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 xfs_trans_t **tp, /* transaction pointer addr */
71 xfs_bmap_free_t *flist, /* i/o: list extents to free */
72 int *committed) /* xact committed or not */
74 xfs_efd_log_item_t *efd; /* extent free data */
75 xfs_efi_log_item_t *efi; /* extent free intention */
76 int error; /* error return value */
77 xfs_bmap_free_item_t *free; /* free extent item */
78 xfs_mount_t *mp; /* filesystem mount structure */
79 xfs_bmap_free_item_t *next; /* next item on free list */
81 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
82 if (flist->xbf_count == 0) {
86 efi = xfs_trans_get_efi(*tp, flist->xbf_count);
87 for (free = flist->xbf_first; free; free = free->xbfi_next)
88 xfs_trans_log_efi_extent(*tp, efi, free->xbfi_startblock,
89 free->xbfi_blockcount);
91 error = xfs_trans_roll(tp, NULL);
94 * We have a new transaction, so we should return committed=1,
95 * even though we're returning an error.
100 efd = xfs_trans_get_efd(*tp, efi, flist->xbf_count);
101 for (free = flist->xbf_first; free != NULL; free = next) {
102 next = free->xbfi_next;
103 if ((error = xfs_free_extent(*tp, free->xbfi_startblock,
104 free->xbfi_blockcount))) {
106 * The bmap free list will be cleaned up at a
107 * higher level. The EFI will be canceled when
108 * this transaction is aborted.
109 * Need to force shutdown here to make sure it
110 * happens, since this transaction may not be
113 mp = (*tp)->t_mountp;
114 if (!XFS_FORCED_SHUTDOWN(mp))
115 xfs_force_shutdown(mp,
116 (error == -EFSCORRUPTED) ?
117 SHUTDOWN_CORRUPT_INCORE :
118 SHUTDOWN_META_IO_ERROR);
121 xfs_trans_log_efd_extent(*tp, efd, free->xbfi_startblock,
122 free->xbfi_blockcount);
123 xfs_bmap_del_free(flist, NULL, free);
130 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
132 xfs_alloctype_t atype = 0; /* type for allocation routines */
133 int error; /* error return value */
134 xfs_mount_t *mp; /* mount point structure */
135 xfs_extlen_t prod = 0; /* product factor for allocators */
136 xfs_extlen_t ralen = 0; /* realtime allocation length */
137 xfs_extlen_t align; /* minimum allocation alignment */
140 mp = ap->ip->i_mount;
141 align = xfs_get_extsz_hint(ap->ip);
142 prod = align / mp->m_sb.sb_rextsize;
143 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
144 align, 1, ap->eof, 0,
145 ap->conv, &ap->offset, &ap->length);
149 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
152 * If the offset & length are not perfectly aligned
153 * then kill prod, it will just get us in trouble.
155 if (do_mod(ap->offset, align) || ap->length % align)
158 * Set ralen to be the actual requested length in rtextents.
160 ralen = ap->length / mp->m_sb.sb_rextsize;
162 * If the old value was close enough to MAXEXTLEN that
163 * we rounded up to it, cut it back so it's valid again.
164 * Note that if it's a really large request (bigger than
165 * MAXEXTLEN), we don't hear about that number, and can't
166 * adjust the starting point to match it.
168 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
169 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
172 * Lock out other modifications to the RT bitmap inode.
174 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
175 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
178 * If it's an allocation to an empty file at offset 0,
179 * pick an extent that will space things out in the rt area.
181 if (ap->eof && ap->offset == 0) {
182 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
184 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
187 ap->blkno = rtx * mp->m_sb.sb_rextsize;
192 xfs_bmap_adjacent(ap);
195 * Realtime allocation, done through xfs_rtallocate_extent.
197 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
198 do_div(ap->blkno, mp->m_sb.sb_rextsize);
201 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
202 &ralen, atype, ap->wasdel, prod, &rtb)))
204 if (rtb == NULLFSBLOCK && prod > 1 &&
205 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
206 ap->length, &ralen, atype,
207 ap->wasdel, 1, &rtb)))
210 if (ap->blkno != NULLFSBLOCK) {
211 ap->blkno *= mp->m_sb.sb_rextsize;
212 ralen *= mp->m_sb.sb_rextsize;
214 ap->ip->i_d.di_nblocks += ralen;
215 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
217 ap->ip->i_delayed_blks -= ralen;
219 * Adjust the disk quota also. This was reserved
222 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
223 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
224 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
232 * Check if the endoff is outside the last extent. If so the caller will grow
233 * the allocation to a stripe unit boundary. All offsets are considered outside
234 * the end of file for an empty fork, so 1 is returned in *eof in that case.
