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 xfs_alloctype_t atype = 0; /* type for allocation routines */
92 int error; /* error return value */
93 xfs_mount_t *mp; /* mount point structure */
94 xfs_extlen_t prod = 0; /* product factor for allocators */
95 xfs_extlen_t ralen = 0; /* realtime allocation length */
96 xfs_extlen_t align; /* minimum allocation alignment */
100 align = xfs_get_extsz_hint(ap->ip);
101 prod = align / mp->m_sb.sb_rextsize;
102 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
103 align, 1, ap->eof, 0,
104 ap->conv, &ap->offset, &ap->length);
108 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
111 * If the offset & length are not perfectly aligned
112 * then kill prod, it will just get us in trouble.
114 if (do_mod(ap->offset, align) || ap->length % align)
117 * Set ralen to be the actual requested length in rtextents.
119 ralen = ap->length / mp->m_sb.sb_rextsize;
121 * If the old value was close enough to MAXEXTLEN that
122 * we rounded up to it, cut it back so it's valid again.
123 * Note that if it's a really large request (bigger than
124 * MAXEXTLEN), we don't hear about that number, and can't
125 * adjust the starting point to match it.
127 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
128 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
131 * Lock out modifications to both the RT bitmap and summary inodes
133 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
134 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
135 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
136 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
139 * If it's an allocation to an empty file at offset 0,
140 * pick an extent that will space things out in the rt area.
142 if (ap->eof && ap->offset == 0) {
143 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
145 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
148 ap->blkno = rtx * mp->m_sb.sb_rextsize;
153 xfs_bmap_adjacent(ap);
156 * Realtime allocation, done through xfs_rtallocate_extent.
158 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
159 do_div(ap->blkno, mp->m_sb.sb_rextsize);
162 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
163 &ralen, atype, ap->wasdel, prod, &rtb)))
165 if (rtb == NULLFSBLOCK && prod > 1 &&
166 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
167 ap->length, &ralen, atype,
168 ap->wasdel, 1, &rtb)))
171 if (ap->blkno != NULLFSBLOCK) {
172 ap->blkno *= mp->m_sb.sb_rextsize;
173 ralen *= mp->m_sb.sb_rextsize;
175 ap->ip->i_d.di_nblocks += ralen;
176 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
178 ap->ip->i_delayed_blks -= ralen;
180 * Adjust the disk quota also. This was reserved
183 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
184 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
185 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
187 /* Zero the extent if we were asked to do so */
188 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
189 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
200 * Check if the endoff is outside the last extent. If so the caller will grow
201 * the allocation to a stripe unit boundary. All offsets are considered outside
202 * the end of file for an empty fork, so 1 is returned in *eof in that case.
206 struct xfs_inode *ip,
207 xfs_fileoff_t endoff,
211 struct xfs_bmbt_irec rec;
214 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
218 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
223 * Extent tree block counting routines.
227 * Count leaf blocks given a range of extent records.
230 xfs_bmap_count_leaves(
238 for (b = 0; b < numrecs; b++) {
239 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
240 *count += xfs_bmbt_get_blockcount(frp);
245 * Count leaf blocks given a range of extent records originally
249 xfs_bmap_disk_count_leaves(
250 struct xfs_mount *mp,
251 struct xfs_btree_block *block,
258 for (b = 1; b <= numrecs; b++) {
259 frp = XFS_BMBT_REC_ADDR(mp, block, b);
260 *count += xfs_bmbt_disk_get_blockcount(frp);
265 * Recursively walks each level of a btree
266 * to count total fsblocks in use.
268 STATIC int /* error */
270 xfs_mount_t *mp, /* file system mount point */
271 xfs_trans_t *tp, /* transaction pointer */
272 xfs_ifork_t *ifp, /* inode fork pointer */
273 xfs_fsblock_t blockno, /* file system block number */
274 int levelin, /* level in btree */
275 int *count) /* Count of blocks */
281 xfs_fsblock_t bno = blockno;
282 xfs_fsblock_t nextbno;
283 struct xfs_btree_block *block, *nextblock;
286 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
291 block = XFS_BUF_TO_BLOCK(bp);
294 /* Not at node above leaves, count this level of nodes */
295 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
296 while (nextbno != NULLFSBLOCK) {
297 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
303 nextblock = XFS_BUF_TO_BLOCK(nbp);
304 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
305 xfs_trans_brelse(tp, nbp);
308 /* Dive to the next level */
309 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
310 bno = be64_to_cpu(*pp);
311 if (unlikely((error =
312 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
313 xfs_trans_brelse(tp, bp);
314 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
315 XFS_ERRLEVEL_LOW, mp);
316 return -EFSCORRUPTED;
318 xfs_trans_brelse(tp, bp);
320 /* count all level 1 nodes and their leaves */
322 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
323 numrecs = be16_to_cpu(block->bb_numrecs);
324 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
325 xfs_trans_brelse(tp, bp);
326 if (nextbno == NULLFSBLOCK)
329 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
335 block = XFS_BUF_TO_BLOCK(bp);
342 * Count fsblocks of the given fork.
344 static int /* error */
345 xfs_bmap_count_blocks(
346 xfs_trans_t *tp, /* transaction pointer */
347 xfs_inode_t *ip, /* incore inode */
348 int whichfork, /* data or attr fork */
349 int *count) /* out: count of blocks */
351 struct xfs_btree_block *block; /* current btree block */
352 xfs_fsblock_t bno; /* block # of "block" */
353 xfs_ifork_t *ifp; /* fork structure */
354 int level; /* btree level, for checking */
355 xfs_mount_t *mp; /* file system mount structure */
356 __be64 *pp; /* pointer to block address */
360 ifp = XFS_IFORK_PTR(ip, whichfork);
361 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
362 xfs_bmap_count_leaves(ifp, 0, xfs_iext_count(ifp), count);
367 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
369 block = ifp->if_broot;
370 level = be16_to_cpu(block->bb_level);
372 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
373 bno = be64_to_cpu(*pp);
374 ASSERT(bno != NULLFSBLOCK);
375 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
376 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
378 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
379 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
381 return -EFSCORRUPTED;
388 * returns 1 for success, 0 if we failed to map the extent.
