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 be called at transaction's end by xfs_bmapi, xfs_bunmapi
61 * caller. Frees all the extents that need freeing, which must be done
62 * last due to locking considerations. We never free any extents in
63 * the first transaction.
65 * Return 1 if the given transaction was committed and a new one
66 * started, and 0 otherwise in the committed parameter.
70 struct xfs_trans **tp, /* transaction pointer addr */
71 struct xfs_bmap_free *flist, /* i/o: list extents to free */
72 int *committed)/* xact committed or not */
74 struct xfs_efd_log_item *efd; /* extent free data */
75 struct xfs_efi_log_item *efi; /* extent free intention */
76 int error; /* error return value */
77 struct xfs_bmap_free_item *free; /* free extent item */
78 struct xfs_bmap_free_item *next; /* next item on free list */
80 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
81 if (flist->xbf_count == 0) {
85 efi = xfs_trans_get_efi(*tp, flist->xbf_count);
86 for (free = flist->xbf_first; free; free = free->xbfi_next)
87 xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock,
88 free->xbfi_blockcount);
90 error = __xfs_trans_roll(tp, NULL, committed);
93 * If the transaction was committed, drop the EFD reference
94 * since we're bailing out of here. The other reference is
95 * dropped when the EFI hits the AIL.
97 * If the transaction was not committed, the EFI is freed by the
98 * EFI item unlock handler on abort. Also, we have a new
99 * transaction so we should return committed=1 even though we're
100 * returning an error.
103 xfs_efi_release(efi);
104 xfs_force_shutdown((*tp)->t_mountp,
105 (error == -EFSCORRUPTED) ?
106 SHUTDOWN_CORRUPT_INCORE :
107 SHUTDOWN_META_IO_ERROR);
115 efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count);
116 for (free = flist->xbf_first; free != NULL; free = next) {
117 next = free->xbfi_next;
120 * Free the extent and log the EFD to dirty the transaction
121 * before handling errors. This ensures that the transaction is
124 * 1.) releases the EFI and frees the EFD
125 * 2.) shuts down the filesystem
127 * The bmap free list is cleaned up at a higher level.
129 error = xfs_free_extent(*tp, free->xbfi_startblock,
130 free->xbfi_blockcount);
131 xfs_trans_log_efd_extent(*tp, efd, free->xbfi_startblock,
132 free->xbfi_blockcount);
136 xfs_bmap_del_free(flist, NULL, free);
144 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
146 xfs_alloctype_t atype = 0; /* type for allocation routines */
147 int error; /* error return value */
148 xfs_mount_t *mp; /* mount point structure */
149 xfs_extlen_t prod = 0; /* product factor for allocators */
150 xfs_extlen_t ralen = 0; /* realtime allocation length */
151 xfs_extlen_t align; /* minimum allocation alignment */
154 mp = ap->ip->i_mount;
155 align = xfs_get_extsz_hint(ap->ip);
156 prod = align / mp->m_sb.sb_rextsize;
157 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
158 align, 1, ap->eof, 0,
159 ap->conv, &ap->offset, &ap->length);
163 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
166 * If the offset & length are not perfectly aligned
167 * then kill prod, it will just get us in trouble.
169 if (do_mod(ap->offset, align) || ap->length % align)
172 * Set ralen to be the actual requested length in rtextents.
174 ralen = ap->length / mp->m_sb.sb_rextsize;
176 * If the old value was close enough to MAXEXTLEN that
177 * we rounded up to it, cut it back so it's valid again.
178 * Note that if it's a really large request (bigger than
179 * MAXEXTLEN), we don't hear about that number, and can't
180 * adjust the starting point to match it.
182 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
183 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
186 * Lock out other modifications to the RT bitmap inode.
188 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
189 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
192 * If it's an allocation to an empty file at offset 0,
193 * pick an extent that will space things out in the rt area.
195 if (ap->eof && ap->offset == 0) {
196 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
198 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
201 ap->blkno = rtx * mp->m_sb.sb_rextsize;
206 xfs_bmap_adjacent(ap);
209 * Realtime allocation, done through xfs_rtallocate_extent.
211 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
212 do_div(ap->blkno, mp->m_sb.sb_rextsize);
215 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
216 &ralen, atype, ap->wasdel, prod, &rtb)))
218 if (rtb == NULLFSBLOCK && prod > 1 &&
219 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
220 ap->length, &ralen, atype,
221 ap->wasdel, 1, &rtb)))
224 if (ap->blkno != NULLFSBLOCK) {
225 ap->blkno *= mp->m_sb.sb_rextsize;
226 ralen *= mp->m_sb.sb_rextsize;
228 ap->ip->i_d.di_nblocks += ralen;
229 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
231 ap->ip->i_delayed_blks -= ralen;
233 * Adjust the disk quota also. This was reserved
236 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
237 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
238 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
246 * Check if the endoff is outside the last extent. If so the caller will grow
247 * the allocation to a stripe unit boundary. All offsets are considered outside
248 * the end of file for an empty fork, so 1 is returned in *eof in that case.
252 struct xfs_inode *ip,
253 xfs_fileoff_t endoff,
257 struct xfs_bmbt_irec rec;
260 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
264 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
269 * Extent tree block counting routines.
