2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_defer.h"
29 #include "xfs_inode.h"
30 #include "xfs_btree.h"
31 #include "xfs_trans.h"
32 #include "xfs_extfree_item.h"
33 #include "xfs_alloc.h"
35 #include "xfs_bmap_util.h"
36 #include "xfs_bmap_btree.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_quota.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_trace.h"
42 #include "xfs_icache.h"
44 #include "xfs_rmap_btree.h"
45 #include "xfs_iomap.h"
46 #include "xfs_reflink.h"
47 #include "xfs_refcount.h"
49 /* Kernel only BMAP related definitions and functions */
52 * Convert the given file system block to a disk block. We have to treat it
53 * differently based on whether the file is a real time file or not, because the
57 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
59 return (XFS_IS_REALTIME_INODE(ip) ? \
60 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
61 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
65 * Routine to zero an extent on disk allocated to the specific inode.
67 * The VFS functions take a linearised filesystem block offset, so we have to
68 * convert the sparse xfs fsb to the right format first.
69 * VFS types are real funky, too.
74 xfs_fsblock_t start_fsb,
77 struct xfs_mount *mp = ip->i_mount;
78 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
79 sector_t block = XFS_BB_TO_FSBT(mp, sector);
81 return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
82 block << (mp->m_super->s_blocksize_bits - 9),
83 count_fsb << (mp->m_super->s_blocksize_bits - 9),
89 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
91 int error; /* error return value */
92 xfs_mount_t *mp; /* mount point structure */
93 xfs_extlen_t prod = 0; /* product factor for allocators */
94 xfs_extlen_t ralen = 0; /* realtime allocation length */
95 xfs_extlen_t align; /* minimum allocation alignment */
99 align = xfs_get_extsz_hint(ap->ip);
100 prod = align / mp->m_sb.sb_rextsize;
101 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
102 align, 1, ap->eof, 0,
103 ap->conv, &ap->offset, &ap->length);
107 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
110 * If the offset & length are not perfectly aligned
111 * then kill prod, it will just get us in trouble.
113 if (do_mod(ap->offset, align) || ap->length % align)
116 * Set ralen to be the actual requested length in rtextents.
118 ralen = ap->length / mp->m_sb.sb_rextsize;
120 * If the old value was close enough to MAXEXTLEN that
121 * we rounded up to it, cut it back so it's valid again.
122 * Note that if it's a really large request (bigger than
123 * MAXEXTLEN), we don't hear about that number, and can't
124 * adjust the starting point to match it.
126 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
127 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
130 * Lock out modifications to both the RT bitmap and summary inodes
132 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
133 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
134 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
135 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
138 * If it's an allocation to an empty file at offset 0,
139 * pick an extent that will space things out in the rt area.
141 if (ap->eof && ap->offset == 0) {
142 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
144 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
147 ap->blkno = rtx * mp->m_sb.sb_rextsize;
152 xfs_bmap_adjacent(ap);
155 * Realtime allocation, done through xfs_rtallocate_extent.
157 do_div(ap->blkno, mp->m_sb.sb_rextsize);
160 error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
161 &ralen, ap->wasdel, prod, &rtb);
166 if (ap->blkno != NULLFSBLOCK) {
167 ap->blkno *= mp->m_sb.sb_rextsize;
168 ralen *= mp->m_sb.sb_rextsize;
170 ap->ip->i_d.di_nblocks += ralen;
171 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
173 ap->ip->i_delayed_blks -= ralen;
175 * Adjust the disk quota also. This was reserved
178 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
179 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
180 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
182 /* Zero the extent if we were asked to do so */
183 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
184 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
195 * Check if the endoff is outside the last extent. If so the caller will grow
196 * the allocation to a stripe unit boundary. All offsets are considered outside
197 * the end of file for an empty fork, so 1 is returned in *eof in that case.
201 struct xfs_inode *ip,
202 xfs_fileoff_t endoff,
206 struct xfs_bmbt_irec rec;
209 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
213 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
218 * Extent tree block counting routines.
222 * Count leaf blocks given a range of extent records.
225 xfs_bmap_count_leaves(
233 for (b = 0; b < numrecs; b++) {
234 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
235 *count += xfs_bmbt_get_blockcount(frp);
240 * Count leaf blocks given a range of extent records originally
244 xfs_bmap_disk_count_leaves(
245 struct xfs_mount *mp,
246 struct xfs_btree_block *block,
253 for (b = 1; b <= numrecs; b++) {
254 frp = XFS_BMBT_REC_ADDR(mp, block, b);
255 *count += xfs_bmbt_disk_get_blockcount(frp);
260 * Recursively walks each level of a btree
261 * to count total fsblocks in use.
263 STATIC int /* error */
265 xfs_mount_t *mp, /* file system mount point */
266 xfs_trans_t *tp, /* transaction pointer */
267 xfs_ifork_t *ifp, /* inode fork pointer */
268 xfs_fsblock_t blockno, /* file system block number */
269 int levelin, /* level in btree */
270 int *count) /* Count of blocks */
276 xfs_fsblock_t bno = blockno;
277 xfs_fsblock_t nextbno;
278 struct xfs_btree_block *block, *nextblock;
281 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
286 block = XFS_BUF_TO_BLOCK(bp);
289 /* Not at node above leaves, count this level of nodes */
290 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
291 while (nextbno != NULLFSBLOCK) {
292 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
298 nextblock = XFS_BUF_TO_BLOCK(nbp);
299 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
300 xfs_trans_brelse(tp, nbp);
303 /* Dive to the next level */
304 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
305 bno = be64_to_cpu(*pp);
306 if (unlikely((error =
307 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
308 xfs_trans_brelse(tp, bp);
309 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
310 XFS_ERRLEVEL_LOW, mp);
311 return -EFSCORRUPTED;
313 xfs_trans_brelse(tp, bp);
315 /* count all level 1 nodes and their leaves */
317 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
318 numrecs = be16_to_cpu(block->bb_numrecs);
319 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
320 xfs_trans_brelse(tp, bp);
321 if (nextbno == NULLFSBLOCK)
324 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
330 block = XFS_BUF_TO_BLOCK(bp);
337 * Count fsblocks of the given fork.