391 xfs_getbmapx_fix_eof_hole(
392 xfs_inode_t *ip, /* xfs incore inode pointer */
394 struct getbmapx *out, /* output structure */
395 int prealloced, /* this is a file with
396 * preallocated data space */
397 __int64_t end, /* last block requested */
398 xfs_fsblock_t startblock,
402 xfs_mount_t *mp; /* file system mount point */
403 xfs_ifork_t *ifp; /* inode fork pointer */
404 xfs_extnum_t lastx; /* last extent pointer */
405 xfs_fileoff_t fileblock;
407 if (startblock == HOLESTARTBLOCK) {
410 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
411 fixlen -= out->bmv_offset;
412 if (prealloced && out->bmv_offset + out->bmv_length == end) {
413 /* Came to hole at EOF. Trim it. */
416 out->bmv_length = fixlen;
419 if (startblock == DELAYSTARTBLOCK)
422 out->bmv_block = xfs_fsb_to_db(ip, startblock);
423 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
424 ifp = XFS_IFORK_PTR(ip, whichfork);
426 xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
427 (lastx == xfs_iext_count(ifp) - 1))
428 out->bmv_oflags |= BMV_OF_LAST;
434 /* Adjust the reported bmap around shared/unshared extent transitions. */
436 xfs_getbmap_adjust_shared(
437 struct xfs_inode *ip,
439 struct xfs_bmbt_irec *map,
440 struct getbmapx *out,
441 struct xfs_bmbt_irec *next_map)
443 struct xfs_mount *mp = ip->i_mount;
451 next_map->br_startblock = NULLFSBLOCK;
452 next_map->br_startoff = NULLFILEOFF;
453 next_map->br_blockcount = 0;
455 /* Only written data blocks can be shared. */
456 if (!xfs_is_reflink_inode(ip) || whichfork != XFS_DATA_FORK ||
457 map->br_startblock == DELAYSTARTBLOCK ||
458 map->br_startblock == HOLESTARTBLOCK ||
462 agno = XFS_FSB_TO_AGNO(mp, map->br_startblock);
463 agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock);
464 error = xfs_reflink_find_shared(mp, agno, agbno, map->br_blockcount,
469 if (ebno == NULLAGBLOCK) {
470 /* No shared blocks at all. */
472 } else if (agbno == ebno) {
474 * Shared extent at (agbno, elen). Shrink the reported
475 * extent length and prepare to move the start of map[i]
476 * to agbno+elen, with the aim of (re)formatting the new
477 * map[i] the next time through the inner loop.
479 out->bmv_length = XFS_FSB_TO_BB(mp, elen);
480 out->bmv_oflags |= BMV_OF_SHARED;
481 if (elen != map->br_blockcount) {
483 next_map->br_startblock += elen;
484 next_map->br_startoff += elen;
485 next_map->br_blockcount -= elen;
487 map->br_blockcount -= elen;
490 * There's an unshared extent (agbno, ebno - agbno)
491 * followed by shared extent at (ebno, elen). Shrink
492 * the reported extent length to cover only the unshared
493 * extent and prepare to move up the start of map[i] to
494 * ebno, with the aim of (re)formatting the new map[i]
495 * the next time through the inner loop.
499 out->bmv_length = XFS_FSB_TO_BB(mp, nlen);
500 next_map->br_startblock += nlen;
501 next_map->br_startoff += nlen;
502 next_map->br_blockcount -= nlen;
503 map->br_blockcount -= nlen;
510 * Get inode's extents as described in bmv, and format for output.
511 * Calls formatter to fill the user's buffer until all extents
512 * are mapped, until the passed-in bmv->bmv_count slots have
513 * been filled, or until the formatter short-circuits the loop,
514 * if it is tracking filled-in extents on its own.
