTED);
376 switch (ip->i_d.di_mode & S_IFMT) {
377 case S_IFIFO:
378 case S_IFCHR:
379 case S_IFBLK:
380 case S_IFSOCK:
381 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
382 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
383 ip->i_mount, dip);
384 return XFS_ERROR(EFSCORRUPTED);
386 ip->i_d.di_size = 0;
387 ip->i_size = 0;
388 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
389 break;
391 case S_IFREG:
392 case S_IFLNK:
393 case S_IFDIR:
394 switch (dip->di_core.di_format) {
395 case XFS_DINODE_FMT_LOCAL:
397 * no local regular files yet
399 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
400 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
401 "corrupt inode %Lu "
402 "(local format for regular file).",
403 (unsigned long long) ip->i_ino);
404 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
405 XFS_ERRLEVEL_LOW,
406 ip->i_mount, dip);
407 return XFS_ERROR(EFSCORRUPTED);
410 di_size = be64_to_cpu(dip->di_core.di_size);
411 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
412 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
413 "corrupt inode %Lu "
414 "(bad size %Ld for local inode).",
415 (unsigned long long) ip->i_ino,
416 (long long) di_size);
417 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
418 XFS_ERRLEVEL_LOW,
419 ip->i_mount, dip);
420 return XFS_ERROR(EFSCORRUPTED);
423 size = (int)di_size;
424 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
425 break;
426 case XFS_DINODE_FMT_EXTENTS:
427 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
428 break;
429 case XFS_DINODE_FMT_BTREE:
430 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
431 break;
432 default:
433 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
434 ip->i_mount);
435 return XFS_ERROR(EFSCORRUPTED);
437 break;
439 default:
440 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
441 return XFS_ERROR(EFSCORRUPTED);
443 if (error) {
444 return error;
446 if (!XFS_DFORK_Q(dip))
447 return 0;
448 ASSERT(ip->i_afp == NULL);
449 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
450 ip->i_afp->if_ext_max =
451 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
452 switch (dip->di_core.di_aformat) {
453 case XFS_DINODE_FMT_LOCAL:
454 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
455 size = be16_to_cpu(atp->hdr.totsize);
456 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
457 break;
458 case XFS_DINODE_FMT_EXTENTS:
459 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
460 break;
461 case XFS_DINODE_FMT_BTREE:
462 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
463 break;
464 default:
465 error = XFS_ERROR(EFSCORRUPTED);
466 break;
468 if (error) {
469 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
470 ip->i_afp = NULL;
471 xfs_idestroy_fork(ip, XFS_DATA_FORK);
473 return error;
477 * The file is in-lined in the on-disk inode.
478 * If it fits into if_inline_data, then copy
479 * it there, otherwise allocate a buffer for it
480 * and copy the data there. Either way, set
481 * if_data to point at the data.
482 * If we allocate a buffer for the data, make
483 * sure that its size is a multiple of 4 and
484 * record the real size in i_real_bytes.
486 STATIC int
487 xfs_iformat_local(
488 xfs_inode_t *ip,
489 xfs_dinode_t *dip,
490 int whichfork,
491 int size)
493 xfs_ifork_t *ifp;
494 int real_size;
497 * If the size is unreasonable, then something
498 * is wrong and we just bail out rather than crash in
499 * kmem_alloc() or memcpy() below.
501 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
502 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
503 "corrupt inode %Lu "
504 "(bad size %d for local fork, size = %d).",
505 (unsigned long long) ip->i_ino, size,
506 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
507 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
508 ip->i_mount, dip);
509 return XFS_ERROR(EFSCORRUPTED);
511 ifp = XFS_IFORK_PTR(ip, whichfork);
512 real_size = 0;
513 if (size == 0)
514 ifp->if_u1.if_data = NULL;
515 else if (size <= sizeof(ifp->if_u2.if_inline_data))
516 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
517 else {
518 real_size = roundup(size, 4);
519 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
521 ifp->if_bytes = size;
522 ifp->if_real_bytes = real_size;
523 if (size)
524 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
525 ifp->if_flags &= ~XFS_IFEXTENTS;
526 ifp->if_flags |= XFS_IFINLINE;
527 return 0;
531 * The file consists of a set of extents all
532 * of which fit into the on-disk inode.
533 * If there are few enough extents to fit into
534 * the if_inline_ext, then copy them there.
535 * Otherwise allocate a buffer for them and copy
536 * them into it. Either way, set if_extents
537 * to point at the extents.
539 STATIC int
540 xfs_iformat_extents(
541 xfs_inode_t *ip,
542 xfs_dinode_t *dip,
543 int whichfork)
545 xfs_bmbt_rec_t *dp;
546 xfs_ifork_t *ifp;
547 int nex;
548 int size;
549 int i;
551 ifp = XFS_IFORK_PTR(ip, whichfork);
552 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
553 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
556 * If the number of extents is unreasonable, then something
557 * is wrong and we just bail out rather than crash in
558 * kmem_alloc() or memcpy() below.
560 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
561 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
562 "corrupt inode %Lu ((a)extents = %d).",
563 (unsigned long long) ip->i_ino, nex);
564 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
565 ip->i_mount, dip);
566 return XFS_ERROR(EFSCORRUPTED);
569 ifp->if_real_bytes = 0;
570 if (nex == 0)
571 ifp->if_u1.if_extents = NULL;
572 else if (nex <= XFS_INLINE_EXTS)
573 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
574 else
575 xfs_iext_add(ifp, 0, nex);
577 ifp->if_bytes = size;
578 if (size) {
579 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
580 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
581 for (i = 0; i < nex; i++, dp++) {
582 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
583 ep->l0 = get_unaligned_be64(&dp->l0);
584 ep->l1 = get_unaligned_be64(&dp->l1);
586 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
587 if (whichfork != XFS_DATA_FORK ||
588 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
589 if (unlikely(xfs_check_nostate_extents(
590 ifp, 0, nex))) {
591 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
592 XFS_ERRLEVEL_LOW,
593 ip->i_mount);
594 return XFS_ERROR(EFSCORRUPTED);
597 ifp->if_flags |= XFS_IFEXTENTS;
598 return 0;
602 * The file has too many extents to fit into
603 * the inode, so they are in B-tree format.
604 * Allocate a buffer for the root of the B-tree
605 * and copy the root into it. The i_extents
606 * field will remain NULL until all of the
607 * extents are read in (when they are needed).
609 STATIC int
610 xfs_iformat_btree(
611 xfs_inode_t *ip,
612 xfs_dinode_t *dip,
613 int whichfork)
615 xfs_bmdr_block_t *dfp;
616 xfs_ifork_t *ifp;
617 /* REFERENCED */
618 int nrecs;
619 int size;
621 ifp = XFS_IFORK_PTR(ip, whichfork);
622 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
623 size = XFS_BMAP_BROOT_SPACE(dfp);
624 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
627 * blow out if -- fork has less extents than can fit in
628 * fork (fork shouldn't be a btree format), root btree
629 * block has more records than can fit into the fork,
630 * or the number of extents is greater than the number of
631 * blocks.
633 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
634 || XFS_BMDR_SPACE_CALC(nrecs) >
635 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
636 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
637 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
638 "corrupt inode %Lu (btree).",
639 (unsigned long long) ip->i_ino);
640 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
641 ip->i_mount);
642 return XFS_ERROR(EFSCORRUPTED);
645 ifp->if_broot_bytes = size;
646 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
647 ASSERT(ifp->if_broot != NULL);
649 * Copy and convert from the on-disk structure
650 * to the in-memory structure.
652 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
653 ifp->if_broot, size);
654 ifp->if_flags &= ~XFS_IFEXTENTS;
655 ifp->if_flags |= XFS_IFBROOT;
657 return 0;
660 void
661 xfs_dinode_from_disk(
662 xfs_icdinode_t *to,
663 xfs_dinode_core_t *from)
665 to->di_magic = be16_to_cpu(from->di_magic);
666 to->di_mode = be16_to_cpu(from->di_mode);
667 to->di_version = from ->di_version;
668 to->di_format = from->di_format;
669 to->di_onlink = be16_to_cpu(from->di_onlink);
670 to->di_uid = be32_to_cpu(from->di_uid);
671 to->di_gid = be32_to_cpu(from->di_gid);
672 to->di_nlink = be32_to_cpu(from->di_nlink);
673 to->di_projid = be16_to_cpu(from->di_projid);
674 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
675 to->di_flushiter = be16_to_cpu(from->di_flushiter);
676 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
677 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
678 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
679 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
680 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
681 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
682 to->di_size = be64_to_cpu(from->di_size);
683 to->di_nblocks = be64_to_cpu(from->di_nblocks);
684 to->di_extsize = be32_to_cpu(from->di_extsize);
685 to->di_nextents = be32_to_cpu(from->di_nextents);
686 to->di_anextents = be16_to_cpu(from->di_anextents);
687 to->di_forkoff = from->di_forkoff;
688 to->di_aformat = from->di_aformat;
689 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
690 to->di_dmstate = be16_to_cpu(from->di_dmstate);
691 to->di_flags = be16_to_cpu(from->di_flags);
692 to->di_gen = be32_to_cpu(from->di_gen);
695 void
696 xfs_dinode_to_disk(
697 xfs_dinode_core_t *to,
698 xfs_icdinode_t *from)
700 to->di_magic = cpu_to_be16(from->di_magic);
701 to->di_mode = cpu_to_be16(from->di_mode);
702 to->di_version = from ->di_version;
703 to->di_format = from->di_format;
704 to->di_onlink = cpu_to_be16(from->di_onlink);
705 to->di_uid = cpu_to_be32(from->di_uid);
706 to->di_gid = cpu_to_be32(from->di_gid);
707 to->di_nlink = cpu_to_be32(from->di_nlink);
708 to->di_projid = cpu_to_be16(from->di_projid);
709 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
710 to->di_flushiter = cpu_to_be16(from->di_flushiter);
711 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
712 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
713 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
714 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
715 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
716 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
717 to->di_size = cpu_to_be64(from->di_size);
718 to->di_nblocks = cpu_to_be64(from->di_nblocks);
719 to->di_extsize = cpu_to_be32(from->di_extsize);
720 to->di_nextents = cpu_to_be32(from->di_nextents);
721 to->di_anextents = cpu_to_be16(from->di_anextents);
722 to->di_forkoff = from->di_forkoff;
723 to->di_aformat = from->di_aformat;
724 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
725 to->di_dmstate = cpu_to_be16(from->di_dmstate);
726 to->di_flags = cpu_to_be16(from->di_flags);
727 to->di_gen = cpu_to_be32(from->di_gen);
730 STATIC uint
731 _xfs_dic2xflags(
732 __uint16_t di_flags)
734 uint flags = 0;
736 if (di_flags & XFS_DIFLAG_ANY) {
737 if (di_flags & XFS_DIFLAG_REALTIME)
738 flags |= XFS_XFLAG_REALTIME;
739 if (di_flags & XFS_DIFLAG_PREALLOC)
740 flags |= XFS_XFLAG_PREALLOC;
741 if (di_flags & XFS_DIFLAG_IMMUTABLE)
742 flags |= XFS_XFLAG_IMMUTABLE;
743 if (di_flags & XFS_DIFLAG_APPEND)
744 flags |= XFS_XFLAG_APPEND;
745 if (di_flags & XFS_DIFLAG_SYNC)
746 flags |= XFS_XFLAG_SYNC;
747 if (di_flags & XFS_DIFLAG_NOATIME)
748 flags |= XFS_XFLAG_NOATIME;
749 if (di_flags & XFS_DIFLAG_NODUMP)
750 flags |= XFS_XFLAG_NODUMP;
751 if (di_flags & XFS_DIFLAG_RTINHERIT)
752 flags |= XFS_XFLAG_RTINHERIT;
753 if (di_flags & XFS_DIFLAG_PROJINHERIT)
754 flags |= XFS_XFLAG_PROJINHERIT;
755 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
756 flags |= XFS_XFLAG_NOSYMLINKS;
757 if (di_flags & XFS_DIFLAG_EXTSIZE)
758 flags |= XFS_XFLAG_EXTSIZE;
759 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
760 flags |= XFS_XFLAG_EXTSZINHERIT;
761 if (di_flags & XFS_DIFLAG_NODEFRAG)
762 flags |= XFS_XFLAG_NODEFRAG;
763 if (di_flags & XFS_DIFLAG_FILESTREAM)
764 flags |= XFS_XFLAG_FILESTREAM;
767 return flags;
770 uint
771 xfs_ip2xflags(
772 xfs_inode_t *ip)
774 xfs_icdinode_t *dic = &ip->i_d;
776 return _xfs_dic2xflags(dic->di_flags) |
777 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
780 uint
781 xfs_dic2xflags(
782 xfs_dinode_t *dip)
784 xfs_dinode_core_t *dic = &dip->di_core;
786 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
787 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
791 * Given a mount structure and an inode number, return a pointer
792 * to a newly allocated in-core inode corresponding to the given
793 * inode number.
795 * Initialize the inode's attributes and extent pointers if it
796 * already has them (it will not if the inode has no links).
799 xfs_iread(
800 xfs_mount_t *mp,
801 xfs_trans_t *tp,
802 xfs_ino_t ino,
803 xfs_inode_t **ipp,
804 xfs_daddr_t bno,
805 uint imap_flags)
807 xfs_buf_t *bp;
808 xfs_dinode_t *dip;
809 xfs_inode_t *ip;
810 int error;
812 ASSERT(xfs_inode_zone != NULL);
814 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
815 ip->i_ino = ino;
816 ip->i_mount = mp;
817 atomic_set(&ip->i_iocount, 0);
818 spin_lock_init(&ip->i_flags_lock);
821 * Get pointer's to the on-disk inode and the buffer containing it.
822 * If the inode number refers to a block outside the file system
823 * then xfs_itobp() will return NULL. In this case we should
824 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
825 * know that this is a new incore inode.
827 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
828 if (error) {
829 kmem_zone_free(xfs_inode_zone, ip);
830 return error;
834 * Initialize inode's trace buffers.
835 * Do this before xfs_iformat in case it adds entries.
837 #ifdef XFS_INODE_TRACE
838 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
839 #endif
840 #ifdef XFS_BMAP_TRACE
841 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
842 #endif
843 #ifdef XFS_BMBT_TRACE
844 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
845 #endif
846 #ifdef XFS_RW_TRACE
847 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
848 #endif
849 #ifdef XFS_ILOCK_TRACE
850 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
851 #endif
852 #ifdef XFS_DIR2_TRACE
853 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
854 #endif
857 * If we got something that isn't an inode it means someone
858 * (nfs or dmi) has a stale handle.
860 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
861 kmem_zone_free(xfs_inode_zone, ip);
862 xfs_trans_brelse(tp, bp);
863 #ifdef DEBUG
864 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
865 "dip->di_core.di_magic (0x%x) != "
866 "XFS_DINODE_MAGIC (0x%x)",
867 be16_to_cpu(dip->di_core.di_magic),
868 XFS_DINODE_MAGIC);
869 #endif /* DEBUG */
870 return XFS_ERROR(EINVAL);
874 * If the on-disk inode is already linked to a directory
875 * entry, copy all of the inode into the in-core inode.
876 * xfs_iformat() handles copying in the inode format
877 * specific information.
878 * Otherwise, just get the truly permanent information.
