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_blocks(VFS_I(ip), block, size);
81 * let the block layer decide on the fastest method of
82 * implementing the zeroing.
84 return sb_issue_zeroout(mp->m_super, block, count_fsb, GFP_NOFS);
89 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
90 * caller. Frees all the extents that need freeing, which must be done
91 * last due to locking considerations. We never free any extents in
92 * the first transaction.
94 * If an inode *ip is provided, rejoin it to the transaction if
95 * the transaction was committed.
99 struct xfs_trans **tp, /* transaction pointer addr */
100 struct xfs_bmap_free *flist, /* i/o: list extents to free */
101 struct xfs_inode *ip)
103 struct xfs_efd_log_item *efd; /* extent free data */
104 struct xfs_efi_log_item *efi; /* extent free intention */
105 int error; /* error return value */
106 int committed;/* xact committed or not */
107 struct xfs_bmap_free_item *free; /* free extent item */
108 struct xfs_bmap_free_item *next; /* next item on free list */
110 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
111 if (flist->xbf_count == 0)
114 efi = xfs_trans_get_efi(*tp, flist->xbf_count);
115 for (free = flist->xbf_first; free; free = free->xbfi_next)
116 xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock,
117 free->xbfi_blockcount);
119 error = __xfs_trans_roll(tp, ip, &committed);
122 * If the transaction was committed, drop the EFD reference
123 * since we're bailing out of here. The other reference is
124 * dropped when the EFI hits the AIL.
126 * If the transaction was not committed, the EFI is freed by the
127 * EFI item unlock handler on abort. Also, we have a new
128 * transaction so we should return committed=1 even though we're
129 * returning an error.
132 xfs_efi_release(efi);
133 xfs_force_shutdown((*tp)->t_mountp,
134 (error == -EFSCORRUPTED) ?
135 SHUTDOWN_CORRUPT_INCORE :
136 SHUTDOWN_META_IO_ERROR);
142 * Get an EFD and free each extent in the list, logging to the EFD in
143 * the process. The remaining bmap free list is cleaned up by the caller
146 efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count);
147 for (free = flist->xbf_first; free != NULL; free = next) {
148 next = free->xbfi_next;
150 error = xfs_trans_free_extent(*tp, efd, free->xbfi_startblock,
151 free->xbfi_blockcount);
155 xfs_bmap_del_free(flist, NULL, free);
163 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
165 xfs_alloctype_t atype = 0; /* type for allocation routines */
166 int error; /* error return value */
167 xfs_mount_t *mp; /* mount point structure */
168 xfs_extlen_t prod = 0; /* product factor for allocators */
169 xfs_extlen_t ralen = 0; /* realtime allocation length */
170 xfs_extlen_t align; /* minimum allocation alignment */
173 mp = ap->ip->i_mount;
174 align = xfs_get_extsz_hint(ap->ip);
175 prod = align / mp->m_sb.sb_rextsize;
176 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
177 align, 1, ap->eof, 0,
178 ap->conv, &ap->offset, &ap->length);
182 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
185 * If the offset & length are not perfectly aligned
186 * then kill prod, it will just get us in trouble.
188 if (do_mod(ap->offset, align) || ap->length % align)
191 * Set ralen to be the actual requested length in rtextents.
193 ralen = ap->length / mp->m_sb.sb_rextsize;
195 * If the old value was close enough to MAXEXTLEN that
196 * we rounded up to it, cut it back so it's valid again.
197 * Note that if it's a really large request (bigger than
198 * MAXEXTLEN), we don't hear about that number, and can't
199 * adjust the starting point to match it.
201 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
202 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
205 * Lock out modifications to both the RT bitmap and summary inodes
207 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
208 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
209 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL);
210 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
213 * If it's an allocation to an empty file at offset 0,
214 * pick an extent that will space things out in the rt area.
216 if (ap->eof && ap->offset == 0) {
217 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
219 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
222 ap->blkno = rtx * mp->m_sb.sb_rextsize;
227 xfs_bmap_adjacent(ap);
230 * Realtime allocation, done through xfs_rtallocate_extent.
232 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
233 do_div(ap->blkno, mp->m_sb.sb_rextsize);
236 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
237 &ralen, atype, ap->wasdel, prod, &rtb)))
239 if (rtb == NULLFSBLOCK && prod > 1 &&
240 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
241 ap->length, &ralen, atype,
242 ap->wasdel, 1, &rtb)))
245 if (ap->blkno != NULLFSBLOCK) {
246 ap->blkno *= mp->m_sb.sb_rextsize;
247 ralen *= mp->m_sb.sb_rextsize;
249 ap->ip->i_d.di_nblocks += ralen;
250 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
252 ap->ip->i_delayed_blks -= ralen;
254 * Adjust the disk quota also. This was reserved
257 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
258 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
259 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
261 /* Zero the extent if we were asked to do so */
262 if (ap->userdata & XFS_ALLOC_USERDATA_ZERO) {
263 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
274 * Check if the endoff is outside the last extent. If so the caller will grow
275 * the allocation to a stripe unit boundary. All offsets are considered outside
276 * the end of file for an empty fork, so 1 is returned in *eof in that case.
