2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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>
22 #include "xfs_format.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_btree.h"
39 #include "xfs_alloc.h"
40 #include "xfs_ialloc.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_filestream.h"
45 #include "xfs_cksum.h"
46 #include "xfs_trace.h"
47 #include "xfs_icache.h"
49 kmem_zone_t *xfs_ifork_zone;
51 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
52 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
53 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
57 * Make sure that the extents in the given memory buffer
67 xfs_bmbt_rec_host_t rec;
70 for (i = 0; i < nrecs; i++) {
71 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
72 rec.l0 = get_unaligned(&ep->l0);
73 rec.l1 = get_unaligned(&ep->l1);
74 xfs_bmbt_get_all(&rec, &irec);
75 if (fmt == XFS_EXTFMT_NOSTATE)
76 ASSERT(irec.br_state == XFS_EXT_NORM);
80 #define xfs_validate_extents(ifp, nrecs, fmt)
85 * Move inode type and inode format specific information from the
86 * on-disk inode to the in-core inode. For fifos, devs, and sockets
87 * this means set if_rdev to the proper value. For files, directories,
88 * and symlinks this means to bring in the in-line data or extent
89 * pointers. For a file in B-tree format, only the root is immediately
90 * brought in-core. The rest will be in-lined in if_extents when it
91 * is first referenced (see xfs_iread_extents()).
98 xfs_attr_shortform_t *atp;
103 if (unlikely(be32_to_cpu(dip->di_nextents) +
104 be16_to_cpu(dip->di_anextents) >
105 be64_to_cpu(dip->di_nblocks))) {
106 xfs_warn(ip->i_mount,
107 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
108 (unsigned long long)ip->i_ino,
109 (int)(be32_to_cpu(dip->di_nextents) +
110 be16_to_cpu(dip->di_anextents)),
112 be64_to_cpu(dip->di_nblocks));
113 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
115 return XFS_ERROR(EFSCORRUPTED);
118 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
119 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
120 (unsigned long long)ip->i_ino,
122 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
124 return XFS_ERROR(EFSCORRUPTED);
127 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
128 !ip->i_mount->m_rtdev_targp)) {
129 xfs_warn(ip->i_mount,
130 "corrupt dinode %Lu, has realtime flag set.",
132 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
133 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
134 return XFS_ERROR(EFSCORRUPTED);
137 switch (ip->i_d.di_mode & S_IFMT) {
142 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
143 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
145 return XFS_ERROR(EFSCORRUPTED);
148 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
154 switch (dip->di_format) {
155 case XFS_DINODE_FMT_LOCAL:
157 * no local regular files yet
159 if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
160 xfs_warn(ip->i_mount,
161 "corrupt inode %Lu (local format for regular file).",
162 (unsigned long long) ip->i_ino);
163 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
166 return XFS_ERROR(EFSCORRUPTED);
169 di_size = be64_to_cpu(dip->di_size);
170 if (unlikely(di_size < 0 ||
171 di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
172 xfs_warn(ip->i_mount,
173 "corrupt inode %Lu (bad size %Ld for local inode).",
174 (unsigned long long) ip->i_ino,
175 (long long) di_size);
176 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
179 return XFS_ERROR(EFSCORRUPTED);
183 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
185 case XFS_DINODE_FMT_EXTENTS:
186 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
188 case XFS_DINODE_FMT_BTREE:
189 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
192 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
194 return XFS_ERROR(EFSCORRUPTED);
199 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
200 return XFS_ERROR(EFSCORRUPTED);
205 if (!XFS_DFORK_Q(dip))
208 ASSERT(ip->i_afp == NULL);
209 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
211 switch (dip->di_aformat) {
212 case XFS_DINODE_FMT_LOCAL:
213 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
214 size = be16_to_cpu(atp->hdr.totsize);
216 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
217 xfs_warn(ip->i_mount,
218 "corrupt inode %Lu (bad attr fork size %Ld).",
219 (unsigned long long) ip->i_ino,
221 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
224 return XFS_ERROR(EFSCORRUPTED);
227 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
229 case XFS_DINODE_FMT_EXTENTS:
230 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
232 case XFS_DINODE_FMT_BTREE:
233 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
236 error = XFS_ERROR(EFSCORRUPTED);
240 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
242 xfs_idestroy_fork(ip, XFS_DATA_FORK);
248 * The file is in-lined in the on-disk inode.
249 * If it fits into if_inline_data, then copy
250 * it there, otherwise allocate a buffer for it
251 * and copy the data there. Either way, set
252 * if_data to point at the data.
253 * If we allocate a buffer for the data, make
254 * sure that its size is a multiple of 4 and
255 * record the real size in i_real_bytes.
268 * If the size is unreasonable, then something
269 * is wrong and we just bail out rather than crash in
270 * kmem_alloc() or memcpy() below.
272 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
273 xfs_warn(ip->i_mount,
274 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
275 (unsigned long long) ip->i_ino, size,
276 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
277 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
279 return XFS_ERROR(EFSCORRUPTED);
281 ifp = XFS_IFORK_PTR(ip, whichfork);
284 ifp->if_u1.if_data = NULL;
285 else if (size <= sizeof(ifp->if_u2.if_inline_data))
286 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
288 real_size = roundup(size, 4);
289 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
291 ifp->if_bytes = size;
292 ifp->if_real_bytes = real_size;
294 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
295 ifp->if_flags &= ~XFS_IFEXTENTS;
296 ifp->if_flags |= XFS_IFINLINE;
301 * The file consists of a set of extents all
302 * of which fit into the on-disk inode.
303 * If there are few enough extents to fit into
304 * the if_inline_ext, then copy them there.
305 * Otherwise allocate a buffer for them and copy
306 * them into it. Either way, set if_extents
307 * to point at the extents.
321 ifp = XFS_IFORK_PTR(ip, whichfork);
322 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
323 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
326 * If the number of extents is unreasonable, then something
327 * is wrong and we just bail out rather than crash in
328 * kmem_alloc() or memcpy() below.