339 static int /* error */
340 xfs_bmap_count_blocks(
341 xfs_trans_t *tp, /* transaction pointer */
342 xfs_inode_t *ip, /* incore inode */
343 int whichfork, /* data or attr fork */
344 int *count) /* out: count of blocks */
346 struct xfs_btree_block *block; /* current btree block */
347 xfs_fsblock_t bno; /* block # of "block" */
348 xfs_ifork_t *ifp; /* fork structure */
349 int level; /* btree level, for checking */
350 xfs_mount_t *mp; /* file system mount structure */
351 __be64 *pp; /* pointer to block address */
355 ifp = XFS_IFORK_PTR(ip, whichfork);
356 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
357 xfs_bmap_count_leaves(ifp, 0, xfs_iext_count(ifp), count);
362 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
364 block = ifp->if_broot;
365 level = be16_to_cpu(block->bb_level);
367 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
368 bno = be64_to_cpu(*pp);
369 ASSERT(bno != NULLFSBLOCK);
370 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
371 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
373 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
374 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
376 return -EFSCORRUPTED;
383 * returns 1 for success, 0 if we failed to map the extent.
386 xfs_getbmapx_fix_eof_hole(
387 xfs_inode_t *ip, /* xfs incore inode pointer */
389 struct getbmapx *out, /* output structure */
390 int prealloced, /* this is a file with
391 * preallocated data space */
392 __int64_t end, /* last block requested */
393 xfs_fsblock_t startblock,
397 xfs_mount_t *mp; /* file system mount point */
398 xfs_ifork_t *ifp; /* inode fork pointer */
399 xfs_extnum_t lastx; /* last extent pointer */
400 xfs_fileoff_t fileblock;
402 if (startblock == HOLESTARTBLOCK) {
405 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
406 fixlen -= out->bmv_offset;
407 if (prealloced && out->bmv_offset + out->bmv_length == end) {
408 /* Came to hole at EOF. Trim it. */
411 out->bmv_length = fixlen;
414 if (startblock == DELAYSTARTBLOCK)
417 out->bmv_block = xfs_fsb_to_db(ip, startblock);
418 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
419 ifp = XFS_IFORK_PTR(ip, whichfork);
421 xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
422 (lastx == xfs_iext_count(ifp) - 1))
423 out->bmv_oflags |= BMV_OF_LAST;
429 /* Adjust the reported bmap around shared/unshared extent transitions. */
431 xfs_getbmap_adjust_shared(
432 struct xfs_inode *ip,
434 struct xfs_bmbt_irec *map,
435 struct getbmapx *out,
436 struct xfs_bmbt_irec *next_map)
438 struct xfs_mount *mp = ip->i_mount;
446 next_map->br_startblock = NULLFSBLOCK;
447 next_map->br_startoff = NULLFILEOFF;
448 next_map->br_blockcount = 0;
450 /* Only written data blocks can be shared. */
451 if (!xfs_is_reflink_inode(ip) ||
452 whichfork != XFS_DATA_FORK ||
453 !xfs_bmap_is_real_extent(map))
456 agno = XFS_FSB_TO_AGNO(mp, map->br_startblock);
457 agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock);
458 error = xfs_reflink_find_shared(mp, agno, agbno, map->br_blockcount,
463 if (ebno == NULLAGBLOCK) {
464 /* No shared blocks at all. */
466 } else if (agbno == ebno) {
468 * Shared extent at (agbno, elen). Shrink the reported
469 * extent length and prepare to move the start of map[i]
470 * to agbno+elen, with the aim of (re)formatting the new
471 * map[i] the next time through the inner loop.
473 out->bmv_length = XFS_FSB_TO_BB(mp, elen);
474 out->bmv_oflags |= BMV_OF_SHARED;
475 if (elen != map->br_blockcount) {
477 next_map->br_startblock += elen;
478 next_map->br_startoff += elen;
479 next_map->br_blockcount -= elen;
481 map->br_blockcount -= elen;
484 * There's an unshared extent (agbno, ebno - agbno)
485 * followed by shared extent at (ebno, elen). Shrink
486 * the reported extent length to cover only the unshared
487 * extent and prepare to move up the start of map[i] to
488 * ebno, with the aim of (re)formatting the new map[i]
489 * the next time through the inner loop.
493 out->bmv_length = XFS_FSB_TO_BB(mp, nlen);
494 next_map->br_startblock += nlen;
495 next_map->br_startoff += nlen;
496 next_map->br_blockcount -= nlen;
497 map->br_blockcount -= nlen;
504 * Get inode's extents as described in bmv, and format for output.
505 * Calls formatter to fill the user's buffer until all extents
506 * are mapped, until the passed-in bmv->bmv_count slots have
507 * been filled, or until the formatter short-circuits the loop,
508 * if it is tracking filled-in extents on its own.