280 struct xfs_inode *ip,
281 xfs_fileoff_t endoff,
285 struct xfs_bmbt_irec rec;
288 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
292 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
297 * Extent tree block counting routines.
301 * Count leaf blocks given a range of extent records.
304 xfs_bmap_count_leaves(
312 for (b = 0; b < numrecs; b++) {
313 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
314 *count += xfs_bmbt_get_blockcount(frp);
319 * Count leaf blocks given a range of extent records originally
323 xfs_bmap_disk_count_leaves(
324 struct xfs_mount *mp,
325 struct xfs_btree_block *block,
332 for (b = 1; b <= numrecs; b++) {
333 frp = XFS_BMBT_REC_ADDR(mp, block, b);
334 *count += xfs_bmbt_disk_get_blockcount(frp);
339 * Recursively walks each level of a btree
340 * to count total fsblocks in use.
342 STATIC int /* error */
344 xfs_mount_t *mp, /* file system mount point */
345 xfs_trans_t *tp, /* transaction pointer */
346 xfs_ifork_t *ifp, /* inode fork pointer */
347 xfs_fsblock_t blockno, /* file system block number */
348 int levelin, /* level in btree */
349 int *count) /* Count of blocks */
355 xfs_fsblock_t bno = blockno;
356 xfs_fsblock_t nextbno;
357 struct xfs_btree_block *block, *nextblock;
360 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
365 block = XFS_BUF_TO_BLOCK(bp);
368 /* Not at node above leaves, count this level of nodes */
369 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
370 while (nextbno != NULLFSBLOCK) {
371 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
377 nextblock = XFS_BUF_TO_BLOCK(nbp);
378 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
379 xfs_trans_brelse(tp, nbp);
382 /* Dive to the next level */
383 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
384 bno = be64_to_cpu(*pp);
385 if (unlikely((error =
386 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
387 xfs_trans_brelse(tp, bp);
388 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
389 XFS_ERRLEVEL_LOW, mp);
390 return -EFSCORRUPTED;
392 xfs_trans_brelse(tp, bp);
394 /* count all level 1 nodes and their leaves */
396 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
397 numrecs = be16_to_cpu(block->bb_numrecs);
398 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
399 xfs_trans_brelse(tp, bp);
400 if (nextbno == NULLFSBLOCK)
403 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
409 block = XFS_BUF_TO_BLOCK(bp);
416 * Count fsblocks of the given fork.
419 xfs_bmap_count_blocks(
420 xfs_trans_t *tp, /* transaction pointer */
421 xfs_inode_t *ip, /* incore inode */
422 int whichfork, /* data or attr fork */
423 int *count) /* out: count of blocks */
425 struct xfs_btree_block *block; /* current btree block */
426 xfs_fsblock_t bno; /* block # of "block" */
427 xfs_ifork_t *ifp; /* fork structure */
428 int level; /* btree level, for checking */
429 xfs_mount_t *mp; /* file system mount structure */
430 __be64 *pp; /* pointer to block address */
434 ifp = XFS_IFORK_PTR(ip, whichfork);
435 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
436 xfs_bmap_count_leaves(ifp, 0,
437 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
443 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
445 block = ifp->if_broot;
446 level = be16_to_cpu(block->bb_level);
448 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
449 bno = be64_to_cpu(*pp);
450 ASSERT(bno != NULLFSBLOCK);
451 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
452 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
454 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
455 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
457 return -EFSCORRUPTED;
464 * returns 1 for success, 0 if we failed to map the extent.
467 xfs_getbmapx_fix_eof_hole(
468 xfs_inode_t *ip, /* xfs incore inode pointer */
469 struct getbmapx *out, /* output structure */
470 int prealloced, /* this is a file with
471 * preallocated data space */
472 __int64_t end, /* last block requested */
473 xfs_fsblock_t startblock)
476 xfs_mount_t *mp; /* file system mount point */
477 xfs_ifork_t *ifp; /* inode fork pointer */
478 xfs_extnum_t lastx; /* last extent pointer */
479 xfs_fileoff_t fileblock;
481 if (startblock == HOLESTARTBLOCK) {
484 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
485 fixlen -= out->bmv_offset;
486 if (prealloced && out->bmv_offset + out->bmv_length == end) {
487 /* Came to hole at EOF. Trim it. */
490 out->bmv_length = fixlen;
493 if (startblock == DELAYSTARTBLOCK)
496 out->bmv_block = xfs_fsb_to_db(ip, startblock);
497 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
498 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
499 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
500 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
501 out->bmv_oflags |= BMV_OF_LAST;
508 * Get inode's extents as described in bmv, and format for output.
509 * Calls formatter to fill the user's buffer until all extents
510 * are mapped, until the passed-in bmv->bmv_count slots have
511 * been filled, or until the formatter short-circuits the loop,
512 * if it is tracking filled-in extents on its own.