273 * Count leaf blocks given a range of extent records.
276 xfs_bmap_count_leaves(
284 for (b = 0; b < numrecs; b++) {
285 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
286 *count += xfs_bmbt_get_blockcount(frp);
291 * Count leaf blocks given a range of extent records originally
295 xfs_bmap_disk_count_leaves(
296 struct xfs_mount *mp,
297 struct xfs_btree_block *block,
304 for (b = 1; b <= numrecs; b++) {
305 frp = XFS_BMBT_REC_ADDR(mp, block, b);
306 *count += xfs_bmbt_disk_get_blockcount(frp);
311 * Recursively walks each level of a btree
312 * to count total fsblocks in use.
314 STATIC int /* error */
316 xfs_mount_t *mp, /* file system mount point */
317 xfs_trans_t *tp, /* transaction pointer */
318 xfs_ifork_t *ifp, /* inode fork pointer */
319 xfs_fsblock_t blockno, /* file system block number */
320 int levelin, /* level in btree */
321 int *count) /* Count of blocks */
327 xfs_fsblock_t bno = blockno;
328 xfs_fsblock_t nextbno;
329 struct xfs_btree_block *block, *nextblock;
332 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
337 block = XFS_BUF_TO_BLOCK(bp);
340 /* Not at node above leaves, count this level of nodes */
341 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
342 while (nextbno != NULLFSBLOCK) {
343 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
349 nextblock = XFS_BUF_TO_BLOCK(nbp);
350 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
351 xfs_trans_brelse(tp, nbp);
354 /* Dive to the next level */
355 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
356 bno = be64_to_cpu(*pp);
357 if (unlikely((error =
358 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
359 xfs_trans_brelse(tp, bp);
360 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
361 XFS_ERRLEVEL_LOW, mp);
362 return -EFSCORRUPTED;
364 xfs_trans_brelse(tp, bp);
366 /* count all level 1 nodes and their leaves */
368 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
369 numrecs = be16_to_cpu(block->bb_numrecs);
370 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
371 xfs_trans_brelse(tp, bp);
372 if (nextbno == NULLFSBLOCK)
375 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
381 block = XFS_BUF_TO_BLOCK(bp);
388 * Count fsblocks of the given fork.
391 xfs_bmap_count_blocks(
392 xfs_trans_t *tp, /* transaction pointer */
393 xfs_inode_t *ip, /* incore inode */
394 int whichfork, /* data or attr fork */
395 int *count) /* out: count of blocks */
397 struct xfs_btree_block *block; /* current btree block */
398 xfs_fsblock_t bno; /* block # of "block" */
399 xfs_ifork_t *ifp; /* fork structure */
400 int level; /* btree level, for checking */
401 xfs_mount_t *mp; /* file system mount structure */
402 __be64 *pp; /* pointer to block address */
406 ifp = XFS_IFORK_PTR(ip, whichfork);
407 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
408 xfs_bmap_count_leaves(ifp, 0,
409 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
415 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
417 block = ifp->if_broot;
418 level = be16_to_cpu(block->bb_level);
420 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
421 bno = be64_to_cpu(*pp);
422 ASSERT(bno != NULLFSBLOCK);
423 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
424 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
426 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
427 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
429 return -EFSCORRUPTED;
436 * returns 1 for success, 0 if we failed to map the extent.
439 xfs_getbmapx_fix_eof_hole(
440 xfs_inode_t *ip, /* xfs incore inode pointer */
441 struct getbmapx *out, /* output structure */
442 int prealloced, /* this is a file with
443 * preallocated data space */
444 __int64_t end, /* last block requested */
445 xfs_fsblock_t startblock)
448 xfs_mount_t *mp; /* file system mount point */
449 xfs_ifork_t *ifp; /* inode fork pointer */
450 xfs_extnum_t lastx; /* last extent pointer */
451 xfs_fileoff_t fileblock;
453 if (startblock == HOLESTARTBLOCK) {
456 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
457 fixlen -= out->bmv_offset;
458 if (prealloced && out->bmv_offset + out->bmv_length == end) {
459 /* Came to hole at EOF. Trim it. */
462 out->bmv_length = fixlen;
465 if (startblock == DELAYSTARTBLOCK)
468 out->bmv_block = xfs_fsb_to_db(ip, startblock);
469 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
470 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
471 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
472 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
473 out->bmv_oflags |= BMV_OF_LAST;
480 * Get inode's extents as described in bmv, and format for output.
481 * Calls formatter to fill the user's buffer until all extents
482 * are mapped, until the passed-in bmv->bmv_count slots have
483 * been filled, or until the formatter short-circuits the loop,
484 * if it is tracking filled-in extents on its own.