880 if (dip->di_core.di_mode) {
881 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
882 error = xfs_iformat(ip, dip);
883 if (error) {
884 kmem_zone_free(xfs_inode_zone, ip);
885 xfs_trans_brelse(tp, bp);
886 #ifdef DEBUG
887 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
888 "xfs_iformat() returned error %d",
889 error);
890 #endif /* DEBUG */
891 return error;
893 } else {
894 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
895 ip->i_d.di_version = dip->di_core.di_version;
896 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
897 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
899 * Make sure to pull in the mode here as well in
900 * case the inode is released without being used.
901 * This ensures that xfs_inactive() will see that
902 * the inode is already free and not try to mess
903 * with the uninitialized part of it.
905 ip->i_d.di_mode = 0;
907 * Initialize the per-fork minima and maxima for a new
908 * inode here. xfs_iformat will do it for old inodes.
910 ip->i_df.if_ext_max =
911 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
914 INIT_LIST_HEAD(&ip->i_reclaim);
917 * The inode format changed when we moved the link count and
918 * made it 32 bits long. If this is an old format inode,
919 * convert it in memory to look like a new one. If it gets
920 * flushed to disk we will convert back before flushing or
921 * logging it. We zero out the new projid field and the old link
922 * count field. We'll handle clearing the pad field (the remains
923 * of the old uuid field) when we actually convert the inode to
924 * the new format. We don't change the version number so that we
925 * can distinguish this from a real new format inode.
927 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
928 ip->i_d.di_nlink = ip->i_d.di_onlink;
929 ip->i_d.di_onlink = 0;
930 ip->i_d.di_projid = 0;
933 ip->i_delayed_blks = 0;
934 ip->i_size = ip->i_d.di_size;
937 * Mark the buffer containing the inode as something to keep
938 * around for a while. This helps to keep recently accessed
939 * meta-data in-core longer.
941 XFS_BUF_SET_REF(bp, XFS_INO_REF);
944 * Use xfs_trans_brelse() to release the buffer containing the
945 * on-disk inode, because it was acquired with xfs_trans_read_buf()
946 * in xfs_itobp() above. If tp is NULL, this is just a normal
947 * brelse(). If we're within a transaction, then xfs_trans_brelse()
948 * will only release the buffer if it is not dirty within the
949 * transaction. It will be OK to release the buffer in this case,
950 * because inodes on disk are never destroyed and we will be
951 * locking the new in-core inode before putting it in the hash
952 * table where other processes can find it. Thus we don't have
953 * to worry about the inode being changed just because we released
954 * the buffer.
956 xfs_trans_brelse(tp, bp);
957 *ipp = ip;
958 return 0;
962 * Read in extents from a btree-format inode.
963 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
966 xfs_iread_extents(
967 xfs_trans_t *tp,
968 xfs_inode_t *ip,
969 int whichfork)
971 int error;
972 xfs_ifork_t *ifp;
973 xfs_extnum_t nextents;
974 size_t size;
976 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
977 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
978 ip->i_mount);
979 return XFS_ERROR(EFSCORRUPTED);
981 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
982 size = nextents * sizeof(xfs_bmbt_rec_t);
983 ifp = XFS_IFORK_PTR(ip, whichfork);
986 * We know that the size is valid (it's checked in iformat_btree)
988 ifp->if_lastex = NULLEXTNUM;
989 ifp->if_bytes = ifp->if_real_bytes = 0;
990 ifp->if_flags |= XFS_IFEXTENTS;
991 xfs_iext_add(ifp, 0, nextents);
992 error = xfs_bmap_read_extents(tp, ip, whichfork);
993 if (error) {
994 xfs_iext_destroy(ifp);
995 ifp->if_flags &= ~XFS_IFEXTENTS;
996 return error;
998 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
999 return 0;
1003 * Allocate an inode on disk and return a copy of its in-core version.
1004 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1005 * appropriately within the inode. The uid and gid for the inode are
1006 * set according to the contents of the given cred structure.
1008 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1009 * has a free inode available, call xfs_iget()
1010 * to obtain the in-core version of the allocated inode. Finally,
1011 * fill in the inode and log its initial contents. In this case,
1012 * ialloc_context would be set to NULL and call_again set to false.
1014 * If xfs_dialloc() does not have an available inode,
1015 * it will replenish its supply by doing an allocation. Since we can
1016 * only do one allocation within a transaction without deadlocks, we
1017 * must commit the current transaction before returning the inode itself.
1018 * In this case, therefore, we will set call_again to true and return.
1019 * The caller should then commit the current transaction, start a new
1020 * transaction, and call xfs_ialloc() again to actually get the inode.
1022 * To ensure that some other process does not grab the inode that
1023 * was allocated during the first call to xfs_ialloc(), this routine
1024 * also returns the [locked] bp pointing to the head of the freelist
1025 * as ialloc_context. The caller should hold this buffer across
1026 * the commit and pass it back into this routine on the second call.
1028 * If we are allocating quota inodes, we do not have a parent inode
1029 * to attach to or associate with (i.e. pip == NULL) because they
1030 * are not linked into the directory structure - they are attached
1031 * directly to the superblock - and so have no parent.
1034 xfs_ialloc(
1035 xfs_trans_t *tp,
1036 xfs_inode_t *pip,
1037 mode_t mode,
1038 xfs_nlink_t nlink,
1039 xfs_dev_t rdev,
1040 cred_t *cr,
1041 xfs_prid_t prid,
1042 int okalloc,
1043 xfs_buf_t **ialloc_context,
1044 boolean_t *call_again,
1045 xfs_inode_t **ipp)
1047 xfs_ino_t ino;
1048 xfs_inode_t *ip;
1049 uint flags;
1050 int error;
1051 timespec_t tv;
1054 * Call the space management code to pick
1055 * the on-disk inode to be allocated.
1057 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
1058 ialloc_context, call_again, &ino);
1059 if (error != 0) {
1060 return error;
1062 if (*call_again || ino == NULLFSINO) {
1063 *ipp = NULL;
1064 return 0;
1066 ASSERT(*ialloc_context == NULL);
1069 * Get the in-core inode with the lock held exclusively.
1070 * This is because we're setting fields here we need
1071 * to prevent others from looking at until we're done.
1073 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1074 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1075 if (error != 0) {
1076 return error;
1078 ASSERT(ip != NULL);
1080 ip->i_d.di_mode = (__uint16_t)mode;
1081 ip->i_d.di_onlink = 0;
1082 ip->i_d.di_nlink = nlink;
1083 ASSERT(ip->i_d.di_nlink == nlink);
1084 ip->i_d.di_uid = current_fsuid();
1085 ip->i_d.di_gid = current_fsgid();
1086 ip->i_d.di_projid = prid;
1087 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1090 * If the superblock version is up to where we support new format
1091 * inodes and this is currently an old format inode, then change
1092 * the inode version number now. This way we only do the conversion
1093 * here rather than here and in the flush/logging code.
1095 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1096 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1097 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1099 * We've already zeroed the old link count, the projid field,
1100 * and the pad field.
1105 * Project ids won't be stored on disk if we are using a version 1 inode.
1107 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1108 xfs_bump_ino_vers2(tp, ip);
1110 if (pip && XFS_INHERIT_GID(pip)) {
1111 ip->i_d.di_gid = pip->i_d.di_gid;
1112 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1113 ip->i_d.di_mode |= S_ISGID;
1118 * If the group ID of the new file does not match the effective group
1119 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1120 * (and only if the irix_sgid_inherit compatibility variable is set).
1122 if ((irix_sgid_inherit) &&
1123 (ip->i_d.di_mode & S_ISGID) &&
1124 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1125 ip->i_d.di_mode &= ~S_ISGID;
1128 ip->i_d.di_size = 0;
1129 ip->i_size = 0;
1130 ip->i_d.di_nextents = 0;
1131 ASSERT(ip->i_d.di_nblocks == 0);
1133 nanotime(&tv);
1134 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1135 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1136 ip->i_d.di_atime = ip->i_d.di_mtime;
1137 ip->i_d.di_ctime = ip->i_d.di_mtime;
1140 * di_gen will have been taken care of in xfs_iread.
1142 ip->i_d.di_extsize = 0;
1143 ip->i_d.di_dmevmask = 0;
1144 ip->i_d.di_dmstate = 0;
1145 ip->i_d.di_flags = 0;
1146 flags = XFS_ILOG_CORE;
1147 switch (mode & S_IFMT) {
1148 case S_IFIFO:
1149 case S_IFCHR:
1150 case S_IFBLK:
1151 case S_IFSOCK:
1152 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1153 ip->i_df.if_u2.if_rdev = rdev;
1154 ip->i_df.if_flags = 0;
1155 flags |= XFS_ILOG_DEV;
1156 break;
1157 case S_IFREG:
1158 if (pip && xfs_inode_is_filestream(pip)) {
1159 error = xfs_filestream_associate(pip, ip);
1160 if (error < 0)
1161 return -error;
1162 if (!error)
1163 xfs_iflags_set(ip, XFS_IFILESTREAM);
1165 /* fall through */
1166 case S_IFDIR:
1167 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1168 uint di_flags = 0;
1170 if ((mode & S_IFMT) == S_IFDIR) {
1171 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1172 di_flags |= XFS_DIFLAG_RTINHERIT;
1173 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1174 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1175 ip->i_d.di_extsize = pip->i_d.di_extsize;
1177 } else if ((mode & S_IFMT) == S_IFREG) {
1178 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1179 di_flags |= XFS_DIFLAG_REALTIME;
1180 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1181 di_flags |= XFS_DIFLAG_EXTSIZE;
1182 ip->i_d.di_extsize = pip->i_d.di_extsize;
1185 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1186 xfs_inherit_noatime)
1187 di_flags |= XFS_DIFLAG_NOATIME;
1188 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1189 xfs_inherit_nodump)
1190 di_flags |= XFS_DIFLAG_NODUMP;
1191 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1192 xfs_inherit_sync)
1193 di_flags |= XFS_DIFLAG_SYNC;
1194 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1195 xfs_inherit_nosymlinks)
1196 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1197 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1198 di_flags |= XFS_DIFLAG_PROJINHERIT;
1199 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1200 xfs_inherit_nodefrag)
1201 di_flags |= XFS_DIFLAG_NODEFRAG;
1202 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1203 di_flags |= XFS_DIFLAG_FILESTREAM;
1204 ip->i_d.di_flags |= di_flags;
1206 /* FALLTHROUGH */
1207 case S_IFLNK:
1208 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1209 ip->i_df.if_flags = XFS_IFEXTENTS;
1210 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1211 ip->i_df.if_u1.if_extents = NULL;
1212 break;
1213 default:
1214 ASSERT(0);
1217 * Attribute fork settings for new inode.
1219 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1220 ip->i_d.di_anextents = 0;
1223 * Log the new values stuffed into the inode.
1225 xfs_trans_log_inode(tp, ip, flags);
1227 /* now that we have an i_mode we can setup inode ops and unlock */
1228 xfs_setup_inode(ip);
1230 *ipp = ip;
1231 return 0;
1235 * Check to make sure that there are no blocks allocated to the
1236 * file beyond the size of the file. We don't check this for
1237 * files with fixed size extents or real time extents, but we
1238 * at least do it for regular files.
1240 #ifdef DEBUG
1241 void
1242 xfs_isize_check(
1243 xfs_mount_t *mp,
1244 xfs_inode_t *ip,
1245 xfs_fsize_t isize)
1247 xfs_fileoff_t map_first;
1248 int nimaps;
1249 xfs_bmbt_irec_t imaps[2];
1251 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1252 return;
1254 if (XFS_IS_REALTIME_INODE(ip))
1255 return;
1257 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1258 return;
1260 nimaps = 2;
1261 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1263 * The filesystem could be shutting down, so bmapi may return
1264 * an error.
1266 if (xfs_bmapi(NULL, ip, map_first,
1267 (XFS_B_TO_FSB(mp,
1268 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1269 map_first),
1270 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1271 NULL, NULL))
1272 return;
1273 ASSERT(nimaps == 1);
1274 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1276 #endif /* DEBUG */
1279 * Calculate the last possible buffered byte in a file. This must
1280 * include data that was buffered beyond the EOF by the write code.
1281 * This also needs to deal with overflowing the xfs_fsize_t type
1282 * which can happen for sizes near the limit.
1284 * We also need to take into account any blocks beyond the EOF. It
1285 * may be the case that they were buffered by a write which failed.
1286 * In that case the pages will still be in memory, but the inode size
1287 * will never have been updated.
1289 xfs_fsize_t
1290 xfs_file_last_byte(
1291 xfs_inode_t *ip)
1293 xfs_mount_t *mp;
1294 xfs_fsize_t last_byte;
1295 xfs_fileoff_t last_block;
1296 xfs_fileoff_t size_last_block;
1297 int error;
1299 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
1301 mp = ip->i_mount;
1303 * Only check for blocks beyond the EOF if the extents have
1304 * been read in. This eliminates the need for the inode lock,
1305 * and it also saves us from looking when it really isn't
1306 * necessary.
1308 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1309 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1310 XFS_DATA_FORK);
1311 if (error) {
1312 last_block = 0;
1314 } else {
1315 last_block = 0;
1317 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
1318 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1320 last_byte = XFS_FSB_TO_B(mp, last_block);
1321 if (last_byte < 0) {
1322 return XFS_MAXIOFFSET(mp);
1324 last_byte += (1 << mp->m_writeio_log);
1325 if (last_byte < 0) {
1326 return XFS_MAXIOFFSET(mp);
1328 return last_byte;
1331 #if defined(XFS_RW_TRACE)
1332 STATIC void
1333 xfs_itrunc_trace(
1334 int tag,
1335 xfs_inode_t *ip,
1336 int flag,
1337 xfs_fsize_t new_size,
1338 xfs_off_t toss_start,
1339 xfs_off_t toss_finish)
1341 if (ip->i_rwtrace == NULL) {
1342 return;
1345 ktrace_enter(ip->i_rwtrace,
1346 (void*)((long)tag),
1347 (void*)ip,
1348 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1349 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1350 (void*)((long)flag),
1351 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1352 (void*)(unsigned long)(new_size & 0xffffffff),
1353 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1354 (void*)(unsigned long)(toss_start & 0xffffffff),
1355 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1356 (void*)(unsigned long)(toss_finish & 0xffffffff),
1357 (void*)(unsigned long)current_cpu(),
1358 (void*)(unsigned long)current_pid(),
1359 (void*)NULL,
1360 (void*)NULL,
1361 (void*)NULL);
1363 #else
1364 #define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1365 #endif
1368 * Start the truncation of the file to new_size. The new size
1369 * must be smaller than the current size. This routine will
1370 * clear the buffer and page caches of file data in the removed
1371 * range, and xfs_itruncate_finish() will remove the underlying
1372 * disk blocks.
1374 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1375 * must NOT have the inode lock held at all. This is because we're
1376 * calling into the buffer/page cache code and we can't hold the
1377 * inode lock when we do so.
1379 * We need to wait for any direct I/Os in flight to complete before we
1380 * proceed with the truncate. This is needed to prevent the extents
1381 * being read or written by the direct I/Os from being removed while the
1382 * I/O is in flight as there is no other method of synchronising
1383 * direct I/O with the truncate operation. Also, because we hold
1384 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1385 * started until the truncate completes and drops the lock. Essentially,
1386 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1387 * between direct I/Os and the truncate operation.