330 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
331 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
332 (unsigned long long) ip->i_ino, nex);
333 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
335 return XFS_ERROR(EFSCORRUPTED);
338 ifp->if_real_bytes = 0;
340 ifp->if_u1.if_extents = NULL;
341 else if (nex <= XFS_INLINE_EXTS)
342 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
344 xfs_iext_add(ifp, 0, nex);
346 ifp->if_bytes = size;
348 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
349 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
350 for (i = 0; i < nex; i++, dp++) {
351 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
352 ep->l0 = get_unaligned_be64(&dp->l0);
353 ep->l1 = get_unaligned_be64(&dp->l1);
355 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
356 if (whichfork != XFS_DATA_FORK ||
357 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
358 if (unlikely(xfs_check_nostate_extents(
360 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
363 return XFS_ERROR(EFSCORRUPTED);
366 ifp->if_flags |= XFS_IFEXTENTS;
371 * The file has too many extents to fit into
372 * the inode, so they are in B-tree format.
373 * Allocate a buffer for the root of the B-tree
374 * and copy the root into it. The i_extents
375 * field will remain NULL until all of the
376 * extents are read in (when they are needed).
384 struct xfs_mount *mp = ip->i_mount;
385 xfs_bmdr_block_t *dfp;
391 ifp = XFS_IFORK_PTR(ip, whichfork);
392 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
393 size = XFS_BMAP_BROOT_SPACE(mp, dfp);
394 nrecs = be16_to_cpu(dfp->bb_numrecs);
397 * blow out if -- fork has less extents than can fit in
398 * fork (fork shouldn't be a btree format), root btree
399 * block has more records than can fit into the fork,
400 * or the number of extents is greater than the number of
403 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
404 XFS_IFORK_MAXEXT(ip, whichfork) ||
405 XFS_BMDR_SPACE_CALC(nrecs) >
406 XFS_DFORK_SIZE(dip, mp, whichfork) ||
407 XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
408 xfs_warn(mp, "corrupt inode %Lu (btree).",
409 (unsigned long long) ip->i_ino);
410 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
412 return XFS_ERROR(EFSCORRUPTED);
415 ifp->if_broot_bytes = size;
416 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
417 ASSERT(ifp->if_broot != NULL);
419 * Copy and convert from the on-disk structure
420 * to the in-memory structure.
422 xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
423 ifp->if_broot, size);
424 ifp->if_flags &= ~XFS_IFEXTENTS;
425 ifp->if_flags |= XFS_IFBROOT;
431 * Read in extents from a btree-format inode.
432 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
442 xfs_extnum_t nextents;
444 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
445 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
447 return XFS_ERROR(EFSCORRUPTED);
449 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
450 ifp = XFS_IFORK_PTR(ip, whichfork);
453 * We know that the size is valid (it's checked in iformat_btree)
455 ifp->if_bytes = ifp->if_real_bytes = 0;
456 ifp->if_flags |= XFS_IFEXTENTS;
457 xfs_iext_add(ifp, 0, nextents);
458 error = xfs_bmap_read_extents(tp, ip, whichfork);
460 xfs_iext_destroy(ifp);
461 ifp->if_flags &= ~XFS_IFEXTENTS;
464 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
468 * Reallocate the space for if_broot based on the number of records
469 * being added or deleted as indicated in rec_diff. Move the records
470 * and pointers in if_broot to fit the new size. When shrinking this
471 * will eliminate holes between the records and pointers created by
472 * the caller. When growing this will create holes to be filled in
475 * The caller must not request to add more records than would fit in
476 * the on-disk inode root. If the if_broot is currently NULL, then
477 * if we are adding records, one will be allocated. The caller must also
478 * not request that the number of records go below zero, although
481 * ip -- the inode whose if_broot area is changing
482 * ext_diff -- the change in the number of records, positive or negative,
483 * requested for the if_broot array.
491 struct xfs_mount *mp = ip->i_mount;
494 struct xfs_btree_block *new_broot;
501 * Handle the degenerate case quietly.
507 ifp = XFS_IFORK_PTR(ip, whichfork);
510 * If there wasn't any memory allocated before, just
511 * allocate it now and get out.
513 if (ifp->if_broot_bytes == 0) {
514 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
515 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
516 ifp->if_broot_bytes = (int)new_size;
521 * If there is already an existing if_broot, then we need
522 * to realloc() it and shift the pointers to their new
523 * location. The records don't change location because
524 * they are kept butted up against the btree block header.
526 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
527 new_max = cur_max + rec_diff;
528 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
529 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
530 XFS_BMAP_BROOT_SPACE_CALC(mp, cur_max),
532 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
533 ifp->if_broot_bytes);
534 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
536 ifp->if_broot_bytes = (int)new_size;
537 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
538 XFS_IFORK_SIZE(ip, whichfork));
539 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
544 * rec_diff is less than 0. In this case, we are shrinking the
545 * if_broot buffer. It must already exist. If we go to zero
546 * records, just get rid of the root and clear the status bit.
548 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
549 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
550 new_max = cur_max + rec_diff;
551 ASSERT(new_max >= 0);
553 new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
557 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
559 * First copy over the btree block header.
561 memcpy(new_broot, ifp->if_broot,
562 XFS_BMBT_BLOCK_LEN(ip->i_mount));
565 ifp->if_flags &= ~XFS_IFBROOT;
569 * Only copy the records and pointers if there are any.
573 * First copy the records.
575 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
576 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
577 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
580 * Then copy the pointers.
582 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
583 ifp->if_broot_bytes);
584 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
586 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
588 kmem_free(ifp->if_broot);
589 ifp->if_broot = new_broot;
590 ifp->if_broot_bytes = (int)new_size;
592 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
593 XFS_IFORK_SIZE(ip, whichfork));
599 * This is called when the amount of space needed for if_data
600 * is increased or decreased. The change in size is indicated by
601 * the number of bytes that need to be added or deleted in the
602 * byte_diff parameter.