489 struct getbmapx *bmv, /* user bmap structure */
490 xfs_bmap_format_t formatter, /* format to user */
491 void *arg) /* formatter arg */
493 __int64_t bmvend; /* last block requested */
494 int error = 0; /* return value */
495 __int64_t fixlen; /* length for -1 case */
496 int i; /* extent number */
497 int lock; /* lock state */
498 xfs_bmbt_irec_t *map; /* buffer for user's data */
499 xfs_mount_t *mp; /* file system mount point */
500 int nex; /* # of user extents can do */
501 int nexleft; /* # of user extents left */
502 int subnex; /* # of bmapi's can do */
503 int nmap; /* number of map entries */
504 struct getbmapx *out; /* output structure */
505 int whichfork; /* data or attr fork */
506 int prealloced; /* this is a file with
507 * preallocated data space */
508 int iflags; /* interface flags */
509 int bmapi_flags; /* flags for xfs_bmapi */
513 iflags = bmv->bmv_iflags;
514 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
516 if (whichfork == XFS_ATTR_FORK) {
517 if (XFS_IFORK_Q(ip)) {
518 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
519 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
520 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
523 ip->i_d.di_aformat != 0 &&
524 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
525 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
527 return -EFSCORRUPTED;
533 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
534 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
535 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
538 if (xfs_get_extsz_hint(ip) ||
539 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
541 fixlen = mp->m_super->s_maxbytes;
544 fixlen = XFS_ISIZE(ip);
548 if (bmv->bmv_length == -1) {
549 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
551 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
552 } else if (bmv->bmv_length == 0) {
553 bmv->bmv_entries = 0;
555 } else if (bmv->bmv_length < 0) {
559 nex = bmv->bmv_count - 1;
562 bmvend = bmv->bmv_offset + bmv->bmv_length;
565 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
567 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
571 xfs_ilock(ip, XFS_IOLOCK_SHARED);
572 if (whichfork == XFS_DATA_FORK) {
573 if (!(iflags & BMV_IF_DELALLOC) &&
574 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
575 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
577 goto out_unlock_iolock;
580 * Even after flushing the inode, there can still be
581 * delalloc blocks on the inode beyond EOF due to
582 * speculative preallocation. These are not removed
583 * until the release function is called or the inode
584 * is inactivated. Hence we cannot assert here that
585 * ip->i_delayed_blks == 0.
589 lock = xfs_ilock_data_map_shared(ip);
591 lock = xfs_ilock_attr_map_shared(ip);
595 * Don't let nex be bigger than the number of extents
596 * we can have assuming alternating holes and real extents.
598 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
599 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
601 bmapi_flags = xfs_bmapi_aflag(whichfork);
602 if (!(iflags & BMV_IF_PREALLOC))
603 bmapi_flags |= XFS_BMAPI_IGSTATE;
606 * Allocate enough space to handle "subnex" maps at a time.
610 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
612 goto out_unlock_ilock;
614 bmv->bmv_entries = 0;
616 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
617 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
625 nmap = (nexleft > subnex) ? subnex : nexleft;
626 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
627 XFS_BB_TO_FSB(mp, bmv->bmv_length),
628 map, &nmap, bmapi_flags);
631 ASSERT(nmap <= subnex);
633 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
634 out[cur_ext].bmv_oflags = 0;
635 if (map[i].br_state == XFS_EXT_UNWRITTEN)
636 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
637 else if (map[i].br_startblock == DELAYSTARTBLOCK)
638 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
639 out[cur_ext].bmv_offset =
640 XFS_FSB_TO_BB(mp, map[i].br_startoff);
641 out[cur_ext].bmv_length =
642 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
643 out[cur_ext].bmv_unused1 = 0;
644 out[cur_ext].bmv_unused2 = 0;
647 * delayed allocation extents that start beyond EOF can
648 * occur due to speculative EOF allocation when the
649 * delalloc extent is larger than the largest freespace
650 * extent at conversion time. These extents cannot be
651 * converted by data writeback, so can exist here even
652 * if we are not supposed to be finding delalloc
655 if (map[i].br_startblock == DELAYSTARTBLOCK &&
656 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
657 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
659 if (map[i].br_startblock == HOLESTARTBLOCK &&
660 whichfork == XFS_ATTR_FORK) {
661 /* came to the end of attribute fork */
662 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
666 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
668 map[i].br_startblock))
672 out[cur_ext].bmv_offset +
673 out[cur_ext].bmv_length;
675 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
678 * In case we don't want to return the hole,
679 * don't increase cur_ext so that we can reuse
680 * it in the next loop.
682 if ((iflags & BMV_IF_NO_HOLES) &&
683 map[i].br_startblock == HOLESTARTBLOCK) {
684 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
692 } while (nmap && nexleft && bmv->bmv_length);
697 xfs_iunlock(ip, lock);
699 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
701 for (i = 0; i < cur_ext; i++) {
702 int full = 0; /* user array is full */
704 /* format results & advance arg */
705 error = formatter(&arg, &out[i], &full);
715 * dead simple method of punching delalyed allocation blocks from a range in
716 * the inode. Walks a block at a time so will be slow, but is only executed in
717 * rare error cases so the overhead is not critical. This will always punch out
718 * both the start and end blocks, even if the ranges only partially overlap
719 * them, so it is up to the caller to ensure that partial blocks are not
723 xfs_bmap_punch_delalloc_range(
724 struct xfs_inode *ip,
725 xfs_fileoff_t start_fsb,
726 xfs_fileoff_t length)
728 xfs_fileoff_t remaining = length;
731 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
735 xfs_bmbt_irec_t imap;
737 xfs_fsblock_t firstblock;
738 xfs_bmap_free_t flist;
741 * Map the range first and check that it is a delalloc extent
742 * before trying to unmap the range. Otherwise we will be
743 * trying to remove a real extent (which requires a
744 * transaction) or a hole, which is probably a bad idea...