513 struct getbmapx *bmv, /* user bmap structure */
514 xfs_bmap_format_t formatter, /* format to user */
515 void *arg) /* formatter arg */
517 __int64_t bmvend; /* last block requested */
518 int error = 0; /* return value */
519 __int64_t fixlen; /* length for -1 case */
520 int i; /* extent number */
521 int lock; /* lock state */
522 xfs_bmbt_irec_t *map; /* buffer for user's data */
523 xfs_mount_t *mp; /* file system mount point */
524 int nex; /* # of user extents can do */
525 int subnex; /* # of bmapi's can do */
526 int nmap; /* number of map entries */
527 struct getbmapx *out; /* output structure */
528 int whichfork; /* data or attr fork */
529 int prealloced; /* this is a file with
530 * preallocated data space */
531 int iflags; /* interface flags */
532 int bmapi_flags; /* flags for xfs_bmapi */
534 struct xfs_bmbt_irec inject_map;
537 iflags = bmv->bmv_iflags;
540 /* Only allow CoW fork queries if we're debugging. */
541 if (iflags & BMV_IF_COWFORK)
544 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
547 if (iflags & BMV_IF_ATTRFORK)
548 whichfork = XFS_ATTR_FORK;
549 else if (iflags & BMV_IF_COWFORK)
550 whichfork = XFS_COW_FORK;
552 whichfork = XFS_DATA_FORK;
556 if (XFS_IFORK_Q(ip)) {
557 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
558 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
559 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
562 ip->i_d.di_aformat != 0 &&
563 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
564 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
566 return -EFSCORRUPTED;
573 if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS)
576 if (xfs_get_cowextsz_hint(ip)) {
578 fixlen = mp->m_super->s_maxbytes;
581 fixlen = XFS_ISIZE(ip);
585 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
586 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
587 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
590 if (xfs_get_extsz_hint(ip) ||
591 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
593 fixlen = mp->m_super->s_maxbytes;
596 fixlen = XFS_ISIZE(ip);
601 if (bmv->bmv_length == -1) {
602 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
604 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
605 } else if (bmv->bmv_length == 0) {
606 bmv->bmv_entries = 0;
608 } else if (bmv->bmv_length < 0) {
612 nex = bmv->bmv_count - 1;
615 bmvend = bmv->bmv_offset + bmv->bmv_length;
618 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
620 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
624 xfs_ilock(ip, XFS_IOLOCK_SHARED);
627 if (!(iflags & BMV_IF_DELALLOC) &&
628 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
629 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
631 goto out_unlock_iolock;
634 * Even after flushing the inode, there can still be
635 * delalloc blocks on the inode beyond EOF due to
636 * speculative preallocation. These are not removed
637 * until the release function is called or the inode
638 * is inactivated. Hence we cannot assert here that
639 * ip->i_delayed_blks == 0.
643 lock = xfs_ilock_data_map_shared(ip);
646 lock = XFS_ILOCK_SHARED;
650 lock = xfs_ilock_attr_map_shared(ip);
655 * Don't let nex be bigger than the number of extents
656 * we can have assuming alternating holes and real extents.
658 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
659 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
661 bmapi_flags = xfs_bmapi_aflag(whichfork);
662 if (!(iflags & BMV_IF_PREALLOC))
663 bmapi_flags |= XFS_BMAPI_IGSTATE;
666 * Allocate enough space to handle "subnex" maps at a time.
670 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
672 goto out_unlock_ilock;
674 bmv->bmv_entries = 0;
676 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
677 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
683 nmap = (nex> subnex) ? subnex : nex;
684 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
685 XFS_BB_TO_FSB(mp, bmv->bmv_length),
686 map, &nmap, bmapi_flags);
689 ASSERT(nmap <= subnex);
691 for (i = 0; i < nmap && bmv->bmv_length &&
692 cur_ext < bmv->bmv_count - 1; i++) {
693 out[cur_ext].bmv_oflags = 0;
694 if (map[i].br_state == XFS_EXT_UNWRITTEN)
695 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
696 else if (map[i].br_startblock == DELAYSTARTBLOCK)
697 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
698 out[cur_ext].bmv_offset =
699 XFS_FSB_TO_BB(mp, map[i].br_startoff);
700 out[cur_ext].bmv_length =
701 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
702 out[cur_ext].bmv_unused1 = 0;
703 out[cur_ext].bmv_unused2 = 0;
706 * delayed allocation extents that start beyond EOF can
707 * occur due to speculative EOF allocation when the
708 * delalloc extent is larger than the largest freespace
709 * extent at conversion time. These extents cannot be
710 * converted by data writeback, so can exist here even
711 * if we are not supposed to be finding delalloc
714 if (map[i].br_startblock == DELAYSTARTBLOCK &&
715 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
716 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
718 if (map[i].br_startblock == HOLESTARTBLOCK &&
719 whichfork == XFS_ATTR_FORK) {
720 /* came to the end of attribute fork */
721 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
725 /* Is this a shared block? */
726 error = xfs_getbmap_adjust_shared(ip, whichfork,
727 &map[i], &out[cur_ext], &inject_map);
731 if (!xfs_getbmapx_fix_eof_hole(ip, whichfork,
732 &out[cur_ext], prealloced, bmvend,
733 map[i].br_startblock,
734 inject_map.br_startblock != NULLFSBLOCK))
738 out[cur_ext].bmv_offset +
739 out[cur_ext].bmv_length;
741 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
744 * In case we don't want to return the hole,
745 * don't increase cur_ext so that we can reuse
746 * it in the next loop.
748 if ((iflags & BMV_IF_NO_HOLES) &&
749 map[i].br_startblock == HOLESTARTBLOCK) {
750 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
755 * In order to report shared extents accurately,
756 * we report each distinct shared/unshared part
757 * of a single bmbt record using multiple bmap
758 * extents. To make that happen, we iterate the
759 * same map array item multiple times, each
760 * time trimming out the subextent that we just
763 * Because of this, we must check the out array
764 * index (cur_ext) directly against bmv_count-1
765 * to avoid overflows.