1389 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1390 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1391 * in the case that the caller is locking things out of order and
1392 * may not be able to call xfs_itruncate_finish() with the inode lock
1393 * held without dropping the I/O lock. If the caller must drop the
1394 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1395 * must be called again with all the same restrictions as the initial
1396 * call.
1399 xfs_itruncate_start(
1400 xfs_inode_t *ip,
1401 uint flags,
1402 xfs_fsize_t new_size)
1404 xfs_fsize_t last_byte;
1405 xfs_off_t toss_start;
1406 xfs_mount_t *mp;
1407 int error = 0;
1409 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1410 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1411 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1412 (flags == XFS_ITRUNC_MAYBE));
1414 mp = ip->i_mount;
1416 /* wait for the completion of any pending DIOs */
1417 if (new_size == 0 || new_size < ip->i_size)
1418 vn_iowait(ip);
1421 * Call toss_pages or flushinval_pages to get rid of pages
1422 * overlapping the region being removed. We have to use
1423 * the less efficient flushinval_pages in the case that the
1424 * caller may not be able to finish the truncate without
1425 * dropping the inode's I/O lock. Make sure
1426 * to catch any pages brought in by buffers overlapping
1427 * the EOF by searching out beyond the isize by our
1428 * block size. We round new_size up to a block boundary
1429 * so that we don't toss things on the same block as
1430 * new_size but before it.
1432 * Before calling toss_page or flushinval_pages, make sure to
1433 * call remapf() over the same region if the file is mapped.
1434 * This frees up mapped file references to the pages in the
1435 * given range and for the flushinval_pages case it ensures
1436 * that we get the latest mapped changes flushed out.
1438 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1439 toss_start = XFS_FSB_TO_B(mp, toss_start);
1440 if (toss_start < 0) {
1442 * The place to start tossing is beyond our maximum
1443 * file size, so there is no way that the data extended
1444 * out there.
1446 return 0;
1448 last_byte = xfs_file_last_byte(ip);
1449 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1450 last_byte);
1451 if (last_byte > toss_start) {
1452 if (flags & XFS_ITRUNC_DEFINITE) {
1453 xfs_tosspages(ip, toss_start,
1454 -1, FI_REMAPF_LOCKED);
1455 } else {
1456 error = xfs_flushinval_pages(ip, toss_start,
1457 -1, FI_REMAPF_LOCKED);
1461 #ifdef DEBUG
1462 if (new_size == 0) {
1463 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
1465 #endif
1466 return error;
1470 * Shrink the file to the given new_size. The new size must be smaller than
1471 * the current size. This will free up the underlying blocks in the removed
1472 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
1474 * The transaction passed to this routine must have made a permanent log
1475 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1476 * given transaction and start new ones, so make sure everything involved in
1477 * the transaction is tidy before calling here. Some transaction will be
1478 * returned to the caller to be committed. The incoming transaction must
1479 * already include the inode, and both inode locks must be held exclusively.
1480 * The inode must also be "held" within the transaction. On return the inode
1481 * will be "held" within the returned transaction. This routine does NOT
1482 * require any disk space to be reserved for it within the transaction.
1484 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1485 * indicates the fork which is to be truncated. For the attribute fork we only
1486 * support truncation to size 0.
1488 * We use the sync parameter to indicate whether or not the first transaction
1489 * we perform might have to be synchronous. For the attr fork, it needs to be
1490 * so if the unlink of the inode is not yet known to be permanent in the log.
1491 * This keeps us from freeing and reusing the blocks of the attribute fork
1492 * before the unlink of the inode becomes permanent.
1494 * For the data fork, we normally have to run synchronously if we're being
1495 * called out of the inactive path or we're being called out of the create path
1496 * where we're truncating an existing file. Either way, the truncate needs to
1497 * be sync so blocks don't reappear in the file with altered data in case of a
1498 * crash. wsync filesystems can run the first case async because anything that
1499 * shrinks the inode has to run sync so by the time we're called here from
1500 * inactive, the inode size is permanently set to 0.
1502 * Calls from the truncate path always need to be sync unless we're in a wsync
1503 * filesystem and the file has already been unlinked.
1505 * The caller is responsible for correctly setting the sync parameter. It gets
1506 * too hard for us to guess here which path we're being called out of just
1507 * based on inode state.
1509 * If we get an error, we must return with the inode locked and linked into the
1510 * current transaction. This keeps things simple for the higher level code,
1511 * because it always knows that the inode is locked and held in the transaction
1512 * that returns to it whether errors occur or not. We don't mark the inode
1513 * dirty on error so that transactions can be easily aborted if possible.
1516 xfs_itruncate_finish(
1517 xfs_trans_t **tp,
1518 xfs_inode_t *ip,
1519 xfs_fsize_t new_size,
1520 int fork,
1521 int sync)
1523 xfs_fsblock_t first_block;
1524 xfs_fileoff_t first_unmap_block;
1525 xfs_fileoff_t last_block;
1526 xfs_filblks_t unmap_len=0;
1527 xfs_mount_t *mp;
1528 xfs_trans_t *ntp;
1529 int done;
1530 int committed;
1531 xfs_bmap_free_t free_list;
1532 int error;
1534 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1535 ASSERT((new_size == 0) || (new_size <= ip->i_size));
1536 ASSERT(*tp != NULL);
1537 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1538 ASSERT(ip->i_transp == *tp);
1539 ASSERT(ip->i_itemp != NULL);
1540 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1543 ntp = *tp;
1544 mp = (ntp)->t_mountp;
1545 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1548 * We only support truncating the entire attribute fork.
1550 if (fork == XFS_ATTR_FORK) {
1551 new_size = 0LL;
1553 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1554 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1556 * The first thing we do is set the size to new_size permanently
1557 * on disk. This way we don't have to worry about anyone ever
1558 * being able to look at the data being freed even in the face
1559 * of a crash. What we're getting around here is the case where
1560 * we free a block, it is allocated to another file, it is written
1561 * to, and then we crash. If the new data gets written to the
1562 * file but the log buffers containing the free and reallocation
1563 * don't, then we'd end up with garbage in the blocks being freed.
1564 * As long as we make the new_size permanent before actually
1565 * freeing any blocks it doesn't matter if they get writtten to.
1567 * The callers must signal into us whether or not the size
1568 * setting here must be synchronous. There are a few cases
1569 * where it doesn't have to be synchronous. Those cases
1570 * occur if the file is unlinked and we know the unlink is
1571 * permanent or if the blocks being truncated are guaranteed
1572 * to be beyond the inode eof (regardless of the link count)
1573 * and the eof value is permanent. Both of these cases occur
1574 * only on wsync-mounted filesystems. In those cases, we're
1575 * guaranteed that no user will ever see the data in the blocks
1576 * that are being truncated so the truncate can run async.
1577 * In the free beyond eof case, the file may wind up with
1578 * more blocks allocated to it than it needs if we crash
1579 * and that won't get fixed until the next time the file
1580 * is re-opened and closed but that's ok as that shouldn't
1581 * be too many blocks.
1583 * However, we can't just make all wsync xactions run async
1584 * because there's one call out of the create path that needs
1585 * to run sync where it's truncating an existing file to size
1586 * 0 whose size is > 0.
1588 * It's probably possible to come up with a test in this
1589 * routine that would correctly distinguish all the above
1590 * cases from the values of the function parameters and the
1591 * inode state but for sanity's sake, I've decided to let the
1592 * layers above just tell us. It's simpler to correctly figure
1593 * out in the layer above exactly under what conditions we
1594 * can run async and I think it's easier for others read and
1595 * follow the logic in case something has to be changed.
1596 * cscope is your friend -- rcc.
1598 * The attribute fork is much simpler.
1600 * For the attribute fork we allow the caller to tell us whether
1601 * the unlink of the inode that led to this call is yet permanent
1602 * in the on disk log. If it is not and we will be freeing extents
1603 * in this inode then we make the first transaction synchronous
1604 * to make sure that the unlink is permanent by the time we free
1605 * the blocks.
1607 if (fork == XFS_DATA_FORK) {
1608 if (ip->i_d.di_nextents > 0) {
1610 * If we are not changing the file size then do
1611 * not update the on-disk file size - we may be
1612 * called from xfs_inactive_free_eofblocks(). If we
1613 * update the on-disk file size and then the system
1614 * crashes before the contents of the file are
1615 * flushed to disk then the files may be full of
1616 * holes (ie NULL files bug).
1618 if (ip->i_size != new_size) {
1619 ip->i_d.di_size = new_size;
1620 ip->i_size = new_size;
1621 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1624 } else if (sync) {
1625 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1626 if (ip->i_d.di_anextents > 0)
1627 xfs_trans_set_sync(ntp);
1629 ASSERT(fork == XFS_DATA_FORK ||
1630 (fork == XFS_ATTR_FORK &&
1631 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1632 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1635 * Since it is possible for space to become allocated beyond
1636 * the end of the file (in a crash where the space is allocated
1637 * but the inode size is not yet updated), simply remove any
1638 * blocks which show up between the new EOF and the maximum
1639 * possible file size. If the first block to be removed is
1640 * beyond the maximum file size (ie it is the same as last_block),
1641 * then there is nothing to do.
1643 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1644 ASSERT(first_unmap_block <= last_block);
1645 done = 0;
1646 if (last_block == first_unmap_block) {
1647 done = 1;
1648 } else {
1649 unmap_len = last_block - first_unmap_block + 1;
1651 while (!done) {
1653 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1654 * will tell us whether it freed the entire range or
1655 * not. If this is a synchronous mount (wsync),
1656 * then we can tell bunmapi to keep all the
1657 * transactions asynchronous since the unlink
1658 * transaction that made this inode inactive has
1659 * already hit the disk. There's no danger of
1660 * the freed blocks being reused, there being a
1661 * crash, and the reused blocks suddenly reappearing
1662 * in this file with garbage in them once recovery
1663 * runs.
1665 XFS_BMAP_INIT(&free_list, &first_block);
1666 error = xfs_bunmapi(ntp, ip,
1667 first_unmap_block, unmap_len,
1668 XFS_BMAPI_AFLAG(fork) |
1669 (sync ? 0 : XFS_BMAPI_ASYNC),
1670 XFS_ITRUNC_MAX_EXTENTS,
1671 &first_block, &free_list,
1672 NULL, &done);
1673 if (error) {
1675 * If the bunmapi call encounters an error,
1676 * return to the caller where the transaction
1677 * can be properly aborted. We just need to
1678 * make sure we're not holding any resources
1679 * that we were not when we came in.
1681 xfs_bmap_cancel(&free_list);
1682 return error;
1686 * Duplicate the transaction that has the permanent
1687 * reservation and commit the old transaction.
1689 error = xfs_bmap_finish(tp, &free_list, &committed);
1690 ntp = *tp;
1691 if (committed) {
1692 /* link the inode into the next xact in the chain */
1693 xfs_trans_ijoin(ntp, ip,
1694 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1695 xfs_trans_ihold(ntp, ip);
1698 if (error) {
1700 * If the bmap finish call encounters an error, return
1701 * to the caller where the transaction can be properly
1702 * aborted. We just need to make sure we're not
1703 * holding any resources that we were not when we came
1704 * in.
1706 * Aborting from this point might lose some blocks in
1707 * the file system, but oh well.
1709 xfs_bmap_cancel(&free_list);
1710 return error;
1713 if (committed) {
1715 * Mark the inode dirty so it will be logged and
1716 * moved forward in the log as part of every commit.
1718 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1721 ntp = xfs_trans_dup(ntp);
1722 error = xfs_trans_commit(*tp, 0);
1723 *tp = ntp;
1725 /* link the inode into the next transaction in the chain */
1726 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1727 xfs_trans_ihold(ntp, ip);
1729 if (!error)
1730 error = xfs_trans_reserve(ntp, 0,
1731 XFS_ITRUNCATE_LOG_RES(mp), 0,
1732 XFS_TRANS_PERM_LOG_RES,
1733 XFS_ITRUNCATE_LOG_COUNT);
1734 if (error)
1735 return error;
1738 * Only update the size in the case of the data fork, but
1739 * always re-log the inode so that our permanent transaction
1740 * can keep on rolling it forward in the log.
1742 if (fork == XFS_DATA_FORK) {
1743 xfs_isize_check(mp, ip, new_size);
1745 * If we are not changing the file size then do
1746 * not update the on-disk file size - we may be
1747 * called from xfs_inactive_free_eofblocks(). If we
1748 * update the on-disk file size and then the system
1749 * crashes before the contents of the file are
1750 * flushed to disk then the files may be full of
1751 * holes (ie NULL files bug).
1753 if (ip->i_size != new_size) {
1754 ip->i_d.di_size = new_size;
1755 ip->i_size = new_size;
1758 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1759 ASSERT((new_size != 0) ||
1760 (fork == XFS_ATTR_FORK) ||
1761 (ip->i_delayed_blks == 0));
1762 ASSERT((new_size != 0) ||
1763 (fork == XFS_ATTR_FORK) ||
1764 (ip->i_d.di_nextents == 0));
1765 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1766 return 0;
1770 * This is called when the inode's link count goes to 0.
1771 * We place the on-disk inode on a list in the AGI. It
1772 * will be pulled from this list when the inode is freed.
1775 xfs_iunlink(
1776 xfs_trans_t *tp,
1777 xfs_inode_t *ip)
1779 xfs_mount_t *mp;
1780 xfs_agi_t *agi;
1781 xfs_dinode_t *dip;
1782 xfs_buf_t *agibp;
1783 xfs_buf_t *ibp;
1784 xfs_agnumber_t agno;
1785 xfs_daddr_t agdaddr;
1786 xfs_agino_t agino;
1787 short bucket_index;
1788 int offset;
1789 int error;
1790 int agi_ok;
1792 ASSERT(ip->i_d.di_nlink == 0);
1793 ASSERT(ip->i_d.di_mode != 0);
1794 ASSERT(ip->i_transp == tp);
1796 mp = tp->t_mountp;
1798 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1799 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1802 * Get the agi buffer first. It ensures lock ordering
1803 * on the list.
1805 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1806 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1807 if (error)
1808 return error;
1811 * Validate the magic number of the agi block.
1813 agi = XFS_BUF_TO_AGI(agibp);
1814 agi_ok =
1815 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1816 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1817 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1818 XFS_RANDOM_IUNLINK))) {
1819 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1820 xfs_trans_brelse(tp, agibp);
1821 return XFS_ERROR(EFSCORRUPTED);
1824 * Get the index into the agi hash table for the
1825 * list this inode will go on.
1827 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1828 ASSERT(agino != 0);
1829 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1830 ASSERT(agi->agi_unlinked[bucket_index]);
1831 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1833 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
1835 * There is already another inode in the bucket we need
1836 * to add ourselves to. Add us at the front of the list.
1837 * Here we put the head pointer into our next pointer,
1838 * and then we fall through to point the head at us.
1840 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1841 if (error)
1842 return error;
1844 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
1845 /* both on-disk, don't endian flip twice */
1846 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1847 offset = ip->i_boffset +
1848 offsetof(xfs_dinode_t, di_next_unlinked);
1849 xfs_trans_inode_buf(tp, ibp);
1850 xfs_trans_log_buf(tp, ibp, offset,
1851 (offset + sizeof(xfs_agino_t) - 1));
1852 xfs_inobp_check(mp, ibp);
1856 * Point the bucket head pointer at the inode being inserted.