519 struct getbmapx *bmv, /* user bmap structure */
520 xfs_bmap_format_t formatter, /* format to user */
521 void *arg) /* formatter arg */
523 __int64_t bmvend; /* last block requested */
524 int error = 0; /* return value */
525 __int64_t fixlen; /* length for -1 case */
526 int i; /* extent number */
527 int lock; /* lock state */
528 xfs_bmbt_irec_t *map; /* buffer for user's data */
529 xfs_mount_t *mp; /* file system mount point */
530 int nex; /* # of user extents can do */
531 int subnex; /* # of bmapi's can do */
532 int nmap; /* number of map entries */
533 struct getbmapx *out; /* output structure */
534 int whichfork; /* data or attr fork */
535 int prealloced; /* this is a file with
536 * preallocated data space */
537 int iflags; /* interface flags */
538 int bmapi_flags; /* flags for xfs_bmapi */
540 struct xfs_bmbt_irec inject_map;
543 iflags = bmv->bmv_iflags;
546 /* Only allow CoW fork queries if we're debugging. */
547 if (iflags & BMV_IF_COWFORK)
550 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
553 if (iflags & BMV_IF_ATTRFORK)
554 whichfork = XFS_ATTR_FORK;
555 else if (iflags & BMV_IF_COWFORK)
556 whichfork = XFS_COW_FORK;
558 whichfork = XFS_DATA_FORK;
562 if (XFS_IFORK_Q(ip)) {
563 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
564 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
565 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
568 ip->i_d.di_aformat != 0 &&
569 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
570 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
572 return -EFSCORRUPTED;
579 if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS)
582 if (xfs_get_cowextsz_hint(ip)) {
584 fixlen = mp->m_super->s_maxbytes;
587 fixlen = XFS_ISIZE(ip);
591 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
592 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
593 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
596 if (xfs_get_extsz_hint(ip) ||
597 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
599 fixlen = mp->m_super->s_maxbytes;
602 fixlen = XFS_ISIZE(ip);
607 if (bmv->bmv_length == -1) {
608 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
610 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
611 } else if (bmv->bmv_length == 0) {
612 bmv->bmv_entries = 0;
614 } else if (bmv->bmv_length < 0) {
618 nex = bmv->bmv_count - 1;
621 bmvend = bmv->bmv_offset + bmv->bmv_length;
624 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
626 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
630 xfs_ilock(ip, XFS_IOLOCK_SHARED);
633 if (!(iflags & BMV_IF_DELALLOC) &&
634 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
635 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
637 goto out_unlock_iolock;
640 * Even after flushing the inode, there can still be
641 * delalloc blocks on the inode beyond EOF due to
642 * speculative preallocation. These are not removed
643 * until the release function is called or the inode
644 * is inactivated. Hence we cannot assert here that
645 * ip->i_delayed_blks == 0.
649 lock = xfs_ilock_data_map_shared(ip);
652 lock = XFS_ILOCK_SHARED;
656 lock = xfs_ilock_attr_map_shared(ip);
661 * Don't let nex be bigger than the number of extents
662 * we can have assuming alternating holes and real extents.
664 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
665 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
667 bmapi_flags = xfs_bmapi_aflag(whichfork);
668 if (!(iflags & BMV_IF_PREALLOC))
669 bmapi_flags |= XFS_BMAPI_IGSTATE;
672 * Allocate enough space to handle "subnex" maps at a time.
676 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
678 goto out_unlock_ilock;
680 bmv->bmv_entries = 0;
682 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
683 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
689 nmap = (nex> subnex) ? subnex : nex;
690 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
691 XFS_BB_TO_FSB(mp, bmv->bmv_length),
692 map, &nmap, bmapi_flags);
695 ASSERT(nmap <= subnex);
697 for (i = 0; i < nmap && bmv->bmv_length &&
698 cur_ext < bmv->bmv_count - 1; i++) {
699 out[cur_ext].bmv_oflags = 0;
700 if (map[i].br_state == XFS_EXT_UNWRITTEN)
701 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
702 else if (map[i].br_startblock == DELAYSTARTBLOCK)
703 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
704 out[cur_ext].bmv_offset =
705 XFS_FSB_TO_BB(mp, map[i].br_startoff);
706 out[cur_ext].bmv_length =
707 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
708 out[cur_ext].bmv_unused1 = 0;
709 out[cur_ext].bmv_unused2 = 0;
712 * delayed allocation extents that start beyond EOF can
713 * occur due to speculative EOF allocation when the
714 * delalloc extent is larger than the largest freespace
715 * extent at conversion time. These extents cannot be
716 * converted by data writeback, so can exist here even
717 * if we are not supposed to be finding delalloc
720 if (map[i].br_startblock == DELAYSTARTBLOCK &&
721 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
722 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
724 if (map[i].br_startblock == HOLESTARTBLOCK &&
725 whichfork == XFS_ATTR_FORK) {
726 /* came to the end of attribute fork */
727 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
731 /* Is this a shared block? */
732 error = xfs_getbmap_adjust_shared(ip, whichfork,
733 &map[i], &out[cur_ext], &inject_map);
737 if (!xfs_getbmapx_fix_eof_hole(ip, whichfork,
738 &out[cur_ext], prealloced, bmvend,
739 map[i].br_startblock,
740 inject_map.br_startblock != NULLFSBLOCK))
744 out[cur_ext].bmv_offset +
745 out[cur_ext].bmv_length;
747 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
750 * In case we don't want to return the hole,
751 * don't increase cur_ext so that we can reuse
752 * it in the next loop.
754 if ((iflags & BMV_IF_NO_HOLES) &&
755 map[i].br_startblock == HOLESTARTBLOCK) {
756 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
761 * In order to report shared extents accurately,
762 * we report each distinct shared/unshared part
763 * of a single bmbt record using multiple bmap
764 * extents. To make that happen, we iterate the
765 * same map array item multiple times, each
766 * time trimming out the subextent that we just
769 * Because of this, we must check the out array
770 * index (cur_ext) directly against bmv_count-1
771 * to avoid overflows.
773 if (inject_map.br_startblock != NULLFSBLOCK) {
780 } while (nmap && bmv->bmv_length && cur_ext < bmv->bmv_count - 1);
785 xfs_iunlock(ip, lock);
787 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
789 for (i = 0; i < cur_ext; i++) {
790 int full = 0; /* user array is full */
792 /* format results & advance arg */
793 error = formatter(&arg, &out[i], &full);
803 * dead simple method of punching delalyed allocation blocks from a range in
804 * the inode. Walks a block at a time so will be slow, but is only executed in
805 * rare error cases so the overhead is not critical. This will always punch out
806 * both the start and end blocks, even if the ranges only partially overlap
807 * them, so it is up to the caller to ensure that partial blocks are not
811 xfs_bmap_punch_delalloc_range(
812 struct xfs_inode *ip,
813 xfs_fileoff_t start_fsb,
814 xfs_fileoff_t length)
816 xfs_fileoff_t remaining = length;
819 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
823 xfs_bmbt_irec_t imap;
825 xfs_fsblock_t firstblock;
826 struct xfs_defer_ops dfops;
829 * Map the range first and check that it is a delalloc extent
830 * before trying to unmap the range. Otherwise we will be
831 * trying to remove a real extent (which requires a
832 * transaction) or a hole, which is probably a bad idea...