517 struct getbmapx *bmv, /* user bmap structure */
518 xfs_bmap_format_t formatter, /* format to user */
519 void *arg) /* formatter arg */
521 __int64_t bmvend; /* last block requested */
522 int error = 0; /* return value */
523 __int64_t fixlen; /* length for -1 case */
524 int i; /* extent number */
525 int lock; /* lock state */
526 xfs_bmbt_irec_t *map; /* buffer for user's data */
527 xfs_mount_t *mp; /* file system mount point */
528 int nex; /* # of user extents can do */
529 int nexleft; /* # of user extents left */
530 int subnex; /* # of bmapi's can do */
531 int nmap; /* number of map entries */
532 struct getbmapx *out; /* output structure */
533 int whichfork; /* data or attr fork */
534 int prealloced; /* this is a file with
535 * preallocated data space */
536 int iflags; /* interface flags */
537 int bmapi_flags; /* flags for xfs_bmapi */
541 iflags = bmv->bmv_iflags;
542 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
544 if (whichfork == XFS_ATTR_FORK) {
545 if (XFS_IFORK_Q(ip)) {
546 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
547 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
548 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
551 ip->i_d.di_aformat != 0 &&
552 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
553 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
555 return -EFSCORRUPTED;
561 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
562 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
563 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
566 if (xfs_get_extsz_hint(ip) ||
567 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
569 fixlen = mp->m_super->s_maxbytes;
572 fixlen = XFS_ISIZE(ip);
576 if (bmv->bmv_length == -1) {
577 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
579 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
580 } else if (bmv->bmv_length == 0) {
581 bmv->bmv_entries = 0;
583 } else if (bmv->bmv_length < 0) {
587 nex = bmv->bmv_count - 1;
590 bmvend = bmv->bmv_offset + bmv->bmv_length;
593 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
595 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
599 xfs_ilock(ip, XFS_IOLOCK_SHARED);
600 if (whichfork == XFS_DATA_FORK) {
601 if (!(iflags & BMV_IF_DELALLOC) &&
602 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
603 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
605 goto out_unlock_iolock;
608 * Even after flushing the inode, there can still be
609 * delalloc blocks on the inode beyond EOF due to
610 * speculative preallocation. These are not removed
611 * until the release function is called or the inode
612 * is inactivated. Hence we cannot assert here that
613 * ip->i_delayed_blks == 0.
617 lock = xfs_ilock_data_map_shared(ip);
619 lock = xfs_ilock_attr_map_shared(ip);
623 * Don't let nex be bigger than the number of extents
624 * we can have assuming alternating holes and real extents.
626 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
627 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
629 bmapi_flags = xfs_bmapi_aflag(whichfork);
630 if (!(iflags & BMV_IF_PREALLOC))
631 bmapi_flags |= XFS_BMAPI_IGSTATE;
634 * Allocate enough space to handle "subnex" maps at a time.
638 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
640 goto out_unlock_ilock;
642 bmv->bmv_entries = 0;
644 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
645 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
653 nmap = (nexleft > subnex) ? subnex : nexleft;
654 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
655 XFS_BB_TO_FSB(mp, bmv->bmv_length),
656 map, &nmap, bmapi_flags);
659 ASSERT(nmap <= subnex);
661 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
662 out[cur_ext].bmv_oflags = 0;
663 if (map[i].br_state == XFS_EXT_UNWRITTEN)
664 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
665 else if (map[i].br_startblock == DELAYSTARTBLOCK)
666 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
667 out[cur_ext].bmv_offset =
668 XFS_FSB_TO_BB(mp, map[i].br_startoff);
669 out[cur_ext].bmv_length =
670 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
671 out[cur_ext].bmv_unused1 = 0;
672 out[cur_ext].bmv_unused2 = 0;
675 * delayed allocation extents that start beyond EOF can
676 * occur due to speculative EOF allocation when the
677 * delalloc extent is larger than the largest freespace
678 * extent at conversion time. These extents cannot be
679 * converted by data writeback, so can exist here even
680 * if we are not supposed to be finding delalloc
683 if (map[i].br_startblock == DELAYSTARTBLOCK &&
684 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
685 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
687 if (map[i].br_startblock == HOLESTARTBLOCK &&
688 whichfork == XFS_ATTR_FORK) {
689 /* came to the end of attribute fork */
690 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
694 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
696 map[i].br_startblock))
700 out[cur_ext].bmv_offset +
701 out[cur_ext].bmv_length;
703 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
706 * In case we don't want to return the hole,
707 * don't increase cur_ext so that we can reuse
708 * it in the next loop.
710 if ((iflags & BMV_IF_NO_HOLES) &&
711 map[i].br_startblock == HOLESTARTBLOCK) {
712 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
720 } while (nmap && nexleft && bmv->bmv_length);
725 xfs_iunlock(ip, lock);
727 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
729 for (i = 0; i < cur_ext; i++) {
730 int full = 0; /* user array is full */
732 /* format results & advance arg */
733 error = formatter(&arg, &out[i], &full);
743 * dead simple method of punching delalyed allocation blocks from a range in
744 * the inode. Walks a block at a time so will be slow, but is only executed in
745 * rare error cases so the overhead is not critical. This will always punch out
746 * both the start and end blocks, even if the ranges only partially overlap
747 * them, so it is up to the caller to ensure that partial blocks are not
751 xfs_bmap_punch_delalloc_range(
752 struct xfs_inode *ip,
753 xfs_fileoff_t start_fsb,
754 xfs_fileoff_t length)
756 xfs_fileoff_t remaining = length;
759 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
763 xfs_bmbt_irec_t imap;
765 xfs_fsblock_t firstblock;
766 xfs_bmap_free_t flist;
769 * Map the range first and check that it is a delalloc extent
770 * before trying to unmap the range. Otherwise we will be
771 * trying to remove a real extent (which requires a
772 * transaction) or a hole, which is probably a bad idea...