1858 ASSERT(agino != 0);
1859 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1860 offset = offsetof(xfs_agi_t, agi_unlinked) +
1861 (sizeof(xfs_agino_t) * bucket_index);
1862 xfs_trans_log_buf(tp, agibp, offset,
1863 (offset + sizeof(xfs_agino_t) - 1));
1864 return 0;
1868 * Pull the on-disk inode from the AGI unlinked list.
1870 STATIC int
1871 xfs_iunlink_remove(
1872 xfs_trans_t *tp,
1873 xfs_inode_t *ip)
1875 xfs_ino_t next_ino;
1876 xfs_mount_t *mp;
1877 xfs_agi_t *agi;
1878 xfs_dinode_t *dip;
1879 xfs_buf_t *agibp;
1880 xfs_buf_t *ibp;
1881 xfs_agnumber_t agno;
1882 xfs_daddr_t agdaddr;
1883 xfs_agino_t agino;
1884 xfs_agino_t next_agino;
1885 xfs_buf_t *last_ibp;
1886 xfs_dinode_t *last_dip = NULL;
1887 short bucket_index;
1888 int offset, last_offset = 0;
1889 int error;
1890 int agi_ok;
1893 * First pull the on-disk inode from the AGI unlinked list.
1895 mp = tp->t_mountp;
1897 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1898 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1901 * Get the agi buffer first. It ensures lock ordering
1902 * on the list.
1904 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1905 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1906 if (error) {
1907 cmn_err(CE_WARN,
1908 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1909 error, mp->m_fsname);
1910 return error;
1913 * Validate the magic number of the agi block.
1915 agi = XFS_BUF_TO_AGI(agibp);
1916 agi_ok =
1917 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1918 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
1919 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1920 XFS_RANDOM_IUNLINK_REMOVE))) {
1921 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1922 mp, agi);
1923 xfs_trans_brelse(tp, agibp);
1924 cmn_err(CE_WARN,
1925 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1926 mp->m_fsname);
1927 return XFS_ERROR(EFSCORRUPTED);
1930 * Get the index into the agi hash table for the
1931 * list this inode will go on.
1933 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1934 ASSERT(agino != 0);
1935 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1936 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
1937 ASSERT(agi->agi_unlinked[bucket_index]);
1939 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1941 * We're at the head of the list. Get the inode's
1942 * on-disk buffer to see if there is anyone after us
1943 * on the list. Only modify our next pointer if it
1944 * is not already NULLAGINO. This saves us the overhead
1945 * of dealing with the buffer when there is no need to
1946 * change it.
1948 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
1949 if (error) {
1950 cmn_err(CE_WARN,
1951 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1952 error, mp->m_fsname);
1953 return error;
1955 next_agino = be32_to_cpu(dip->di_next_unlinked);
1956 ASSERT(next_agino != 0);
1957 if (next_agino != NULLAGINO) {
1958 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1959 offset = ip->i_boffset +
1960 offsetof(xfs_dinode_t, di_next_unlinked);
1961 xfs_trans_inode_buf(tp, ibp);
1962 xfs_trans_log_buf(tp, ibp, offset,
1963 (offset + sizeof(xfs_agino_t) - 1));
1964 xfs_inobp_check(mp, ibp);
1965 } else {
1966 xfs_trans_brelse(tp, ibp);
1969 * Point the bucket head pointer at the next inode.
1971 ASSERT(next_agino != 0);
1972 ASSERT(next_agino != agino);
1973 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1974 offset = offsetof(xfs_agi_t, agi_unlinked) +
1975 (sizeof(xfs_agino_t) * bucket_index);
1976 xfs_trans_log_buf(tp, agibp, offset,
1977 (offset + sizeof(xfs_agino_t) - 1));
1978 } else {
1980 * We need to search the list for the inode being freed.
1982 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1983 last_ibp = NULL;
1984 while (next_agino != agino) {
1986 * If the last inode wasn't the one pointing to
1987 * us, then release its buffer since we're not
1988 * going to do anything with it.
1990 if (last_ibp != NULL) {
1991 xfs_trans_brelse(tp, last_ibp);
1993 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1994 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1995 &last_ibp, &last_offset);
1996 if (error) {
1997 cmn_err(CE_WARN,
1998 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1999 error, mp->m_fsname);
2000 return error;
2002 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2003 ASSERT(next_agino != NULLAGINO);
2004 ASSERT(next_agino != 0);
2007 * Now last_ibp points to the buffer previous to us on
2008 * the unlinked list. Pull us from the list.
2010 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2011 if (error) {
2012 cmn_err(CE_WARN,
2013 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2014 error, mp->m_fsname);
2015 return error;
2017 next_agino = be32_to_cpu(dip->di_next_unlinked);
2018 ASSERT(next_agino != 0);
2019 ASSERT(next_agino != agino);
2020 if (next_agino != NULLAGINO) {
2021 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2022 offset = ip->i_boffset +
2023 offsetof(xfs_dinode_t, di_next_unlinked);
2024 xfs_trans_inode_buf(tp, ibp);
2025 xfs_trans_log_buf(tp, ibp, offset,
2026 (offset + sizeof(xfs_agino_t) - 1));
2027 xfs_inobp_check(mp, ibp);
2028 } else {
2029 xfs_trans_brelse(tp, ibp);
2032 * Point the previous inode on the list to the next inode.
2034 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2035 ASSERT(next_agino != 0);
2036 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2037 xfs_trans_inode_buf(tp, last_ibp);
2038 xfs_trans_log_buf(tp, last_ibp, offset,
2039 (offset + sizeof(xfs_agino_t) - 1));
2040 xfs_inobp_check(mp, last_ibp);
2042 return 0;
2045 STATIC void
2046 xfs_ifree_cluster(
2047 xfs_inode_t *free_ip,
2048 xfs_trans_t *tp,
2049 xfs_ino_t inum)
2051 xfs_mount_t *mp = free_ip->i_mount;
2052 int blks_per_cluster;
2053 int nbufs;
2054 int ninodes;
2055 int i, j, found, pre_flushed;
2056 xfs_daddr_t blkno;
2057 xfs_buf_t *bp;
2058 xfs_inode_t *ip, **ip_found;
2059 xfs_inode_log_item_t *iip;
2060 xfs_log_item_t *lip;
2061 xfs_perag_t *pag = xfs_get_perag(mp, inum);
2063 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2064 blks_per_cluster = 1;
2065 ninodes = mp->m_sb.sb_inopblock;
2066 nbufs = XFS_IALLOC_BLOCKS(mp);
2067 } else {
2068 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2069 mp->m_sb.sb_blocksize;
2070 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2071 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2074 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2076 for (j = 0; j < nbufs; j++, inum += ninodes) {
2077 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2078 XFS_INO_TO_AGBNO(mp, inum));
2082 * Look for each inode in memory and attempt to lock it,
2083 * we can be racing with flush and tail pushing here.
2084 * any inode we get the locks on, add to an array of
2085 * inode items to process later.
2087 * The get the buffer lock, we could beat a flush
2088 * or tail pushing thread to the lock here, in which
2089 * case they will go looking for the inode buffer
2090 * and fail, we need some other form of interlock
2091 * here.
2093 found = 0;
2094 for (i = 0; i < ninodes; i++) {
2095 read_lock(&pag->pag_ici_lock);
2096 ip = radix_tree_lookup(&pag->pag_ici_root,
2097 XFS_INO_TO_AGINO(mp, (inum + i)));
2099 /* Inode not in memory or we found it already,
2100 * nothing to do
2102 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
2103 read_unlock(&pag->pag_ici_lock);
2104 continue;
2107 if (xfs_inode_clean(ip)) {
2108 read_unlock(&pag->pag_ici_lock);
2109 continue;
2112 /* If we can get the locks then add it to the
2113 * list, otherwise by the time we get the bp lock
2114 * below it will already be attached to the
2115 * inode buffer.
2118 /* This inode will already be locked - by us, lets
2119 * keep it that way.
2122 if (ip == free_ip) {
2123 if (xfs_iflock_nowait(ip)) {
2124 xfs_iflags_set(ip, XFS_ISTALE);
2125 if (xfs_inode_clean(ip)) {
2126 xfs_ifunlock(ip);
2127 } else {
2128 ip_found[found++] = ip;
2131 read_unlock(&pag->pag_ici_lock);
2132 continue;
2135 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2136 if (xfs_iflock_nowait(ip)) {
2137 xfs_iflags_set(ip, XFS_ISTALE);
2139 if (xfs_inode_clean(ip)) {
2140 xfs_ifunlock(ip);
2141 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2142 } else {
2143 ip_found[found++] = ip;
2145 } else {
2146 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2149 read_unlock(&pag->pag_ici_lock);
2152 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2153 mp->m_bsize * blks_per_cluster,
2154 XFS_BUF_LOCK);
2156 pre_flushed = 0;
2157 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2158 while (lip) {
2159 if (lip->li_type == XFS_LI_INODE) {
2160 iip = (xfs_inode_log_item_t *)lip;
2161 ASSERT(iip->ili_logged == 1);
2162 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2163 spin_lock(&mp->m_ail_lock);
2164 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2165 spin_unlock(&mp->m_ail_lock);
2166 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2167 pre_flushed++;
2169 lip = lip->li_bio_list;
2172 for (i = 0; i < found; i++) {
2173 ip = ip_found[i];
2174 iip = ip->i_itemp;
2176 if (!iip) {
2177 ip->i_update_core = 0;
2178 xfs_ifunlock(ip);
2179 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2180 continue;
2183 iip->ili_last_fields = iip->ili_format.ilf_fields;
2184 iip->ili_format.ilf_fields = 0;
2185 iip->ili_logged = 1;
2186 spin_lock(&mp->m_ail_lock);
2187 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2188 spin_unlock(&mp->m_ail_lock);
2190 xfs_buf_attach_iodone(bp,
2191 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2192 xfs_istale_done, (xfs_log_item_t *)iip);
2193 if (ip != free_ip) {
2194 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2198 if (found || pre_flushed)
2199 xfs_trans_stale_inode_buf(tp, bp);
2200 xfs_trans_binval(tp, bp);
2203 kmem_free(ip_found);
2204 xfs_put_perag(mp, pag);
2208 * This is called to return an inode to the inode free list.
2209 * The inode should already be truncated to 0 length and have
2210 * no pages associated with it. This routine also assumes that
2211 * the inode is already a part of the transaction.
2213 * The on-disk copy of the inode will have been added to the list
2214 * of unlinked inodes in the AGI. We need to remove the inode from
2215 * that list atomically with respect to freeing it here.
2218 xfs_ifree(
2219 xfs_trans_t *tp,
2220 xfs_inode_t *ip,
2221 xfs_bmap_free_t *flist)
2223 int error;
2224 int delete;
2225 xfs_ino_t first_ino;
2226 xfs_dinode_t *dip;
2227 xfs_buf_t *ibp;
2229 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2230 ASSERT(ip->i_transp == tp);
2231 ASSERT(ip->i_d.di_nlink == 0);
2232 ASSERT(ip->i_d.di_nextents == 0);
2233 ASSERT(ip->i_d.di_anextents == 0);
2234 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
2235 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2236 ASSERT(ip->i_d.di_nblocks == 0);
2239 * Pull the on-disk inode from the AGI unlinked list.
2241 error = xfs_iunlink_remove(tp, ip);
2242 if (error != 0) {
2243 return error;
2246 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2247 if (error != 0) {
2248 return error;
2250 ip->i_d.di_mode = 0; /* mark incore inode as free */
2251 ip->i_d.di_flags = 0;
2252 ip->i_d.di_dmevmask = 0;
2253 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2254 ip->i_df.if_ext_max =
2255 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2256 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2257 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2259 * Bump the generation count so no one will be confused
2260 * by reincarnations of this inode.
2262 ip->i_d.di_gen++;
2264 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2266 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
2267 if (error)
2268 return error;
2271 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2272 * from picking up this inode when it is reclaimed (its incore state
2273 * initialzed but not flushed to disk yet). The in-core di_mode is
2274 * already cleared and a corresponding transaction logged.
2275 * The hack here just synchronizes the in-core to on-disk
2276 * di_mode value in advance before the actual inode sync to disk.
2277 * This is OK because the inode is already unlinked and would never
2278 * change its di_mode again for this inode generation.
2279 * This is a temporary hack that would require a proper fix
2280 * in the future.
2282 dip->di_core.di_mode = 0;
2284 if (delete) {
2285 xfs_ifree_cluster(ip, tp, first_ino);
2288 return 0;
2292 * Reallocate the space for if_broot based on the number of records
2293 * being added or deleted as indicated in rec_diff. Move the records
2294 * and pointers in if_broot to fit the new size. When shrinking this
2295 * will eliminate holes between the records and pointers created by
2296 * the caller. When growing this will create holes to be filled in
2297 * by the caller.
2299 * The caller must not request to add more records than would fit in
2300 * the on-disk inode root. If the if_broot is currently NULL, then
2301 * if we adding records one will be allocated. The caller must also
2302 * not request that the number of records go below zero, although
2303 * it can go to zero.
2305 * ip -- the inode whose if_broot area is changing
2306 * ext_diff -- the change in the number of records, positive or negative,
2307 * requested for the if_broot array.
2309 void
2310 xfs_iroot_realloc(
2311 xfs_inode_t *ip,
2312 int rec_diff,
2313 int whichfork)
2315 int cur_max;
2316 xfs_ifork_t *ifp;
2317 xfs_bmbt_block_t *new_broot;
2318 int new_max;
2319 size_t new_size;
2320 char *np;
2321 char *op;
2324 * Handle the degenerate case quietly.
2326 if (rec_diff == 0) {
2327 return;
2330 ifp = XFS_IFORK_PTR(ip, whichfork);
2331 if (rec_diff > 0) {
2333 * If there wasn't any memory allocated before, just
2334 * allocate it now and get out.
2336 if (ifp->if_broot_bytes == 0) {
2337 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2338 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2339 KM_SLEEP);
2340 ifp->if_broot_bytes = (int)new_size;
2341 return;
2345 * If there is already an existing if_broot, then we need
2346 * to realloc() it and shift the pointers to their new
2347 * location. The records don't change location because
2348 * they are kept butted up against the btree block header.
2350 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2351 new_max = cur_max + rec_diff;
2352 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2353 ifp->if_broot = (xfs_bmbt_block_t *)
2354 kmem_realloc(ifp->if_broot,
2355 new_size,
2356 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2357 KM_SLEEP);
2358 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2359 ifp->if_broot_bytes);
2360 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2361 (int)new_size);
2362 ifp->if_broot_bytes = (int)new_size;
2363 ASSERT(ifp->if_broot_bytes <=
2364 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2365 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2366 return;
2370 * rec_diff is less than 0. In this case, we are shrinking the
2371 * if_broot buffer. It must already exist. If we go to zero
2372 * records, just get rid of the root and clear the status bit.
2374 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2375 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2376 new_max = cur_max + rec_diff;
2377 ASSERT(new_max >= 0);
2378 if (new_max > 0)
2379 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2380 else
2381 new_size = 0;
2382 if (new_size > 0) {
2383 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2385 * First copy over the btree block header.
2387 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2388 } else {
2389 new_broot = NULL;
2390 ifp->if_flags &= ~XFS_IFBROOT;
2394 * Only copy the records and pointers if there are any.