767 if (inject_map.br_startblock != NULLFSBLOCK) {
774 } while (nmap && bmv->bmv_length && cur_ext < bmv->bmv_count - 1);
779 xfs_iunlock(ip, lock);
781 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
783 for (i = 0; i < cur_ext; i++) {
784 /* format results & advance arg */
785 error = formatter(&arg, &out[i]);
795 * dead simple method of punching delalyed allocation blocks from a range in
796 * the inode. Walks a block at a time so will be slow, but is only executed in
797 * rare error cases so the overhead is not critical. This will always punch out
798 * both the start and end blocks, even if the ranges only partially overlap
799 * them, so it is up to the caller to ensure that partial blocks are not
803 xfs_bmap_punch_delalloc_range(
804 struct xfs_inode *ip,
805 xfs_fileoff_t start_fsb,
806 xfs_fileoff_t length)
808 xfs_fileoff_t remaining = length;
811 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
815 xfs_bmbt_irec_t imap;
817 xfs_fsblock_t firstblock;
818 struct xfs_defer_ops dfops;
821 * Map the range first and check that it is a delalloc extent
822 * before trying to unmap the range. Otherwise we will be
823 * trying to remove a real extent (which requires a
824 * transaction) or a hole, which is probably a bad idea...
826 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
830 /* something screwed, just bail */
831 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
832 xfs_alert(ip->i_mount,
833 "Failed delalloc mapping lookup ino %lld fsb %lld.",
834 ip->i_ino, start_fsb);
842 if (imap.br_startblock != DELAYSTARTBLOCK) {
843 /* been converted, ignore */
846 WARN_ON(imap.br_blockcount == 0);
849 * Note: while we initialise the firstblock/dfops pair, they
850 * should never be used because blocks should never be
851 * allocated or freed for a delalloc extent and hence we need
852 * don't cancel or finish them after the xfs_bunmapi() call.
854 xfs_defer_init(&dfops, &firstblock);
855 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
860 ASSERT(!xfs_defer_has_unfinished_work(&dfops));
864 } while(remaining > 0);
870 * Test whether it is appropriate to check an inode for and free post EOF
871 * blocks. The 'force' parameter determines whether we should also consider
872 * regular files that are marked preallocated or append-only.
875 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
877 /* prealloc/delalloc exists only on regular files */
878 if (!S_ISREG(VFS_I(ip)->i_mode))
882 * Zero sized files with no cached pages and delalloc blocks will not
883 * have speculative prealloc/delalloc blocks to remove.
885 if (VFS_I(ip)->i_size == 0 &&
886 VFS_I(ip)->i_mapping->nrpages == 0 &&
887 ip->i_delayed_blks == 0)
890 /* If we haven't read in the extent list, then don't do it now. */
891 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
895 * Do not free real preallocated or append-only files unless the file
896 * has delalloc blocks and we are forced to remove them.
898 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
899 if (!force || ip->i_delayed_blks == 0)
906 * This is called to free any blocks beyond eof. The caller must hold
907 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
908 * reference to the inode.
912 struct xfs_inode *ip)
914 struct xfs_trans *tp;
916 xfs_fileoff_t end_fsb;
917 xfs_fileoff_t last_fsb;
918 xfs_filblks_t map_len;
920 struct xfs_bmbt_irec imap;
921 struct xfs_mount *mp = ip->i_mount;
924 * Figure out if there are any blocks beyond the end
925 * of the file. If not, then there is nothing to do.
927 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
928 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
929 if (last_fsb <= end_fsb)
931 map_len = last_fsb - end_fsb;
934 xfs_ilock(ip, XFS_ILOCK_SHARED);
935 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
936 xfs_iunlock(ip, XFS_ILOCK_SHARED);
939 * If there are blocks after the end of file, truncate the file to its
940 * current size to free them up.
942 if (!error && (nimaps != 0) &&
943 (imap.br_startblock != HOLESTARTBLOCK ||
944 ip->i_delayed_blks)) {
946 * Attach the dquots to the inode up front.
948 error = xfs_qm_dqattach(ip, 0);
952 /* wait on dio to ensure i_size has settled */
953 inode_dio_wait(VFS_I(ip));
955 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
958 ASSERT(XFS_FORCED_SHUTDOWN(mp));
962 xfs_ilock(ip, XFS_ILOCK_EXCL);
963 xfs_trans_ijoin(tp, ip, 0);
966 * Do not update the on-disk file size. If we update the
967 * on-disk file size and then the system crashes before the
968 * contents of the file are flushed to disk then the files
969 * may be full of holes (ie NULL files bug).
971 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
975 * If we get an error at this point we simply don't
976 * bother truncating the file.
978 xfs_trans_cancel(tp);
980 error = xfs_trans_commit(tp);
982 xfs_inode_clear_eofblocks_tag(ip);
985 xfs_iunlock(ip, XFS_ILOCK_EXCL);
991 xfs_alloc_file_space(
992 struct xfs_inode *ip,
997 xfs_mount_t *mp = ip->i_mount;
999 xfs_filblks_t allocated_fsb;
1000 xfs_filblks_t allocatesize_fsb;
1001 xfs_extlen_t extsz, temp;
1002 xfs_fileoff_t startoffset_fsb;
1003 xfs_fsblock_t firstfsb;
1008 xfs_bmbt_irec_t imaps[1], *imapp;
1009 struct xfs_defer_ops dfops;
1010 uint qblocks, resblks, resrtextents;
1013 trace_xfs_alloc_file_space(ip);
1015 if (XFS_FORCED_SHUTDOWN(mp))
1018 error = xfs_qm_dqattach(ip, 0);
1025 rt = XFS_IS_REALTIME_INODE(ip);
1026 extsz = xfs_get_extsz_hint(ip);
1031 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1032 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1035 * Allocate file space until done or until there is an error
1037 while (allocatesize_fsb && !error) {
1041 * Determine space reservations for data/realtime.
