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_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_trans.h"
31 #include "xfs_extfree_item.h"
32 #include "xfs_alloc.h"
34 #include "xfs_bmap_util.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_quota.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_trace.h"
41 #include "xfs_icache.h"
44 /* Kernel only BMAP related definitions and functions */
47 * Convert the given file system block to a disk block. We have to treat it
48 * differently based on whether the file is a real time file or not, because the
52 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
54 return (XFS_IS_REALTIME_INODE(ip) ? \
55 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
56 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
60 * Routine to zero an extent on disk allocated to the specific inode.
62 * The VFS functions take a linearised filesystem block offset, so we have to
63 * convert the sparse xfs fsb to the right format first.
64 * VFS types are real funky, too.
69 xfs_fsblock_t start_fsb,
72 struct xfs_mount *mp = ip->i_mount;
73 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
74 sector_t block = XFS_BB_TO_FSBT(mp, sector);
75 ssize_t size = XFS_FSB_TO_B(mp, count_fsb);
77 if (IS_DAX(VFS_I(ip)))
78 return dax_clear_sectors(xfs_find_bdev_for_inode(VFS_I(ip)),
82 * let the block layer decide on the fastest method of
83 * implementing the zeroing.
85 return sb_issue_zeroout(mp->m_super, block, count_fsb, GFP_NOFS);
90 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
91 * caller. Frees all the extents that need freeing, which must be done
92 * last due to locking considerations. We never free any extents in
93 * the first transaction.
95 * If an inode *ip is provided, rejoin it to the transaction if
96 * the transaction was committed.
100 struct xfs_trans **tp, /* transaction pointer addr */
101 struct xfs_bmap_free *flist, /* i/o: list extents to free */
102 struct xfs_inode *ip)
104 struct xfs_efd_log_item *efd; /* extent free data */
105 struct xfs_efi_log_item *efi; /* extent free intention */
106 int error; /* error return value */
107 int committed;/* xact committed or not */
108 struct xfs_bmap_free_item *free; /* free extent item */
109 struct xfs_bmap_free_item *next; /* next item on free list */
111 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
112 if (flist->xbf_count == 0)
115 efi = xfs_trans_get_efi(*tp, flist->xbf_count);
116 for (free = flist->xbf_first; free; free = free->xbfi_next)
117 xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock,
118 free->xbfi_blockcount);
120 error = __xfs_trans_roll(tp, ip, &committed);
123 * If the transaction was committed, drop the EFD reference
124 * since we're bailing out of here. The other reference is
125 * dropped when the EFI hits the AIL.
127 * If the transaction was not committed, the EFI is freed by the
128 * EFI item unlock handler on abort. Also, we have a new
129 * transaction so we should return committed=1 even though we're
130 * returning an error.
133 xfs_efi_release(efi);
134 xfs_force_shutdown((*tp)->t_mountp,
135 (error == -EFSCORRUPTED) ?
136 SHUTDOWN_CORRUPT_INCORE :
137 SHUTDOWN_META_IO_ERROR);
143 * Get an EFD and free each extent in the list, logging to the EFD in
144 * the process. The remaining bmap free list is cleaned up by the caller
147 efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count);
148 for (free = flist->xbf_first; free != NULL; free = next) {
149 next = free->xbfi_next;
151 error = xfs_trans_free_extent(*tp, efd, free->xbfi_startblock,
152 free->xbfi_blockcount);
156 xfs_bmap_del_free(flist, NULL, free);
164 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
166 xfs_alloctype_t atype = 0; /* type for allocation routines */
167 int error; /* error return value */
168 xfs_mount_t *mp; /* mount point structure */
169 xfs_extlen_t prod = 0; /* product factor for allocators */
170 xfs_extlen_t ralen = 0; /* realtime allocation length */
171 xfs_extlen_t align; /* minimum allocation alignment */
174 mp = ap->ip->i_mount;
175 align = xfs_get_extsz_hint(ap->ip);
176 prod = align / mp->m_sb.sb_rextsize;
177 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
178 align, 1, ap->eof, 0,
179 ap->conv, &ap->offset, &ap->length);
183 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
186 * If the offset & length are not perfectly aligned
187 * then kill prod, it will just get us in trouble.
189 if (do_mod(ap->offset, align) || ap->length % align)
192 * Set ralen to be the actual requested length in rtextents.
194 ralen = ap->length / mp->m_sb.sb_rextsize;
196 * If the old value was close enough to MAXEXTLEN that
197 * we rounded up to it, cut it back so it's valid again.
198 * Note that if it's a really large request (bigger than
199 * MAXEXTLEN), we don't hear about that number, and can't
200 * adjust the starting point to match it.
202 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
203 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
206 * Lock out other modifications to the RT bitmap inode.
208 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
209 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
212 * If it's an allocation to an empty file at offset 0,
213 * pick an extent that will space things out in the rt area.
215 if (ap->eof && ap->offset == 0) {
216 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
218 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
221 ap->blkno = rtx * mp->m_sb.sb_rextsize;
226 xfs_bmap_adjacent(ap);
229 * Realtime allocation, done through xfs_rtallocate_extent.