774 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
778 /* something screwed, just bail */
779 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
780 xfs_alert(ip->i_mount,
781 "Failed delalloc mapping lookup ino %lld fsb %lld.",
782 ip->i_ino, start_fsb);
790 if (imap.br_startblock != DELAYSTARTBLOCK) {
791 /* been converted, ignore */
794 WARN_ON(imap.br_blockcount == 0);
797 * Note: while we initialise the firstblock/flist pair, they
798 * should never be used because blocks should never be
799 * allocated or freed for a delalloc extent and hence we need
800 * don't cancel or finish them after the xfs_bunmapi() call.
802 xfs_bmap_init(&flist, &firstblock);
803 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
808 ASSERT(!flist.xbf_count && !flist.xbf_first);
812 } while(remaining > 0);
818 * Test whether it is appropriate to check an inode for and free post EOF
819 * blocks. The 'force' parameter determines whether we should also consider
820 * regular files that are marked preallocated or append-only.
823 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
825 /* prealloc/delalloc exists only on regular files */
826 if (!S_ISREG(ip->i_d.di_mode))
830 * Zero sized files with no cached pages and delalloc blocks will not
831 * have speculative prealloc/delalloc blocks to remove.
833 if (VFS_I(ip)->i_size == 0 &&
834 VFS_I(ip)->i_mapping->nrpages == 0 &&
835 ip->i_delayed_blks == 0)
838 /* If we haven't read in the extent list, then don't do it now. */
839 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
843 * Do not free real preallocated or append-only files unless the file
844 * has delalloc blocks and we are forced to remove them.
846 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
847 if (!force || ip->i_delayed_blks == 0)
854 * This is called by xfs_inactive to free any blocks beyond eof
855 * when the link count isn't zero and by xfs_dm_punch_hole() when
856 * punching a hole to EOF.
866 xfs_fileoff_t end_fsb;
867 xfs_fileoff_t last_fsb;
868 xfs_filblks_t map_len;
870 xfs_bmbt_irec_t imap;
873 * Figure out if there are any blocks beyond the end
874 * of the file. If not, then there is nothing to do.
876 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
877 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
878 if (last_fsb <= end_fsb)
880 map_len = last_fsb - end_fsb;
883 xfs_ilock(ip, XFS_ILOCK_SHARED);
884 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
885 xfs_iunlock(ip, XFS_ILOCK_SHARED);
887 if (!error && (nimaps != 0) &&
888 (imap.br_startblock != HOLESTARTBLOCK ||
889 ip->i_delayed_blks)) {
891 * Attach the dquots to the inode up front.
893 error = xfs_qm_dqattach(ip, 0);
898 * There are blocks after the end of file.
899 * Free them up now by truncating the file to
902 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
905 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
906 xfs_trans_cancel(tp);
911 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
913 ASSERT(XFS_FORCED_SHUTDOWN(mp));
914 xfs_trans_cancel(tp);
916 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
920 xfs_ilock(ip, XFS_ILOCK_EXCL);
921 xfs_trans_ijoin(tp, ip, 0);
924 * Do not update the on-disk file size. If we update the
925 * on-disk file size and then the system crashes before the
926 * contents of the file are flushed to disk then the files
927 * may be full of holes (ie NULL files bug).
929 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
933 * If we get an error at this point we simply don't
934 * bother truncating the file.
936 xfs_trans_cancel(tp);
938 error = xfs_trans_commit(tp);
940 xfs_inode_clear_eofblocks_tag(ip);
943 xfs_iunlock(ip, XFS_ILOCK_EXCL);
945 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
951 xfs_alloc_file_space(
952 struct xfs_inode *ip,
957 xfs_mount_t *mp = ip->i_mount;
959 xfs_filblks_t allocated_fsb;
960 xfs_filblks_t allocatesize_fsb;
961 xfs_extlen_t extsz, temp;
962 xfs_fileoff_t startoffset_fsb;
963 xfs_fsblock_t firstfsb;
968 xfs_bmbt_irec_t imaps[1], *imapp;
969 xfs_bmap_free_t free_list;
970 uint qblocks, resblks, resrtextents;
973 trace_xfs_alloc_file_space(ip);
975 if (XFS_FORCED_SHUTDOWN(mp))
978 error = xfs_qm_dqattach(ip, 0);
985 rt = XFS_IS_REALTIME_INODE(ip);
986 extsz = xfs_get_extsz_hint(ip);
991 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
992 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
995 * Allocate file space until done or until there is an error
997 while (allocatesize_fsb && !error) {
1001 * Determine space reservations for data/realtime.