746 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
750 /* something screwed, just bail */
751 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
752 xfs_alert(ip->i_mount,
753 "Failed delalloc mapping lookup ino %lld fsb %lld.",
754 ip->i_ino, start_fsb);
762 if (imap.br_startblock != DELAYSTARTBLOCK) {
763 /* been converted, ignore */
766 WARN_ON(imap.br_blockcount == 0);
769 * Note: while we initialise the firstblock/flist pair, they
770 * should never be used because blocks should never be
771 * allocated or freed for a delalloc extent and hence we need
772 * don't cancel or finish them after the xfs_bunmapi() call.
774 xfs_bmap_init(&flist, &firstblock);
775 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
780 ASSERT(!flist.xbf_count && !flist.xbf_first);
784 } while(remaining > 0);
790 * Test whether it is appropriate to check an inode for and free post EOF
791 * blocks. The 'force' parameter determines whether we should also consider
792 * regular files that are marked preallocated or append-only.
795 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
797 /* prealloc/delalloc exists only on regular files */
798 if (!S_ISREG(ip->i_d.di_mode))
802 * Zero sized files with no cached pages and delalloc blocks will not
803 * have speculative prealloc/delalloc blocks to remove.
805 if (VFS_I(ip)->i_size == 0 &&
806 VFS_I(ip)->i_mapping->nrpages == 0 &&
807 ip->i_delayed_blks == 0)
810 /* If we haven't read in the extent list, then don't do it now. */
811 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
815 * Do not free real preallocated or append-only files unless the file
816 * has delalloc blocks and we are forced to remove them.
818 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
819 if (!force || ip->i_delayed_blks == 0)
826 * This is called by xfs_inactive to free any blocks beyond eof
827 * when the link count isn't zero and by xfs_dm_punch_hole() when
828 * punching a hole to EOF.
838 xfs_fileoff_t end_fsb;
839 xfs_fileoff_t last_fsb;
840 xfs_filblks_t map_len;
842 xfs_bmbt_irec_t imap;
845 * Figure out if there are any blocks beyond the end
846 * of the file. If not, then there is nothing to do.
848 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
849 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
850 if (last_fsb <= end_fsb)
852 map_len = last_fsb - end_fsb;
855 xfs_ilock(ip, XFS_ILOCK_SHARED);
856 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
857 xfs_iunlock(ip, XFS_ILOCK_SHARED);
859 if (!error && (nimaps != 0) &&
860 (imap.br_startblock != HOLESTARTBLOCK ||
861 ip->i_delayed_blks)) {
863 * Attach the dquots to the inode up front.
865 error = xfs_qm_dqattach(ip, 0);
870 * There are blocks after the end of file.
871 * Free them up now by truncating the file to
874 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
877 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
878 xfs_trans_cancel(tp);
883 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
885 ASSERT(XFS_FORCED_SHUTDOWN(mp));
886 xfs_trans_cancel(tp);
888 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
892 xfs_ilock(ip, XFS_ILOCK_EXCL);
893 xfs_trans_ijoin(tp, ip, 0);
896 * Do not update the on-disk file size. If we update the
897 * on-disk file size and then the system crashes before the
898 * contents of the file are flushed to disk then the files
899 * may be full of holes (ie NULL files bug).
901 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
905 * If we get an error at this point we simply don't
906 * bother truncating the file.
908 xfs_trans_cancel(tp);
910 error = xfs_trans_commit(tp);
912 xfs_inode_clear_eofblocks_tag(ip);
915 xfs_iunlock(ip, XFS_ILOCK_EXCL);
917 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
923 xfs_alloc_file_space(
924 struct xfs_inode *ip,
929 xfs_mount_t *mp = ip->i_mount;
931 xfs_filblks_t allocated_fsb;
932 xfs_filblks_t allocatesize_fsb;
933 xfs_extlen_t extsz, temp;
934 xfs_fileoff_t startoffset_fsb;
935 xfs_fsblock_t firstfsb;
940 xfs_bmbt_irec_t imaps[1], *imapp;
941 xfs_bmap_free_t free_list;
942 uint qblocks, resblks, resrtextents;
946 trace_xfs_alloc_file_space(ip);
948 if (XFS_FORCED_SHUTDOWN(mp))
951 error = xfs_qm_dqattach(ip, 0);
958 rt = XFS_IS_REALTIME_INODE(ip);
959 extsz = xfs_get_extsz_hint(ip);
964 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
965 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
968 * Allocate file space until done or until there is an error
970 while (allocatesize_fsb && !error) {
974 * Determine space reservations for data/realtime.