231 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
232 do_div(ap->blkno, mp->m_sb.sb_rextsize);
235 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
236 &ralen, atype, ap->wasdel, prod, &rtb)))
238 if (rtb == NULLFSBLOCK && prod > 1 &&
239 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
240 ap->length, &ralen, atype,
241 ap->wasdel, 1, &rtb)))
244 if (ap->blkno != NULLFSBLOCK) {
245 ap->blkno *= mp->m_sb.sb_rextsize;
246 ralen *= mp->m_sb.sb_rextsize;
248 ap->ip->i_d.di_nblocks += ralen;
249 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
251 ap->ip->i_delayed_blks -= ralen;
253 * Adjust the disk quota also. This was reserved
256 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
257 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
258 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
260 /* Zero the extent if we were asked to do so */
261 if (ap->userdata & XFS_ALLOC_USERDATA_ZERO) {
262 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
273 * Check if the endoff is outside the last extent. If so the caller will grow
274 * the allocation to a stripe unit boundary. All offsets are considered outside
275 * the end of file for an empty fork, so 1 is returned in *eof in that case.
279 struct xfs_inode *ip,
280 xfs_fileoff_t endoff,
284 struct xfs_bmbt_irec rec;
287 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
291 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
296 * Extent tree block counting routines.
300 * Count leaf blocks given a range of extent records.
303 xfs_bmap_count_leaves(
311 for (b = 0; b < numrecs; b++) {
312 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
313 *count += xfs_bmbt_get_blockcount(frp);
318 * Count leaf blocks given a range of extent records originally
322 xfs_bmap_disk_count_leaves(
323 struct xfs_mount *mp,
324 struct xfs_btree_block *block,
331 for (b = 1; b <= numrecs; b++) {
332 frp = XFS_BMBT_REC_ADDR(mp, block, b);
333 *count += xfs_bmbt_disk_get_blockcount(frp);
338 * Recursively walks each level of a btree
339 * to count total fsblocks in use.
341 STATIC int /* error */
343 xfs_mount_t *mp, /* file system mount point */
344 xfs_trans_t *tp, /* transaction pointer */
345 xfs_ifork_t *ifp, /* inode fork pointer */
346 xfs_fsblock_t blockno, /* file system block number */
347 int levelin, /* level in btree */
348 int *count) /* Count of blocks */
354 xfs_fsblock_t bno = blockno;
355 xfs_fsblock_t nextbno;
356 struct xfs_btree_block *block, *nextblock;
359 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
364 block = XFS_BUF_TO_BLOCK(bp);
367 /* Not at node above leaves, count this level of nodes */
368 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
369 while (nextbno != NULLFSBLOCK) {
370 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
376 nextblock = XFS_BUF_TO_BLOCK(nbp);
377 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
378 xfs_trans_brelse(tp, nbp);
381 /* Dive to the next level */
382 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
383 bno = be64_to_cpu(*pp);
384 if (unlikely((error =
385 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
386 xfs_trans_brelse(tp, bp);
387 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
388 XFS_ERRLEVEL_LOW, mp);
389 return -EFSCORRUPTED;
391 xfs_trans_brelse(tp, bp);
393 /* count all level 1 nodes and their leaves */
395 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
396 numrecs = be16_to_cpu(block->bb_numrecs);
397 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
398 xfs_trans_brelse(tp, bp);
399 if (nextbno == NULLFSBLOCK)
402 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
408 block = XFS_BUF_TO_BLOCK(bp);
415 * Count fsblocks of the given fork.
418 xfs_bmap_count_blocks(
419 xfs_trans_t *tp, /* transaction pointer */
420 xfs_inode_t *ip, /* incore inode */
421 int whichfork, /* data or attr fork */
422 int *count) /* out: count of blocks */
424 struct xfs_btree_block *block; /* current btree block */
425 xfs_fsblock_t bno; /* block # of "block" */
426 xfs_ifork_t *ifp; /* fork structure */
427 int level; /* btree level, for checking */
428 xfs_mount_t *mp; /* file system mount structure */
429 __be64 *pp; /* pointer to block address */
433 ifp = XFS_IFORK_PTR(ip, whichfork);
434 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
435 xfs_bmap_count_leaves(ifp, 0,
436 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
442 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
444 block = ifp->if_broot;
445 level = be16_to_cpu(block->bb_level);
447 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
448 bno = be64_to_cpu(*pp);
449 ASSERT(bno != NULLFSBLOCK);
450 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
451 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
453 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
454 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
456 return -EFSCORRUPTED;
463 * returns 1 for success, 0 if we failed to map the extent.
466 xfs_getbmapx_fix_eof_hole(
467 xfs_inode_t *ip, /* xfs incore inode pointer */
468 struct getbmapx *out, /* output structure */
469 int prealloced, /* this is a file with
470 * preallocated data space */
471 __int64_t end, /* last block requested */
472 xfs_fsblock_t startblock)
475 xfs_mount_t *mp; /* file system mount point */
476 xfs_ifork_t *ifp; /* inode fork pointer */
477 xfs_extnum_t lastx; /* last extent pointer */
478 xfs_fileoff_t fileblock;
480 if (startblock == HOLESTARTBLOCK) {
483 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
484 fixlen -= out->bmv_offset;
485 if (prealloced && out->bmv_offset + out->bmv_length == end) {
486 /* Came to hole at EOF. Trim it. */
489 out->bmv_length = fixlen;
492 if (startblock == DELAYSTARTBLOCK)
495 out->bmv_block = xfs_fsb_to_db(ip, startblock);
496 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
497 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
498 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
499 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
500 out->bmv_oflags |= BMV_OF_LAST;
507 * Get inode's extents as described in bmv, and format for output.