834 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
838 /* something screwed, just bail */
839 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
840 xfs_alert(ip->i_mount,
841 "Failed delalloc mapping lookup ino %lld fsb %lld.",
842 ip->i_ino, start_fsb);
850 if (imap.br_startblock != DELAYSTARTBLOCK) {
851 /* been converted, ignore */
854 WARN_ON(imap.br_blockcount == 0);
857 * Note: while we initialise the firstblock/dfops pair, they
858 * should never be used because blocks should never be
859 * allocated or freed for a delalloc extent and hence we need
860 * don't cancel or finish them after the xfs_bunmapi() call.
862 xfs_defer_init(&dfops, &firstblock);
863 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
868 ASSERT(!xfs_defer_has_unfinished_work(&dfops));
872 } while(remaining > 0);
878 * Test whether it is appropriate to check an inode for and free post EOF
879 * blocks. The 'force' parameter determines whether we should also consider
880 * regular files that are marked preallocated or append-only.
883 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
885 /* prealloc/delalloc exists only on regular files */
886 if (!S_ISREG(VFS_I(ip)->i_mode))
890 * Zero sized files with no cached pages and delalloc blocks will not
891 * have speculative prealloc/delalloc blocks to remove.
893 if (VFS_I(ip)->i_size == 0 &&
894 VFS_I(ip)->i_mapping->nrpages == 0 &&
895 ip->i_delayed_blks == 0)
898 /* If we haven't read in the extent list, then don't do it now. */
899 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
903 * Do not free real preallocated or append-only files unless the file
904 * has delalloc blocks and we are forced to remove them.
906 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
907 if (!force || ip->i_delayed_blks == 0)
914 * This is called by xfs_inactive to free any blocks beyond eof
915 * when the link count isn't zero and by xfs_dm_punch_hole() when
916 * punching a hole to EOF.
926 xfs_fileoff_t end_fsb;
927 xfs_fileoff_t last_fsb;
928 xfs_filblks_t map_len;
930 xfs_bmbt_irec_t imap;
933 * Figure out if there are any blocks beyond the end
934 * of the file. If not, then there is nothing to do.
936 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
937 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
938 if (last_fsb <= end_fsb)
940 map_len = last_fsb - end_fsb;
943 xfs_ilock(ip, XFS_ILOCK_SHARED);
944 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
945 xfs_iunlock(ip, XFS_ILOCK_SHARED);
947 if (!error && (nimaps != 0) &&
948 (imap.br_startblock != HOLESTARTBLOCK ||
949 ip->i_delayed_blks)) {
951 * Attach the dquots to the inode up front.
953 error = xfs_qm_dqattach(ip, 0);
958 * There are blocks after the end of file.
959 * Free them up now by truncating the file to
963 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
967 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
970 ASSERT(XFS_FORCED_SHUTDOWN(mp));
972 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
976 xfs_ilock(ip, XFS_ILOCK_EXCL);
977 xfs_trans_ijoin(tp, ip, 0);
980 * Do not update the on-disk file size. If we update the
981 * on-disk file size and then the system crashes before the
982 * contents of the file are flushed to disk then the files
983 * may be full of holes (ie NULL files bug).
985 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
989 * If we get an error at this point we simply don't
990 * bother truncating the file.
992 xfs_trans_cancel(tp);
994 error = xfs_trans_commit(tp);
996 xfs_inode_clear_eofblocks_tag(ip);
999 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1001 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1007 xfs_alloc_file_space(
1008 struct xfs_inode *ip,
1013 xfs_mount_t *mp = ip->i_mount;
1015 xfs_filblks_t allocated_fsb;
1016 xfs_filblks_t allocatesize_fsb;
1017 xfs_extlen_t extsz, temp;
1018 xfs_fileoff_t startoffset_fsb;
1019 xfs_fsblock_t firstfsb;
1024 xfs_bmbt_irec_t imaps[1], *imapp;
1025 struct xfs_defer_ops dfops;
1026 uint qblocks, resblks, resrtextents;
1029 trace_xfs_alloc_file_space(ip);
1031 if (XFS_FORCED_SHUTDOWN(mp))
1034 error = xfs_qm_dqattach(ip, 0);
1041 rt = XFS_IS_REALTIME_INODE(ip);
1042 extsz = xfs_get_extsz_hint(ip);
1047 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1048 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1051 * Allocate file space until done or until there is an error
1053 while (allocatesize_fsb && !error) {
1057 * Determine space reservations for data/realtime.
1059 if (unlikely(extsz)) {
1060 s = startoffset_fsb;
1063 e = startoffset_fsb + allocatesize_fsb;
1064 if ((temp = do_mod(startoffset_fsb, extsz)))
1066 if ((temp = do_mod(e, extsz)))
1070 e = allocatesize_fsb;
1074 * The transaction reservation is limited to a 32-bit block
1075 * count, hence we need to limit the number of blocks we are
1076 * trying to reserve to avoid an overflow. We can't allocate
1077 * more than @nimaps extents, and an extent is limited on disk
1078 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1080 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1082 resrtextents = qblocks = resblks;
1083 resrtextents /= mp->m_sb.sb_rextsize;
1084 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1085 quota_flag = XFS_QMOPT_RES_RTBLKS;
1088 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1089 quota_flag = XFS_QMOPT_RES_REGBLKS;
1093 * Allocate and setup the transaction.