1043 if (unlikely(extsz)) {
1044 s = startoffset_fsb;
1047 e = startoffset_fsb + allocatesize_fsb;
1048 if ((temp = do_mod(startoffset_fsb, extsz)))
1050 if ((temp = do_mod(e, extsz)))
1054 e = allocatesize_fsb;
1058 * The transaction reservation is limited to a 32-bit block
1059 * count, hence we need to limit the number of blocks we are
1060 * trying to reserve to avoid an overflow. We can't allocate
1061 * more than @nimaps extents, and an extent is limited on disk
1062 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1064 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1066 resrtextents = qblocks = resblks;
1067 resrtextents /= mp->m_sb.sb_rextsize;
1068 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1069 quota_flag = XFS_QMOPT_RES_RTBLKS;
1072 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1073 quota_flag = XFS_QMOPT_RES_REGBLKS;
1077 * Allocate and setup the transaction.
1079 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1080 resrtextents, 0, &tp);
1083 * Check for running out of space
1087 * Free the transaction structure.
1089 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1092 xfs_ilock(ip, XFS_ILOCK_EXCL);
1093 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1098 xfs_trans_ijoin(tp, ip, 0);
1100 xfs_defer_init(&dfops, &firstfsb);
1101 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1102 allocatesize_fsb, alloc_type, &firstfsb,
1103 resblks, imapp, &nimaps, &dfops);
1108 * Complete the transaction
1110 error = xfs_defer_finish(&tp, &dfops, NULL);
1114 error = xfs_trans_commit(tp);
1115 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1119 allocated_fsb = imapp->br_blockcount;
1126 startoffset_fsb += allocated_fsb;
1127 allocatesize_fsb -= allocated_fsb;
1132 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1133 xfs_defer_cancel(&dfops);
1134 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1136 error1: /* Just cancel transaction */
1137 xfs_trans_cancel(tp);
1138 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1144 struct xfs_inode *ip,
1145 xfs_fileoff_t startoffset_fsb,
1146 xfs_filblks_t len_fsb,
1149 struct xfs_mount *mp = ip->i_mount;
1150 struct xfs_trans *tp;
1151 struct xfs_defer_ops dfops;
1152 xfs_fsblock_t firstfsb;
1153 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1156 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1158 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1162 xfs_ilock(ip, XFS_ILOCK_EXCL);
1163 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1164 ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1166 goto out_trans_cancel;
1168 xfs_trans_ijoin(tp, ip, 0);
1170 xfs_defer_init(&dfops, &firstfsb);
1171 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb,
1174 goto out_bmap_cancel;
1176 error = xfs_defer_finish(&tp, &dfops, ip);
1178 goto out_bmap_cancel;
1180 error = xfs_trans_commit(tp);
1182 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1186 xfs_defer_cancel(&dfops);
1188 xfs_trans_cancel(tp);
1193 xfs_adjust_extent_unmap_boundaries(
1194 struct xfs_inode *ip,
1195 xfs_fileoff_t *startoffset_fsb,
1196 xfs_fileoff_t *endoffset_fsb)
1198 struct xfs_mount *mp = ip->i_mount;
1199 struct xfs_bmbt_irec imap;
1201 xfs_extlen_t mod = 0;
1204 error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0);
1208 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1209 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1210 mod = do_mod(imap.br_startblock, mp->m_sb.sb_rextsize);
1212 *startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1216 error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0);
1220 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1221 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1223 if (mod && mod != mp->m_sb.sb_rextsize)
1224 *endoffset_fsb -= mod;
1231 xfs_flush_unmap_range(
1232 struct xfs_inode *ip,
1236 struct xfs_mount *mp = ip->i_mount;
1237 struct inode *inode = VFS_I(ip);
1238 xfs_off_t rounding, start, end;
1241 /* wait for the completion of any pending DIOs */
1242 inode_dio_wait(inode);
1244 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1245 start = round_down(offset, rounding);
1246 end = round_up(offset + len, rounding) - 1;
1248 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1251 truncate_pagecache_range(inode, start, end);
1256 xfs_free_file_space(
1257 struct xfs_inode *ip,
1261 struct xfs_mount *mp = ip->i_mount;
1262 xfs_fileoff_t startoffset_fsb;
1263 xfs_fileoff_t endoffset_fsb;
1264 int done = 0, error;
1266 trace_xfs_free_file_space(ip);
1268 error = xfs_qm_dqattach(ip, 0);
1272 if (len <= 0) /* if nothing being freed */
1275 error = xfs_flush_unmap_range(ip, offset, len);
1279 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1280 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1283 * Need to zero the stuff we're not freeing, on disk. If it's a RT file
1284 * and we can't use unwritten extents then we actually need to ensure
1285 * to zero the whole extent, otherwise we just need to take of block
1286 * boundaries, and xfs_bunmapi will handle the rest.
1288 if (XFS_IS_REALTIME_INODE(ip) &&
1289 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1290 error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb,
1296 if (endoffset_fsb > startoffset_fsb) {
1298 error = xfs_unmap_extent(ip, startoffset_fsb,
1299 endoffset_fsb - startoffset_fsb, &done);
1306 * Now that we've unmap all full blocks we'll have to zero out any
1307 * partial block at the beginning and/or end. xfs_zero_range is
1308 * smart enough to skip any holes, including those we just created,
1309 * but we must take care not to zero beyond EOF and enlarge i_size.
1312 if (offset >= XFS_ISIZE(ip))
1315 if (offset + len > XFS_ISIZE(ip))
1316 len = XFS_ISIZE(ip) - offset;
1318 return xfs_zero_range(ip, offset, len, NULL);
1322 * Preallocate and zero a range of a file. This mechanism has the allocation
1323 * semantics of fallocate and in addition converts data in the range to zeroes.
1326 xfs_zero_file_space(
1327 struct xfs_inode *ip,
1331 struct xfs_mount *mp = ip->i_mount;
1335 trace_xfs_zero_file_space(ip);
1337 blksize = 1 << mp->m_sb.sb_blocklog;
1340 * Punch a hole and prealloc the range. We use hole punch rather than
1341 * unwritten extent conversion for two reasons:
1343 * 1.) Hole punch handles partial block zeroing for us.