508 * Calls formatter to fill the user's buffer until all extents
509 * are mapped, until the passed-in bmv->bmv_count slots have
510 * been filled, or until the formatter short-circuits the loop,
511 * if it is tracking filled-in extents on its own.
516 struct getbmapx *bmv, /* user bmap structure */
517 xfs_bmap_format_t formatter, /* format to user */
518 void *arg) /* formatter arg */
520 __int64_t bmvend; /* last block requested */
521 int error = 0; /* return value */
522 __int64_t fixlen; /* length for -1 case */
523 int i; /* extent number */
524 int lock; /* lock state */
525 xfs_bmbt_irec_t *map; /* buffer for user's data */
526 xfs_mount_t *mp; /* file system mount point */
527 int nex; /* # of user extents can do */
528 int nexleft; /* # of user extents left */
529 int subnex; /* # of bmapi's can do */
530 int nmap; /* number of map entries */
531 struct getbmapx *out; /* output structure */
532 int whichfork; /* data or attr fork */
533 int prealloced; /* this is a file with
534 * preallocated data space */
535 int iflags; /* interface flags */
536 int bmapi_flags; /* flags for xfs_bmapi */
540 iflags = bmv->bmv_iflags;
541 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
543 if (whichfork == XFS_ATTR_FORK) {
544 if (XFS_IFORK_Q(ip)) {
545 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
546 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
547 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
550 ip->i_d.di_aformat != 0 &&
551 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
552 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
554 return -EFSCORRUPTED;
560 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
561 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
562 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
565 if (xfs_get_extsz_hint(ip) ||
566 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
568 fixlen = mp->m_super->s_maxbytes;
571 fixlen = XFS_ISIZE(ip);
575 if (bmv->bmv_length == -1) {
576 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
578 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
579 } else if (bmv->bmv_length == 0) {
580 bmv->bmv_entries = 0;
582 } else if (bmv->bmv_length < 0) {
586 nex = bmv->bmv_count - 1;
589 bmvend = bmv->bmv_offset + bmv->bmv_length;
592 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
594 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
598 xfs_ilock(ip, XFS_IOLOCK_SHARED);
599 if (whichfork == XFS_DATA_FORK) {
600 if (!(iflags & BMV_IF_DELALLOC) &&
601 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
602 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
604 goto out_unlock_iolock;
607 * Even after flushing the inode, there can still be
608 * delalloc blocks on the inode beyond EOF due to
609 * speculative preallocation. These are not removed
610 * until the release function is called or the inode
611 * is inactivated. Hence we cannot assert here that
612 * ip->i_delayed_blks == 0.
616 lock = xfs_ilock_data_map_shared(ip);
618 lock = xfs_ilock_attr_map_shared(ip);
622 * Don't let nex be bigger than the number of extents
623 * we can have assuming alternating holes and real extents.
625 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
626 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
628 bmapi_flags = xfs_bmapi_aflag(whichfork);
629 if (!(iflags & BMV_IF_PREALLOC))
630 bmapi_flags |= XFS_BMAPI_IGSTATE;
633 * Allocate enough space to handle "subnex" maps at a time.
637 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
639 goto out_unlock_ilock;
641 bmv->bmv_entries = 0;
643 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
644 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
652 nmap = (nexleft > subnex) ? subnex : nexleft;
653 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
654 XFS_BB_TO_FSB(mp, bmv->bmv_length),
655 map, &nmap, bmapi_flags);
658 ASSERT(nmap <= subnex);
660 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
661 out[cur_ext].bmv_oflags = 0;
662 if (map[i].br_state == XFS_EXT_UNWRITTEN)
663 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
664 else if (map[i].br_startblock == DELAYSTARTBLOCK)
665 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
666 out[cur_ext].bmv_offset =
667 XFS_FSB_TO_BB(mp, map[i].br_startoff);
668 out[cur_ext].bmv_length =
669 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
670 out[cur_ext].bmv_unused1 = 0;
671 out[cur_ext].bmv_unused2 = 0;
674 * delayed allocation extents that start beyond EOF can
675 * occur due to speculative EOF allocation when the
676 * delalloc extent is larger than the largest freespace
677 * extent at conversion time. These extents cannot be
678 * converted by data writeback, so can exist here even
679 * if we are not supposed to be finding delalloc
682 if (map[i].br_startblock == DELAYSTARTBLOCK &&
683 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
684 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
686 if (map[i].br_startblock == HOLESTARTBLOCK &&
687 whichfork == XFS_ATTR_FORK) {
688 /* came to the end of attribute fork */
689 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
693 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
695 map[i].br_startblock))
699 out[cur_ext].bmv_offset +
700 out[cur_ext].bmv_length;
702 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
705 * In case we don't want to return the hole,
706 * don't increase cur_ext so that we can reuse
707 * it in the next loop.