1095 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1096 resrtextents, 0, &tp);
1099 * Check for running out of space
1103 * Free the transaction structure.
1105 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1108 xfs_ilock(ip, XFS_ILOCK_EXCL);
1109 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1114 xfs_trans_ijoin(tp, ip, 0);
1116 xfs_defer_init(&dfops, &firstfsb);
1117 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1118 allocatesize_fsb, alloc_type, &firstfsb,
1119 resblks, imapp, &nimaps, &dfops);
1124 * Complete the transaction
1126 error = xfs_defer_finish(&tp, &dfops, NULL);
1130 error = xfs_trans_commit(tp);
1131 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1135 allocated_fsb = imapp->br_blockcount;
1142 startoffset_fsb += allocated_fsb;
1143 allocatesize_fsb -= allocated_fsb;
1148 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1149 xfs_defer_cancel(&dfops);
1150 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1152 error1: /* Just cancel transaction */
1153 xfs_trans_cancel(tp);
1154 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1160 struct xfs_inode *ip,
1161 xfs_fileoff_t startoffset_fsb,
1162 xfs_filblks_t len_fsb,
1165 struct xfs_mount *mp = ip->i_mount;
1166 struct xfs_trans *tp;
1167 struct xfs_defer_ops dfops;
1168 xfs_fsblock_t firstfsb;
1169 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1172 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1174 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1178 xfs_ilock(ip, XFS_ILOCK_EXCL);
1179 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1180 ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1182 goto out_trans_cancel;
1184 xfs_trans_ijoin(tp, ip, 0);
1186 xfs_defer_init(&dfops, &firstfsb);
1187 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb,
1190 goto out_bmap_cancel;
1192 error = xfs_defer_finish(&tp, &dfops, ip);
1194 goto out_bmap_cancel;
1196 error = xfs_trans_commit(tp);
1198 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1202 xfs_defer_cancel(&dfops);
1204 xfs_trans_cancel(tp);
1209 xfs_adjust_extent_unmap_boundaries(
1210 struct xfs_inode *ip,
1211 xfs_fileoff_t *startoffset_fsb,
1212 xfs_fileoff_t *endoffset_fsb)
1214 struct xfs_mount *mp = ip->i_mount;
1215 struct xfs_bmbt_irec imap;
1217 xfs_extlen_t mod = 0;
1220 error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0);
1224 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1227 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1228 block = imap.br_startblock;
1229 mod = do_div(block, mp->m_sb.sb_rextsize);
1231 *startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1235 error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0);
1239 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1240 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1242 if (mod && mod != mp->m_sb.sb_rextsize)
1243 *endoffset_fsb -= mod;
1250 xfs_flush_unmap_range(
1251 struct xfs_inode *ip,
1255 struct xfs_mount *mp = ip->i_mount;
1256 struct inode *inode = VFS_I(ip);
1257 xfs_off_t rounding, start, end;
1260 /* wait for the completion of any pending DIOs */
1261 inode_dio_wait(inode);
1263 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1264 start = round_down(offset, rounding);
1265 end = round_up(offset + len, rounding) - 1;
1267 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1270 truncate_pagecache_range(inode, start, end);
1275 xfs_free_file_space(
1276 struct xfs_inode *ip,
1280 struct xfs_mount *mp = ip->i_mount;
1281 xfs_fileoff_t startoffset_fsb;
1282 xfs_fileoff_t endoffset_fsb;
1283 int done = 0, error;
1285 trace_xfs_free_file_space(ip);
1287 error = xfs_qm_dqattach(ip, 0);
1291 if (len <= 0) /* if nothing being freed */
1294 error = xfs_flush_unmap_range(ip, offset, len);
1298 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1299 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1302 * Need to zero the stuff we're not freeing, on disk. If it's a RT file
1303 * and we can't use unwritten extents then we actually need to ensure
1304 * to zero the whole extent, otherwise we just need to take of block
1305 * boundaries, and xfs_bunmapi will handle the rest.
1307 if (XFS_IS_REALTIME_INODE(ip) &&
1308 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1309 error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb,
1315 if (endoffset_fsb > startoffset_fsb) {
1317 error = xfs_unmap_extent(ip, startoffset_fsb,
1318 endoffset_fsb - startoffset_fsb, &done);
1325 * Now that we've unmap all full blocks we'll have to zero out any
1326 * partial block at the beginning and/or end. xfs_zero_range is
1327 * smart enough to skip any holes, including those we just created.
1329 return xfs_zero_range(ip, offset, len, NULL);
1333 * Preallocate and zero a range of a file. This mechanism has the allocation
1334 * semantics of fallocate and in addition converts data in the range to zeroes.
1337 xfs_zero_file_space(
1338 struct xfs_inode *ip,
1342 struct xfs_mount *mp = ip->i_mount;
1346 trace_xfs_zero_file_space(ip);
1348 blksize = 1 << mp->m_sb.sb_blocklog;
1351 * Punch a hole and prealloc the range. We use hole punch rather than
1352 * unwritten extent conversion for two reasons:
1354 * 1.) Hole punch handles partial block zeroing for us.
1356 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1357 * by virtue of the hole punch.
1359 error = xfs_free_file_space(ip, offset, len);
1363 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1364 round_up(offset + len, blksize) -
1365 round_down(offset, blksize),
1366 XFS_BMAPI_PREALLOC);
1373 * @next_fsb will keep track of the extent currently undergoing shift.
1374 * @stop_fsb will keep track of the extent at which we have to stop.