238 struct xfs_inode *ip,
239 xfs_fileoff_t endoff,
243 struct xfs_bmbt_irec rec;
246 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
250 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
255 * Extent tree block counting routines.
259 * Count leaf blocks given a range of extent records.
262 xfs_bmap_count_leaves(
270 for (b = 0; b < numrecs; b++) {
271 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
272 *count += xfs_bmbt_get_blockcount(frp);
277 * Count leaf blocks given a range of extent records originally
281 xfs_bmap_disk_count_leaves(
282 struct xfs_mount *mp,
283 struct xfs_btree_block *block,
290 for (b = 1; b <= numrecs; b++) {
291 frp = XFS_BMBT_REC_ADDR(mp, block, b);
292 *count += xfs_bmbt_disk_get_blockcount(frp);
297 * Recursively walks each level of a btree
298 * to count total fsblocks in use.
300 STATIC int /* error */
302 xfs_mount_t *mp, /* file system mount point */
303 xfs_trans_t *tp, /* transaction pointer */
304 xfs_ifork_t *ifp, /* inode fork pointer */
305 xfs_fsblock_t blockno, /* file system block number */
306 int levelin, /* level in btree */
307 int *count) /* Count of blocks */
313 xfs_fsblock_t bno = blockno;
314 xfs_fsblock_t nextbno;
315 struct xfs_btree_block *block, *nextblock;
318 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
323 block = XFS_BUF_TO_BLOCK(bp);
326 /* Not at node above leaves, count this level of nodes */
327 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
328 while (nextbno != NULLFSBLOCK) {
329 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
335 nextblock = XFS_BUF_TO_BLOCK(nbp);
336 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
337 xfs_trans_brelse(tp, nbp);
340 /* Dive to the next level */
341 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
342 bno = be64_to_cpu(*pp);
343 if (unlikely((error =
344 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
345 xfs_trans_brelse(tp, bp);
346 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
347 XFS_ERRLEVEL_LOW, mp);
348 return -EFSCORRUPTED;
350 xfs_trans_brelse(tp, bp);
352 /* count all level 1 nodes and their leaves */
354 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
355 numrecs = be16_to_cpu(block->bb_numrecs);
356 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
357 xfs_trans_brelse(tp, bp);
358 if (nextbno == NULLFSBLOCK)
361 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
367 block = XFS_BUF_TO_BLOCK(bp);
374 * Count fsblocks of the given fork.
377 xfs_bmap_count_blocks(
378 xfs_trans_t *tp, /* transaction pointer */
379 xfs_inode_t *ip, /* incore inode */
380 int whichfork, /* data or attr fork */
381 int *count) /* out: count of blocks */
383 struct xfs_btree_block *block; /* current btree block */
384 xfs_fsblock_t bno; /* block # of "block" */
385 xfs_ifork_t *ifp; /* fork structure */
386 int level; /* btree level, for checking */
387 xfs_mount_t *mp; /* file system mount structure */
388 __be64 *pp; /* pointer to block address */
392 ifp = XFS_IFORK_PTR(ip, whichfork);
393 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
394 xfs_bmap_count_leaves(ifp, 0,
395 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
401 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
403 block = ifp->if_broot;
404 level = be16_to_cpu(block->bb_level);
406 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
407 bno = be64_to_cpu(*pp);
408 ASSERT(bno != NULLFSBLOCK);
409 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
410 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
412 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
413 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
415 return -EFSCORRUPTED;
422 * returns 1 for success, 0 if we failed to map the extent.