709 if ((iflags & BMV_IF_NO_HOLES) &&
710 map[i].br_startblock == HOLESTARTBLOCK) {
711 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
719 } while (nmap && nexleft && bmv->bmv_length);
724 xfs_iunlock(ip, lock);
726 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
728 for (i = 0; i < cur_ext; i++) {
729 int full = 0; /* user array is full */
731 /* format results & advance arg */
732 error = formatter(&arg, &out[i], &full);
742 * dead simple method of punching delalyed allocation blocks from a range in
743 * the inode. Walks a block at a time so will be slow, but is only executed in
744 * rare error cases so the overhead is not critical. This will always punch out
745 * both the start and end blocks, even if the ranges only partially overlap
746 * them, so it is up to the caller to ensure that partial blocks are not
750 xfs_bmap_punch_delalloc_range(
751 struct xfs_inode *ip,
752 xfs_fileoff_t start_fsb,
753 xfs_fileoff_t length)
755 xfs_fileoff_t remaining = length;
758 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
762 xfs_bmbt_irec_t imap;
764 xfs_fsblock_t firstblock;
765 xfs_bmap_free_t flist;
768 * Map the range first and check that it is a delalloc extent
769 * before trying to unmap the range. Otherwise we will be
770 * trying to remove a real extent (which requires a
771 * transaction) or a hole, which is probably a bad idea...
773 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
777 /* something screwed, just bail */
778 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
779 xfs_alert(ip->i_mount,
780 "Failed delalloc mapping lookup ino %lld fsb %lld.",
781 ip->i_ino, start_fsb);
789 if (imap.br_startblock != DELAYSTARTBLOCK) {
790 /* been converted, ignore */
793 WARN_ON(imap.br_blockcount == 0);
796 * Note: while we initialise the firstblock/flist pair, they
797 * should never be used because blocks should never be
798 * allocated or freed for a delalloc extent and hence we need
799 * don't cancel or finish them after the xfs_bunmapi() call.
801 xfs_bmap_init(&flist, &firstblock);
802 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
807 ASSERT(!flist.xbf_count && !flist.xbf_first);
811 } while(remaining > 0);
817 * Test whether it is appropriate to check an inode for and free post EOF
818 * blocks. The 'force' parameter determines whether we should also consider
819 * regular files that are marked preallocated or append-only.
822 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
824 /* prealloc/delalloc exists only on regular files */
825 if (!S_ISREG(ip->i_d.di_mode))
829 * Zero sized files with no cached pages and delalloc blocks will not
830 * have speculative prealloc/delalloc blocks to remove.
832 if (VFS_I(ip)->i_size == 0 &&
833 VFS_I(ip)->i_mapping->nrpages == 0 &&
834 ip->i_delayed_blks == 0)
837 /* If we haven't read in the extent list, then don't do it now. */
838 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
842 * Do not free real preallocated or append-only files unless the file
843 * has delalloc blocks and we are forced to remove them.
845 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
846 if (!force || ip->i_delayed_blks == 0)
853 * This is called by xfs_inactive to free any blocks beyond eof
854 * when the link count isn't zero and by xfs_dm_punch_hole() when
855 * punching a hole to EOF.
865 xfs_fileoff_t end_fsb;
866 xfs_fileoff_t last_fsb;
867 xfs_filblks_t map_len;
869 xfs_bmbt_irec_t imap;
872 * Figure out if there are any blocks beyond the end
873 * of the file. If not, then there is nothing to do.
875 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
876 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
877 if (last_fsb <= end_fsb)
879 map_len = last_fsb - end_fsb;
882 xfs_ilock(ip, XFS_ILOCK_SHARED);
883 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
884 xfs_iunlock(ip, XFS_ILOCK_SHARED);
886 if (!error && (nimaps != 0) &&
887 (imap.br_startblock != HOLESTARTBLOCK ||
888 ip->i_delayed_blks)) {
890 * Attach the dquots to the inode up front.
892 error = xfs_qm_dqattach(ip, 0);
897 * There are blocks after the end of file.
898 * Free them up now by truncating the file to
901 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
904 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
905 xfs_trans_cancel(tp);
910 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
912 ASSERT(XFS_FORCED_SHUTDOWN(mp));
913 xfs_trans_cancel(tp);
915 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
919 xfs_ilock(ip, XFS_ILOCK_EXCL);
920 xfs_trans_ijoin(tp, ip, 0);
923 * Do not update the on-disk file size. If we update the
924 * on-disk file size and then the system crashes before the
925 * contents of the file are flushed to disk then the files
926 * may be full of holes (ie NULL files bug).
928 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
932 * If we get an error at this point we simply don't
933 * bother truncating the file.
935 xfs_trans_cancel(tp);
937 error = xfs_trans_commit(tp);
939 xfs_inode_clear_eofblocks_tag(ip);
942 xfs_iunlock(ip, XFS_ILOCK_EXCL);
944 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
950 xfs_alloc_file_space(
951 struct xfs_inode *ip,
956 xfs_mount_t *mp = ip->i_mount;
958 xfs_filblks_t allocated_fsb;
959 xfs_filblks_t allocatesize_fsb;
960 xfs_extlen_t extsz, temp;
961 xfs_fileoff_t startoffset_fsb;
962 xfs_fsblock_t firstfsb;
967 xfs_bmbt_irec_t imaps[1], *imapp;
968 xfs_bmap_free_t free_list;
969 uint qblocks, resblks, resrtextents;
972 trace_xfs_alloc_file_space(ip);
974 if (XFS_FORCED_SHUTDOWN(mp))
977 error = xfs_qm_dqattach(ip, 0);
984 rt = XFS_IS_REALTIME_INODE(ip);
985 extsz = xfs_get_extsz_hint(ip);
990 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
991 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
994 * Allocate file space until done or until there is an error
996 while (allocatesize_fsb && !error) {
1000 * Determine space reservations for data/realtime.