604 * If the amount of space needed has decreased below the size of the
605 * inline buffer, then switch to using the inline buffer. Otherwise,
606 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
609 * ip -- the inode whose if_data area is changing
610 * byte_diff -- the change in the number of bytes, positive or negative,
611 * requested for the if_data array.
623 if (byte_diff == 0) {
627 ifp = XFS_IFORK_PTR(ip, whichfork);
628 new_size = (int)ifp->if_bytes + byte_diff;
629 ASSERT(new_size >= 0);
632 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
633 kmem_free(ifp->if_u1.if_data);
635 ifp->if_u1.if_data = NULL;
637 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
639 * If the valid extents/data can fit in if_inline_ext/data,
640 * copy them from the malloc'd vector and free it.
642 if (ifp->if_u1.if_data == NULL) {
643 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
644 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
645 ASSERT(ifp->if_real_bytes != 0);
646 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
648 kmem_free(ifp->if_u1.if_data);
649 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
654 * Stuck with malloc/realloc.
655 * For inline data, the underlying buffer must be
656 * a multiple of 4 bytes in size so that it can be
657 * logged and stay on word boundaries. We enforce
660 real_size = roundup(new_size, 4);
661 if (ifp->if_u1.if_data == NULL) {
662 ASSERT(ifp->if_real_bytes == 0);
663 ifp->if_u1.if_data = kmem_alloc(real_size,
665 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
667 * Only do the realloc if the underlying size
668 * is really changing.
670 if (ifp->if_real_bytes != real_size) {
672 kmem_realloc(ifp->if_u1.if_data,
678 ASSERT(ifp->if_real_bytes == 0);
679 ifp->if_u1.if_data = kmem_alloc(real_size,
681 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
685 ifp->if_real_bytes = real_size;
686 ifp->if_bytes = new_size;
687 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
697 ifp = XFS_IFORK_PTR(ip, whichfork);
698 if (ifp->if_broot != NULL) {
699 kmem_free(ifp->if_broot);
700 ifp->if_broot = NULL;
704 * If the format is local, then we can't have an extents
705 * array so just look for an inline data array. If we're
706 * not local then we may or may not have an extents list,
707 * so check and free it up if we do.
709 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
710 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
711 (ifp->if_u1.if_data != NULL)) {
712 ASSERT(ifp->if_real_bytes != 0);
713 kmem_free(ifp->if_u1.if_data);
714 ifp->if_u1.if_data = NULL;
715 ifp->if_real_bytes = 0;
717 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
718 ((ifp->if_flags & XFS_IFEXTIREC) ||
719 ((ifp->if_u1.if_extents != NULL) &&
720 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
721 ASSERT(ifp->if_real_bytes != 0);
722 xfs_iext_destroy(ifp);
724 ASSERT(ifp->if_u1.if_extents == NULL ||
725 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
726 ASSERT(ifp->if_real_bytes == 0);
727 if (whichfork == XFS_ATTR_FORK) {
728 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
734 * xfs_iextents_copy()
736 * This is called to copy the REAL extents (as opposed to the delayed
737 * allocation extents) from the inode into the given buffer. It
738 * returns the number of bytes copied into the buffer.
740 * If there are no delayed allocation extents, then we can just
741 * memcpy() the extents into the buffer. Otherwise, we need to
742 * examine each extent in turn and skip those which are delayed.
754 xfs_fsblock_t start_block;
756 ifp = XFS_IFORK_PTR(ip, whichfork);
757 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
758 ASSERT(ifp->if_bytes > 0);
760 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
761 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
765 * There are some delayed allocation extents in the
766 * inode, so copy the extents one at a time and skip
767 * the delayed ones. There must be at least one
768 * non-delayed extent.
771 for (i = 0; i < nrecs; i++) {
772 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
773 start_block = xfs_bmbt_get_startblock(ep);
774 if (isnullstartblock(start_block)) {
776 * It's a delayed allocation extent, so skip it.
781 /* Translate to on disk format */
782 put_unaligned_be64(ep->l0, &dp->l0);
783 put_unaligned_be64(ep->l1, &dp->l1);
788 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
790 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
794 * Each of the following cases stores data into the same region
795 * of the on-disk inode, so only one of them can be valid at
796 * any given time. While it is possible to have conflicting formats
797 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
798 * in EXTENTS format, this can only happen when the fork has
799 * changed formats after being modified but before being flushed.
800 * In these cases, the format always takes precedence, because the
801 * format indicates the current state of the fork.
807 xfs_inode_log_item_t *iip,
814 static const short brootflag[2] =
815 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
816 static const short dataflag[2] =
817 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
818 static const short extflag[2] =
819 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
823 ifp = XFS_IFORK_PTR(ip, whichfork);
825 * This can happen if we gave up in iformat in an error path,
826 * for the attribute fork.
829 ASSERT(whichfork == XFS_ATTR_FORK);
832 cp = XFS_DFORK_PTR(dip, whichfork);
834 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
835 case XFS_DINODE_FMT_LOCAL:
836 if ((iip->ili_fields & dataflag[whichfork]) &&
837 (ifp->if_bytes > 0)) {
838 ASSERT(ifp->if_u1.if_data != NULL);
839 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
840 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
844 case XFS_DINODE_FMT_EXTENTS:
845 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
846 !(iip->ili_fields & extflag[whichfork]));
847 if ((iip->ili_fields & extflag[whichfork]) &&
848 (ifp->if_bytes > 0)) {
849 ASSERT(xfs_iext_get_ext(ifp, 0));
850 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
851 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
856 case XFS_DINODE_FMT_BTREE:
857 if ((iip->ili_fields & brootflag[whichfork]) &&
858 (ifp->if_broot_bytes > 0)) {
859 ASSERT(ifp->if_broot != NULL);
860 ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
861 XFS_IFORK_SIZE(ip, whichfork));
862 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
863 (xfs_bmdr_block_t *)cp,
864 XFS_DFORK_SIZE(dip, mp, whichfork));
868 case XFS_DINODE_FMT_DEV:
869 if (iip->ili_fields & XFS_ILOG_DEV) {
870 ASSERT(whichfork == XFS_DATA_FORK);
871 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
875 case XFS_DINODE_FMT_UUID:
876 if (iip->ili_fields & XFS_ILOG_UUID) {
877 ASSERT(whichfork == XFS_DATA_FORK);
878 memcpy(XFS_DFORK_DPTR(dip),
879 &ip->i_df.if_u2.if_uuid,
891 * Return a pointer to the extent record at file index idx.