425 xfs_getbmapx_fix_eof_hole(
426 xfs_inode_t *ip, /* xfs incore inode pointer */
427 struct getbmapx *out, /* output structure */
428 int prealloced, /* this is a file with
429 * preallocated data space */
430 __int64_t end, /* last block requested */
431 xfs_fsblock_t startblock)
434 xfs_mount_t *mp; /* file system mount point */
435 xfs_ifork_t *ifp; /* inode fork pointer */
436 xfs_extnum_t lastx; /* last extent pointer */
437 xfs_fileoff_t fileblock;
439 if (startblock == HOLESTARTBLOCK) {
442 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
443 fixlen -= out->bmv_offset;
444 if (prealloced && out->bmv_offset + out->bmv_length == end) {
445 /* Came to hole at EOF. Trim it. */
448 out->bmv_length = fixlen;
451 if (startblock == DELAYSTARTBLOCK)
454 out->bmv_block = xfs_fsb_to_db(ip, startblock);
455 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
456 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
457 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
458 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
459 out->bmv_oflags |= BMV_OF_LAST;
466 * Get inode's extents as described in bmv, and format for output.
467 * Calls formatter to fill the user's buffer until all extents
468 * are mapped, until the passed-in bmv->bmv_count slots have
469 * been filled, or until the formatter short-circuits the loop,
470 * if it is tracking filled-in extents on its own.
475 struct getbmapx *bmv, /* user bmap structure */
476 xfs_bmap_format_t formatter, /* format to user */
477 void *arg) /* formatter arg */
479 __int64_t bmvend; /* last block requested */
480 int error = 0; /* return value */
481 __int64_t fixlen; /* length for -1 case */
482 int i; /* extent number */
483 int lock; /* lock state */
484 xfs_bmbt_irec_t *map; /* buffer for user's data */
485 xfs_mount_t *mp; /* file system mount point */
486 int nex; /* # of user extents can do */
487 int nexleft; /* # of user extents left */
488 int subnex; /* # of bmapi's can do */
489 int nmap; /* number of map entries */
490 struct getbmapx *out; /* output structure */
491 int whichfork; /* data or attr fork */
492 int prealloced; /* this is a file with
493 * preallocated data space */
494 int iflags; /* interface flags */
495 int bmapi_flags; /* flags for xfs_bmapi */
499 iflags = bmv->bmv_iflags;
500 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
502 if (whichfork == XFS_ATTR_FORK) {
503 if (XFS_IFORK_Q(ip)) {
504 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
505 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
506 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
509 ip->i_d.di_aformat != 0 &&
510 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
511 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
513 return -EFSCORRUPTED;
519 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
520 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
521 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
524 if (xfs_get_extsz_hint(ip) ||
525 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
527 fixlen = mp->m_super->s_maxbytes;
530 fixlen = XFS_ISIZE(ip);
534 if (bmv->bmv_length == -1) {
535 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
537 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
538 } else if (bmv->bmv_length == 0) {
539 bmv->bmv_entries = 0;
541 } else if (bmv->bmv_length < 0) {
545 nex = bmv->bmv_count - 1;
548 bmvend = bmv->bmv_offset + bmv->bmv_length;
551 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
553 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
557 xfs_ilock(ip, XFS_IOLOCK_SHARED);
558 if (whichfork == XFS_DATA_FORK) {
559 if (!(iflags & BMV_IF_DELALLOC) &&
560 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
561 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
563 goto out_unlock_iolock;
566 * Even after flushing the inode, there can still be
567 * delalloc blocks on the inode beyond EOF due to
568 * speculative preallocation. These are not removed
569 * until the release function is called or the inode
570 * is inactivated. Hence we cannot assert here that
571 * ip->i_delayed_blks == 0.
575 lock = xfs_ilock_data_map_shared(ip);
577 lock = xfs_ilock_attr_map_shared(ip);
581 * Don't let nex be bigger than the number of extents
582 * we can have assuming alternating holes and real extents.
584 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
585 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
587 bmapi_flags = xfs_bmapi_aflag(whichfork);
588 if (!(iflags & BMV_IF_PREALLOC))
589 bmapi_flags |= XFS_BMAPI_IGSTATE;
592 * Allocate enough space to handle "subnex" maps at a time.