1002 if (unlikely(extsz)) {
1003 s = startoffset_fsb;
1006 e = startoffset_fsb + allocatesize_fsb;
1007 if ((temp = do_mod(startoffset_fsb, extsz)))
1009 if ((temp = do_mod(e, extsz)))
1013 e = allocatesize_fsb;
1017 * The transaction reservation is limited to a 32-bit block
1018 * count, hence we need to limit the number of blocks we are
1019 * trying to reserve to avoid an overflow. We can't allocate
1020 * more than @nimaps extents, and an extent is limited on disk
1021 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1023 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1025 resrtextents = qblocks = resblks;
1026 resrtextents /= mp->m_sb.sb_rextsize;
1027 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1028 quota_flag = XFS_QMOPT_RES_RTBLKS;
1031 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1032 quota_flag = XFS_QMOPT_RES_REGBLKS;
1036 * Allocate and setup the transaction.
1038 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1039 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1040 resblks, resrtextents);
1042 * Check for running out of space
1046 * Free the transaction structure.
1048 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1049 xfs_trans_cancel(tp);
1052 xfs_ilock(ip, XFS_ILOCK_EXCL);
1053 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1058 xfs_trans_ijoin(tp, ip, 0);
1060 xfs_bmap_init(&free_list, &firstfsb);
1061 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1062 allocatesize_fsb, alloc_type, &firstfsb,
1063 resblks, imapp, &nimaps, &free_list);
1068 * Complete the transaction
1070 error = xfs_bmap_finish(&tp, &free_list, NULL);
1074 error = xfs_trans_commit(tp);
1075 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1079 allocated_fsb = imapp->br_blockcount;
1086 startoffset_fsb += allocated_fsb;
1087 allocatesize_fsb -= allocated_fsb;
1092 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1093 xfs_bmap_cancel(&free_list);
1094 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1096 error1: /* Just cancel transaction */
1097 xfs_trans_cancel(tp);
1098 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1103 * Zero file bytes between startoff and endoff inclusive.
1104 * The iolock is held exclusive and no blocks are buffered.
1106 * This function is used by xfs_free_file_space() to zero
1107 * partial blocks when the range to free is not block aligned.
1108 * When unreserving space with boundaries that are not block
1109 * aligned we round up the start and round down the end
1110 * boundaries and then use this function to zero the parts of
1111 * the blocks that got dropped during the rounding.
1114 xfs_zero_remaining_bytes(
1119 xfs_bmbt_irec_t imap;
1120 xfs_fileoff_t offset_fsb;
1121 xfs_off_t lastoffset;
1124 xfs_mount_t *mp = ip->i_mount;
1129 * Avoid doing I/O beyond eof - it's not necessary
1130 * since nothing can read beyond eof. The space will
1131 * be zeroed when the file is extended anyway.
1133 if (startoff >= XFS_ISIZE(ip))
1136 if (endoff > XFS_ISIZE(ip))
1137 endoff = XFS_ISIZE(ip);
1139 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1142 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1145 lock_mode = xfs_ilock_data_map_shared(ip);
1146 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1147 xfs_iunlock(ip, lock_mode);
1149 if (error || nimap < 1)
1151 ASSERT(imap.br_blockcount >= 1);
1152 ASSERT(imap.br_startoff == offset_fsb);
1153 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1155 if (imap.br_startblock == HOLESTARTBLOCK ||
1156 imap.br_state == XFS_EXT_UNWRITTEN) {
1157 /* skip the entire extent */
1158 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff +
1159 imap.br_blockcount) - 1;
1163 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1164 if (lastoffset > endoff)
1165 lastoffset = endoff;
1167 /* DAX can just zero the backing device directly */
1168 if (IS_DAX(VFS_I(ip))) {
1169 error = dax_zero_page_range(VFS_I(ip), offset,
1170 lastoffset - offset + 1,
1171 xfs_get_blocks_direct);
1177 error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ?
1178 mp->m_rtdev_targp : mp->m_ddev_targp,
1179 xfs_fsb_to_db(ip, imap.br_startblock),
1180 BTOBB(mp->m_sb.sb_blocksize),
1186 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1187 0, lastoffset - offset + 1);
1189 error = xfs_bwrite(bp);
1198 xfs_free_file_space(
1199 struct xfs_inode *ip,
1204 xfs_fileoff_t endoffset_fsb;
1206 xfs_fsblock_t firstfsb;
1207 xfs_bmap_free_t free_list;
1208 xfs_bmbt_irec_t imap;
1210 xfs_off_t iendoffset;
1217 xfs_fileoff_t startoffset_fsb;
1222 trace_xfs_free_file_space(ip);
1224 error = xfs_qm_dqattach(ip, 0);
1229 if (len <= 0) /* if nothing being freed */
1231 rt = XFS_IS_REALTIME_INODE(ip);
1232 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1233 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1235 /* wait for the completion of any pending DIOs */
1236 inode_dio_wait(VFS_I(ip));
1238 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1239 ioffset = round_down(offset, rounding);
1240 iendoffset = round_up(offset + len, rounding) - 1;
1241 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, ioffset,
1245 truncate_pagecache_range(VFS_I(ip), ioffset, iendoffset);
1248 * Need to zero the stuff we're not freeing, on disk.