893 xfs_bmbt_rec_host_t *
895 xfs_ifork_t *ifp, /* inode fork pointer */
896 xfs_extnum_t idx) /* index of target extent */
899 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
901 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
902 return ifp->if_u1.if_ext_irec->er_extbuf;
903 } else if (ifp->if_flags & XFS_IFEXTIREC) {
904 xfs_ext_irec_t *erp; /* irec pointer */
905 int erp_idx = 0; /* irec index */
906 xfs_extnum_t page_idx = idx; /* ext index in target list */
908 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
909 return &erp->er_extbuf[page_idx];
910 } else if (ifp->if_bytes) {
911 return &ifp->if_u1.if_extents[idx];
918 * Insert new item(s) into the extent records for incore inode
919 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
923 xfs_inode_t *ip, /* incore inode pointer */
924 xfs_extnum_t idx, /* starting index of new items */
925 xfs_extnum_t count, /* number of inserted items */
926 xfs_bmbt_irec_t *new, /* items to insert */
927 int state) /* type of extent conversion */
929 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
930 xfs_extnum_t i; /* extent record index */
932 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
934 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
935 xfs_iext_add(ifp, idx, count);
936 for (i = idx; i < idx + count; i++, new++)
937 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
941 * This is called when the amount of space required for incore file
942 * extents needs to be increased. The ext_diff parameter stores the
943 * number of new extents being added and the idx parameter contains
944 * the extent index where the new extents will be added. If the new
945 * extents are being appended, then we just need to (re)allocate and
946 * initialize the space. Otherwise, if the new extents are being
947 * inserted into the middle of the existing entries, a bit more work
948 * is required to make room for the new extents to be inserted. The
949 * caller is responsible for filling in the new extent entries upon
954 xfs_ifork_t *ifp, /* inode fork pointer */
955 xfs_extnum_t idx, /* index to begin adding exts */
956 int ext_diff) /* number of extents to add */
958 int byte_diff; /* new bytes being added */
959 int new_size; /* size of extents after adding */
960 xfs_extnum_t nextents; /* number of extents in file */
962 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
963 ASSERT((idx >= 0) && (idx <= nextents));
964 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
965 new_size = ifp->if_bytes + byte_diff;
967 * If the new number of extents (nextents + ext_diff)
968 * fits inside the inode, then continue to use the inline
971 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
972 if (idx < nextents) {
973 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
974 &ifp->if_u2.if_inline_ext[idx],
975 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
976 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
978 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
979 ifp->if_real_bytes = 0;
982 * Otherwise use a linear (direct) extent list.
983 * If the extents are currently inside the inode,
984 * xfs_iext_realloc_direct will switch us from
985 * inline to direct extent allocation mode.
987 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
988 xfs_iext_realloc_direct(ifp, new_size);
989 if (idx < nextents) {
990 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
991 &ifp->if_u1.if_extents[idx],
992 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
993 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
996 /* Indirection array */
1002 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
1003 if (ifp->if_flags & XFS_IFEXTIREC) {
1004 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
1006 xfs_iext_irec_init(ifp);
1007 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1008 erp = ifp->if_u1.if_ext_irec;
1010 /* Extents fit in target extent page */
1011 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
1012 if (page_idx < erp->er_extcount) {
1013 memmove(&erp->er_extbuf[page_idx + ext_diff],
1014 &erp->er_extbuf[page_idx],
1015 (erp->er_extcount - page_idx) *
1016 sizeof(xfs_bmbt_rec_t));
1017 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
1019 erp->er_extcount += ext_diff;
1020 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
1022 /* Insert a new extent page */
1024 xfs_iext_add_indirect_multi(ifp,
1025 erp_idx, page_idx, ext_diff);
1028 * If extent(s) are being appended to the last page in
1029 * the indirection array and the new extent(s) don't fit
1030 * in the page, then erp is NULL and erp_idx is set to
1031 * the next index needed in the indirection array.
1034 int count = ext_diff;
1037 erp = xfs_iext_irec_new(ifp, erp_idx);
1038 erp->er_extcount = count;
1039 count -= MIN(count, (int)XFS_LINEAR_EXTS);
1046 ifp->if_bytes = new_size;
1050 * This is called when incore extents are being added to the indirection
1051 * array and the new extents do not fit in the target extent list. The
1052 * erp_idx parameter contains the irec index for the target extent list
1053 * in the indirection array, and the idx parameter contains the extent
1054 * index within the list. The number of extents being added is stored
1055 * in the count parameter.
1057 * |-------| |-------|
1058 * | | | | idx - number of extents before idx
1060 * | | | | count - number of extents being inserted at idx
1061 * |-------| |-------|
1062 * | count | | nex2 | nex2 - number of extents after idx + count
1063 * |-------| |-------|
1066 xfs_iext_add_indirect_multi(
1067 xfs_ifork_t *ifp, /* inode fork pointer */
1068 int erp_idx, /* target extent irec index */
1069 xfs_extnum_t idx, /* index within target list */
1070 int count) /* new extents being added */
1072 int byte_diff; /* new bytes being added */
1073 xfs_ext_irec_t *erp; /* pointer to irec entry */
1074 xfs_extnum_t ext_diff; /* number of extents to add */
1075 xfs_extnum_t ext_cnt; /* new extents still needed */
1076 xfs_extnum_t nex2; /* extents after idx + count */
1077 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
1078 int nlists; /* number of irec's (lists) */
1080 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1081 erp = &ifp->if_u1.if_ext_irec[erp_idx];
1082 nex2 = erp->er_extcount - idx;
1083 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1086 * Save second part of target extent list
1087 * (all extents past */
1089 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
1090 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
1091 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
1092 erp->er_extcount -= nex2;
1093 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
1094 memset(&erp->er_extbuf[idx], 0, byte_diff);
1098 * Add the new extents to the end of the target
1099 * list, then allocate new irec record(s) and
1100 * extent buffer(s) as needed to store the rest
1101 * of the new extents.