596 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
598 goto out_unlock_ilock;
600 bmv->bmv_entries = 0;
602 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
603 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
611 nmap = (nexleft > subnex) ? subnex : nexleft;
612 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
613 XFS_BB_TO_FSB(mp, bmv->bmv_length),
614 map, &nmap, bmapi_flags);
617 ASSERT(nmap <= subnex);
619 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
620 out[cur_ext].bmv_oflags = 0;
621 if (map[i].br_state == XFS_EXT_UNWRITTEN)
622 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
623 else if (map[i].br_startblock == DELAYSTARTBLOCK)
624 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
625 out[cur_ext].bmv_offset =
626 XFS_FSB_TO_BB(mp, map[i].br_startoff);
627 out[cur_ext].bmv_length =
628 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
629 out[cur_ext].bmv_unused1 = 0;
630 out[cur_ext].bmv_unused2 = 0;
633 * delayed allocation extents that start beyond EOF can
634 * occur due to speculative EOF allocation when the
635 * delalloc extent is larger than the largest freespace
636 * extent at conversion time. These extents cannot be
637 * converted by data writeback, so can exist here even
638 * if we are not supposed to be finding delalloc
641 if (map[i].br_startblock == DELAYSTARTBLOCK &&
642 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
643 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
645 if (map[i].br_startblock == HOLESTARTBLOCK &&
646 whichfork == XFS_ATTR_FORK) {
647 /* came to the end of attribute fork */
648 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
652 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
654 map[i].br_startblock))
658 out[cur_ext].bmv_offset +
659 out[cur_ext].bmv_length;
661 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
664 * In case we don't want to return the hole,
665 * don't increase cur_ext so that we can reuse
666 * it in the next loop.
668 if ((iflags & BMV_IF_NO_HOLES) &&
669 map[i].br_startblock == HOLESTARTBLOCK) {
670 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
678 } while (nmap && nexleft && bmv->bmv_length);
683 xfs_iunlock(ip, lock);
685 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
687 for (i = 0; i < cur_ext; i++) {
688 int full = 0; /* user array is full */
690 /* format results & advance arg */
691 error = formatter(&arg, &out[i], &full);
701 * dead simple method of punching delalyed allocation blocks from a range in
702 * the inode. Walks a block at a time so will be slow, but is only executed in
703 * rare error cases so the overhead is not critical. This will always punch out
704 * both the start and end blocks, even if the ranges only partially overlap
705 * them, so it is up to the caller to ensure that partial blocks are not
709 xfs_bmap_punch_delalloc_range(
710 struct xfs_inode *ip,
711 xfs_fileoff_t start_fsb,
712 xfs_fileoff_t length)
714 xfs_fileoff_t remaining = length;
717 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
721 xfs_bmbt_irec_t imap;
723 xfs_fsblock_t firstblock;
724 xfs_bmap_free_t flist;
727 * Map the range first and check that it is a delalloc extent
728 * before trying to unmap the range. Otherwise we will be
729 * trying to remove a real extent (which requires a
730 * transaction) or a hole, which is probably a bad idea...
732 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
736 /* something screwed, just bail */
737 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
738 xfs_alert(ip->i_mount,
739 "Failed delalloc mapping lookup ino %lld fsb %lld.",
740 ip->i_ino, start_fsb);
748 if (imap.br_startblock != DELAYSTARTBLOCK) {
749 /* been converted, ignore */
752 WARN_ON(imap.br_blockcount == 0);
755 * Note: while we initialise the firstblock/flist pair, they
756 * should never be used because blocks should never be
757 * allocated or freed for a delalloc extent and hence we need
758 * don't cancel or finish them after the xfs_bunmapi() call.
760 xfs_bmap_init(&flist, &firstblock);
761 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
766 ASSERT(!flist.xbf_count && !flist.xbf_first);
770 } while(remaining > 0);
776 * Test whether it is appropriate to check an inode for and free post EOF
777 * blocks. The 'force' parameter determines whether we should also consider
778 * regular files that are marked preallocated or append-only.
781 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
783 /* prealloc/delalloc exists only on regular files */
784 if (!S_ISREG(ip->i_d.di_mode))
788 * Zero sized files with no cached pages and delalloc blocks will not
789 * have speculative prealloc/delalloc blocks to remove.
791 if (VFS_I(ip)->i_size == 0 &&
792 VFS_I(ip)->i_mapping->nrpages == 0 &&
793 ip->i_delayed_blks == 0)
796 /* If we haven't read in the extent list, then don't do it now. */
797 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
801 * Do not free real preallocated or append-only files unless the file
802 * has delalloc blocks and we are forced to remove them.
804 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
805 if (!force || ip->i_delayed_blks == 0)
812 * This is called by xfs_inactive to free any blocks beyond eof
813 * when the link count isn't zero and by xfs_dm_punch_hole() when
814 * punching a hole to EOF.