1345 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1346 * by virtue of the hole punch.
1348 error = xfs_free_file_space(ip, offset, len);
1352 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1353 round_up(offset + len, blksize) -
1354 round_down(offset, blksize),
1355 XFS_BMAPI_PREALLOC);
1362 * @next_fsb will keep track of the extent currently undergoing shift.
1363 * @stop_fsb will keep track of the extent at which we have to stop.
1364 * If we are shifting left, we will start with block (offset + len) and
1365 * shift each extent till last extent.
1366 * If we are shifting right, we will start with last extent inside file space
1367 * and continue until we reach the block corresponding to offset.
1370 xfs_shift_file_space(
1371 struct xfs_inode *ip,
1374 enum shift_direction direction)
1377 struct xfs_mount *mp = ip->i_mount;
1378 struct xfs_trans *tp;
1380 struct xfs_defer_ops dfops;
1381 xfs_fsblock_t first_block;
1382 xfs_fileoff_t stop_fsb;
1383 xfs_fileoff_t next_fsb;
1384 xfs_fileoff_t shift_fsb;
1387 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1389 if (direction == SHIFT_LEFT) {
1391 * Reserve blocks to cover potential extent merges after left
1394 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1395 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1396 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1399 * If right shift, delegate the work of initialization of
1400 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1403 next_fsb = NULLFSBLOCK;
1404 stop_fsb = XFS_B_TO_FSB(mp, offset);
1407 shift_fsb = XFS_B_TO_FSB(mp, len);
1410 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1411 * into the accessible region of the file.
1413 if (xfs_can_free_eofblocks(ip, true)) {
1414 error = xfs_free_eofblocks(ip);
1420 * Writeback and invalidate cache for the remainder of the file as we're
1421 * about to shift down every extent from offset to EOF.
1423 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1427 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1428 offset >> PAGE_SHIFT, -1);
1433 * The extent shiting code works on extent granularity. So, if
1434 * stop_fsb is not the starting block of extent, we need to split
1435 * the extent at stop_fsb.
1437 if (direction == SHIFT_RIGHT) {
1438 error = xfs_bmap_split_extent(ip, stop_fsb);
1443 while (!error && !done) {
1444 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0,
1449 xfs_ilock(ip, XFS_ILOCK_EXCL);
1450 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1451 ip->i_gdquot, ip->i_pdquot, resblks, 0,
1452 XFS_QMOPT_RES_REGBLKS);
1454 goto out_trans_cancel;
1456 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1458 xfs_defer_init(&dfops, &first_block);
1461 * We are using the write transaction in which max 2 bmbt
1462 * updates are allowed
1464 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1465 &done, stop_fsb, &first_block, &dfops,
1466 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1468 goto out_bmap_cancel;
1470 error = xfs_defer_finish(&tp, &dfops, NULL);
1472 goto out_bmap_cancel;
1474 error = xfs_trans_commit(tp);
1480 xfs_defer_cancel(&dfops);
1482 xfs_trans_cancel(tp);
1487 * xfs_collapse_file_space()
1488 * This routine frees disk space and shift extent for the given file.
1489 * The first thing we do is to free data blocks in the specified range
1490 * by calling xfs_free_file_space(). It would also sync dirty data
1491 * and invalidate page cache over the region on which collapse range
1492 * is working. And Shift extent records to the left to cover a hole.
1499 xfs_collapse_file_space(
1500 struct xfs_inode *ip,
1506 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1507 trace_xfs_collapse_file_space(ip);
1509 error = xfs_free_file_space(ip, offset, len);
1513 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1517 * xfs_insert_file_space()
1518 * This routine create hole space by shifting extents for the given file.
1519 * The first thing we do is to sync dirty data and invalidate page cache
1520 * over the region on which insert range is working. And split an extent
1521 * to two extents at given offset by calling xfs_bmap_split_extent.
1522 * And shift all extent records which are laying between [offset,
1523 * last allocated extent] to the right to reserve hole range.
1529 xfs_insert_file_space(
1530 struct xfs_inode *ip,
1534 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1535 trace_xfs_insert_file_space(ip);
1537 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1541 * We need to check that the format of the data fork in the temporary inode is
1542 * valid for the target inode before doing the swap. This is not a problem with
1543 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1544 * data fork depending on the space the attribute fork is taking so we can get
1545 * invalid formats on the target inode.
1547 * E.g. target has space for 7 extents in extent format, temp inode only has
1548 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1549 * btree, but when swapped it needs to be in extent format. Hence we can't just
1550 * blindly swap data forks on attr2 filesystems.
1552 * Note that we check the swap in both directions so that we don't end up with
1553 * a corrupt temporary inode, either.
1555 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1556 * inode will prevent this situation from occurring, so all we do here is
1557 * reject and log the attempt. basically we are putting the responsibility on
1558 * userspace to get this right.
1561 xfs_swap_extents_check_format(
1562 struct xfs_inode *ip, /* target inode */
1563 struct xfs_inode *tip) /* tmp inode */
1566 /* Should never get a local format */
1567 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1568 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1572 * if the target inode has less extents that then temporary inode then
1573 * why did userspace call us?
1575 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1579 * If we have to use the (expensive) rmap swap method, we can
1580 * handle any number of extents and any format.
1582 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1586 * if the target inode is in extent form and the temp inode is in btree
1587 * form then we will end up with the target inode in the wrong format
1588 * as we already know there are less extents in the temp inode.