1104 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
1106 erp->er_extcount += ext_diff;
1107 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
1108 ext_cnt -= ext_diff;
1112 erp = xfs_iext_irec_new(ifp, erp_idx);
1113 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
1114 erp->er_extcount = ext_diff;
1115 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
1116 ext_cnt -= ext_diff;
1119 /* Add nex2 extents back to indirection array */
1121 xfs_extnum_t ext_avail;
1124 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
1125 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
1128 * If nex2 extents fit in the current page, append
1129 * nex2_ep after the new extents.
1131 if (nex2 <= ext_avail) {
1132 i = erp->er_extcount;
1135 * Otherwise, check if space is available in the
1138 else if ((erp_idx < nlists - 1) &&
1139 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
1140 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
1143 /* Create a hole for nex2 extents */
1144 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
1145 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
1148 * Final choice, create a new extent page for
1153 erp = xfs_iext_irec_new(ifp, erp_idx);
1155 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
1157 erp->er_extcount += nex2;
1158 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
1163 * This is called when the amount of space required for incore file
1164 * extents needs to be decreased. The ext_diff parameter stores the
1165 * number of extents to be removed and the idx parameter contains
1166 * the extent index where the extents will be removed from.
1168 * If the amount of space needed has decreased below the linear
1169 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
1170 * extent array. Otherwise, use kmem_realloc() to adjust the
1171 * size to what is needed.
1175 xfs_inode_t *ip, /* incore inode pointer */
1176 xfs_extnum_t idx, /* index to begin removing exts */
1177 int ext_diff, /* number of extents to remove */
1178 int state) /* type of extent conversion */
1180 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
1181 xfs_extnum_t nextents; /* number of extents in file */
1182 int new_size; /* size of extents after removal */
1184 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
1186 ASSERT(ext_diff > 0);
1187 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1188 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
1190 if (new_size == 0) {
1191 xfs_iext_destroy(ifp);
1192 } else if (ifp->if_flags & XFS_IFEXTIREC) {
1193 xfs_iext_remove_indirect(ifp, idx, ext_diff);
1194 } else if (ifp->if_real_bytes) {
1195 xfs_iext_remove_direct(ifp, idx, ext_diff);
1197 xfs_iext_remove_inline(ifp, idx, ext_diff);
1199 ifp->if_bytes = new_size;
1203 * This removes ext_diff extents from the inline buffer, beginning
1204 * at extent index idx.
1207 xfs_iext_remove_inline(
1208 xfs_ifork_t *ifp, /* inode fork pointer */
1209 xfs_extnum_t idx, /* index to begin removing exts */
1210 int ext_diff) /* number of extents to remove */
1212 int nextents; /* number of extents in file */
1214 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
1215 ASSERT(idx < XFS_INLINE_EXTS);
1216 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1217 ASSERT(((nextents - ext_diff) > 0) &&
1218 (nextents - ext_diff) < XFS_INLINE_EXTS);
1220 if (idx + ext_diff < nextents) {
1221 memmove(&ifp->if_u2.if_inline_ext[idx],
1222 &ifp->if_u2.if_inline_ext[idx + ext_diff],
1223 (nextents - (idx + ext_diff)) *
1224 sizeof(xfs_bmbt_rec_t));
1225 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
1226 0, ext_diff * sizeof(xfs_bmbt_rec_t));
1228 memset(&ifp->if_u2.if_inline_ext[idx], 0,
1229 ext_diff * sizeof(xfs_bmbt_rec_t));
1234 * This removes ext_diff extents from a linear (direct) extent list,
1235 * beginning at extent index idx. If the extents are being removed
1236 * from the end of the list (ie. truncate) then we just need to re-
1237 * allocate the list to remove the extra space. Otherwise, if the
1238 * extents are being removed from the middle of the existing extent
1239 * entries, then we first need to move the extent records beginning
1240 * at idx + ext_diff up in the list to overwrite the records being
1241 * removed, then remove the extra space via kmem_realloc.
1244 xfs_iext_remove_direct(
1245 xfs_ifork_t *ifp, /* inode fork pointer */
1246 xfs_extnum_t idx, /* index to begin removing exts */
1247 int ext_diff) /* number of extents to remove */
1249 xfs_extnum_t nextents; /* number of extents in file */
1250 int new_size; /* size of extents after removal */
1252 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
1253 new_size = ifp->if_bytes -
1254 (ext_diff * sizeof(xfs_bmbt_rec_t));
1255 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1257 if (new_size == 0) {
1258 xfs_iext_destroy(ifp);
1261 /* Move extents up in the list (if needed) */
1262 if (idx + ext_diff < nextents) {
1263 memmove(&ifp->if_u1.if_extents[idx],
1264 &ifp->if_u1.if_extents[idx + ext_diff],
1265 (nextents - (idx + ext_diff)) *
1266 sizeof(xfs_bmbt_rec_t));
1268 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
1269 0, ext_diff * sizeof(xfs_bmbt_rec_t));
1271 * Reallocate the direct extent list. If the extents
1272 * will fit inside the inode then xfs_iext_realloc_direct
1273 * will switch from direct to inline extent allocation
1276 xfs_iext_realloc_direct(ifp, new_size);
1277 ifp->if_bytes = new_size;
1281 * This is called when incore extents are being removed from the
1282 * indirection array and the extents being removed span multiple extent
1283 * buffers. The idx parameter contains the file extent index where we
1284 * want to begin removing extents, and the count parameter contains
1285 * how many extents need to be removed.