824 xfs_fileoff_t end_fsb;
825 xfs_fileoff_t last_fsb;
826 xfs_filblks_t map_len;
828 xfs_bmbt_irec_t imap;
831 * Figure out if there are any blocks beyond the end
832 * of the file. If not, then there is nothing to do.
834 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
835 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
836 if (last_fsb <= end_fsb)
838 map_len = last_fsb - end_fsb;
841 xfs_ilock(ip, XFS_ILOCK_SHARED);
842 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
843 xfs_iunlock(ip, XFS_ILOCK_SHARED);
845 if (!error && (nimaps != 0) &&
846 (imap.br_startblock != HOLESTARTBLOCK ||
847 ip->i_delayed_blks)) {
849 * Attach the dquots to the inode up front.
851 error = xfs_qm_dqattach(ip, 0);
856 * There are blocks after the end of file.
857 * Free them up now by truncating the file to
860 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
863 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
864 xfs_trans_cancel(tp);
869 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
871 ASSERT(XFS_FORCED_SHUTDOWN(mp));
872 xfs_trans_cancel(tp);
874 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
878 xfs_ilock(ip, XFS_ILOCK_EXCL);
879 xfs_trans_ijoin(tp, ip, 0);
882 * Do not update the on-disk file size. If we update the
883 * on-disk file size and then the system crashes before the
884 * contents of the file are flushed to disk then the files
885 * may be full of holes (ie NULL files bug).
887 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
891 * If we get an error at this point we simply don't
892 * bother truncating the file.
894 xfs_trans_cancel(tp);
896 error = xfs_trans_commit(tp,
897 XFS_TRANS_RELEASE_LOG_RES);
899 xfs_inode_clear_eofblocks_tag(ip);
902 xfs_iunlock(ip, XFS_ILOCK_EXCL);
904 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
910 xfs_alloc_file_space(
911 struct xfs_inode *ip,
916 xfs_mount_t *mp = ip->i_mount;
918 xfs_filblks_t allocated_fsb;
919 xfs_filblks_t allocatesize_fsb;
920 xfs_extlen_t extsz, temp;
921 xfs_fileoff_t startoffset_fsb;
922 xfs_fsblock_t firstfsb;
927 xfs_bmbt_irec_t imaps[1], *imapp;
928 xfs_bmap_free_t free_list;
929 uint qblocks, resblks, resrtextents;
933 trace_xfs_alloc_file_space(ip);
935 if (XFS_FORCED_SHUTDOWN(mp))
938 error = xfs_qm_dqattach(ip, 0);
945 rt = XFS_IS_REALTIME_INODE(ip);
946 extsz = xfs_get_extsz_hint(ip);
951 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
952 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
955 * Allocate file space until done or until there is an error
957 while (allocatesize_fsb && !error) {
961 * Determine space reservations for data/realtime.
963 if (unlikely(extsz)) {
967 e = startoffset_fsb + allocatesize_fsb;
968 if ((temp = do_mod(startoffset_fsb, extsz)))
970 if ((temp = do_mod(e, extsz)))
974 e = allocatesize_fsb;
978 * The transaction reservation is limited to a 32-bit block
979 * count, hence we need to limit the number of blocks we are
980 * trying to reserve to avoid an overflow. We can't allocate
981 * more than @nimaps extents, and an extent is limited on disk
982 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
984 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
986 resrtextents = qblocks = resblks;
987 resrtextents /= mp->m_sb.sb_rextsize;
988 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
989 quota_flag = XFS_QMOPT_RES_RTBLKS;
992 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
993 quota_flag = XFS_QMOPT_RES_REGBLKS;
997 * Allocate and setup the transaction.
999 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1000 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1001 resblks, resrtextents);
1003 * Check for running out of space
1007 * Free the transaction structure.
1009 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1010 xfs_trans_cancel(tp);
1013 xfs_ilock(ip, XFS_ILOCK_EXCL);
1014 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1019 xfs_trans_ijoin(tp, ip, 0);
1021 xfs_bmap_init(&free_list, &firstfsb);
1022 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1023 allocatesize_fsb, alloc_type, &firstfsb,
1024 0, imapp, &nimaps, &free_list);
1030 * Complete the transaction
1032 error = xfs_bmap_finish(&tp, &free_list, &committed);
1037 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1038 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1043 allocated_fsb = imapp->br_blockcount;
1050 startoffset_fsb += allocated_fsb;
1051 allocatesize_fsb -= allocated_fsb;
1056 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1057 xfs_bmap_cancel(&free_list);
1058 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1060 error1: /* Just cancel transaction */
1061 xfs_trans_cancel(tp);
1062 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1067 * Zero file bytes between startoff and endoff inclusive.