1249 * If it's a realtime file & can't use unwritten extents then we
1250 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1251 * will take care of it for us.
1253 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1255 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1259 ASSERT(nimap == 0 || nimap == 1);
1260 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1263 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1264 block = imap.br_startblock;
1265 mod = do_div(block, mp->m_sb.sb_rextsize);
1267 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1270 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1274 ASSERT(nimap == 0 || nimap == 1);
1275 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1276 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1278 if (mod && (mod != mp->m_sb.sb_rextsize))
1279 endoffset_fsb -= mod;
1282 if ((done = (endoffset_fsb <= startoffset_fsb)))
1284 * One contiguous piece to clear
1286 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1289 * Some full blocks, possibly two pieces to clear
1291 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1292 error = xfs_zero_remaining_bytes(ip, offset,
1293 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1295 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1296 error = xfs_zero_remaining_bytes(ip,
1297 XFS_FSB_TO_B(mp, endoffset_fsb),
1302 * free file space until done or until there is an error
1304 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1305 while (!error && !done) {
1308 * allocate and setup the transaction. Allow this
1309 * transaction to dip into the reserve blocks to ensure
1310 * the freeing of the space succeeds at ENOSPC.
1312 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1313 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1316 * check for running out of space
1320 * Free the transaction structure.
1322 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1323 xfs_trans_cancel(tp);
1326 xfs_ilock(ip, XFS_ILOCK_EXCL);
1327 error = xfs_trans_reserve_quota(tp, mp,
1328 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1329 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1333 xfs_trans_ijoin(tp, ip, 0);
1336 * issue the bunmapi() call to free the blocks
1338 xfs_bmap_init(&free_list, &firstfsb);
1339 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1340 endoffset_fsb - startoffset_fsb,
1341 0, 2, &firstfsb, &free_list, &done);
1346 * complete the transaction
1348 error = xfs_bmap_finish(&tp, &free_list, NULL);
1352 error = xfs_trans_commit(tp);
1353 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1360 xfs_bmap_cancel(&free_list);
1362 xfs_trans_cancel(tp);
1363 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1368 * Preallocate and zero a range of a file. This mechanism has the allocation
1369 * semantics of fallocate and in addition converts data in the range to zeroes.
1372 xfs_zero_file_space(
1373 struct xfs_inode *ip,
1377 struct xfs_mount *mp = ip->i_mount;
1381 trace_xfs_zero_file_space(ip);
1383 blksize = 1 << mp->m_sb.sb_blocklog;
1386 * Punch a hole and prealloc the range. We use hole punch rather than
1387 * unwritten extent conversion for two reasons:
1389 * 1.) Hole punch handles partial block zeroing for us.
1391 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1392 * by virtue of the hole punch.
1394 error = xfs_free_file_space(ip, offset, len);
1398 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1399 round_up(offset + len, blksize) -
1400 round_down(offset, blksize),
1401 XFS_BMAPI_PREALLOC);
1408 * @next_fsb will keep track of the extent currently undergoing shift.
1409 * @stop_fsb will keep track of the extent at which we have to stop.
1410 * If we are shifting left, we will start with block (offset + len) and
1411 * shift each extent till last extent.
1412 * If we are shifting right, we will start with last extent inside file space
1413 * and continue until we reach the block corresponding to offset.
1416 xfs_shift_file_space(
1417 struct xfs_inode *ip,
1420 enum shift_direction direction)
1423 struct xfs_mount *mp = ip->i_mount;
1424 struct xfs_trans *tp;
1426 struct xfs_bmap_free free_list;
1427 xfs_fsblock_t first_block;
1428 xfs_fileoff_t stop_fsb;
1429 xfs_fileoff_t next_fsb;
1430 xfs_fileoff_t shift_fsb;
1432 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1434 if (direction == SHIFT_LEFT) {
1435 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1436 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1439 * If right shift, delegate the work of initialization of
1440 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1442 next_fsb = NULLFSBLOCK;
1443 stop_fsb = XFS_B_TO_FSB(mp, offset);
1446 shift_fsb = XFS_B_TO_FSB(mp, len);
1449 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1450 * into the accessible region of the file.
1452 if (xfs_can_free_eofblocks(ip, true)) {
1453 error = xfs_free_eofblocks(mp, ip, false);
1459 * Writeback and invalidate cache for the remainder of the file as we're
1460 * about to shift down every extent from offset to EOF.
1462 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1466 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1467 offset >> PAGE_CACHE_SHIFT, -1);
1472 * The extent shiting code works on extent granularity. So, if
1473 * stop_fsb is not the starting block of extent, we need to split
1474 * the extent at stop_fsb.
1476 if (direction == SHIFT_RIGHT) {
1477 error = xfs_bmap_split_extent(ip, stop_fsb);
1482 while (!error && !done) {
1483 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1485 * We would need to reserve permanent block for transaction.
1486 * This will come into picture when after shifting extent into
1487 * hole we found that adjacent extents can be merged which
1488 * may lead to freeing of a block during record update.