1287 * |-------| |-------|
1288 * | nex1 | | | nex1 - number of extents before idx
1289 * |-------| | count |
1290 * | | | | count - number of extents being removed at idx
1291 * | count | |-------|
1292 * | | | nex2 | nex2 - number of extents after idx + count
1293 * |-------| |-------|
1296 xfs_iext_remove_indirect(
1297 xfs_ifork_t *ifp, /* inode fork pointer */
1298 xfs_extnum_t idx, /* index to begin removing extents */
1299 int count) /* number of extents to remove */
1301 xfs_ext_irec_t *erp; /* indirection array pointer */
1302 int erp_idx = 0; /* indirection array index */
1303 xfs_extnum_t ext_cnt; /* extents left to remove */
1304 xfs_extnum_t ext_diff; /* extents to remove in current list */
1305 xfs_extnum_t nex1; /* number of extents before idx */
1306 xfs_extnum_t nex2; /* extents after idx + count */
1307 int page_idx = idx; /* index in target extent list */
1309 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1310 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
1311 ASSERT(erp != NULL);
1315 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
1316 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
1318 * Check for deletion of entire list;
1319 * xfs_iext_irec_remove() updates extent offsets.
1321 if (ext_diff == erp->er_extcount) {
1322 xfs_iext_irec_remove(ifp, erp_idx);
1323 ext_cnt -= ext_diff;
1326 ASSERT(erp_idx < ifp->if_real_bytes /
1328 erp = &ifp->if_u1.if_ext_irec[erp_idx];
1335 /* Move extents up (if needed) */
1337 memmove(&erp->er_extbuf[nex1],
1338 &erp->er_extbuf[nex1 + ext_diff],
1339 nex2 * sizeof(xfs_bmbt_rec_t));
1341 /* Zero out rest of page */
1342 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
1343 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
1344 /* Update remaining counters */
1345 erp->er_extcount -= ext_diff;
1346 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
1347 ext_cnt -= ext_diff;
1352 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
1353 xfs_iext_irec_compact(ifp);
1357 * Create, destroy, or resize a linear (direct) block of extents.
1360 xfs_iext_realloc_direct(
1361 xfs_ifork_t *ifp, /* inode fork pointer */
1362 int new_size) /* new size of extents */
1364 int rnew_size; /* real new size of extents */
1366 rnew_size = new_size;
1368 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
1369 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
1370 (new_size != ifp->if_real_bytes)));
1372 /* Free extent records */
1373 if (new_size == 0) {
1374 xfs_iext_destroy(ifp);
1376 /* Resize direct extent list and zero any new bytes */
1377 else if (ifp->if_real_bytes) {
1378 /* Check if extents will fit inside the inode */
1379 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
1380 xfs_iext_direct_to_inline(ifp, new_size /
1381 (uint)sizeof(xfs_bmbt_rec_t));
1382 ifp->if_bytes = new_size;
1385 if (!is_power_of_2(new_size)){
1386 rnew_size = roundup_pow_of_two(new_size);
1388 if (rnew_size != ifp->if_real_bytes) {
1389 ifp->if_u1.if_extents =
1390 kmem_realloc(ifp->if_u1.if_extents,
1392 ifp->if_real_bytes, KM_NOFS);
1394 if (rnew_size > ifp->if_real_bytes) {
1395 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
1396 (uint)sizeof(xfs_bmbt_rec_t)], 0,
1397 rnew_size - ifp->if_real_bytes);
1401 * Switch from the inline extent buffer to a direct
1402 * extent list. Be sure to include the inline extent
1403 * bytes in new_size.
1406 new_size += ifp->if_bytes;
1407 if (!is_power_of_2(new_size)) {
1408 rnew_size = roundup_pow_of_two(new_size);
1410 xfs_iext_inline_to_direct(ifp, rnew_size);
1412 ifp->if_real_bytes = rnew_size;
1413 ifp->if_bytes = new_size;
1417 * Switch from linear (direct) extent records to inline buffer.
1420 xfs_iext_direct_to_inline(
1421 xfs_ifork_t *ifp, /* inode fork pointer */
1422 xfs_extnum_t nextents) /* number of extents in file */
1424 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1425 ASSERT(nextents <= XFS_INLINE_EXTS);
1427 * The inline buffer was zeroed when we switched
1428 * from inline to direct extent allocation mode,
1429 * so we don't need to clear it here.
1431 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
1432 nextents * sizeof(xfs_bmbt_rec_t));
1433 kmem_free(ifp->if_u1.if_extents);
1434 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
1435 ifp->if_real_bytes = 0;
1439 * Switch from inline buffer to linear (direct) extent records.
1440 * new_size should already be rounded up to the next power of 2
1441 * by the caller (when appropriate), so use new_size as it is.
1442 * However, since new_size may be rounded up, we can't update
1443 * if_bytes here. It is the caller's responsibility to update
1444 * if_bytes upon return.
1447 xfs_iext_inline_to_direct(
1448 xfs_ifork_t *ifp, /* inode fork pointer */
1449 int new_size) /* number of extents in file */
1451 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
1452 memset(ifp->if_u1.if_extents, 0, new_size);
1453 if (ifp->if_bytes) {
1454 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
1456 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
1457 sizeof(xfs_bmbt_rec_t));
1459 ifp->if_real_bytes = new_size;
1463 * Resize an extent indirection array to new_size bytes.
1466 xfs_iext_realloc_indirect(
1467 xfs_ifork_t *ifp, /* inode fork pointer */
1468 int new_size) /* new indirection array size */
1470 int nlists; /* number of irec's (ex lists) */
1471 int size; /* current indirection array size */
1473 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1474 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1475 size = nlists * sizeof(xfs_ext_irec_t);
1476 ASSERT(ifp->if_real_bytes);
1477 ASSERT((new_size >= 0) && (new_size != size));
1478 if (new_size == 0) {
1479 xfs_iext_destroy(ifp);
1481 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
1482 kmem_realloc(ifp->if_u1.if_ext_irec,
1483 new_size, size, KM_NOFS);
1488 * Switch from indirection array to linear (direct) extent allocations.