976 if (unlikely(extsz)) {
980 e = startoffset_fsb + allocatesize_fsb;
981 if ((temp = do_mod(startoffset_fsb, extsz)))
983 if ((temp = do_mod(e, extsz)))
987 e = allocatesize_fsb;
991 * The transaction reservation is limited to a 32-bit block
992 * count, hence we need to limit the number of blocks we are
993 * trying to reserve to avoid an overflow. We can't allocate
994 * more than @nimaps extents, and an extent is limited on disk
995 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
997 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
999 resrtextents = qblocks = resblks;
1000 resrtextents /= mp->m_sb.sb_rextsize;
1001 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1002 quota_flag = XFS_QMOPT_RES_RTBLKS;
1005 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1006 quota_flag = XFS_QMOPT_RES_REGBLKS;
1010 * Allocate and setup the transaction.
1012 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1013 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1014 resblks, resrtextents);
1016 * Check for running out of space
1020 * Free the transaction structure.
1022 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1023 xfs_trans_cancel(tp);
1026 xfs_ilock(ip, XFS_ILOCK_EXCL);
1027 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1032 xfs_trans_ijoin(tp, ip, 0);
1034 xfs_bmap_init(&free_list, &firstfsb);
1035 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1036 allocatesize_fsb, alloc_type, &firstfsb,
1037 0, imapp, &nimaps, &free_list);
1043 * Complete the transaction
1045 error = xfs_bmap_finish(&tp, &free_list, &committed);
1050 error = xfs_trans_commit(tp);
1051 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1056 allocated_fsb = imapp->br_blockcount;
1063 startoffset_fsb += allocated_fsb;
1064 allocatesize_fsb -= allocated_fsb;
1069 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1070 xfs_bmap_cancel(&free_list);
1071 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1073 error1: /* Just cancel transaction */
1074 xfs_trans_cancel(tp);
1075 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1080 * Zero file bytes between startoff and endoff inclusive.
1081 * The iolock is held exclusive and no blocks are buffered.
1083 * This function is used by xfs_free_file_space() to zero
1084 * partial blocks when the range to free is not block aligned.
1085 * When unreserving space with boundaries that are not block
1086 * aligned we round up the start and round down the end
1087 * boundaries and then use this function to zero the parts of
1088 * the blocks that got dropped during the rounding.
1091 xfs_zero_remaining_bytes(
1096 xfs_bmbt_irec_t imap;
1097 xfs_fileoff_t offset_fsb;
1098 xfs_off_t lastoffset;
1101 xfs_mount_t *mp = ip->i_mount;
1106 * Avoid doing I/O beyond eof - it's not necessary
1107 * since nothing can read beyond eof. The space will
1108 * be zeroed when the file is extended anyway.
1110 if (startoff >= XFS_ISIZE(ip))
1113 if (endoff > XFS_ISIZE(ip))
1114 endoff = XFS_ISIZE(ip);
1116 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1119 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1122 lock_mode = xfs_ilock_data_map_shared(ip);
1123 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1124 xfs_iunlock(ip, lock_mode);
1126 if (error || nimap < 1)
1128 ASSERT(imap.br_blockcount >= 1);
1129 ASSERT(imap.br_startoff == offset_fsb);
1130 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1132 if (imap.br_startblock == HOLESTARTBLOCK ||
1133 imap.br_state == XFS_EXT_UNWRITTEN) {
1134 /* skip the entire extent */
1135 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff +
1136 imap.br_blockcount) - 1;
1140 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1141 if (lastoffset > endoff)
1142 lastoffset = endoff;
1144 /* DAX can just zero the backing device directly */
1145 if (IS_DAX(VFS_I(ip))) {
1146 error = dax_zero_page_range(VFS_I(ip), offset,
1147 lastoffset - offset + 1,
1148 xfs_get_blocks_direct);
1154 error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ?
1155 mp->m_rtdev_targp : mp->m_ddev_targp,
1156 xfs_fsb_to_db(ip, imap.br_startblock),
1157 BTOBB(mp->m_sb.sb_blocksize),
1163 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1164 0, lastoffset - offset + 1);
1166 error = xfs_bwrite(bp);
1175 xfs_free_file_space(
1176 struct xfs_inode *ip,
1182 xfs_fileoff_t endoffset_fsb;
1184 xfs_fsblock_t firstfsb;
1185 xfs_bmap_free_t free_list;
1186 xfs_bmbt_irec_t imap;
1188 xfs_off_t iendoffset;
1195 xfs_fileoff_t startoffset_fsb;
1200 trace_xfs_free_file_space(ip);
1202 error = xfs_qm_dqattach(ip, 0);
1207 if (len <= 0) /* if nothing being freed */
1209 rt = XFS_IS_REALTIME_INODE(ip);
1210 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1211 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1213 /* wait for the completion of any pending DIOs */
1214 inode_dio_wait(VFS_I(ip));
1216 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1217 ioffset = round_down(offset, rounding);
1218 iendoffset = round_up(offset + len, rounding) - 1;
1219 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, ioffset,
1223 truncate_pagecache_range(VFS_I(ip), ioffset, iendoffset);
1226 * Need to zero the stuff we're not freeing, on disk.
1227 * If it's a realtime file & can't use unwritten extents then we
1228 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1229 * will take care of it for us.