2396 if (new_max > 0) {
2398 * First copy the records.
2400 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2401 ifp->if_broot_bytes);
2402 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2403 (int)new_size);
2404 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2407 * Then copy the pointers.
2409 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2410 ifp->if_broot_bytes);
2411 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2412 (int)new_size);
2413 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2415 kmem_free(ifp->if_broot);
2416 ifp->if_broot = new_broot;
2417 ifp->if_broot_bytes = (int)new_size;
2418 ASSERT(ifp->if_broot_bytes <=
2419 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2420 return;
2425 * This is called when the amount of space needed for if_data
2426 * is increased or decreased. The change in size is indicated by
2427 * the number of bytes that need to be added or deleted in the
2428 * byte_diff parameter.
2430 * If the amount of space needed has decreased below the size of the
2431 * inline buffer, then switch to using the inline buffer. Otherwise,
2432 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2433 * to what is needed.
2435 * ip -- the inode whose if_data area is changing
2436 * byte_diff -- the change in the number of bytes, positive or negative,
2437 * requested for the if_data array.
2439 void
2440 xfs_idata_realloc(
2441 xfs_inode_t *ip,
2442 int byte_diff,
2443 int whichfork)
2445 xfs_ifork_t *ifp;
2446 int new_size;
2447 int real_size;
2449 if (byte_diff == 0) {
2450 return;
2453 ifp = XFS_IFORK_PTR(ip, whichfork);
2454 new_size = (int)ifp->if_bytes + byte_diff;
2455 ASSERT(new_size >= 0);
2457 if (new_size == 0) {
2458 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2459 kmem_free(ifp->if_u1.if_data);
2461 ifp->if_u1.if_data = NULL;
2462 real_size = 0;
2463 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2465 * If the valid extents/data can fit in if_inline_ext/data,
2466 * copy them from the malloc'd vector and free it.
2468 if (ifp->if_u1.if_data == NULL) {
2469 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2470 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2471 ASSERT(ifp->if_real_bytes != 0);
2472 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2473 new_size);
2474 kmem_free(ifp->if_u1.if_data);
2475 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2477 real_size = 0;
2478 } else {
2480 * Stuck with malloc/realloc.
2481 * For inline data, the underlying buffer must be
2482 * a multiple of 4 bytes in size so that it can be
2483 * logged and stay on word boundaries. We enforce
2484 * that here.
2486 real_size = roundup(new_size, 4);
2487 if (ifp->if_u1.if_data == NULL) {
2488 ASSERT(ifp->if_real_bytes == 0);
2489 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2490 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2492 * Only do the realloc if the underlying size
2493 * is really changing.
2495 if (ifp->if_real_bytes != real_size) {
2496 ifp->if_u1.if_data =
2497 kmem_realloc(ifp->if_u1.if_data,
2498 real_size,
2499 ifp->if_real_bytes,
2500 KM_SLEEP);
2502 } else {
2503 ASSERT(ifp->if_real_bytes == 0);
2504 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2505 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2506 ifp->if_bytes);
2509 ifp->if_real_bytes = real_size;
2510 ifp->if_bytes = new_size;
2511 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2518 * Map inode to disk block and offset.
2520 * mp -- the mount point structure for the current file system
2521 * tp -- the current transaction
2522 * ino -- the inode number of the inode to be located
2523 * imap -- this structure is filled in with the information necessary
2524 * to retrieve the given inode from disk
2525 * flags -- flags to pass to xfs_dilocate indicating whether or not
2526 * lookups in the inode btree were OK or not
2529 xfs_imap(
2530 xfs_mount_t *mp,
2531 xfs_trans_t *tp,
2532 xfs_ino_t ino,
2533 xfs_imap_t *imap,
2534 uint flags)
2536 xfs_fsblock_t fsbno;
2537 int len;
2538 int off;
2539 int error;
2541 fsbno = imap->im_blkno ?
2542 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2543 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2544 if (error)
2545 return error;
2547 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2548 imap->im_len = XFS_FSB_TO_BB(mp, len);
2549 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2550 imap->im_ioffset = (ushort)off;
2551 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2554 * If the inode number maps to a block outside the bounds
2555 * of the file system then return NULL rather than calling
2556 * read_buf and panicing when we get an error from the
2557 * driver.
2559 if ((imap->im_blkno + imap->im_len) >
2560 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2561 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2562 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2563 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2564 (unsigned long long) imap->im_blkno,
2565 (unsigned long long) imap->im_len,
2566 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2567 return EINVAL;
2569 return 0;
2572 void
2573 xfs_idestroy_fork(
2574 xfs_inode_t *ip,
2575 int whichfork)
2577 xfs_ifork_t *ifp;
2579 ifp = XFS_IFORK_PTR(ip, whichfork);
2580 if (ifp->if_broot != NULL) {
2581 kmem_free(ifp->if_broot);
2582 ifp->if_broot = NULL;
2586 * If the format is local, then we can't have an extents
2587 * array so just look for an inline data array. If we're
2588 * not local then we may or may not have an extents list,
2589 * so check and free it up if we do.
2591 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2592 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2593 (ifp->if_u1.if_data != NULL)) {
2594 ASSERT(ifp->if_real_bytes != 0);
2595 kmem_free(ifp->if_u1.if_data);
2596 ifp->if_u1.if_data = NULL;
2597 ifp->if_real_bytes = 0;
2599 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2600 ((ifp->if_flags & XFS_IFEXTIREC) ||
2601 ((ifp->if_u1.if_extents != NULL) &&
2602 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2603 ASSERT(ifp->if_real_bytes != 0);
2604 xfs_iext_destroy(ifp);
2606 ASSERT(ifp->if_u1.if_extents == NULL ||
2607 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2608 ASSERT(ifp->if_real_bytes == 0);
2609 if (whichfork == XFS_ATTR_FORK) {
2610 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2611 ip->i_afp = NULL;
2616 * This is called free all the memory associated with an inode.
2617 * It must free the inode itself and any buffers allocated for
2618 * if_extents/if_data and if_broot. It must also free the lock
2619 * associated with the inode.
2621 void
2622 xfs_idestroy(
2623 xfs_inode_t *ip)
2625 switch (ip->i_d.di_mode & S_IFMT) {
2626 case S_IFREG:
2627 case S_IFDIR:
2628 case S_IFLNK:
2629 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2630 break;
2632 if (ip->i_afp)
2633 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2634 mrfree(&ip->i_lock);
2635 mrfree(&ip->i_iolock);
2637 #ifdef XFS_INODE_TRACE
2638 ktrace_free(ip->i_trace);
2639 #endif
2640 #ifdef XFS_BMAP_TRACE
2641 ktrace_free(ip->i_xtrace);
2642 #endif
2643 #ifdef XFS_BMBT_TRACE
2644 ktrace_free(ip->i_btrace);
2645 #endif
2646 #ifdef XFS_RW_TRACE
2647 ktrace_free(ip->i_rwtrace);
2648 #endif
2649 #ifdef XFS_ILOCK_TRACE
2650 ktrace_free(ip->i_lock_trace);
2651 #endif
2652 #ifdef XFS_DIR2_TRACE
2653 ktrace_free(ip->i_dir_trace);
2654 #endif
2655 if (ip->i_itemp) {
2657 * Only if we are shutting down the fs will we see an
2658 * inode still in the AIL. If it is there, we should remove
2659 * it to prevent a use-after-free from occurring.
2661 xfs_mount_t *mp = ip->i_mount;
2662 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
2664 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2665 XFS_FORCED_SHUTDOWN(ip->i_mount));
2666 if (lip->li_flags & XFS_LI_IN_AIL) {
2667 spin_lock(&mp->m_ail_lock);
2668 if (lip->li_flags & XFS_LI_IN_AIL)
2669 xfs_trans_delete_ail(mp, lip);
2670 else
2671 spin_unlock(&mp->m_ail_lock);
2673 xfs_inode_item_destroy(ip);
2675 kmem_zone_free(xfs_inode_zone, ip);
2680 * Increment the pin count of the given buffer.
2681 * This value is protected by ipinlock spinlock in the mount structure.
2683 void
2684 xfs_ipin(
2685 xfs_inode_t *ip)
2687 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2689 atomic_inc(&ip->i_pincount);
2693 * Decrement the pin count of the given inode, and wake up
2694 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2695 * inode must have been previously pinned with a call to xfs_ipin().
2697 void
2698 xfs_iunpin(
2699 xfs_inode_t *ip)
2701 ASSERT(atomic_read(&ip->i_pincount) > 0);
2703 if (atomic_dec_and_test(&ip->i_pincount))
2704 wake_up(&ip->i_ipin_wait);
2708 * This is called to unpin an inode. It can be directed to wait or to return
2709 * immediately without waiting for the inode to be unpinned. The caller must
2710 * have the inode locked in at least shared mode so that the buffer cannot be
2711 * subsequently pinned once someone is waiting for it to be unpinned.
2713 STATIC void
2714 __xfs_iunpin_wait(
2715 xfs_inode_t *ip,
2716 int wait)
2718 xfs_inode_log_item_t *iip = ip->i_itemp;
2720 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2721 if (atomic_read(&ip->i_pincount) == 0)
2722 return;
2724 /* Give the log a push to start the unpinning I/O */
2725 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2726 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2727 if (wait)
2728 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2731 static inline void
2732 xfs_iunpin_wait(
2733 xfs_inode_t *ip)
2735 __xfs_iunpin_wait(ip, 1);
2738 static inline void
2739 xfs_iunpin_nowait(
2740 xfs_inode_t *ip)
2742 __xfs_iunpin_wait(ip, 0);
2747 * xfs_iextents_copy()
2749 * This is called to copy the REAL extents (as opposed to the delayed
2750 * allocation extents) from the inode into the given buffer. It
2751 * returns the number of bytes copied into the buffer.
2753 * If there are no delayed allocation extents, then we can just
2754 * memcpy() the extents into the buffer. Otherwise, we need to
2755 * examine each extent in turn and skip those which are delayed.
2758 xfs_iextents_copy(
2759 xfs_inode_t *ip,
2760 xfs_bmbt_rec_t *dp,
2761 int whichfork)
2763 int copied;
2764 int i;
2765 xfs_ifork_t *ifp;
2766 int nrecs;
2767 xfs_fsblock_t start_block;
2769 ifp = XFS_IFORK_PTR(ip, whichfork);
2770 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2771 ASSERT(ifp->if_bytes > 0);
2773 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2774 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2775 ASSERT(nrecs > 0);
2778 * There are some delayed allocation extents in the
2779 * inode, so copy the extents one at a time and skip
2780 * the delayed ones. There must be at least one
2781 * non-delayed extent.
2783 copied = 0;
2784 for (i = 0; i < nrecs; i++) {
2785 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2786 start_block = xfs_bmbt_get_startblock(ep);
2787 if (ISNULLSTARTBLOCK(start_block)) {
2789 * It's a delayed allocation extent, so skip it.
2791 continue;
2794 /* Translate to on disk format */
2795 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2796 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2797 dp++;
2798 copied++;
2800 ASSERT(copied != 0);
2801 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2803 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2807 * Each of the following cases stores data into the same region
2808 * of the on-disk inode, so only one of them can be valid at
2809 * any given time. While it is possible to have conflicting formats
2810 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2811 * in EXTENTS format, this can only happen when the fork has
2812 * changed formats after being modified but before being flushed.
2813 * In these cases, the format always takes precedence, because the
2814 * format indicates the current state of the fork.
2816 /*ARGSUSED*/
2817 STATIC void
2818 xfs_iflush_fork(
2819 xfs_inode_t *ip,
2820 xfs_dinode_t *dip,
2821 xfs_inode_log_item_t *iip,
2822 int whichfork,
2823 xfs_buf_t *bp)
2825 char *cp;
2826 xfs_ifork_t *ifp;
2827 xfs_mount_t *mp;
2828 #ifdef XFS_TRANS_DEBUG
2829 int first;
2830 #endif
2831 static const short brootflag[2] =
2832 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2833 static const short dataflag[2] =
2834 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2835 static const short extflag[2] =
2836 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2838 if (!iip)
2839 return;
2840 ifp = XFS_IFORK_PTR(ip, whichfork);
2842 * This can happen if we gave up in iformat in an error path,
2843 * for the attribute fork.
2845 if (!ifp) {
2846 ASSERT(whichfork == XFS_ATTR_FORK);
2847 return;
2849 cp = XFS_DFORK_PTR(dip, whichfork);
2850 mp = ip->i_mount;
2851 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2852 case XFS_DINODE_FMT_LOCAL:
2853 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2854 (ifp->if_bytes > 0)) {
2855 ASSERT(ifp->if_u1.if_data != NULL);
2856 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2857 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2859 break;
2861 case XFS_DINODE_FMT_EXTENTS:
2862 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2863 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2864 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2865 (ifp->if_bytes == 0));
2866 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2867 (ifp->if_bytes > 0));
2868 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2869 (ifp->if_bytes > 0)) {
2870 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2871 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2872 whichfork);
2874 break;
2876 case XFS_DINODE_FMT_BTREE:
2877 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2878 (ifp->if_broot_bytes > 0)) {
2879 ASSERT(ifp->if_broot != NULL);
2880 ASSERT(ifp->if_broot_bytes <=
2881 (XFS_IFORK_SIZE(ip, whichfork) +
2882 XFS_BROOT_SIZE_ADJ));
2883 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2884 (xfs_bmdr_block_t *)cp,
2885 XFS_DFORK_SIZE(dip, mp, whichfork));
2887 break;
2889 case XFS_DINODE_FMT_DEV:
2890 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2891 ASSERT(whichfork == XFS_DATA_FORK);
2892 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
2894 break;
2896 case XFS_DINODE_FMT_UUID:
2897 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2898 ASSERT(whichfork == XFS_DATA_FORK);
2899 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2900 sizeof(uuid_t));
2902 break;
2904 default:
2905 ASSERT(0);
2906 break;
2910 STATIC int
2911 xfs_iflush_cluster(
2912 xfs_inode_t *ip,
2913 xfs_buf_t *bp)
2915 xfs_mount_t *mp = ip->i_mount;
2916 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2917 unsigned long first_index, mask;
2918 unsigned long inodes_per_cluster;
2919 int ilist_size;
2920 xfs_inode_t **ilist;
2921 xfs_inode_t *iq;
2922 int nr_found;
2923 int clcount = 0;
2924 int bufwasdelwri;
2925 int i;
2927 ASSERT(pag->pagi_inodeok);
2928 ASSERT(pag->pag_ici_init);
2930 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2931 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2932 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2933 if (!ilist)
2934 return 0;
2936 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2937 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2938 read_lock(&pag->pag_ici_lock);
2939 /* really need a gang lookup range call here */
2940 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2941 first_index, inodes_per_cluster);
2942 if (nr_found == 0)
2943 goto out_free;
2945 for (i = 0; i < nr_found; i++) {
2946 iq = ilist[i];
2947 if (iq == ip)
2948 continue;
2949 /* if the inode lies outside this cluster, we're done. */
2950 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2951 break;
2953 * Do an un-protected check to see if the inode is dirty and
2954 * is a candidate for flushing. These checks will be repeated
2955 * later after the appropriate locks are acquired.
2957 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2958 continue;
2961 * Try to get locks. If any are unavailable or it is pinned,
2962 * then this inode cannot be flushed and is skipped.