1491 xfs_iext_indirect_to_direct(
1492 xfs_ifork_t *ifp) /* inode fork pointer */
1494 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1495 xfs_extnum_t nextents; /* number of extents in file */
1496 int size; /* size of file extents */
1498 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1499 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1500 ASSERT(nextents <= XFS_LINEAR_EXTS);
1501 size = nextents * sizeof(xfs_bmbt_rec_t);
1503 xfs_iext_irec_compact_pages(ifp);
1504 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
1506 ep = ifp->if_u1.if_ext_irec->er_extbuf;
1507 kmem_free(ifp->if_u1.if_ext_irec);
1508 ifp->if_flags &= ~XFS_IFEXTIREC;
1509 ifp->if_u1.if_extents = ep;
1510 ifp->if_bytes = size;
1511 if (nextents < XFS_LINEAR_EXTS) {
1512 xfs_iext_realloc_direct(ifp, size);
1517 * Free incore file extents.
1521 xfs_ifork_t *ifp) /* inode fork pointer */
1523 if (ifp->if_flags & XFS_IFEXTIREC) {
1527 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1528 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
1529 xfs_iext_irec_remove(ifp, erp_idx);
1531 ifp->if_flags &= ~XFS_IFEXTIREC;
1532 } else if (ifp->if_real_bytes) {
1533 kmem_free(ifp->if_u1.if_extents);
1534 } else if (ifp->if_bytes) {
1535 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
1536 sizeof(xfs_bmbt_rec_t));
1538 ifp->if_u1.if_extents = NULL;
1539 ifp->if_real_bytes = 0;
1544 * Return a pointer to the extent record for file system block bno.
1546 xfs_bmbt_rec_host_t * /* pointer to found extent record */
1547 xfs_iext_bno_to_ext(
1548 xfs_ifork_t *ifp, /* inode fork pointer */
1549 xfs_fileoff_t bno, /* block number to search for */
1550 xfs_extnum_t *idxp) /* index of target extent */
1552 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
1553 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
1554 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
1555 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
1556 int high; /* upper boundary in search */
1557 xfs_extnum_t idx = 0; /* index of target extent */
1558 int low; /* lower boundary in search */
1559 xfs_extnum_t nextents; /* number of file extents */
1560 xfs_fileoff_t startoff = 0; /* start offset of extent */
1562 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1563 if (nextents == 0) {
1568 if (ifp->if_flags & XFS_IFEXTIREC) {
1569 /* Find target extent list */
1571 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
1572 base = erp->er_extbuf;
1573 high = erp->er_extcount - 1;
1575 base = ifp->if_u1.if_extents;
1576 high = nextents - 1;
1578 /* Binary search extent records */
1579 while (low <= high) {
1580 idx = (low + high) >> 1;
1582 startoff = xfs_bmbt_get_startoff(ep);
1583 blockcount = xfs_bmbt_get_blockcount(ep);
1584 if (bno < startoff) {
1586 } else if (bno >= startoff + blockcount) {
1589 /* Convert back to file-based extent index */
1590 if (ifp->if_flags & XFS_IFEXTIREC) {
1591 idx += erp->er_extoff;
1597 /* Convert back to file-based extent index */
1598 if (ifp->if_flags & XFS_IFEXTIREC) {
1599 idx += erp->er_extoff;
1601 if (bno >= startoff + blockcount) {
1602 if (++idx == nextents) {
1605 ep = xfs_iext_get_ext(ifp, idx);
1613 * Return a pointer to the indirection array entry containing the
1614 * extent record for filesystem block bno. Store the index of the
1615 * target irec in *erp_idxp.
1617 xfs_ext_irec_t * /* pointer to found extent record */
1618 xfs_iext_bno_to_irec(
1619 xfs_ifork_t *ifp, /* inode fork pointer */
1620 xfs_fileoff_t bno, /* block number to search for */
1621 int *erp_idxp) /* irec index of target ext list */
1623 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
1624 xfs_ext_irec_t *erp_next; /* next indirection array entry */
1625 int erp_idx; /* indirection array index */
1626 int nlists; /* number of extent irec's (lists) */
1627 int high; /* binary search upper limit */
1628 int low; /* binary search lower limit */
1630 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1631 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1635 while (low <= high) {
1636 erp_idx = (low + high) >> 1;
1637 erp = &ifp->if_u1.if_ext_irec[erp_idx];
1638 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
1639 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
1641 } else if (erp_next && bno >=
1642 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
1648 *erp_idxp = erp_idx;
1653 * Return a pointer to the indirection array entry containing the
1654 * extent record at file extent index *idxp. Store the index of the
1655 * target irec in *erp_idxp and store the page index of the target
1656 * extent record in *idxp.
1659 xfs_iext_idx_to_irec(
1660 xfs_ifork_t *ifp, /* inode fork pointer */
1661 xfs_extnum_t *idxp, /* extent index (file -> page) */
1662 int *erp_idxp, /* pointer to target irec */
1663 int realloc) /* new bytes were just added */
1665 xfs_ext_irec_t *prev; /* pointer to previous irec */
1666 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
1667 int erp_idx; /* indirection array index */
1668 int nlists; /* number of irec's (ex lists) */
1669 int high; /* binary search upper limit */
1670 int low; /* binary search lower limit */
1671 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
1673 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1674 ASSERT(page_idx >= 0);
1675 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
1676 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
1678 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1683 /* Binary search extent irec's */
1684 while (low <= high) {
1685 erp_idx = (low + high) >> 1;
1686 erp = &ifp->if_u1.if_ext_irec[erp_idx];
1687 prev = erp_idx > 0 ? erp - 1 : NULL;
1688 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
1689 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
1691 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
1692 (page_idx == erp->er_extoff + erp->er_extcount &&
1695 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
1696 erp->er_extcount == XFS_LINEAR_EXTS) {
1700 erp = erp_idx < nlists ? erp + 1 : NULL;
1703 page_idx -= erp->er_extoff;
1708 *erp_idxp = erp_idx;
1713 * Allocate and initialize an indirection array once the space needed
1714 * for incore extents increases above XFS_IEXT_BUFSZ.