1375 * If we are shifting left, we will start with block (offset + len) and
1376 * shift each extent till last extent.
1377 * If we are shifting right, we will start with last extent inside file space
1378 * and continue until we reach the block corresponding to offset.
1381 xfs_shift_file_space(
1382 struct xfs_inode *ip,
1385 enum shift_direction direction)
1388 struct xfs_mount *mp = ip->i_mount;
1389 struct xfs_trans *tp;
1391 struct xfs_defer_ops dfops;
1392 xfs_fsblock_t first_block;
1393 xfs_fileoff_t stop_fsb;
1394 xfs_fileoff_t next_fsb;
1395 xfs_fileoff_t shift_fsb;
1397 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1399 if (direction == SHIFT_LEFT) {
1400 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1401 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1404 * If right shift, delegate the work of initialization of
1405 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1407 next_fsb = NULLFSBLOCK;
1408 stop_fsb = XFS_B_TO_FSB(mp, offset);
1411 shift_fsb = XFS_B_TO_FSB(mp, len);
1414 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1415 * into the accessible region of the file.
1417 if (xfs_can_free_eofblocks(ip, true)) {
1418 error = xfs_free_eofblocks(mp, ip, false);
1424 * Writeback and invalidate cache for the remainder of the file as we're
1425 * about to shift down every extent from offset to EOF.
1427 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1431 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1432 offset >> PAGE_SHIFT, -1);
1437 * The extent shiting code works on extent granularity. So, if
1438 * stop_fsb is not the starting block of extent, we need to split
1439 * the extent at stop_fsb.
1441 if (direction == SHIFT_RIGHT) {
1442 error = xfs_bmap_split_extent(ip, stop_fsb);
1447 while (!error && !done) {
1449 * We would need to reserve permanent block for transaction.
1450 * This will come into picture when after shifting extent into
1451 * hole we found that adjacent extents can be merged which
1452 * may lead to freeing of a block during record update.
1454 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1455 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1459 xfs_ilock(ip, XFS_ILOCK_EXCL);
1460 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1461 ip->i_gdquot, ip->i_pdquot,
1462 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1463 XFS_QMOPT_RES_REGBLKS);
1465 goto out_trans_cancel;
1467 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1469 xfs_defer_init(&dfops, &first_block);
1472 * We are using the write transaction in which max 2 bmbt
1473 * updates are allowed
1475 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1476 &done, stop_fsb, &first_block, &dfops,
1477 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1479 goto out_bmap_cancel;
1481 error = xfs_defer_finish(&tp, &dfops, NULL);
1483 goto out_bmap_cancel;
1485 error = xfs_trans_commit(tp);
1491 xfs_defer_cancel(&dfops);
1493 xfs_trans_cancel(tp);
1498 * xfs_collapse_file_space()
1499 * This routine frees disk space and shift extent for the given file.
1500 * The first thing we do is to free data blocks in the specified range
1501 * by calling xfs_free_file_space(). It would also sync dirty data
1502 * and invalidate page cache over the region on which collapse range
1503 * is working. And Shift extent records to the left to cover a hole.
1510 xfs_collapse_file_space(
1511 struct xfs_inode *ip,
1517 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1518 trace_xfs_collapse_file_space(ip);
1520 error = xfs_free_file_space(ip, offset, len);
1524 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1528 * xfs_insert_file_space()
1529 * This routine create hole space by shifting extents for the given file.
1530 * The first thing we do is to sync dirty data and invalidate page cache
1531 * over the region on which insert range is working. And split an extent
1532 * to two extents at given offset by calling xfs_bmap_split_extent.
1533 * And shift all extent records which are laying between [offset,
1534 * last allocated extent] to the right to reserve hole range.
1540 xfs_insert_file_space(
1541 struct xfs_inode *ip,
1545 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1546 trace_xfs_insert_file_space(ip);
1548 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1552 * We need to check that the format of the data fork in the temporary inode is
1553 * valid for the target inode before doing the swap. This is not a problem with
1554 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1555 * data fork depending on the space the attribute fork is taking so we can get
1556 * invalid formats on the target inode.
1558 * E.g. target has space for 7 extents in extent format, temp inode only has
1559 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1560 * btree, but when swapped it needs to be in extent format. Hence we can't just
1561 * blindly swap data forks on attr2 filesystems.
1563 * Note that we check the swap in both directions so that we don't end up with
1564 * a corrupt temporary inode, either.
1566 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1567 * inode will prevent this situation from occurring, so all we do here is
1568 * reject and log the attempt. basically we are putting the responsibility on
1569 * userspace to get this right.
1572 xfs_swap_extents_check_format(
1573 struct xfs_inode *ip, /* target inode */
1574 struct xfs_inode *tip) /* tmp inode */
1577 /* Should never get a local format */
1578 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1579 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1583 * if the target inode has less extents that then temporary inode then
1584 * why did userspace call us?
1586 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1590 * If we have to use the (expensive) rmap swap method, we can
1591 * handle any number of extents and any format.
1593 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1597 * if the target inode is in extent form and the temp inode is in btree
1598 * form then we will end up with the target inode in the wrong format
1599 * as we already know there are less extents in the temp inode.
1601 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1602 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1605 /* Check temp in extent form to max in target */
1606 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1607 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1608 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1611 /* Check target in extent form to max in temp */
1612 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1613 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1614 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1618 * If we are in a btree format, check that the temp root block will fit
1619 * in the target and that it has enough extents to be in btree format
1622 * Note that we have to be careful to allow btree->extent conversions
1623 * (a common defrag case) which will occur when the temp inode is in
1626 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1627 if (XFS_IFORK_BOFF(ip) &&
1628 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1630 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1631 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1635 /* Reciprocal target->temp btree format checks */
1636 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1637 if (XFS_IFORK_BOFF(tip) &&
1638 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1640 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1641 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1649 xfs_swap_extent_flush(
1650 struct xfs_inode *ip)
1654 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1657 truncate_pagecache_range(VFS_I(ip), 0, -1);
1659 /* Verify O_DIRECT for ftmp */
1660 if (VFS_I(ip)->i_mapping->nrpages)
1666 * Move extents from one file to another, when rmap is enabled.