1003 if (unlikely(extsz)) {
1004 s = startoffset_fsb;
1007 e = startoffset_fsb + allocatesize_fsb;
1008 if ((temp = do_mod(startoffset_fsb, extsz)))
1010 if ((temp = do_mod(e, extsz)))
1014 e = allocatesize_fsb;
1018 * The transaction reservation is limited to a 32-bit block
1019 * count, hence we need to limit the number of blocks we are
1020 * trying to reserve to avoid an overflow. We can't allocate
1021 * more than @nimaps extents, and an extent is limited on disk
1022 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1024 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1026 resrtextents = qblocks = resblks;
1027 resrtextents /= mp->m_sb.sb_rextsize;
1028 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1029 quota_flag = XFS_QMOPT_RES_RTBLKS;
1032 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1033 quota_flag = XFS_QMOPT_RES_REGBLKS;
1037 * Allocate and setup the transaction.
1039 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1040 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1041 resblks, resrtextents);
1043 * Check for running out of space
1047 * Free the transaction structure.
1049 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1050 xfs_trans_cancel(tp);
1053 xfs_ilock(ip, XFS_ILOCK_EXCL);
1054 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1059 xfs_trans_ijoin(tp, ip, 0);
1061 xfs_bmap_init(&free_list, &firstfsb);
1062 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1063 allocatesize_fsb, alloc_type, &firstfsb,
1064 resblks, imapp, &nimaps, &free_list);
1069 * Complete the transaction
1071 error = xfs_bmap_finish(&tp, &free_list, NULL);
1075 error = xfs_trans_commit(tp);
1076 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1080 allocated_fsb = imapp->br_blockcount;
1087 startoffset_fsb += allocated_fsb;
1088 allocatesize_fsb -= allocated_fsb;
1093 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1094 xfs_bmap_cancel(&free_list);
1095 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1097 error1: /* Just cancel transaction */
1098 xfs_trans_cancel(tp);
1099 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1104 * Zero file bytes between startoff and endoff inclusive.
1105 * The iolock is held exclusive and no blocks are buffered.
1107 * This function is used by xfs_free_file_space() to zero
1108 * partial blocks when the range to free is not block aligned.
1109 * When unreserving space with boundaries that are not block
1110 * aligned we round up the start and round down the end
1111 * boundaries and then use this function to zero the parts of
1112 * the blocks that got dropped during the rounding.
1115 xfs_zero_remaining_bytes(
1120 xfs_bmbt_irec_t imap;
1121 xfs_fileoff_t offset_fsb;
1122 xfs_off_t lastoffset;
1125 xfs_mount_t *mp = ip->i_mount;
1130 * Avoid doing I/O beyond eof - it's not necessary
1131 * since nothing can read beyond eof. The space will
1132 * be zeroed when the file is extended anyway.
1134 if (startoff >= XFS_ISIZE(ip))
1137 if (endoff > XFS_ISIZE(ip))
1138 endoff = XFS_ISIZE(ip);
1140 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1143 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1146 lock_mode = xfs_ilock_data_map_shared(ip);
1147 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1148 xfs_iunlock(ip, lock_mode);
1150 if (error || nimap < 1)
1152 ASSERT(imap.br_blockcount >= 1);
1153 ASSERT(imap.br_startoff == offset_fsb);
1154 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1156 if (imap.br_startblock == HOLESTARTBLOCK ||
1157 imap.br_state == XFS_EXT_UNWRITTEN) {
1158 /* skip the entire extent */
1159 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff +
1160 imap.br_blockcount) - 1;
1164 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1165 if (lastoffset > endoff)
1166 lastoffset = endoff;
1168 /* DAX can just zero the backing device directly */
1169 if (IS_DAX(VFS_I(ip))) {
1170 error = dax_zero_page_range(VFS_I(ip), offset,
1171 lastoffset - offset + 1,
1172 xfs_get_blocks_direct);
1178 error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ?
1179 mp->m_rtdev_targp : mp->m_ddev_targp,
1180 xfs_fsb_to_db(ip, imap.br_startblock),
1181 BTOBB(mp->m_sb.sb_blocksize),
1187 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1188 0, lastoffset - offset + 1);
1190 error = xfs_bwrite(bp);
1199 xfs_free_file_space(
1200 struct xfs_inode *ip,
1205 xfs_fileoff_t endoffset_fsb;
1207 xfs_fsblock_t firstfsb;
1208 xfs_bmap_free_t free_list;
1209 xfs_bmbt_irec_t imap;
1211 xfs_off_t iendoffset;
1218 xfs_fileoff_t startoffset_fsb;
1223 trace_xfs_free_file_space(ip);
1225 error = xfs_qm_dqattach(ip, 0);
1230 if (len <= 0) /* if nothing being freed */
1232 rt = XFS_IS_REALTIME_INODE(ip);
1233 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1234 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1236 /* wait for the completion of any pending DIOs */
1237 inode_dio_wait(VFS_I(ip));
1239 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1240 ioffset = round_down(offset, rounding);
1241 iendoffset = round_up(offset + len, rounding) - 1;
1242 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, ioffset,
1246 truncate_pagecache_range(VFS_I(ip), ioffset, iendoffset);
1249 * Need to zero the stuff we're not freeing, on disk.
1250 * If it's a realtime file & can't use unwritten extents then we
1251 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1252 * will take care of it for us.