1231 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1233 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1237 ASSERT(nimap == 0 || nimap == 1);
1238 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1241 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1242 block = imap.br_startblock;
1243 mod = do_div(block, mp->m_sb.sb_rextsize);
1245 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1248 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1252 ASSERT(nimap == 0 || nimap == 1);
1253 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1254 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1256 if (mod && (mod != mp->m_sb.sb_rextsize))
1257 endoffset_fsb -= mod;
1260 if ((done = (endoffset_fsb <= startoffset_fsb)))
1262 * One contiguous piece to clear
1264 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1267 * Some full blocks, possibly two pieces to clear
1269 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1270 error = xfs_zero_remaining_bytes(ip, offset,
1271 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1273 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1274 error = xfs_zero_remaining_bytes(ip,
1275 XFS_FSB_TO_B(mp, endoffset_fsb),
1280 * free file space until done or until there is an error
1282 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1283 while (!error && !done) {
1286 * allocate and setup the transaction. Allow this
1287 * transaction to dip into the reserve blocks to ensure
1288 * the freeing of the space succeeds at ENOSPC.
1290 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1291 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1294 * check for running out of space
1298 * Free the transaction structure.
1300 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1301 xfs_trans_cancel(tp);
1304 xfs_ilock(ip, XFS_ILOCK_EXCL);
1305 error = xfs_trans_reserve_quota(tp, mp,
1306 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1307 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1311 xfs_trans_ijoin(tp, ip, 0);
1314 * issue the bunmapi() call to free the blocks
1316 xfs_bmap_init(&free_list, &firstfsb);
1317 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1318 endoffset_fsb - startoffset_fsb,
1319 0, 2, &firstfsb, &free_list, &done);
1325 * complete the transaction
1327 error = xfs_bmap_finish(&tp, &free_list, &committed);
1332 error = xfs_trans_commit(tp);
1333 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1340 xfs_bmap_cancel(&free_list);
1342 xfs_trans_cancel(tp);
1343 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1348 * Preallocate and zero a range of a file. This mechanism has the allocation
1349 * semantics of fallocate and in addition converts data in the range to zeroes.
1352 xfs_zero_file_space(
1353 struct xfs_inode *ip,
1357 struct xfs_mount *mp = ip->i_mount;
1361 trace_xfs_zero_file_space(ip);
1363 blksize = 1 << mp->m_sb.sb_blocklog;
1366 * Punch a hole and prealloc the range. We use hole punch rather than
1367 * unwritten extent conversion for two reasons:
1369 * 1.) Hole punch handles partial block zeroing for us.
1371 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1372 * by virtue of the hole punch.
1374 error = xfs_free_file_space(ip, offset, len);
1378 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1379 round_up(offset + len, blksize) -
1380 round_down(offset, blksize),
1381 XFS_BMAPI_PREALLOC);
1388 * @next_fsb will keep track of the extent currently undergoing shift.
1389 * @stop_fsb will keep track of the extent at which we have to stop.
1390 * If we are shifting left, we will start with block (offset + len) and
1391 * shift each extent till last extent.
1392 * If we are shifting right, we will start with last extent inside file space
1393 * and continue until we reach the block corresponding to offset.
1396 xfs_shift_file_space(
1397 struct xfs_inode *ip,
1400 enum shift_direction direction)
1403 struct xfs_mount *mp = ip->i_mount;
1404 struct xfs_trans *tp;
1406 struct xfs_bmap_free free_list;
1407 xfs_fsblock_t first_block;
1409 xfs_fileoff_t stop_fsb;
1410 xfs_fileoff_t next_fsb;
1411 xfs_fileoff_t shift_fsb;
1413 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1415 if (direction == SHIFT_LEFT) {
1416 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1417 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1420 * If right shift, delegate the work of initialization of
1421 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1423 next_fsb = NULLFSBLOCK;
1424 stop_fsb = XFS_B_TO_FSB(mp, offset);
1427 shift_fsb = XFS_B_TO_FSB(mp, len);
1430 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1431 * into the accessible region of the file.
1433 if (xfs_can_free_eofblocks(ip, true)) {
1434 error = xfs_free_eofblocks(mp, ip, false);
1440 * Writeback and invalidate cache for the remainder of the file as we're
1441 * about to shift down every extent from offset to EOF.
1443 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1447 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1448 offset >> PAGE_CACHE_SHIFT, -1);
1453 * The extent shiting code works on extent granularity. So, if
1454 * stop_fsb is not the starting block of extent, we need to split
1455 * the extent at stop_fsb.
1457 if (direction == SHIFT_RIGHT) {
1458 error = xfs_bmap_split_extent(ip, stop_fsb);
1463 while (!error && !done) {
1464 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1466 * We would need to reserve permanent block for transaction.
1467 * This will come into picture when after shifting extent into
1468 * hole we found that adjacent extents can be merged which
1469 * may lead to freeing of a block during record update.
1471 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1472 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1474 xfs_trans_cancel(tp);
1478 xfs_ilock(ip, XFS_ILOCK_EXCL);
1479 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1480 ip->i_gdquot, ip->i_pdquot,
1481 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1482 XFS_QMOPT_RES_REGBLKS);
1486 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1488 xfs_bmap_init(&free_list, &first_block);
1491 * We are using the write transaction in which max 2 bmbt
1492 * updates are allowed
1494 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1495 &done, stop_fsb, &first_block, &free_list,
1496 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1500 error = xfs_bmap_finish(&tp, &free_list, &committed);
1504 error = xfs_trans_commit(tp);
1510 xfs_trans_cancel(tp);
1515 * xfs_collapse_file_space()
1516 * This routine frees disk space and shift extent for the given file.