1490 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1491 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1493 xfs_trans_cancel(tp);
1497 xfs_ilock(ip, XFS_ILOCK_EXCL);
1498 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1499 ip->i_gdquot, ip->i_pdquot,
1500 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1501 XFS_QMOPT_RES_REGBLKS);
1503 goto out_trans_cancel;
1505 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1507 xfs_bmap_init(&free_list, &first_block);
1510 * We are using the write transaction in which max 2 bmbt
1511 * updates are allowed
1513 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1514 &done, stop_fsb, &first_block, &free_list,
1515 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1517 goto out_bmap_cancel;
1519 error = xfs_bmap_finish(&tp, &free_list, NULL);
1521 goto out_bmap_cancel;
1523 error = xfs_trans_commit(tp);
1529 xfs_bmap_cancel(&free_list);
1531 xfs_trans_cancel(tp);
1536 * xfs_collapse_file_space()
1537 * This routine frees disk space and shift extent for the given file.
1538 * The first thing we do is to free data blocks in the specified range
1539 * by calling xfs_free_file_space(). It would also sync dirty data
1540 * and invalidate page cache over the region on which collapse range
1541 * is working. And Shift extent records to the left to cover a hole.
1548 xfs_collapse_file_space(
1549 struct xfs_inode *ip,
1555 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1556 trace_xfs_collapse_file_space(ip);
1558 error = xfs_free_file_space(ip, offset, len);
1562 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1566 * xfs_insert_file_space()
1567 * This routine create hole space by shifting extents for the given file.
1568 * The first thing we do is to sync dirty data and invalidate page cache
1569 * over the region on which insert range is working. And split an extent
1570 * to two extents at given offset by calling xfs_bmap_split_extent.
1571 * And shift all extent records which are laying between [offset,
1572 * last allocated extent] to the right to reserve hole range.
1578 xfs_insert_file_space(
1579 struct xfs_inode *ip,
1583 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1584 trace_xfs_insert_file_space(ip);
1586 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1590 * We need to check that the format of the data fork in the temporary inode is
1591 * valid for the target inode before doing the swap. This is not a problem with
1592 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1593 * data fork depending on the space the attribute fork is taking so we can get
1594 * invalid formats on the target inode.
1596 * E.g. target has space for 7 extents in extent format, temp inode only has
1597 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1598 * btree, but when swapped it needs to be in extent format. Hence we can't just
1599 * blindly swap data forks on attr2 filesystems.
1601 * Note that we check the swap in both directions so that we don't end up with
1602 * a corrupt temporary inode, either.
1604 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1605 * inode will prevent this situation from occurring, so all we do here is
1606 * reject and log the attempt. basically we are putting the responsibility on
1607 * userspace to get this right.
1610 xfs_swap_extents_check_format(
1611 xfs_inode_t *ip, /* target inode */
1612 xfs_inode_t *tip) /* tmp inode */
1615 /* Should never get a local format */
1616 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1617 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1621 * if the target inode has less extents that then temporary inode then
1622 * why did userspace call us?
1624 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1628 * if the target inode is in extent form and the temp inode is in btree
1629 * form then we will end up with the target inode in the wrong format
1630 * as we already know there are less extents in the temp inode.
1632 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1633 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1636 /* Check temp in extent form to max in target */
1637 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1638 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1639 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1642 /* Check target in extent form to max in temp */
1643 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1644 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1645 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1649 * If we are in a btree format, check that the temp root block will fit
1650 * in the target and that it has enough extents to be in btree format
1653 * Note that we have to be careful to allow btree->extent conversions
1654 * (a common defrag case) which will occur when the temp inode is in
1657 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1658 if (XFS_IFORK_BOFF(ip) &&
1659 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1661 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1662 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1666 /* Reciprocal target->temp btree format checks */
1667 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1668 if (XFS_IFORK_BOFF(tip) &&
1669 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1671 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1672 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1680 xfs_swap_extent_flush(
1681 struct xfs_inode *ip)
1685 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1688 truncate_pagecache_range(VFS_I(ip), 0, -1);
1690 /* Verify O_DIRECT for ftmp */
1691 if (VFS_I(ip)->i_mapping->nrpages)
1698 xfs_inode_t *ip, /* target inode */
1699 xfs_inode_t *tip, /* tmp inode */
1702 xfs_mount_t *mp = ip->i_mount;
1704 xfs_bstat_t *sbp = &sxp->sx_stat;
1705 xfs_ifork_t *tempifp, *ifp, *tifp;
1706 int src_log_flags, target_log_flags;
1713 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1720 * Lock the inodes against other IO, page faults and truncate to
1721 * begin with. Then we can ensure the inodes are flushed and have no
1722 * page cache safely. Once we have done this we can take the ilocks and
1723 * do the rest of the checks.
1725 lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1726 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1727 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1729 /* Verify that both files have the same format */
1730 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1735 /* Verify both files are either real-time or non-realtime */
1736 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1741 error = xfs_swap_extent_flush(ip);
1744 error = xfs_swap_extent_flush(tip);
1748 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1749 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1751 xfs_trans_cancel(tp);
1756 * Lock and join the inodes to the tansaction so that transaction commit
1757 * or cancel will unlock the inodes from this point onwards.