1254 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1256 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1260 ASSERT(nimap == 0 || nimap == 1);
1261 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1264 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1265 block = imap.br_startblock;
1266 mod = do_div(block, mp->m_sb.sb_rextsize);
1268 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1271 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1275 ASSERT(nimap == 0 || nimap == 1);
1276 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1277 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1279 if (mod && (mod != mp->m_sb.sb_rextsize))
1280 endoffset_fsb -= mod;
1283 if ((done = (endoffset_fsb <= startoffset_fsb)))
1285 * One contiguous piece to clear
1287 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1290 * Some full blocks, possibly two pieces to clear
1292 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1293 error = xfs_zero_remaining_bytes(ip, offset,
1294 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1296 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1297 error = xfs_zero_remaining_bytes(ip,
1298 XFS_FSB_TO_B(mp, endoffset_fsb),
1303 * free file space until done or until there is an error
1305 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1306 while (!error && !done) {
1309 * allocate and setup the transaction. Allow this
1310 * transaction to dip into the reserve blocks to ensure
1311 * the freeing of the space succeeds at ENOSPC.
1313 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1314 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1317 * check for running out of space
1321 * Free the transaction structure.
1323 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1324 xfs_trans_cancel(tp);
1327 xfs_ilock(ip, XFS_ILOCK_EXCL);
1328 error = xfs_trans_reserve_quota(tp, mp,
1329 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1330 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1334 xfs_trans_ijoin(tp, ip, 0);
1337 * issue the bunmapi() call to free the blocks
1339 xfs_bmap_init(&free_list, &firstfsb);
1340 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1341 endoffset_fsb - startoffset_fsb,
1342 0, 2, &firstfsb, &free_list, &done);
1347 * complete the transaction
1349 error = xfs_bmap_finish(&tp, &free_list, NULL);
1353 error = xfs_trans_commit(tp);
1354 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1361 xfs_bmap_cancel(&free_list);
1363 xfs_trans_cancel(tp);
1364 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1369 * Preallocate and zero a range of a file. This mechanism has the allocation
1370 * semantics of fallocate and in addition converts data in the range to zeroes.
1373 xfs_zero_file_space(
1374 struct xfs_inode *ip,
1378 struct xfs_mount *mp = ip->i_mount;
1382 trace_xfs_zero_file_space(ip);
1384 blksize = 1 << mp->m_sb.sb_blocklog;
1387 * Punch a hole and prealloc the range. We use hole punch rather than
1388 * unwritten extent conversion for two reasons:
1390 * 1.) Hole punch handles partial block zeroing for us.
1392 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1393 * by virtue of the hole punch.
1395 error = xfs_free_file_space(ip, offset, len);
1399 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1400 round_up(offset + len, blksize) -
1401 round_down(offset, blksize),
1402 XFS_BMAPI_PREALLOC);
1409 * @next_fsb will keep track of the extent currently undergoing shift.
1410 * @stop_fsb will keep track of the extent at which we have to stop.
1411 * If we are shifting left, we will start with block (offset + len) and
1412 * shift each extent till last extent.
1413 * If we are shifting right, we will start with last extent inside file space
1414 * and continue until we reach the block corresponding to offset.
1417 xfs_shift_file_space(
1418 struct xfs_inode *ip,
1421 enum shift_direction direction)
1424 struct xfs_mount *mp = ip->i_mount;
1425 struct xfs_trans *tp;
1427 struct xfs_bmap_free free_list;
1428 xfs_fsblock_t first_block;
1429 xfs_fileoff_t stop_fsb;
1430 xfs_fileoff_t next_fsb;
1431 xfs_fileoff_t shift_fsb;
1433 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1435 if (direction == SHIFT_LEFT) {
1436 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1437 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1440 * If right shift, delegate the work of initialization of
1441 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1443 next_fsb = NULLFSBLOCK;
1444 stop_fsb = XFS_B_TO_FSB(mp, offset);
1447 shift_fsb = XFS_B_TO_FSB(mp, len);
1450 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1451 * into the accessible region of the file.
1453 if (xfs_can_free_eofblocks(ip, true)) {
1454 error = xfs_free_eofblocks(mp, ip, false);
1460 * Writeback and invalidate cache for the remainder of the file as we're
1461 * about to shift down every extent from offset to EOF.
1463 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1467 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1468 offset >> PAGE_CACHE_SHIFT, -1);
1473 * The extent shiting code works on extent granularity. So, if
1474 * stop_fsb is not the starting block of extent, we need to split
1475 * the extent at stop_fsb.
1477 if (direction == SHIFT_RIGHT) {
1478 error = xfs_bmap_split_extent(ip, stop_fsb);
1483 while (!error && !done) {
1484 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1486 * We would need to reserve permanent block for transaction.
1487 * This will come into picture when after shifting extent into
1488 * hole we found that adjacent extents can be merged which
1489 * may lead to freeing of a block during record update.
1491 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1492 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1494 xfs_trans_cancel(tp);
1498 xfs_ilock(ip, XFS_ILOCK_EXCL);
1499 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1500 ip->i_gdquot, ip->i_pdquot,
1501 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1502 XFS_QMOPT_RES_REGBLKS);
1504 goto out_trans_cancel;
1506 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1508 xfs_bmap_init(&free_list, &first_block);
1511 * We are using the write transaction in which max 2 bmbt
1512 * updates are allowed
1514 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1515 &done, stop_fsb, &first_block, &free_list,
1516 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1518 goto out_bmap_cancel;
1520 error = xfs_bmap_finish(&tp, &free_list, NULL);
1522 goto out_bmap_cancel;
1524 error = xfs_trans_commit(tp);
1530 xfs_bmap_cancel(&free_list);
1532 xfs_trans_cancel(tp);
1537 * xfs_collapse_file_space()
1538 * This routine frees disk space and shift extent for the given file.