2965 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2966 continue;
2967 if (!xfs_iflock_nowait(iq)) {
2968 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2969 continue;
2971 if (xfs_ipincount(iq)) {
2972 xfs_ifunlock(iq);
2973 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2974 continue;
2978 * arriving here means that this inode can be flushed. First
2979 * re-check that it's dirty before flushing.
2981 if (!xfs_inode_clean(iq)) {
2982 int error;
2983 error = xfs_iflush_int(iq, bp);
2984 if (error) {
2985 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2986 goto cluster_corrupt_out;
2988 clcount++;
2989 } else {
2990 xfs_ifunlock(iq);
2992 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2995 if (clcount) {
2996 XFS_STATS_INC(xs_icluster_flushcnt);
2997 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3000 out_free:
3001 read_unlock(&pag->pag_ici_lock);
3002 kmem_free(ilist);
3003 return 0;
3006 cluster_corrupt_out:
3008 * Corruption detected in the clustering loop. Invalidate the
3009 * inode buffer and shut down the filesystem.
3011 read_unlock(&pag->pag_ici_lock);
3013 * Clean up the buffer. If it was B_DELWRI, just release it --
3014 * brelse can handle it with no problems. If not, shut down the
3015 * filesystem before releasing the buffer.
3017 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3018 if (bufwasdelwri)
3019 xfs_buf_relse(bp);
3021 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3023 if (!bufwasdelwri) {
3025 * Just like incore_relse: if we have b_iodone functions,
3026 * mark the buffer as an error and call them. Otherwise
3027 * mark it as stale and brelse.
3029 if (XFS_BUF_IODONE_FUNC(bp)) {
3030 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3031 XFS_BUF_UNDONE(bp);
3032 XFS_BUF_STALE(bp);
3033 XFS_BUF_SHUT(bp);
3034 XFS_BUF_ERROR(bp,EIO);
3035 xfs_biodone(bp);
3036 } else {
3037 XFS_BUF_STALE(bp);
3038 xfs_buf_relse(bp);
3043 * Unlocks the flush lock
3045 xfs_iflush_abort(iq);
3046 kmem_free(ilist);
3047 return XFS_ERROR(EFSCORRUPTED);
3051 * xfs_iflush() will write a modified inode's changes out to the
3052 * inode's on disk home. The caller must have the inode lock held
3053 * in at least shared mode and the inode flush completion must be
3054 * active as well. The inode lock will still be held upon return from
3055 * the call and the caller is free to unlock it.
3056 * The inode flush will be completed when the inode reaches the disk.
3057 * The flags indicate how the inode's buffer should be written out.
3060 xfs_iflush(
3061 xfs_inode_t *ip,
3062 uint flags)
3064 xfs_inode_log_item_t *iip;
3065 xfs_buf_t *bp;
3066 xfs_dinode_t *dip;
3067 xfs_mount_t *mp;
3068 int error;
3069 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
3070 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3072 XFS_STATS_INC(xs_iflush_count);
3074 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3075 ASSERT(!completion_done(&ip->i_flush));
3076 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3077 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3079 iip = ip->i_itemp;
3080 mp = ip->i_mount;
3083 * If the inode isn't dirty, then just release the inode
3084 * flush lock and do nothing.
3086 if (xfs_inode_clean(ip)) {
3087 xfs_ifunlock(ip);
3088 return 0;
3092 * We can't flush the inode until it is unpinned, so wait for it if we
3093 * are allowed to block. We know noone new can pin it, because we are
3094 * holding the inode lock shared and you need to hold it exclusively to
3095 * pin the inode.
3097 * If we are not allowed to block, force the log out asynchronously so
3098 * that when we come back the inode will be unpinned. If other inodes
3099 * in the same cluster are dirty, they will probably write the inode
3100 * out for us if they occur after the log force completes.
3102 if (noblock && xfs_ipincount(ip)) {
3103 xfs_iunpin_nowait(ip);
3104 xfs_ifunlock(ip);
3105 return EAGAIN;
3107 xfs_iunpin_wait(ip);
3110 * This may have been unpinned because the filesystem is shutting
3111 * down forcibly. If that's the case we must not write this inode
3112 * to disk, because the log record didn't make it to disk!
3114 if (XFS_FORCED_SHUTDOWN(mp)) {
3115 ip->i_update_core = 0;
3116 if (iip)
3117 iip->ili_format.ilf_fields = 0;
3118 xfs_ifunlock(ip);
3119 return XFS_ERROR(EIO);
3123 * Decide how buffer will be flushed out. This is done before
3124 * the call to xfs_iflush_int because this field is zeroed by it.
3126 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3128 * Flush out the inode buffer according to the directions
3129 * of the caller. In the cases where the caller has given
3130 * us a choice choose the non-delwri case. This is because
3131 * the inode is in the AIL and we need to get it out soon.
3133 switch (flags) {
3134 case XFS_IFLUSH_SYNC:
3135 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3136 flags = 0;
3137 break;
3138 case XFS_IFLUSH_ASYNC_NOBLOCK:
3139 case XFS_IFLUSH_ASYNC:
3140 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3141 flags = INT_ASYNC;
3142 break;
3143 case XFS_IFLUSH_DELWRI:
3144 flags = INT_DELWRI;
3145 break;
3146 default:
3147 ASSERT(0);
3148 flags = 0;
3149 break;
3151 } else {
3152 switch (flags) {
3153 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3154 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3155 case XFS_IFLUSH_DELWRI:
3156 flags = INT_DELWRI;
3157 break;
3158 case XFS_IFLUSH_ASYNC_NOBLOCK:
3159 case XFS_IFLUSH_ASYNC:
3160 flags = INT_ASYNC;
3161 break;
3162 case XFS_IFLUSH_SYNC:
3163 flags = 0;
3164 break;
3165 default:
3166 ASSERT(0);
3167 flags = 0;
3168 break;
3173 * Get the buffer containing the on-disk inode.
3175 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3176 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3177 if (error || !bp) {
3178 xfs_ifunlock(ip);
3179 return error;
3183 * First flush out the inode that xfs_iflush was called with.
3185 error = xfs_iflush_int(ip, bp);
3186 if (error)
3187 goto corrupt_out;
3190 * If the buffer is pinned then push on the log now so we won't
3191 * get stuck waiting in the write for too long.
3193 if (XFS_BUF_ISPINNED(bp))
3194 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3197 * inode clustering:
3198 * see if other inodes can be gathered into this write
3200 error = xfs_iflush_cluster(ip, bp);
3201 if (error)
3202 goto cluster_corrupt_out;
3204 if (flags & INT_DELWRI) {
3205 xfs_bdwrite(mp, bp);
3206 } else if (flags & INT_ASYNC) {
3207 error = xfs_bawrite(mp, bp);
3208 } else {
3209 error = xfs_bwrite(mp, bp);
3211 return error;
3213 corrupt_out:
3214 xfs_buf_relse(bp);
3215 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3216 cluster_corrupt_out:
3218 * Unlocks the flush lock
3220 xfs_iflush_abort(ip);
3221 return XFS_ERROR(EFSCORRUPTED);
3225 STATIC int
3226 xfs_iflush_int(
3227 xfs_inode_t *ip,
3228 xfs_buf_t *bp)
3230 xfs_inode_log_item_t *iip;
3231 xfs_dinode_t *dip;
3232 xfs_mount_t *mp;
3233 #ifdef XFS_TRANS_DEBUG
3234 int first;
3235 #endif
3237 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3238 ASSERT(!completion_done(&ip->i_flush));
3239 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3240 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3242 iip = ip->i_itemp;
3243 mp = ip->i_mount;
3247 * If the inode isn't dirty, then just release the inode
3248 * flush lock and do nothing.
3250 if (xfs_inode_clean(ip)) {
3251 xfs_ifunlock(ip);
3252 return 0;
3255 /* set *dip = inode's place in the buffer */
3256 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3259 * Clear i_update_core before copying out the data.
3260 * This is for coordination with our timestamp updates
3261 * that don't hold the inode lock. They will always
3262 * update the timestamps BEFORE setting i_update_core,
3263 * so if we clear i_update_core after they set it we
3264 * are guaranteed to see their updates to the timestamps.
3265 * I believe that this depends on strongly ordered memory
3266 * semantics, but we have that. We use the SYNCHRONIZE
3267 * macro to make sure that the compiler does not reorder
3268 * the i_update_core access below the data copy below.
3270 ip->i_update_core = 0;
3271 SYNCHRONIZE();
3274 * Make sure to get the latest atime from the Linux inode.
3276 xfs_synchronize_atime(ip);
3278 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
3279 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3280 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3281 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3282 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
3283 goto corrupt_out;
3285 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3286 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3287 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3288 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3289 ip->i_ino, ip, ip->i_d.di_magic);
3290 goto corrupt_out;
3292 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3293 if (XFS_TEST_ERROR(
3294 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3295 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3296 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3297 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3298 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3299 ip->i_ino, ip);
3300 goto corrupt_out;
3302 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3303 if (XFS_TEST_ERROR(
3304 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3305 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3306 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3307 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3308 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3309 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3310 ip->i_ino, ip);
3311 goto corrupt_out;
3314 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3315 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3316 XFS_RANDOM_IFLUSH_5)) {
3317 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3318 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3319 ip->i_ino,
3320 ip->i_d.di_nextents + ip->i_d.di_anextents,
3321 ip->i_d.di_nblocks,
3322 ip);
3323 goto corrupt_out;
3325 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3326 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3327 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3328 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3329 ip->i_ino, ip->i_d.di_forkoff, ip);
3330 goto corrupt_out;
3333 * bump the flush iteration count, used to detect flushes which
3334 * postdate a log record during recovery.
3337 ip->i_d.di_flushiter++;
3340 * Copy the dirty parts of the inode into the on-disk
3341 * inode. We always copy out the core of the inode,
3342 * because if the inode is dirty at all the core must
3343 * be.
3345 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
3347 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3348 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3349 ip->i_d.di_flushiter = 0;
3352 * If this is really an old format inode and the superblock version
3353 * has not been updated to support only new format inodes, then
3354 * convert back to the old inode format. If the superblock version
3355 * has been updated, then make the conversion permanent.
3357 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3358 xfs_sb_version_hasnlink(&mp->m_sb));
3359 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3360 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3362 * Convert it back.
3364 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3365 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3366 } else {
3368 * The superblock version has already been bumped,
3369 * so just make the conversion to the new inode
3370 * format permanent.
3372 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3373 dip->di_core.di_version = XFS_DINODE_VERSION_2;
3374 ip->i_d.di_onlink = 0;
3375 dip->di_core.di_onlink = 0;
3376 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3377 memset(&(dip->di_core.di_pad[0]), 0,
3378 sizeof(dip->di_core.di_pad));
3379 ASSERT(ip->i_d.di_projid == 0);
3383 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3384 if (XFS_IFORK_Q(ip))
3385 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3386 xfs_inobp_check(mp, bp);
3389 * We've recorded everything logged in the inode, so we'd
3390 * like to clear the ilf_fields bits so we don't log and
3391 * flush things unnecessarily. However, we can't stop
3392 * logging all this information until the data we've copied
3393 * into the disk buffer is written to disk. If we did we might
3394 * overwrite the copy of the inode in the log with all the
3395 * data after re-logging only part of it, and in the face of
3396 * a crash we wouldn't have all the data we need to recover.
3398 * What we do is move the bits to the ili_last_fields field.
3399 * When logging the inode, these bits are moved back to the
3400 * ilf_fields field. In the xfs_iflush_done() routine we
3401 * clear ili_last_fields, since we know that the information
3402 * those bits represent is permanently on disk. As long as
3403 * the flush completes before the inode is logged again, then
3404 * both ilf_fields and ili_last_fields will be cleared.
3406 * We can play with the ilf_fields bits here, because the inode
3407 * lock must be held exclusively in order to set bits there
3408 * and the flush lock protects the ili_last_fields bits.
3409 * Set ili_logged so the flush done
3410 * routine can tell whether or not to look in the AIL.
3411 * Also, store the current LSN of the inode so that we can tell
3412 * whether the item has moved in the AIL from xfs_iflush_done().
3413 * In order to read the lsn we need the AIL lock, because
3414 * it is a 64 bit value that cannot be read atomically.
3416 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3417 iip->ili_last_fields = iip->ili_format.ilf_fields;
3418 iip->ili_format.ilf_fields = 0;
3419 iip->ili_logged = 1;
3421 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3422 spin_lock(&mp->m_ail_lock);
3423 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3424 spin_unlock(&mp->m_ail_lock);
3427 * Attach the function xfs_iflush_done to the inode's
3428 * buffer. This will remove the inode from the AIL
3429 * and unlock the inode's flush lock when the inode is
3430 * completely written to disk.
3432 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3433 xfs_iflush_done, (xfs_log_item_t *)iip);
3435 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3436 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3437 } else {
3439 * We're flushing an inode which is not in the AIL and has
3440 * not been logged but has i_update_core set. For this
3441 * case we can use a B_DELWRI flush and immediately drop
3442 * the inode flush lock because we can avoid the whole
3443 * AIL state thing. It's OK to drop the flush lock now,
3444 * because we've already locked the buffer and to do anything
3445 * you really need both.
3447 if (iip != NULL) {
3448 ASSERT(iip->ili_logged == 0);
3449 ASSERT(iip->ili_last_fields == 0);
3450 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3452 xfs_ifunlock(ip);
3455 return 0;
3457 corrupt_out:
3458 return XFS_ERROR(EFSCORRUPTED);
3463 * Flush all inactive inodes in mp.
3465 void
3466 xfs_iflush_all(
3467 xfs_mount_t *mp)
3469 xfs_inode_t *ip;
3471 again:
3472 XFS_MOUNT_ILOCK(mp);
3473 ip = mp->m_inodes;
3474 if (ip == NULL)
3475 goto out;
3477 do {
3478 /* Make sure we skip markers inserted by sync */
3479 if (ip->i_mount == NULL) {
3480 ip = ip->i_mnext;
3481 continue;
3484 if (!VFS_I(ip)) {
3485 XFS_MOUNT_IUNLOCK(mp);
3486 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3487 goto again;
3490 ASSERT(vn_count(VFS_I(ip)) == 0);
3492 ip = ip->i_mnext;
3493 } while (ip != mp->m_inodes);
3494 out:
3495 XFS_MOUNT_IUNLOCK(mp);
3498 #ifdef XFS_ILOCK_TRACE
3499 ktrace_t *xfs_ilock_trace_buf;
3501 void
3502 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3504 ktrace_enter(ip->i_lock_trace,
3505 (void *)ip,
3506 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3507 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3508 (void *)ra, /* caller of ilock */
3509 (void *)(unsigned long)current_cpu(),
3510 (void *)(unsigned long)current_pid(),
3511 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3513 #endif
3516 * Return a pointer to the extent record at file index idx.