1590 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1591 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1594 /* Check temp in extent form to max in target */
1595 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1596 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1597 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1600 /* Check target in extent form to max in temp */
1601 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1602 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1603 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1607 * If we are in a btree format, check that the temp root block will fit
1608 * in the target and that it has enough extents to be in btree format
1611 * Note that we have to be careful to allow btree->extent conversions
1612 * (a common defrag case) which will occur when the temp inode is in
1615 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1616 if (XFS_IFORK_BOFF(ip) &&
1617 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1619 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1620 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1624 /* Reciprocal target->temp btree format checks */
1625 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1626 if (XFS_IFORK_BOFF(tip) &&
1627 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1629 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1630 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1638 xfs_swap_extent_flush(
1639 struct xfs_inode *ip)
1643 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1646 truncate_pagecache_range(VFS_I(ip), 0, -1);
1648 /* Verify O_DIRECT for ftmp */
1649 if (VFS_I(ip)->i_mapping->nrpages)
1655 * Move extents from one file to another, when rmap is enabled.
1658 xfs_swap_extent_rmap(
1659 struct xfs_trans **tpp,
1660 struct xfs_inode *ip,
1661 struct xfs_inode *tip)
1663 struct xfs_bmbt_irec irec;
1664 struct xfs_bmbt_irec uirec;
1665 struct xfs_bmbt_irec tirec;
1666 xfs_fileoff_t offset_fsb;
1667 xfs_fileoff_t end_fsb;
1668 xfs_filblks_t count_fsb;
1669 xfs_fsblock_t firstfsb;
1670 struct xfs_defer_ops dfops;
1675 __uint64_t tip_flags2;
1678 * If the source file has shared blocks, we must flag the donor
1679 * file as having shared blocks so that we get the shared-block
1680 * rmap functions when we go to fix up the rmaps. The flags
1681 * will be switch for reals later.
1683 tip_flags2 = tip->i_d.di_flags2;
1684 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1685 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1688 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1689 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1692 /* Read extent from the donor file */
1694 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1698 ASSERT(nimaps == 1);
1699 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1701 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1702 ilen = tirec.br_blockcount;
1704 /* Unmap the old blocks in the source file. */
1705 while (tirec.br_blockcount) {
1706 xfs_defer_init(&dfops, &firstfsb);
1707 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1709 /* Read extent from the source file */
1711 error = xfs_bmapi_read(ip, tirec.br_startoff,
1712 tirec.br_blockcount, &irec,
1716 ASSERT(nimaps == 1);
1717 ASSERT(tirec.br_startoff == irec.br_startoff);
1718 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1720 /* Trim the extent. */
1722 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1723 tirec.br_blockcount,
1724 irec.br_blockcount);
1725 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1727 /* Remove the mapping from the donor file. */
1728 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1733 /* Remove the mapping from the source file. */
1734 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1739 /* Map the donor file's blocks into the source file. */
1740 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1745 /* Map the source file's blocks into the donor file. */
1746 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1751 error = xfs_defer_finish(tpp, &dfops, ip);
1755 tirec.br_startoff += rlen;
1756 if (tirec.br_startblock != HOLESTARTBLOCK &&
1757 tirec.br_startblock != DELAYSTARTBLOCK)
1758 tirec.br_startblock += rlen;
1759 tirec.br_blockcount -= rlen;
1767 tip->i_d.di_flags2 = tip_flags2;
1771 xfs_defer_cancel(&dfops);
1773 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1774 tip->i_d.di_flags2 = tip_flags2;
1778 /* Swap the extents of two files by swapping data forks. */
1780 xfs_swap_extent_forks(
1781 struct xfs_trans *tp,
1782 struct xfs_inode *ip,
1783 struct xfs_inode *tip,
1785 int *target_log_flags)
1787 struct xfs_ifork tempifp, *ifp, *tifp;
1790 xfs_extnum_t nextents;
1795 * Count the number of extended attribute blocks
1797 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1798 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1799 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK,
1804 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1805 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1806 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1813 * Before we've swapped the forks, lets set the owners of the forks
1814 * appropriately. We have to do this as we are demand paging the btree
1815 * buffers, and so the validation done on read will expect the owner
1816 * field to be correctly set. Once we change the owners, we can swap the
1819 if (ip->i_d.di_version == 3 &&
1820 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1821 (*target_log_flags) |= XFS_ILOG_DOWNER;
1822 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1828 if (tip->i_d.di_version == 3 &&
1829 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1830 (*src_log_flags) |= XFS_ILOG_DOWNER;
1831 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1838 * Swap the data forks of the inodes
1842 tempifp = *ifp; /* struct copy */
1843 *ifp = *tifp; /* struct copy */
1844 *tifp = tempifp; /* struct copy */
1847 * Fix the on-disk inode values
1849 tmp = (__uint64_t)ip->i_d.di_nblocks;
1850 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1851 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1853 tmp = (__uint64_t) ip->i_d.di_nextents;
1854 ip->i_d.di_nextents = tip->i_d.di_nextents;
1855 tip->i_d.di_nextents = tmp;
1857 tmp = (__uint64_t) ip->i_d.di_format;
1858 ip->i_d.di_format = tip->i_d.di_format;
1859 tip->i_d.di_format = tmp;
1862 * The extents in the source inode could still contain speculative
1863 * preallocation beyond EOF (e.g. the file is open but not modified
1864 * while defrag is in progress). In that case, we need to copy over the
1865 * number of delalloc blocks the data fork in the source inode is
1866 * tracking beyond EOF so that when the fork is truncated away when the
1867 * temporary inode is unlinked we don't underrun the i_delayed_blks
1868 * counter on that inode.
1870 ASSERT(tip->i_delayed_blks == 0);
1871 tip->i_delayed_blks = ip->i_delayed_blks;
1872 ip->i_delayed_blks = 0;
1874 switch (ip->i_d.di_format) {
1875 case XFS_DINODE_FMT_EXTENTS:
1877 * If the extents fit in the inode, fix the pointer. Otherwise
1878 * it's already NULL or pointing to the extent.