1517 * The first thing we do is to free data blocks in the specified range
1518 * by calling xfs_free_file_space(). It would also sync dirty data
1519 * and invalidate page cache over the region on which collapse range
1520 * is working. And Shift extent records to the left to cover a hole.
1527 xfs_collapse_file_space(
1528 struct xfs_inode *ip,
1534 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1535 trace_xfs_collapse_file_space(ip);
1537 error = xfs_free_file_space(ip, offset, len);
1541 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1545 * xfs_insert_file_space()
1546 * This routine create hole space by shifting extents for the given file.
1547 * The first thing we do is to sync dirty data and invalidate page cache
1548 * over the region on which insert range is working. And split an extent
1549 * to two extents at given offset by calling xfs_bmap_split_extent.
1550 * And shift all extent records which are laying between [offset,
1551 * last allocated extent] to the right to reserve hole range.
1557 xfs_insert_file_space(
1558 struct xfs_inode *ip,
1562 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1563 trace_xfs_insert_file_space(ip);
1565 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1569 * We need to check that the format of the data fork in the temporary inode is
1570 * valid for the target inode before doing the swap. This is not a problem with
1571 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1572 * data fork depending on the space the attribute fork is taking so we can get
1573 * invalid formats on the target inode.
1575 * E.g. target has space for 7 extents in extent format, temp inode only has
1576 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1577 * btree, but when swapped it needs to be in extent format. Hence we can't just
1578 * blindly swap data forks on attr2 filesystems.
1580 * Note that we check the swap in both directions so that we don't end up with
1581 * a corrupt temporary inode, either.
1583 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1584 * inode will prevent this situation from occurring, so all we do here is
1585 * reject and log the attempt. basically we are putting the responsibility on
1586 * userspace to get this right.
1589 xfs_swap_extents_check_format(
1590 xfs_inode_t *ip, /* target inode */
1591 xfs_inode_t *tip) /* tmp inode */
1594 /* Should never get a local format */
1595 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1596 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1600 * if the target inode has less extents that then temporary inode then
1601 * why did userspace call us?
1603 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1607 * if the target inode is in extent form and the temp inode is in btree
1608 * form then we will end up with the target inode in the wrong format
1609 * as we already know there are less extents in the temp inode.
1611 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1612 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1615 /* Check temp in extent form to max in target */
1616 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1617 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1618 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1621 /* Check target in extent form to max in temp */
1622 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1623 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1624 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1628 * If we are in a btree format, check that the temp root block will fit
1629 * in the target and that it has enough extents to be in btree format
1632 * Note that we have to be careful to allow btree->extent conversions
1633 * (a common defrag case) which will occur when the temp inode is in
1636 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1637 if (XFS_IFORK_BOFF(ip) &&
1638 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1640 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1641 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1645 /* Reciprocal target->temp btree format checks */
1646 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1647 if (XFS_IFORK_BOFF(tip) &&
1648 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1650 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1651 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1659 xfs_swap_extent_flush(
1660 struct xfs_inode *ip)
1664 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1667 truncate_pagecache_range(VFS_I(ip), 0, -1);
1669 /* Verify O_DIRECT for ftmp */
1670 if (VFS_I(ip)->i_mapping->nrpages)
1677 xfs_inode_t *ip, /* target inode */
1678 xfs_inode_t *tip, /* tmp inode */
1681 xfs_mount_t *mp = ip->i_mount;
1683 xfs_bstat_t *sbp = &sxp->sx_stat;
1684 xfs_ifork_t *tempifp, *ifp, *tifp;
1685 int src_log_flags, target_log_flags;
1692 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1699 * Lock the inodes against other IO, page faults and truncate to
1700 * begin with. Then we can ensure the inodes are flushed and have no
1701 * page cache safely. Once we have done this we can take the ilocks and
1702 * do the rest of the checks.
1704 lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1705 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1706 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1708 /* Verify that both files have the same format */
1709 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1714 /* Verify both files are either real-time or non-realtime */
1715 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1720 error = xfs_swap_extent_flush(ip);
1723 error = xfs_swap_extent_flush(tip);
1727 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1728 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1730 xfs_trans_cancel(tp);
1735 * Lock and join the inodes to the tansaction so that transaction commit
1736 * or cancel will unlock the inodes from this point onwards.
1738 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1739 lock_flags |= XFS_ILOCK_EXCL;
1740 xfs_trans_ijoin(tp, ip, lock_flags);
1741 xfs_trans_ijoin(tp, tip, lock_flags);
1744 /* Verify all data are being swapped */
1745 if (sxp->sx_offset != 0 ||
1746 sxp->sx_length != ip->i_d.di_size ||
1747 sxp->sx_length != tip->i_d.di_size) {
1749 goto out_trans_cancel;
1752 trace_xfs_swap_extent_before(ip, 0);
1753 trace_xfs_swap_extent_before(tip, 1);
1755 /* check inode formats now that data is flushed */
1756 error = xfs_swap_extents_check_format(ip, tip);
1759 "%s: inode 0x%llx format is incompatible for exchanging.",
1760 __func__, ip->i_ino);
1761 goto out_trans_cancel;
1765 * Compare the current change & modify times with that
1766 * passed in. If they differ, we abort this swap.