3518 xfs_bmbt_rec_host_t *
3519 xfs_iext_get_ext(
3520 xfs_ifork_t *ifp, /* inode fork pointer */
3521 xfs_extnum_t idx) /* index of target extent */
3523 ASSERT(idx >= 0);
3524 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3525 return ifp->if_u1.if_ext_irec->er_extbuf;
3526 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3527 xfs_ext_irec_t *erp; /* irec pointer */
3528 int erp_idx = 0; /* irec index */
3529 xfs_extnum_t page_idx = idx; /* ext index in target list */
3531 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3532 return &erp->er_extbuf[page_idx];
3533 } else if (ifp->if_bytes) {
3534 return &ifp->if_u1.if_extents[idx];
3535 } else {
3536 return NULL;
3541 * Insert new item(s) into the extent records for incore inode
3542 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3544 void
3545 xfs_iext_insert(
3546 xfs_ifork_t *ifp, /* inode fork pointer */
3547 xfs_extnum_t idx, /* starting index of new items */
3548 xfs_extnum_t count, /* number of inserted items */
3549 xfs_bmbt_irec_t *new) /* items to insert */
3551 xfs_extnum_t i; /* extent record index */
3553 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3554 xfs_iext_add(ifp, idx, count);
3555 for (i = idx; i < idx + count; i++, new++)
3556 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
3560 * This is called when the amount of space required for incore file
3561 * extents needs to be increased. The ext_diff parameter stores the
3562 * number of new extents being added and the idx parameter contains
3563 * the extent index where the new extents will be added. If the new
3564 * extents are being appended, then we just need to (re)allocate and
3565 * initialize the space. Otherwise, if the new extents are being
3566 * inserted into the middle of the existing entries, a bit more work
3567 * is required to make room for the new extents to be inserted. The
3568 * caller is responsible for filling in the new extent entries upon
3569 * return.
3571 void
3572 xfs_iext_add(
3573 xfs_ifork_t *ifp, /* inode fork pointer */
3574 xfs_extnum_t idx, /* index to begin adding exts */
3575 int ext_diff) /* number of extents to add */
3577 int byte_diff; /* new bytes being added */
3578 int new_size; /* size of extents after adding */
3579 xfs_extnum_t nextents; /* number of extents in file */
3581 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3582 ASSERT((idx >= 0) && (idx <= nextents));
3583 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3584 new_size = ifp->if_bytes + byte_diff;
3586 * If the new number of extents (nextents + ext_diff)
3587 * fits inside the inode, then continue to use the inline
3588 * extent buffer.
3590 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3591 if (idx < nextents) {
3592 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3593 &ifp->if_u2.if_inline_ext[idx],
3594 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3595 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3597 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3598 ifp->if_real_bytes = 0;
3599 ifp->if_lastex = nextents + ext_diff;
3602 * Otherwise use a linear (direct) extent list.
3603 * If the extents are currently inside the inode,
3604 * xfs_iext_realloc_direct will switch us from
3605 * inline to direct extent allocation mode.
3607 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
3608 xfs_iext_realloc_direct(ifp, new_size);
3609 if (idx < nextents) {
3610 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3611 &ifp->if_u1.if_extents[idx],
3612 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3613 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3616 /* Indirection array */
3617 else {
3618 xfs_ext_irec_t *erp;
3619 int erp_idx = 0;
3620 int page_idx = idx;
3622 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3623 if (ifp->if_flags & XFS_IFEXTIREC) {
3624 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3625 } else {
3626 xfs_iext_irec_init(ifp);
3627 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3628 erp = ifp->if_u1.if_ext_irec;
3630 /* Extents fit in target extent page */
3631 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3632 if (page_idx < erp->er_extcount) {
3633 memmove(&erp->er_extbuf[page_idx + ext_diff],
3634 &erp->er_extbuf[page_idx],
3635 (erp->er_extcount - page_idx) *
3636 sizeof(xfs_bmbt_rec_t));
3637 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3639 erp->er_extcount += ext_diff;
3640 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3642 /* Insert a new extent page */
3643 else if (erp) {
3644 xfs_iext_add_indirect_multi(ifp,
3645 erp_idx, page_idx, ext_diff);
3648 * If extent(s) are being appended to the last page in
3649 * the indirection array and the new extent(s) don't fit
3650 * in the page, then erp is NULL and erp_idx is set to
3651 * the next index needed in the indirection array.
3653 else {
3654 int count = ext_diff;
3656 while (count) {
3657 erp = xfs_iext_irec_new(ifp, erp_idx);
3658 erp->er_extcount = count;
3659 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3660 if (count) {
3661 erp_idx++;
3666 ifp->if_bytes = new_size;
3670 * This is called when incore extents are being added to the indirection
3671 * array and the new extents do not fit in the target extent list. The
3672 * erp_idx parameter contains the irec index for the target extent list
3673 * in the indirection array, and the idx parameter contains the extent
3674 * index within the list. The number of extents being added is stored
3675 * in the count parameter.
3677 * |-------| |-------|
3678 * | | | | idx - number of extents before idx
3679 * | idx | | count |
3680 * | | | | count - number of extents being inserted at idx
3681 * |-------| |-------|
3682 * | count | | nex2 | nex2 - number of extents after idx + count
3683 * |-------| |-------|
3685 void
3686 xfs_iext_add_indirect_multi(
3687 xfs_ifork_t *ifp, /* inode fork pointer */
3688 int erp_idx, /* target extent irec index */
3689 xfs_extnum_t idx, /* index within target list */
3690 int count) /* new extents being added */
3692 int byte_diff; /* new bytes being added */
3693 xfs_ext_irec_t *erp; /* pointer to irec entry */
3694 xfs_extnum_t ext_diff; /* number of extents to add */
3695 xfs_extnum_t ext_cnt; /* new extents still needed */
3696 xfs_extnum_t nex2; /* extents after idx + count */
3697 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3698 int nlists; /* number of irec's (lists) */
3700 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3701 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3702 nex2 = erp->er_extcount - idx;
3703 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3706 * Save second part of target extent list
3707 * (all extents past */
3708 if (nex2) {
3709 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3710 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3711 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3712 erp->er_extcount -= nex2;
3713 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3714 memset(&erp->er_extbuf[idx], 0, byte_diff);
3718 * Add the new extents to the end of the target
3719 * list, then allocate new irec record(s) and
3720 * extent buffer(s) as needed to store the rest
3721 * of the new extents.
3723 ext_cnt = count;
3724 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3725 if (ext_diff) {
3726 erp->er_extcount += ext_diff;
3727 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3728 ext_cnt -= ext_diff;
3730 while (ext_cnt) {
3731 erp_idx++;
3732 erp = xfs_iext_irec_new(ifp, erp_idx);
3733 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3734 erp->er_extcount = ext_diff;
3735 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3736 ext_cnt -= ext_diff;
3739 /* Add nex2 extents back to indirection array */
3740 if (nex2) {
3741 xfs_extnum_t ext_avail;
3742 int i;
3744 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3745 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3746 i = 0;
3748 * If nex2 extents fit in the current page, append
3749 * nex2_ep after the new extents.
3751 if (nex2 <= ext_avail) {
3752 i = erp->er_extcount;
3755 * Otherwise, check if space is available in the
3756 * next page.
3758 else if ((erp_idx < nlists - 1) &&
3759 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3760 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3761 erp_idx++;
3762 erp++;
3763 /* Create a hole for nex2 extents */
3764 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3765 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3768 * Final choice, create a new extent page for
3769 * nex2 extents.
3771 else {
3772 erp_idx++;
3773 erp = xfs_iext_irec_new(ifp, erp_idx);
3775 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3776 kmem_free(nex2_ep);
3777 erp->er_extcount += nex2;
3778 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3783 * This is called when the amount of space required for incore file
3784 * extents needs to be decreased. The ext_diff parameter stores the
3785 * number of extents to be removed and the idx parameter contains
3786 * the extent index where the extents will be removed from.
3788 * If the amount of space needed has decreased below the linear
3789 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3790 * extent array. Otherwise, use kmem_realloc() to adjust the
3791 * size to what is needed.
3793 void
3794 xfs_iext_remove(
3795 xfs_ifork_t *ifp, /* inode fork pointer */
3796 xfs_extnum_t idx, /* index to begin removing exts */
3797 int ext_diff) /* number of extents to remove */
3799 xfs_extnum_t nextents; /* number of extents in file */
3800 int new_size; /* size of extents after removal */
3802 ASSERT(ext_diff > 0);
3803 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3804 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3806 if (new_size == 0) {
3807 xfs_iext_destroy(ifp);
3808 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3809 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3810 } else if (ifp->if_real_bytes) {
3811 xfs_iext_remove_direct(ifp, idx, ext_diff);
3812 } else {
3813 xfs_iext_remove_inline(ifp, idx, ext_diff);
3815 ifp->if_bytes = new_size;
3819 * This removes ext_diff extents from the inline buffer, beginning
3820 * at extent index idx.
3822 void
3823 xfs_iext_remove_inline(
3824 xfs_ifork_t *ifp, /* inode fork pointer */
3825 xfs_extnum_t idx, /* index to begin removing exts */
3826 int ext_diff) /* number of extents to remove */
3828 int nextents; /* number of extents in file */
3830 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3831 ASSERT(idx < XFS_INLINE_EXTS);
3832 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3833 ASSERT(((nextents - ext_diff) > 0) &&
3834 (nextents - ext_diff) < XFS_INLINE_EXTS);
3836 if (idx + ext_diff < nextents) {
3837 memmove(&ifp->if_u2.if_inline_ext[idx],
3838 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3839 (nextents - (idx + ext_diff)) *
3840 sizeof(xfs_bmbt_rec_t));
3841 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3842 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3843 } else {
3844 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3845 ext_diff * sizeof(xfs_bmbt_rec_t));
3850 * This removes ext_diff extents from a linear (direct) extent list,
3851 * beginning at extent index idx. If the extents are being removed
3852 * from the end of the list (ie. truncate) then we just need to re-
3853 * allocate the list to remove the extra space. Otherwise, if the
3854 * extents are being removed from the middle of the existing extent
3855 * entries, then we first need to move the extent records beginning
3856 * at idx + ext_diff up in the list to overwrite the records being
3857 * removed, then remove the extra space via kmem_realloc.
3859 void
3860 xfs_iext_remove_direct(
3861 xfs_ifork_t *ifp, /* inode fork pointer */
3862 xfs_extnum_t idx, /* index to begin removing exts */
3863 int ext_diff) /* number of extents to remove */
3865 xfs_extnum_t nextents; /* number of extents in file */
3866 int new_size; /* size of extents after removal */
3868 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3869 new_size = ifp->if_bytes -
3870 (ext_diff * sizeof(xfs_bmbt_rec_t));
3871 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3873 if (new_size == 0) {
3874 xfs_iext_destroy(ifp);
3875 return;
3877 /* Move extents up in the list (if needed) */
3878 if (idx + ext_diff < nextents) {
3879 memmove(&ifp->if_u1.if_extents[idx],
3880 &ifp->if_u1.if_extents[idx + ext_diff],
3881 (nextents - (idx + ext_diff)) *
3882 sizeof(xfs_bmbt_rec_t));
3884 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3885 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3887 * Reallocate the direct extent list. If the extents
3888 * will fit inside the inode then xfs_iext_realloc_direct
3889 * will switch from direct to inline extent allocation
3890 * mode for us.
3892 xfs_iext_realloc_direct(ifp, new_size);
3893 ifp->if_bytes = new_size;
3897 * This is called when incore extents are being removed from the
3898 * indirection array and the extents being removed span multiple extent
3899 * buffers. The idx parameter contains the file extent index where we
3900 * want to begin removing extents, and the count parameter contains
3901 * how many extents need to be removed.
3903 * |-------| |-------|
3904 * | nex1 | | | nex1 - number of extents before idx
3905 * |-------| | count |
3906 * | | | | count - number of extents being removed at idx
3907 * | count | |-------|
3908 * | | | nex2 | nex2 - number of extents after idx + count
3909 * |-------| |-------|
3911 void
3912 xfs_iext_remove_indirect(
3913 xfs_ifork_t *ifp, /* inode fork pointer */
3914 xfs_extnum_t idx, /* index to begin removing extents */
3915 int count) /* number of extents to remove */
3917 xfs_ext_irec_t *erp; /* indirection array pointer */
3918 int erp_idx = 0; /* indirection array index */
3919 xfs_extnum_t ext_cnt; /* extents left to remove */
3920 xfs_extnum_t ext_diff; /* extents to remove in current list */
3921 xfs_extnum_t nex1; /* number of extents before idx */
3922 xfs_extnum_t nex2; /* extents after idx + count */
3923 int nlists; /* entries in indirection array */
3924 int page_idx = idx; /* index in target extent list */
3926 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3927 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3928 ASSERT(erp != NULL);
3929 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3930 nex1 = page_idx;
3931 ext_cnt = count;
3932 while (ext_cnt) {
3933 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3934 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3936 * Check for deletion of entire list;
3937 * xfs_iext_irec_remove() updates extent offsets.
3939 if (ext_diff == erp->er_extcount) {
3940 xfs_iext_irec_remove(ifp, erp_idx);
3941 ext_cnt -= ext_diff;
3942 nex1 = 0;
3943 if (ext_cnt) {
3944 ASSERT(erp_idx < ifp->if_real_bytes /
3945 XFS_IEXT_BUFSZ);
3946 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3947 nex1 = 0;
3948 continue;
3949 } else {
3950 break;
3953 /* Move extents up (if needed) */
3954 if (nex2) {
3955 memmove(&erp->er_extbuf[nex1],
3956 &erp->er_extbuf[nex1 + ext_diff],
3957 nex2 * sizeof(xfs_bmbt_rec_t));
3959 /* Zero out rest of page */
3960 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3961 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3962 /* Update remaining counters */
3963 erp->er_extcount -= ext_diff;
3964 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3965 ext_cnt -= ext_diff;
3966 nex1 = 0;
3967 erp_idx++;
3968 erp++;
3970 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3971 xfs_iext_irec_compact(ifp);
3975 * Create, destroy, or resize a linear (direct) block of extents.
3977 void
3978 xfs_iext_realloc_direct(
3979 xfs_ifork_t *ifp, /* inode fork pointer */
3980 int new_size) /* new size of extents */
3982 int rnew_size; /* real new size of extents */
3984 rnew_size = new_size;
3986 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3987 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3988 (new_size != ifp->if_real_bytes)));
3990 /* Free extent records */
3991 if (new_size == 0) {
3992 xfs_iext_destroy(ifp);
3994 /* Resize direct extent list and zero any new bytes */
3995 else if (ifp->if_real_bytes) {
3996 /* Check if extents will fit inside the inode */
3997 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3998 xfs_iext_direct_to_inline(ifp, new_size /
3999 (uint)sizeof(xfs_bmbt_rec_t));
4000 ifp->if_bytes = new_size;
4001 return;
4003 if (!is_power_of_2(new_size)){
4004 rnew_size = roundup_pow_of_two(new_size);
4006 if (rnew_size != ifp->if_real_bytes) {
4007 ifp->if_u1.if_extents =
4008 kmem_realloc(ifp->if_u1.if_extents,
4009 rnew_size,
4010 ifp->if_real_bytes, KM_NOFS);
4012 if (rnew_size > ifp->if_real_bytes) {
4013 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4014 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4015 rnew_size - ifp->if_real_bytes);
4019 * Switch from the inline extent buffer to a direct
4020 * extent list. Be sure to include the inline extent
4021 * bytes in new_size.
4023 else {
4024 new_size += ifp->if_bytes;
4025 if (!is_power_of_2(new_size)) {
4026 rnew_size = roundup_pow_of_two(new_size);
4028 xfs_iext_inline_to_direct(ifp, rnew_size);
4030 ifp->if_real_bytes = rnew_size;
4031 ifp->if_bytes = new_size;
4035 * Switch from linear (direct) extent records to inline buffer.