1880 nextents = xfs_iext_count(&ip->i_df);
1881 if (nextents <= XFS_INLINE_EXTS)
1882 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
1883 (*src_log_flags) |= XFS_ILOG_DEXT;
1885 case XFS_DINODE_FMT_BTREE:
1886 ASSERT(ip->i_d.di_version < 3 ||
1887 (*src_log_flags & XFS_ILOG_DOWNER));
1888 (*src_log_flags) |= XFS_ILOG_DBROOT;
1892 switch (tip->i_d.di_format) {
1893 case XFS_DINODE_FMT_EXTENTS:
1895 * If the extents fit in the inode, fix the pointer. Otherwise
1896 * it's already NULL or pointing to the extent.
1898 nextents = xfs_iext_count(&tip->i_df);
1899 if (nextents <= XFS_INLINE_EXTS)
1900 tifp->if_u1.if_extents = tifp->if_u2.if_inline_ext;
1901 (*target_log_flags) |= XFS_ILOG_DEXT;
1903 case XFS_DINODE_FMT_BTREE:
1904 (*target_log_flags) |= XFS_ILOG_DBROOT;
1905 ASSERT(tip->i_d.di_version < 3 ||
1906 (*target_log_flags & XFS_ILOG_DOWNER));
1915 struct xfs_inode *ip, /* target inode */
1916 struct xfs_inode *tip, /* tmp inode */
1917 struct xfs_swapext *sxp)
1919 struct xfs_mount *mp = ip->i_mount;
1920 struct xfs_trans *tp;
1921 struct xfs_bstat *sbp = &sxp->sx_stat;
1922 int src_log_flags, target_log_flags;
1925 struct xfs_ifork *cowfp;
1930 * Lock the inodes against other IO, page faults and truncate to
1931 * begin with. Then we can ensure the inodes are flushed and have no
1932 * page cache safely. Once we have done this we can take the ilocks and
1933 * do the rest of the checks.
1935 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1936 lock_flags = XFS_MMAPLOCK_EXCL;
1937 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1939 /* Verify that both files have the same format */
1940 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1945 /* Verify both files are either real-time or non-realtime */
1946 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1951 error = xfs_swap_extent_flush(ip);
1954 error = xfs_swap_extent_flush(tip);
1959 * Extent "swapping" with rmap requires a permanent reservation and
1960 * a block reservation because it's really just a remap operation
1961 * performed with log redo items!
1963 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1965 * Conceptually this shouldn't affect the shape of either
1966 * bmbt, but since we atomically move extents one by one,
1967 * we reserve enough space to rebuild both trees.
1969 resblks = XFS_SWAP_RMAP_SPACE_RES(mp,
1970 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK),
1972 XFS_SWAP_RMAP_SPACE_RES(mp,
1973 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK),
1975 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1978 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0,
1984 * Lock and join the inodes to the tansaction so that transaction commit
1985 * or cancel will unlock the inodes from this point onwards.
1987 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1988 lock_flags |= XFS_ILOCK_EXCL;
1989 xfs_trans_ijoin(tp, ip, 0);
1990 xfs_trans_ijoin(tp, tip, 0);
1993 /* Verify all data are being swapped */
1994 if (sxp->sx_offset != 0 ||
1995 sxp->sx_length != ip->i_d.di_size ||
1996 sxp->sx_length != tip->i_d.di_size) {
1998 goto out_trans_cancel;
2001 trace_xfs_swap_extent_before(ip, 0);
2002 trace_xfs_swap_extent_before(tip, 1);
2004 /* check inode formats now that data is flushed */
2005 error = xfs_swap_extents_check_format(ip, tip);
2008 "%s: inode 0x%llx format is incompatible for exchanging.",
2009 __func__, ip->i_ino);
2010 goto out_trans_cancel;
2014 * Compare the current change & modify times with that
2015 * passed in. If they differ, we abort this swap.
2016 * This is the mechanism used to ensure the calling
2017 * process that the file was not changed out from
2020 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
2021 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
2022 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
2023 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
2025 goto out_trans_cancel;
2029 * Note the trickiness in setting the log flags - we set the owner log
2030 * flag on the opposite inode (i.e. the inode we are setting the new
2031 * owner to be) because once we swap the forks and log that, log
2032 * recovery is going to see the fork as owned by the swapped inode,
2033 * not the pre-swapped inodes.
2035 src_log_flags = XFS_ILOG_CORE;
2036 target_log_flags = XFS_ILOG_CORE;
2038 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2039 error = xfs_swap_extent_rmap(&tp, ip, tip);
2041 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
2044 goto out_trans_cancel;
2046 /* Do we have to swap reflink flags? */
2047 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
2048 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
2049 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2050 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2051 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2052 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2053 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
2054 cowfp = ip->i_cowfp;
2055 ip->i_cowfp = tip->i_cowfp;
2056 tip->i_cowfp = cowfp;
2057 xfs_inode_set_cowblocks_tag(ip);
2058 xfs_inode_set_cowblocks_tag(tip);
2061 xfs_trans_log_inode(tp, ip, src_log_flags);
2062 xfs_trans_log_inode(tp, tip, target_log_flags);
2065 * If this is a synchronous mount, make sure that the
2066 * transaction goes to disk before returning to the user.
2068 if (mp->m_flags & XFS_MOUNT_WSYNC)
2069 xfs_trans_set_sync(tp);
2071 error = xfs_trans_commit(tp);
2073 trace_xfs_swap_extent_after(ip, 0);
2074 trace_xfs_swap_extent_after(tip, 1);
2077 xfs_iunlock(ip, lock_flags);
2078 xfs_iunlock(tip, lock_flags);
2079 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
2083 xfs_trans_cancel(tp);