1759 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1760 lock_flags |= XFS_ILOCK_EXCL;
1761 xfs_trans_ijoin(tp, ip, lock_flags);
1762 xfs_trans_ijoin(tp, tip, lock_flags);
1765 /* Verify all data are being swapped */
1766 if (sxp->sx_offset != 0 ||
1767 sxp->sx_length != ip->i_d.di_size ||
1768 sxp->sx_length != tip->i_d.di_size) {
1770 goto out_trans_cancel;
1773 trace_xfs_swap_extent_before(ip, 0);
1774 trace_xfs_swap_extent_before(tip, 1);
1776 /* check inode formats now that data is flushed */
1777 error = xfs_swap_extents_check_format(ip, tip);
1780 "%s: inode 0x%llx format is incompatible for exchanging.",
1781 __func__, ip->i_ino);
1782 goto out_trans_cancel;
1786 * Compare the current change & modify times with that
1787 * passed in. If they differ, we abort this swap.
1788 * This is the mechanism used to ensure the calling
1789 * process that the file was not changed out from
1792 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1793 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1794 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1795 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1797 goto out_trans_cancel;
1800 * Count the number of extended attribute blocks
1802 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1803 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1804 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1806 goto out_trans_cancel;
1808 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1809 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1810 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1813 goto out_trans_cancel;
1817 * Before we've swapped the forks, lets set the owners of the forks
1818 * appropriately. We have to do this as we are demand paging the btree
1819 * buffers, and so the validation done on read will expect the owner
1820 * field to be correctly set. Once we change the owners, we can swap the
1823 * Note the trickiness in setting the log flags - we set the owner log
1824 * flag on the opposite inode (i.e. the inode we are setting the new
1825 * owner to be) because once we swap the forks and log that, log
1826 * recovery is going to see the fork as owned by the swapped inode,
1827 * not the pre-swapped inodes.
1829 src_log_flags = XFS_ILOG_CORE;
1830 target_log_flags = XFS_ILOG_CORE;
1831 if (ip->i_d.di_version == 3 &&
1832 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1833 target_log_flags |= XFS_ILOG_DOWNER;
1834 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1837 goto out_trans_cancel;
1840 if (tip->i_d.di_version == 3 &&
1841 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1842 src_log_flags |= XFS_ILOG_DOWNER;
1843 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1846 goto out_trans_cancel;
1850 * Swap the data forks of the inodes
1854 *tempifp = *ifp; /* struct copy */
1855 *ifp = *tifp; /* struct copy */
1856 *tifp = *tempifp; /* struct copy */
1859 * Fix the on-disk inode values
1861 tmp = (__uint64_t)ip->i_d.di_nblocks;
1862 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1863 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1865 tmp = (__uint64_t) ip->i_d.di_nextents;
1866 ip->i_d.di_nextents = tip->i_d.di_nextents;
1867 tip->i_d.di_nextents = tmp;
1869 tmp = (__uint64_t) ip->i_d.di_format;
1870 ip->i_d.di_format = tip->i_d.di_format;
1871 tip->i_d.di_format = tmp;
1874 * The extents in the source inode could still contain speculative
1875 * preallocation beyond EOF (e.g. the file is open but not modified
1876 * while defrag is in progress). In that case, we need to copy over the
1877 * number of delalloc blocks the data fork in the source inode is
1878 * tracking beyond EOF so that when the fork is truncated away when the
1879 * temporary inode is unlinked we don't underrun the i_delayed_blks
1880 * counter on that inode.
1882 ASSERT(tip->i_delayed_blks == 0);
1883 tip->i_delayed_blks = ip->i_delayed_blks;
1884 ip->i_delayed_blks = 0;
1886 switch (ip->i_d.di_format) {
1887 case XFS_DINODE_FMT_EXTENTS:
1888 /* If the extents fit in the inode, fix the
1889 * pointer. Otherwise it's already NULL or
1890 * pointing to the extent.
1892 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1893 ifp->if_u1.if_extents =
1894 ifp->if_u2.if_inline_ext;
1896 src_log_flags |= XFS_ILOG_DEXT;
1898 case XFS_DINODE_FMT_BTREE:
1899 ASSERT(ip->i_d.di_version < 3 ||
1900 (src_log_flags & XFS_ILOG_DOWNER));
1901 src_log_flags |= XFS_ILOG_DBROOT;
1905 switch (tip->i_d.di_format) {
1906 case XFS_DINODE_FMT_EXTENTS:
1907 /* If the extents fit in the inode, fix the
1908 * pointer. Otherwise it's already NULL or
1909 * pointing to the extent.
1911 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1912 tifp->if_u1.if_extents =
1913 tifp->if_u2.if_inline_ext;
1915 target_log_flags |= XFS_ILOG_DEXT;
1917 case XFS_DINODE_FMT_BTREE:
1918 target_log_flags |= XFS_ILOG_DBROOT;
1919 ASSERT(tip->i_d.di_version < 3 ||
1920 (target_log_flags & XFS_ILOG_DOWNER));
1924 xfs_trans_log_inode(tp, ip, src_log_flags);
1925 xfs_trans_log_inode(tp, tip, target_log_flags);
1928 * If this is a synchronous mount, make sure that the
1929 * transaction goes to disk before returning to the user.
1931 if (mp->m_flags & XFS_MOUNT_WSYNC)
1932 xfs_trans_set_sync(tp);
1934 error = xfs_trans_commit(tp);
1936 trace_xfs_swap_extent_after(ip, 0);
1937 trace_xfs_swap_extent_after(tip, 1);
1943 xfs_iunlock(ip, lock_flags);
1944 xfs_iunlock(tip, lock_flags);
1948 xfs_trans_cancel(tp);