1669 xfs_swap_extent_rmap(
1670 struct xfs_trans **tpp,
1671 struct xfs_inode *ip,
1672 struct xfs_inode *tip)
1674 struct xfs_bmbt_irec irec;
1675 struct xfs_bmbt_irec uirec;
1676 struct xfs_bmbt_irec tirec;
1677 xfs_fileoff_t offset_fsb;
1678 xfs_fileoff_t end_fsb;
1679 xfs_filblks_t count_fsb;
1680 xfs_fsblock_t firstfsb;
1681 struct xfs_defer_ops dfops;
1686 __uint64_t tip_flags2;
1689 * If the source file has shared blocks, we must flag the donor
1690 * file as having shared blocks so that we get the shared-block
1691 * rmap functions when we go to fix up the rmaps. The flags
1692 * will be switch for reals later.
1694 tip_flags2 = tip->i_d.di_flags2;
1695 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1696 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1699 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1700 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1703 /* Read extent from the donor file */
1705 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1709 ASSERT(nimaps == 1);
1710 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1712 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1713 ilen = tirec.br_blockcount;
1715 /* Unmap the old blocks in the source file. */
1716 while (tirec.br_blockcount) {
1717 xfs_defer_init(&dfops, &firstfsb);
1718 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1720 /* Read extent from the source file */
1722 error = xfs_bmapi_read(ip, tirec.br_startoff,
1723 tirec.br_blockcount, &irec,
1727 ASSERT(nimaps == 1);
1728 ASSERT(tirec.br_startoff == irec.br_startoff);
1729 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1731 /* Trim the extent. */
1733 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1734 tirec.br_blockcount,
1735 irec.br_blockcount);
1736 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1738 /* Remove the mapping from the donor file. */
1739 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1744 /* Remove the mapping from the source file. */
1745 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops,
1750 /* Map the donor file's blocks into the source file. */
1751 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1756 /* Map the source file's blocks into the donor file. */
1757 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops,
1762 error = xfs_defer_finish(tpp, &dfops, ip);
1766 tirec.br_startoff += rlen;
1767 if (tirec.br_startblock != HOLESTARTBLOCK &&
1768 tirec.br_startblock != DELAYSTARTBLOCK)
1769 tirec.br_startblock += rlen;
1770 tirec.br_blockcount -= rlen;
1778 tip->i_d.di_flags2 = tip_flags2;
1782 xfs_defer_cancel(&dfops);
1784 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1785 tip->i_d.di_flags2 = tip_flags2;
1789 /* Swap the extents of two files by swapping data forks. */
1791 xfs_swap_extent_forks(
1792 struct xfs_trans *tp,
1793 struct xfs_inode *ip,
1794 struct xfs_inode *tip,
1796 int *target_log_flags)
1798 struct xfs_ifork tempifp, *ifp, *tifp;
1801 xfs_extnum_t nextents;
1806 * Count the number of extended attribute blocks
1808 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1809 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1810 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK,
1815 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1816 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1817 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1824 * Before we've swapped the forks, lets set the owners of the forks
1825 * appropriately. We have to do this as we are demand paging the btree
1826 * buffers, and so the validation done on read will expect the owner
1827 * field to be correctly set. Once we change the owners, we can swap the
1830 if (ip->i_d.di_version == 3 &&
1831 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1832 (*target_log_flags) |= XFS_ILOG_DOWNER;
1833 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1839 if (tip->i_d.di_version == 3 &&
1840 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1841 (*src_log_flags) |= XFS_ILOG_DOWNER;
1842 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1849 * Swap the data forks of the inodes
1853 tempifp = *ifp; /* struct copy */
1854 *ifp = *tifp; /* struct copy */
1855 *tifp = tempifp; /* struct copy */
1858 * Fix the on-disk inode values
1860 tmp = (__uint64_t)ip->i_d.di_nblocks;
1861 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1862 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1864 tmp = (__uint64_t) ip->i_d.di_nextents;
1865 ip->i_d.di_nextents = tip->i_d.di_nextents;
1866 tip->i_d.di_nextents = tmp;
1868 tmp = (__uint64_t) ip->i_d.di_format;
1869 ip->i_d.di_format = tip->i_d.di_format;
1870 tip->i_d.di_format = tmp;
1873 * The extents in the source inode could still contain speculative
1874 * preallocation beyond EOF (e.g. the file is open but not modified
1875 * while defrag is in progress). In that case, we need to copy over the
1876 * number of delalloc blocks the data fork in the source inode is
1877 * tracking beyond EOF so that when the fork is truncated away when the
1878 * temporary inode is unlinked we don't underrun the i_delayed_blks
1879 * counter on that inode.
1881 ASSERT(tip->i_delayed_blks == 0);
1882 tip->i_delayed_blks = ip->i_delayed_blks;
1883 ip->i_delayed_blks = 0;
1885 switch (ip->i_d.di_format) {
1886 case XFS_DINODE_FMT_EXTENTS:
1888 * If the extents fit in the inode, fix the pointer. Otherwise
1889 * it's already NULL or pointing to the extent.