1767 * This is the mechanism used to ensure the calling
1768 * process that the file was not changed out from
1771 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1772 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1773 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1774 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1776 goto out_trans_cancel;
1779 * Count the number of extended attribute blocks
1781 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1782 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1783 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1785 goto out_trans_cancel;
1787 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1788 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1789 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1792 goto out_trans_cancel;
1796 * Before we've swapped the forks, lets set the owners of the forks
1797 * appropriately. We have to do this as we are demand paging the btree
1798 * buffers, and so the validation done on read will expect the owner
1799 * field to be correctly set. Once we change the owners, we can swap the
1802 * Note the trickiness in setting the log flags - we set the owner log
1803 * flag on the opposite inode (i.e. the inode we are setting the new
1804 * owner to be) because once we swap the forks and log that, log
1805 * recovery is going to see the fork as owned by the swapped inode,
1806 * not the pre-swapped inodes.
1808 src_log_flags = XFS_ILOG_CORE;
1809 target_log_flags = XFS_ILOG_CORE;
1810 if (ip->i_d.di_version == 3 &&
1811 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1812 target_log_flags |= XFS_ILOG_DOWNER;
1813 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1816 goto out_trans_cancel;
1819 if (tip->i_d.di_version == 3 &&
1820 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1821 src_log_flags |= XFS_ILOG_DOWNER;
1822 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1825 goto out_trans_cancel;
1829 * Swap the data forks of the inodes
1833 *tempifp = *ifp; /* struct copy */
1834 *ifp = *tifp; /* struct copy */
1835 *tifp = *tempifp; /* struct copy */
1838 * Fix the on-disk inode values
1840 tmp = (__uint64_t)ip->i_d.di_nblocks;
1841 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1842 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1844 tmp = (__uint64_t) ip->i_d.di_nextents;
1845 ip->i_d.di_nextents = tip->i_d.di_nextents;
1846 tip->i_d.di_nextents = tmp;
1848 tmp = (__uint64_t) ip->i_d.di_format;
1849 ip->i_d.di_format = tip->i_d.di_format;
1850 tip->i_d.di_format = tmp;
1853 * The extents in the source inode could still contain speculative
1854 * preallocation beyond EOF (e.g. the file is open but not modified
1855 * while defrag is in progress). In that case, we need to copy over the
1856 * number of delalloc blocks the data fork in the source inode is
1857 * tracking beyond EOF so that when the fork is truncated away when the
1858 * temporary inode is unlinked we don't underrun the i_delayed_blks
1859 * counter on that inode.
1861 ASSERT(tip->i_delayed_blks == 0);
1862 tip->i_delayed_blks = ip->i_delayed_blks;
1863 ip->i_delayed_blks = 0;
1865 switch (ip->i_d.di_format) {
1866 case XFS_DINODE_FMT_EXTENTS:
1867 /* If the extents fit in the inode, fix the
1868 * pointer. Otherwise it's already NULL or
1869 * pointing to the extent.
1871 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1872 ifp->if_u1.if_extents =
1873 ifp->if_u2.if_inline_ext;
1875 src_log_flags |= XFS_ILOG_DEXT;
1877 case XFS_DINODE_FMT_BTREE:
1878 ASSERT(ip->i_d.di_version < 3 ||
1879 (src_log_flags & XFS_ILOG_DOWNER));
1880 src_log_flags |= XFS_ILOG_DBROOT;
1884 switch (tip->i_d.di_format) {
1885 case XFS_DINODE_FMT_EXTENTS:
1886 /* If the extents fit in the inode, fix the
1887 * pointer. Otherwise it's already NULL or
1888 * pointing to the extent.
1890 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1891 tifp->if_u1.if_extents =
1892 tifp->if_u2.if_inline_ext;
1894 target_log_flags |= XFS_ILOG_DEXT;
1896 case XFS_DINODE_FMT_BTREE:
1897 target_log_flags |= XFS_ILOG_DBROOT;
1898 ASSERT(tip->i_d.di_version < 3 ||
1899 (target_log_flags & XFS_ILOG_DOWNER));
1903 xfs_trans_log_inode(tp, ip, src_log_flags);
1904 xfs_trans_log_inode(tp, tip, target_log_flags);
1907 * If this is a synchronous mount, make sure that the
1908 * transaction goes to disk before returning to the user.
1910 if (mp->m_flags & XFS_MOUNT_WSYNC)
1911 xfs_trans_set_sync(tp);
1913 error = xfs_trans_commit(tp);
1915 trace_xfs_swap_extent_after(ip, 0);
1916 trace_xfs_swap_extent_after(tip, 1);
1922 xfs_iunlock(ip, lock_flags);
1923 xfs_iunlock(tip, lock_flags);
1927 xfs_trans_cancel(tp);