1539 * The first thing we do is to free data blocks in the specified range
1540 * by calling xfs_free_file_space(). It would also sync dirty data
1541 * and invalidate page cache over the region on which collapse range
1542 * is working. And Shift extent records to the left to cover a hole.
1549 xfs_collapse_file_space(
1550 struct xfs_inode *ip,
1556 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1557 trace_xfs_collapse_file_space(ip);
1559 error = xfs_free_file_space(ip, offset, len);
1563 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1567 * xfs_insert_file_space()
1568 * This routine create hole space by shifting extents for the given file.
1569 * The first thing we do is to sync dirty data and invalidate page cache
1570 * over the region on which insert range is working. And split an extent
1571 * to two extents at given offset by calling xfs_bmap_split_extent.
1572 * And shift all extent records which are laying between [offset,
1573 * last allocated extent] to the right to reserve hole range.
1579 xfs_insert_file_space(
1580 struct xfs_inode *ip,
1584 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1585 trace_xfs_insert_file_space(ip);
1587 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1591 * We need to check that the format of the data fork in the temporary inode is
1592 * valid for the target inode before doing the swap. This is not a problem with
1593 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1594 * data fork depending on the space the attribute fork is taking so we can get
1595 * invalid formats on the target inode.
1597 * E.g. target has space for 7 extents in extent format, temp inode only has
1598 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1599 * btree, but when swapped it needs to be in extent format. Hence we can't just
1600 * blindly swap data forks on attr2 filesystems.
1602 * Note that we check the swap in both directions so that we don't end up with
1603 * a corrupt temporary inode, either.
1605 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1606 * inode will prevent this situation from occurring, so all we do here is
1607 * reject and log the attempt. basically we are putting the responsibility on
1608 * userspace to get this right.
1611 xfs_swap_extents_check_format(
1612 xfs_inode_t *ip, /* target inode */
1613 xfs_inode_t *tip) /* tmp inode */
1616 /* Should never get a local format */
1617 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1618 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1622 * if the target inode has less extents that then temporary inode then
1623 * why did userspace call us?
1625 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1629 * if the target inode is in extent form and the temp inode is in btree
1630 * form then we will end up with the target inode in the wrong format
1631 * as we already know there are less extents in the temp inode.
1633 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1634 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1637 /* Check temp in extent form to max in target */
1638 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1639 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1640 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1643 /* Check target in extent form to max in temp */
1644 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1645 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1646 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1650 * If we are in a btree format, check that the temp root block will fit
1651 * in the target and that it has enough extents to be in btree format
1654 * Note that we have to be careful to allow btree->extent conversions
1655 * (a common defrag case) which will occur when the temp inode is in
1658 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1659 if (XFS_IFORK_BOFF(ip) &&
1660 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1662 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1663 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1667 /* Reciprocal target->temp btree format checks */
1668 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1669 if (XFS_IFORK_BOFF(tip) &&
1670 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1672 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1673 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1681 xfs_swap_extent_flush(
1682 struct xfs_inode *ip)
1686 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1689 truncate_pagecache_range(VFS_I(ip), 0, -1);
1691 /* Verify O_DIRECT for ftmp */
1692 if (VFS_I(ip)->i_mapping->nrpages)
1699 xfs_inode_t *ip, /* target inode */
1700 xfs_inode_t *tip, /* tmp inode */
1703 xfs_mount_t *mp = ip->i_mount;
1705 xfs_bstat_t *sbp = &sxp->sx_stat;
1706 xfs_ifork_t *tempifp, *ifp, *tifp;
1707 int src_log_flags, target_log_flags;
1714 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1721 * Lock the inodes against other IO, page faults and truncate to
1722 * begin with. Then we can ensure the inodes are flushed and have no
1723 * page cache safely. Once we have done this we can take the ilocks and
1724 * do the rest of the checks.
1726 lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1727 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1728 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1730 /* Verify that both files have the same format */
1731 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1736 /* Verify both files are either real-time or non-realtime */
1737 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1742 error = xfs_swap_extent_flush(ip);
1745 error = xfs_swap_extent_flush(tip);
1749 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1750 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1752 xfs_trans_cancel(tp);
1757 * Lock and join the inodes to the tansaction so that transaction commit
1758 * or cancel will unlock the inodes from this point onwards.
1760 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1761 lock_flags |= XFS_ILOCK_EXCL;
1762 xfs_trans_ijoin(tp, ip, lock_flags);
1763 xfs_trans_ijoin(tp, tip, lock_flags);
1766 /* Verify all data are being swapped */
1767 if (sxp->sx_offset != 0 ||
1768 sxp->sx_length != ip->i_d.di_size ||
1769 sxp->sx_length != tip->i_d.di_size) {
1771 goto out_trans_cancel;
1774 trace_xfs_swap_extent_before(ip, 0);
1775 trace_xfs_swap_extent_before(tip, 1);
1777 /* check inode formats now that data is flushed */
1778 error = xfs_swap_extents_check_format(ip, tip);
1781 "%s: inode 0x%llx format is incompatible for exchanging.",
1782 __func__, ip->i_ino);
1783 goto out_trans_cancel;
1787 * Compare the current change & modify times with that
1788 * passed in. If they differ, we abort this swap.