1718 xfs_ifork_t *ifp) /* inode fork pointer */
1720 xfs_ext_irec_t *erp; /* indirection array pointer */
1721 xfs_extnum_t nextents; /* number of extents in file */
1723 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
1724 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1725 ASSERT(nextents <= XFS_LINEAR_EXTS);
1727 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
1729 if (nextents == 0) {
1730 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
1731 } else if (!ifp->if_real_bytes) {
1732 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
1733 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
1734 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
1736 erp->er_extbuf = ifp->if_u1.if_extents;
1737 erp->er_extcount = nextents;
1740 ifp->if_flags |= XFS_IFEXTIREC;
1741 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
1742 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
1743 ifp->if_u1.if_ext_irec = erp;
1749 * Allocate and initialize a new entry in the indirection array.
1753 xfs_ifork_t *ifp, /* inode fork pointer */
1754 int erp_idx) /* index for new irec */
1756 xfs_ext_irec_t *erp; /* indirection array pointer */
1757 int i; /* loop counter */
1758 int nlists; /* number of irec's (ex lists) */
1760 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1761 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1763 /* Resize indirection array */
1764 xfs_iext_realloc_indirect(ifp, ++nlists *
1765 sizeof(xfs_ext_irec_t));
1767 * Move records down in the array so the
1768 * new page can use erp_idx.
1770 erp = ifp->if_u1.if_ext_irec;
1771 for (i = nlists - 1; i > erp_idx; i--) {
1772 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
1774 ASSERT(i == erp_idx);
1776 /* Initialize new extent record */
1777 erp = ifp->if_u1.if_ext_irec;
1778 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
1779 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
1780 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
1781 erp[erp_idx].er_extcount = 0;
1782 erp[erp_idx].er_extoff = erp_idx > 0 ?
1783 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
1784 return (&erp[erp_idx]);
1788 * Remove a record from the indirection array.
1791 xfs_iext_irec_remove(
1792 xfs_ifork_t *ifp, /* inode fork pointer */
1793 int erp_idx) /* irec index to remove */
1795 xfs_ext_irec_t *erp; /* indirection array pointer */
1796 int i; /* loop counter */
1797 int nlists; /* number of irec's (ex lists) */
1799 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1800 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1801 erp = &ifp->if_u1.if_ext_irec[erp_idx];
1802 if (erp->er_extbuf) {
1803 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
1805 kmem_free(erp->er_extbuf);
1807 /* Compact extent records */
1808 erp = ifp->if_u1.if_ext_irec;
1809 for (i = erp_idx; i < nlists - 1; i++) {
1810 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
1813 * Manually free the last extent record from the indirection
1814 * array. A call to xfs_iext_realloc_indirect() with a size
1815 * of zero would result in a call to xfs_iext_destroy() which
1816 * would in turn call this function again, creating a nasty
1820 xfs_iext_realloc_indirect(ifp,
1821 nlists * sizeof(xfs_ext_irec_t));
1823 kmem_free(ifp->if_u1.if_ext_irec);
1825 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
1829 * This is called to clean up large amounts of unused memory allocated
1830 * by the indirection array. Before compacting anything though, verify
1831 * that the indirection array is still needed and switch back to the
1832 * linear extent list (or even the inline buffer) if possible. The
1833 * compaction policy is as follows:
1835 * Full Compaction: Extents fit into a single page (or inline buffer)
1836 * Partial Compaction: Extents occupy less than 50% of allocated space
1837 * No Compaction: Extents occupy at least 50% of allocated space
1840 xfs_iext_irec_compact(
1841 xfs_ifork_t *ifp) /* inode fork pointer */
1843 xfs_extnum_t nextents; /* number of extents in file */
1844 int nlists; /* number of irec's (ex lists) */
1846 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1847 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1848 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1850 if (nextents == 0) {
1851 xfs_iext_destroy(ifp);
1852 } else if (nextents <= XFS_INLINE_EXTS) {
1853 xfs_iext_indirect_to_direct(ifp);
1854 xfs_iext_direct_to_inline(ifp, nextents);
1855 } else if (nextents <= XFS_LINEAR_EXTS) {
1856 xfs_iext_indirect_to_direct(ifp);
1857 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
1858 xfs_iext_irec_compact_pages(ifp);
1863 * Combine extents from neighboring extent pages.
1866 xfs_iext_irec_compact_pages(
1867 xfs_ifork_t *ifp) /* inode fork pointer */
1869 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
1870 int erp_idx = 0; /* indirection array index */
1871 int nlists; /* number of irec's (ex lists) */
1873 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1874 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1875 while (erp_idx < nlists - 1) {
1876 erp = &ifp->if_u1.if_ext_irec[erp_idx];
1878 if (erp_next->er_extcount <=
1879 (XFS_LINEAR_EXTS - erp->er_extcount)) {
1880 memcpy(&erp->er_extbuf[erp->er_extcount],
1881 erp_next->er_extbuf, erp_next->er_extcount *
1882 sizeof(xfs_bmbt_rec_t));
1883 erp->er_extcount += erp_next->er_extcount;
1885 * Free page before removing extent record
1886 * so er_extoffs don't get modified in
1887 * xfs_iext_irec_remove.
1889 kmem_free(erp_next->er_extbuf);
1890 erp_next->er_extbuf = NULL;
1891 xfs_iext_irec_remove(ifp, erp_idx + 1);
1892 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1900 * This is called to update the er_extoff field in the indirection
1901 * array when extents have been added or removed from one of the
1902 * extent lists. erp_idx contains the irec index to begin updating
1903 * at and ext_diff contains the number of extents that were added
1907 xfs_iext_irec_update_extoffs(
1908 xfs_ifork_t *ifp, /* inode fork pointer */
1909 int erp_idx, /* irec index to update */
1910 int ext_diff) /* number of new extents */
1912 int i; /* loop counter */
1913 int nlists; /* number of irec's (ex lists */
1915 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1916 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1917 for (i = erp_idx; i < nlists; i++) {
1918 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;