1891 nextents = xfs_iext_count(&ip->i_df);
1892 if (nextents <= XFS_INLINE_EXTS)
1893 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
1894 (*src_log_flags) |= XFS_ILOG_DEXT;
1896 case XFS_DINODE_FMT_BTREE:
1897 ASSERT(ip->i_d.di_version < 3 ||
1898 (*src_log_flags & XFS_ILOG_DOWNER));
1899 (*src_log_flags) |= XFS_ILOG_DBROOT;
1903 switch (tip->i_d.di_format) {
1904 case XFS_DINODE_FMT_EXTENTS:
1906 * If the extents fit in the inode, fix the pointer. Otherwise
1907 * it's already NULL or pointing to the extent.
1909 nextents = xfs_iext_count(&tip->i_df);
1910 if (nextents <= XFS_INLINE_EXTS)
1911 tifp->if_u1.if_extents = tifp->if_u2.if_inline_ext;
1912 (*target_log_flags) |= XFS_ILOG_DEXT;
1914 case XFS_DINODE_FMT_BTREE:
1915 (*target_log_flags) |= XFS_ILOG_DBROOT;
1916 ASSERT(tip->i_d.di_version < 3 ||
1917 (*target_log_flags & XFS_ILOG_DOWNER));
1926 struct xfs_inode *ip, /* target inode */
1927 struct xfs_inode *tip, /* tmp inode */
1928 struct xfs_swapext *sxp)
1930 struct xfs_mount *mp = ip->i_mount;
1931 struct xfs_trans *tp;
1932 struct xfs_bstat *sbp = &sxp->sx_stat;
1933 int src_log_flags, target_log_flags;
1936 struct xfs_ifork *cowfp;
1941 * Lock the inodes against other IO, page faults and truncate to
1942 * begin with. Then we can ensure the inodes are flushed and have no
1943 * page cache safely. Once we have done this we can take the ilocks and
1944 * do the rest of the checks.
1946 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1947 lock_flags = XFS_MMAPLOCK_EXCL;
1948 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1950 /* Verify that both files have the same format */
1951 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1956 /* Verify both files are either real-time or non-realtime */
1957 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1962 error = xfs_swap_extent_flush(ip);
1965 error = xfs_swap_extent_flush(tip);
1970 * Extent "swapping" with rmap requires a permanent reservation and
1971 * a block reservation because it's really just a remap operation
1972 * performed with log redo items!
1974 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1976 * Conceptually this shouldn't affect the shape of either
1977 * bmbt, but since we atomically move extents one by one,
1978 * we reserve enough space to rebuild both trees.
1980 resblks = XFS_SWAP_RMAP_SPACE_RES(mp,
1981 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK),
1983 XFS_SWAP_RMAP_SPACE_RES(mp,
1984 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK),
1986 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
1989 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0,
1995 * Lock and join the inodes to the tansaction so that transaction commit
1996 * or cancel will unlock the inodes from this point onwards.
1998 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1999 lock_flags |= XFS_ILOCK_EXCL;
2000 xfs_trans_ijoin(tp, ip, 0);
2001 xfs_trans_ijoin(tp, tip, 0);
2004 /* Verify all data are being swapped */
2005 if (sxp->sx_offset != 0 ||
2006 sxp->sx_length != ip->i_d.di_size ||
2007 sxp->sx_length != tip->i_d.di_size) {
2009 goto out_trans_cancel;
2012 trace_xfs_swap_extent_before(ip, 0);
2013 trace_xfs_swap_extent_before(tip, 1);
2015 /* check inode formats now that data is flushed */
2016 error = xfs_swap_extents_check_format(ip, tip);
2019 "%s: inode 0x%llx format is incompatible for exchanging.",
2020 __func__, ip->i_ino);
2021 goto out_trans_cancel;
2025 * Compare the current change & modify times with that
2026 * passed in. If they differ, we abort this swap.
2027 * This is the mechanism used to ensure the calling
2028 * process that the file was not changed out from
2031 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
2032 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
2033 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
2034 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
2036 goto out_trans_cancel;
2040 * Note the trickiness in setting the log flags - we set the owner log
2041 * flag on the opposite inode (i.e. the inode we are setting the new
2042 * owner to be) because once we swap the forks and log that, log
2043 * recovery is going to see the fork as owned by the swapped inode,
2044 * not the pre-swapped inodes.
2046 src_log_flags = XFS_ILOG_CORE;
2047 target_log_flags = XFS_ILOG_CORE;
2049 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
2050 error = xfs_swap_extent_rmap(&tp, ip, tip);
2052 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
2055 goto out_trans_cancel;
2057 /* Do we have to swap reflink flags? */
2058 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
2059 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
2060 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2061 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2062 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
2063 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
2064 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
2065 cowfp = ip->i_cowfp;
2066 ip->i_cowfp = tip->i_cowfp;
2067 tip->i_cowfp = cowfp;
2068 xfs_inode_set_cowblocks_tag(ip);
2069 xfs_inode_set_cowblocks_tag(tip);
2072 xfs_trans_log_inode(tp, ip, src_log_flags);
2073 xfs_trans_log_inode(tp, tip, target_log_flags);
2076 * If this is a synchronous mount, make sure that the
2077 * transaction goes to disk before returning to the user.
2079 if (mp->m_flags & XFS_MOUNT_WSYNC)
2080 xfs_trans_set_sync(tp);
2082 error = xfs_trans_commit(tp);
2084 trace_xfs_swap_extent_after(ip, 0);
2085 trace_xfs_swap_extent_after(tip, 1);
2088 xfs_iunlock(ip, lock_flags);
2089 xfs_iunlock(tip, lock_flags);
2090 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
2094 xfs_trans_cancel(tp);