1789 * This is the mechanism used to ensure the calling
1790 * process that the file was not changed out from
1793 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1794 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1795 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1796 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1798 goto out_trans_cancel;
1801 * Count the number of extended attribute blocks
1803 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1804 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1805 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1807 goto out_trans_cancel;
1809 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1810 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1811 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1814 goto out_trans_cancel;
1818 * Before we've swapped the forks, lets set the owners of the forks
1819 * appropriately. We have to do this as we are demand paging the btree
1820 * buffers, and so the validation done on read will expect the owner
1821 * field to be correctly set. Once we change the owners, we can swap the
1824 * Note the trickiness in setting the log flags - we set the owner log
1825 * flag on the opposite inode (i.e. the inode we are setting the new
1826 * owner to be) because once we swap the forks and log that, log
1827 * recovery is going to see the fork as owned by the swapped inode,
1828 * not the pre-swapped inodes.
1830 src_log_flags = XFS_ILOG_CORE;
1831 target_log_flags = XFS_ILOG_CORE;
1832 if (ip->i_d.di_version == 3 &&
1833 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1834 target_log_flags |= XFS_ILOG_DOWNER;
1835 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1838 goto out_trans_cancel;
1841 if (tip->i_d.di_version == 3 &&
1842 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1843 src_log_flags |= XFS_ILOG_DOWNER;
1844 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1847 goto out_trans_cancel;
1851 * Swap the data forks of the inodes
1855 *tempifp = *ifp; /* struct copy */
1856 *ifp = *tifp; /* struct copy */
1857 *tifp = *tempifp; /* struct copy */
1860 * Fix the on-disk inode values
1862 tmp = (__uint64_t)ip->i_d.di_nblocks;
1863 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1864 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1866 tmp = (__uint64_t) ip->i_d.di_nextents;
1867 ip->i_d.di_nextents = tip->i_d.di_nextents;
1868 tip->i_d.di_nextents = tmp;
1870 tmp = (__uint64_t) ip->i_d.di_format;
1871 ip->i_d.di_format = tip->i_d.di_format;
1872 tip->i_d.di_format = tmp;
1875 * The extents in the source inode could still contain speculative
1876 * preallocation beyond EOF (e.g. the file is open but not modified
1877 * while defrag is in progress). In that case, we need to copy over the
1878 * number of delalloc blocks the data fork in the source inode is
1879 * tracking beyond EOF so that when the fork is truncated away when the
1880 * temporary inode is unlinked we don't underrun the i_delayed_blks
1881 * counter on that inode.
1883 ASSERT(tip->i_delayed_blks == 0);
1884 tip->i_delayed_blks = ip->i_delayed_blks;
1885 ip->i_delayed_blks = 0;
1887 switch (ip->i_d.di_format) {
1888 case XFS_DINODE_FMT_EXTENTS:
1889 /* If the extents fit in the inode, fix the
1890 * pointer. Otherwise it's already NULL or
1891 * pointing to the extent.
1893 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1894 ifp->if_u1.if_extents =
1895 ifp->if_u2.if_inline_ext;
1897 src_log_flags |= XFS_ILOG_DEXT;
1899 case XFS_DINODE_FMT_BTREE:
1900 ASSERT(ip->i_d.di_version < 3 ||
1901 (src_log_flags & XFS_ILOG_DOWNER));
1902 src_log_flags |= XFS_ILOG_DBROOT;
1906 switch (tip->i_d.di_format) {
1907 case XFS_DINODE_FMT_EXTENTS:
1908 /* If the extents fit in the inode, fix the
1909 * pointer. Otherwise it's already NULL or
1910 * pointing to the extent.
1912 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1913 tifp->if_u1.if_extents =
1914 tifp->if_u2.if_inline_ext;
1916 target_log_flags |= XFS_ILOG_DEXT;
1918 case XFS_DINODE_FMT_BTREE:
1919 target_log_flags |= XFS_ILOG_DBROOT;
1920 ASSERT(tip->i_d.di_version < 3 ||
1921 (target_log_flags & XFS_ILOG_DOWNER));
1925 xfs_trans_log_inode(tp, ip, src_log_flags);
1926 xfs_trans_log_inode(tp, tip, target_log_flags);
1929 * If this is a synchronous mount, make sure that the
1930 * transaction goes to disk before returning to the user.
1932 if (mp->m_flags & XFS_MOUNT_WSYNC)
1933 xfs_trans_set_sync(tp);
1935 error = xfs_trans_commit(tp);
1937 trace_xfs_swap_extent_after(ip, 0);
1938 trace_xfs_swap_extent_after(tip, 1);
1944 xfs_iunlock(ip, lock_flags);
1945 xfs_iunlock(tip, lock_flags);
1949 xfs_trans_cancel(tp);