4037 void
4038 xfs_iext_direct_to_inline(
4039 xfs_ifork_t *ifp, /* inode fork pointer */
4040 xfs_extnum_t nextents) /* number of extents in file */
4042 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4043 ASSERT(nextents <= XFS_INLINE_EXTS);
4045 * The inline buffer was zeroed when we switched
4046 * from inline to direct extent allocation mode,
4047 * so we don't need to clear it here.
4049 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4050 nextents * sizeof(xfs_bmbt_rec_t));
4051 kmem_free(ifp->if_u1.if_extents);
4052 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4053 ifp->if_real_bytes = 0;
4057 * Switch from inline buffer to linear (direct) extent records.
4058 * new_size should already be rounded up to the next power of 2
4059 * by the caller (when appropriate), so use new_size as it is.
4060 * However, since new_size may be rounded up, we can't update
4061 * if_bytes here. It is the caller's responsibility to update
4062 * if_bytes upon return.
4064 void
4065 xfs_iext_inline_to_direct(
4066 xfs_ifork_t *ifp, /* inode fork pointer */
4067 int new_size) /* number of extents in file */
4069 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
4070 memset(ifp->if_u1.if_extents, 0, new_size);
4071 if (ifp->if_bytes) {
4072 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4073 ifp->if_bytes);
4074 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4075 sizeof(xfs_bmbt_rec_t));
4077 ifp->if_real_bytes = new_size;
4081 * Resize an extent indirection array to new_size bytes.
4083 void
4084 xfs_iext_realloc_indirect(
4085 xfs_ifork_t *ifp, /* inode fork pointer */
4086 int new_size) /* new indirection array size */
4088 int nlists; /* number of irec's (ex lists) */
4089 int size; /* current indirection array size */
4091 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4092 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4093 size = nlists * sizeof(xfs_ext_irec_t);
4094 ASSERT(ifp->if_real_bytes);
4095 ASSERT((new_size >= 0) && (new_size != size));
4096 if (new_size == 0) {
4097 xfs_iext_destroy(ifp);
4098 } else {
4099 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4100 kmem_realloc(ifp->if_u1.if_ext_irec,
4101 new_size, size, KM_NOFS);
4106 * Switch from indirection array to linear (direct) extent allocations.
4108 void
4109 xfs_iext_indirect_to_direct(
4110 xfs_ifork_t *ifp) /* inode fork pointer */
4112 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
4113 xfs_extnum_t nextents; /* number of extents in file */
4114 int size; /* size of file extents */
4116 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4117 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4118 ASSERT(nextents <= XFS_LINEAR_EXTS);
4119 size = nextents * sizeof(xfs_bmbt_rec_t);
4121 xfs_iext_irec_compact_pages(ifp);
4122 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4124 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4125 kmem_free(ifp->if_u1.if_ext_irec);
4126 ifp->if_flags &= ~XFS_IFEXTIREC;
4127 ifp->if_u1.if_extents = ep;
4128 ifp->if_bytes = size;
4129 if (nextents < XFS_LINEAR_EXTS) {
4130 xfs_iext_realloc_direct(ifp, size);
4135 * Free incore file extents.
4137 void
4138 xfs_iext_destroy(
4139 xfs_ifork_t *ifp) /* inode fork pointer */
4141 if (ifp->if_flags & XFS_IFEXTIREC) {
4142 int erp_idx;
4143 int nlists;
4145 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4146 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4147 xfs_iext_irec_remove(ifp, erp_idx);
4149 ifp->if_flags &= ~XFS_IFEXTIREC;
4150 } else if (ifp->if_real_bytes) {
4151 kmem_free(ifp->if_u1.if_extents);
4152 } else if (ifp->if_bytes) {
4153 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4154 sizeof(xfs_bmbt_rec_t));
4156 ifp->if_u1.if_extents = NULL;
4157 ifp->if_real_bytes = 0;
4158 ifp->if_bytes = 0;
4162 * Return a pointer to the extent record for file system block bno.
4164 xfs_bmbt_rec_host_t * /* pointer to found extent record */
4165 xfs_iext_bno_to_ext(
4166 xfs_ifork_t *ifp, /* inode fork pointer */
4167 xfs_fileoff_t bno, /* block number to search for */
4168 xfs_extnum_t *idxp) /* index of target extent */
4170 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
4171 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4172 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
4173 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4174 int high; /* upper boundary in search */
4175 xfs_extnum_t idx = 0; /* index of target extent */
4176 int low; /* lower boundary in search */
4177 xfs_extnum_t nextents; /* number of file extents */
4178 xfs_fileoff_t startoff = 0; /* start offset of extent */
4180 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4181 if (nextents == 0) {
4182 *idxp = 0;
4183 return NULL;
4185 low = 0;
4186 if (ifp->if_flags & XFS_IFEXTIREC) {
4187 /* Find target extent list */
4188 int erp_idx = 0;
4189 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4190 base = erp->er_extbuf;
4191 high = erp->er_extcount - 1;
4192 } else {
4193 base = ifp->if_u1.if_extents;
4194 high = nextents - 1;
4196 /* Binary search extent records */
4197 while (low <= high) {
4198 idx = (low + high) >> 1;
4199 ep = base + idx;
4200 startoff = xfs_bmbt_get_startoff(ep);
4201 blockcount = xfs_bmbt_get_blockcount(ep);
4202 if (bno < startoff) {
4203 high = idx - 1;
4204 } else if (bno >= startoff + blockcount) {
4205 low = idx + 1;
4206 } else {
4207 /* Convert back to file-based extent index */
4208 if (ifp->if_flags & XFS_IFEXTIREC) {
4209 idx += erp->er_extoff;
4211 *idxp = idx;
4212 return ep;
4215 /* Convert back to file-based extent index */
4216 if (ifp->if_flags & XFS_IFEXTIREC) {
4217 idx += erp->er_extoff;
4219 if (bno >= startoff + blockcount) {
4220 if (++idx == nextents) {
4221 ep = NULL;
4222 } else {
4223 ep = xfs_iext_get_ext(ifp, idx);
4226 *idxp = idx;
4227 return ep;
4231 * Return a pointer to the indirection array entry containing the
4232 * extent record for filesystem block bno. Store the index of the
4233 * target irec in *erp_idxp.
4235 xfs_ext_irec_t * /* pointer to found extent record */
4236 xfs_iext_bno_to_irec(
4237 xfs_ifork_t *ifp, /* inode fork pointer */
4238 xfs_fileoff_t bno, /* block number to search for */
4239 int *erp_idxp) /* irec index of target ext list */
4241 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4242 xfs_ext_irec_t *erp_next; /* next indirection array entry */
4243 int erp_idx; /* indirection array index */
4244 int nlists; /* number of extent irec's (lists) */
4245 int high; /* binary search upper limit */
4246 int low; /* binary search lower limit */
4248 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4249 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4250 erp_idx = 0;
4251 low = 0;
4252 high = nlists - 1;
4253 while (low <= high) {
4254 erp_idx = (low + high) >> 1;
4255 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4256 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4257 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4258 high = erp_idx - 1;
4259 } else if (erp_next && bno >=
4260 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4261 low = erp_idx + 1;
4262 } else {
4263 break;
4266 *erp_idxp = erp_idx;
4267 return erp;
4271 * Return a pointer to the indirection array entry containing the
4272 * extent record at file extent index *idxp. Store the index of the
4273 * target irec in *erp_idxp and store the page index of the target
4274 * extent record in *idxp.
4276 xfs_ext_irec_t *
4277 xfs_iext_idx_to_irec(
4278 xfs_ifork_t *ifp, /* inode fork pointer */
4279 xfs_extnum_t *idxp, /* extent index (file -> page) */
4280 int *erp_idxp, /* pointer to target irec */
4281 int realloc) /* new bytes were just added */
4283 xfs_ext_irec_t *prev; /* pointer to previous irec */
4284 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4285 int erp_idx; /* indirection array index */
4286 int nlists; /* number of irec's (ex lists) */
4287 int high; /* binary search upper limit */
4288 int low; /* binary search lower limit */
4289 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4291 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4292 ASSERT(page_idx >= 0 && page_idx <=
4293 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4294 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4295 erp_idx = 0;
4296 low = 0;
4297 high = nlists - 1;
4299 /* Binary search extent irec's */
4300 while (low <= high) {
4301 erp_idx = (low + high) >> 1;
4302 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4303 prev = erp_idx > 0 ? erp - 1 : NULL;
4304 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4305 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4306 high = erp_idx - 1;
4307 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4308 (page_idx == erp->er_extoff + erp->er_extcount &&
4309 !realloc)) {
4310 low = erp_idx + 1;
4311 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4312 erp->er_extcount == XFS_LINEAR_EXTS) {
4313 ASSERT(realloc);
4314 page_idx = 0;
4315 erp_idx++;
4316 erp = erp_idx < nlists ? erp + 1 : NULL;
4317 break;
4318 } else {
4319 page_idx -= erp->er_extoff;
4320 break;
4323 *idxp = page_idx;
4324 *erp_idxp = erp_idx;
4325 return(erp);
4329 * Allocate and initialize an indirection array once the space needed
4330 * for incore extents increases above XFS_IEXT_BUFSZ.
4332 void
4333 xfs_iext_irec_init(
4334 xfs_ifork_t *ifp) /* inode fork pointer */
4336 xfs_ext_irec_t *erp; /* indirection array pointer */
4337 xfs_extnum_t nextents; /* number of extents in file */
4339 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4340 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4341 ASSERT(nextents <= XFS_LINEAR_EXTS);
4343 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
4345 if (nextents == 0) {
4346 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4347 } else if (!ifp->if_real_bytes) {
4348 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4349 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4350 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4352 erp->er_extbuf = ifp->if_u1.if_extents;
4353 erp->er_extcount = nextents;
4354 erp->er_extoff = 0;
4356 ifp->if_flags |= XFS_IFEXTIREC;
4357 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4358 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4359 ifp->if_u1.if_ext_irec = erp;
4361 return;
4365 * Allocate and initialize a new entry in the indirection array.
4367 xfs_ext_irec_t *
4368 xfs_iext_irec_new(
4369 xfs_ifork_t *ifp, /* inode fork pointer */
4370 int erp_idx) /* index for new irec */
4372 xfs_ext_irec_t *erp; /* indirection array pointer */
4373 int i; /* loop counter */
4374 int nlists; /* number of irec's (ex lists) */
4376 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4377 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4379 /* Resize indirection array */
4380 xfs_iext_realloc_indirect(ifp, ++nlists *
4381 sizeof(xfs_ext_irec_t));
4383 * Move records down in the array so the
4384 * new page can use erp_idx.
4386 erp = ifp->if_u1.if_ext_irec;
4387 for (i = nlists - 1; i > erp_idx; i--) {
4388 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4390 ASSERT(i == erp_idx);
4392 /* Initialize new extent record */
4393 erp = ifp->if_u1.if_ext_irec;
4394 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
4395 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4396 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4397 erp[erp_idx].er_extcount = 0;
4398 erp[erp_idx].er_extoff = erp_idx > 0 ?
4399 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4400 return (&erp[erp_idx]);
4404 * Remove a record from the indirection array.
4406 void
4407 xfs_iext_irec_remove(
4408 xfs_ifork_t *ifp, /* inode fork pointer */
4409 int erp_idx) /* irec index to remove */
4411 xfs_ext_irec_t *erp; /* indirection array pointer */
4412 int i; /* loop counter */
4413 int nlists; /* number of irec's (ex lists) */
4415 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4416 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4417 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4418 if (erp->er_extbuf) {
4419 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4420 -erp->er_extcount);
4421 kmem_free(erp->er_extbuf);
4423 /* Compact extent records */
4424 erp = ifp->if_u1.if_ext_irec;
4425 for (i = erp_idx; i < nlists - 1; i++) {
4426 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4429 * Manually free the last extent record from the indirection
4430 * array. A call to xfs_iext_realloc_indirect() with a size
4431 * of zero would result in a call to xfs_iext_destroy() which
4432 * would in turn call this function again, creating a nasty
4433 * infinite loop.
4435 if (--nlists) {
4436 xfs_iext_realloc_indirect(ifp,
4437 nlists * sizeof(xfs_ext_irec_t));
4438 } else {
4439 kmem_free(ifp->if_u1.if_ext_irec);
4441 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4445 * This is called to clean up large amounts of unused memory allocated
4446 * by the indirection array. Before compacting anything though, verify
4447 * that the indirection array is still needed and switch back to the
4448 * linear extent list (or even the inline buffer) if possible. The
4449 * compaction policy is as follows:
4451 * Full Compaction: Extents fit into a single page (or inline buffer)
4452 * Partial Compaction: Extents occupy less than 50% of allocated space
4453 * No Compaction: Extents occupy at least 50% of allocated space
4455 void
4456 xfs_iext_irec_compact(
4457 xfs_ifork_t *ifp) /* inode fork pointer */
4459 xfs_extnum_t nextents; /* number of extents in file */
4460 int nlists; /* number of irec's (ex lists) */
4462 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4463 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4464 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4466 if (nextents == 0) {
4467 xfs_iext_destroy(ifp);
4468 } else if (nextents <= XFS_INLINE_EXTS) {
4469 xfs_iext_indirect_to_direct(ifp);
4470 xfs_iext_direct_to_inline(ifp, nextents);
4471 } else if (nextents <= XFS_LINEAR_EXTS) {
4472 xfs_iext_indirect_to_direct(ifp);
4473 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4474 xfs_iext_irec_compact_pages(ifp);
4479 * Combine extents from neighboring extent pages.
4481 void
4482 xfs_iext_irec_compact_pages(
4483 xfs_ifork_t *ifp) /* inode fork pointer */
4485 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4486 int erp_idx = 0; /* indirection array index */
4487 int nlists; /* number of irec's (ex lists) */
4489 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4490 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4491 while (erp_idx < nlists - 1) {
4492 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4493 erp_next = erp + 1;
4494 if (erp_next->er_extcount <=
4495 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4496 memcpy(&erp->er_extbuf[erp->er_extcount],
4497 erp_next->er_extbuf, erp_next->er_extcount *
4498 sizeof(xfs_bmbt_rec_t));
4499 erp->er_extcount += erp_next->er_extcount;
4501 * Free page before removing extent record
4502 * so er_extoffs don't get modified in
4503 * xfs_iext_irec_remove.
4505 kmem_free(erp_next->er_extbuf);
4506 erp_next->er_extbuf = NULL;
4507 xfs_iext_irec_remove(ifp, erp_idx + 1);
4508 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4509 } else {
4510 erp_idx++;
4516 * This is called to update the er_extoff field in the indirection
4517 * array when extents have been added or removed from one of the
4518 * extent lists. erp_idx contains the irec index to begin updating
4519 * at and ext_diff contains the number of extents that were added
4520 * or removed.
4522 void
4523 xfs_iext_irec_update_extoffs(
4524 xfs_ifork_t *ifp, /* inode fork pointer */
4525 int erp_idx, /* irec index to update */
4526 int ext_diff) /* number of new extents */
4528 int i; /* loop counter */
4529 int nlists; /* number of irec's (ex lists */
4531 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4532 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4533 for (i = erp_idx; i < nlists; i++) {
4534 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;