1068 * The iolock is held exclusive and no blocks are buffered.
1070 * This function is used by xfs_free_file_space() to zero
1071 * partial blocks when the range to free is not block aligned.
1072 * When unreserving space with boundaries that are not block
1073 * aligned we round up the start and round down the end
1074 * boundaries and then use this function to zero the parts of
1075 * the blocks that got dropped during the rounding.
1078 xfs_zero_remaining_bytes(
1083 xfs_bmbt_irec_t imap;
1084 xfs_fileoff_t offset_fsb;
1085 xfs_off_t lastoffset;
1088 xfs_mount_t *mp = ip->i_mount;
1093 * Avoid doing I/O beyond eof - it's not necessary
1094 * since nothing can read beyond eof. The space will
1095 * be zeroed when the file is extended anyway.
1097 if (startoff >= XFS_ISIZE(ip))
1100 if (endoff > XFS_ISIZE(ip))
1101 endoff = XFS_ISIZE(ip);
1103 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1106 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1109 lock_mode = xfs_ilock_data_map_shared(ip);
1110 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1111 xfs_iunlock(ip, lock_mode);
1113 if (error || nimap < 1)
1115 ASSERT(imap.br_blockcount >= 1);
1116 ASSERT(imap.br_startoff == offset_fsb);
1117 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1118 if (lastoffset > endoff)
1119 lastoffset = endoff;
1120 if (imap.br_startblock == HOLESTARTBLOCK)
1122 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1123 if (imap.br_state == XFS_EXT_UNWRITTEN)
1126 error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ?
1127 mp->m_rtdev_targp : mp->m_ddev_targp,
1128 xfs_fsb_to_db(ip, imap.br_startblock),
1129 BTOBB(mp->m_sb.sb_blocksize),
1135 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1136 0, lastoffset - offset + 1);
1138 error = xfs_bwrite(bp);
1147 xfs_free_file_space(
1148 struct xfs_inode *ip,
1154 xfs_fileoff_t endoffset_fsb;
1156 xfs_fsblock_t firstfsb;
1157 xfs_bmap_free_t free_list;
1158 xfs_bmbt_irec_t imap;
1160 xfs_off_t iendoffset;
1167 xfs_fileoff_t startoffset_fsb;
1172 trace_xfs_free_file_space(ip);
1174 error = xfs_qm_dqattach(ip, 0);
1179 if (len <= 0) /* if nothing being freed */
1181 rt = XFS_IS_REALTIME_INODE(ip);
1182 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1183 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1185 /* wait for the completion of any pending DIOs */
1186 inode_dio_wait(VFS_I(ip));
1188 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1189 ioffset = round_down(offset, rounding);
1190 iendoffset = round_up(offset + len, rounding) - 1;
1191 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, ioffset,
1195 truncate_pagecache_range(VFS_I(ip), ioffset, iendoffset);
1198 * Need to zero the stuff we're not freeing, on disk.
1199 * If it's a realtime file & can't use unwritten extents then we
1200 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1201 * will take care of it for us.
1203 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1205 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1209 ASSERT(nimap == 0 || nimap == 1);
1210 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1213 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1214 block = imap.br_startblock;
1215 mod = do_div(block, mp->m_sb.sb_rextsize);
1217 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1220 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1224 ASSERT(nimap == 0 || nimap == 1);
1225 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1226 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1228 if (mod && (mod != mp->m_sb.sb_rextsize))
1229 endoffset_fsb -= mod;
1232 if ((done = (endoffset_fsb <= startoffset_fsb)))
1234 * One contiguous piece to clear
1236 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1239 * Some full blocks, possibly two pieces to clear
1241 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1242 error = xfs_zero_remaining_bytes(ip, offset,
1243 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1245 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1246 error = xfs_zero_remaining_bytes(ip,
1247 XFS_FSB_TO_B(mp, endoffset_fsb),
1252 * free file space until done or until there is an error
1254 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1255 while (!error && !done) {
1258 * allocate and setup the transaction. Allow this
1259 * transaction to dip into the reserve blocks to ensure
1260 * the freeing of the space succeeds at ENOSPC.
1262 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1263 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1266 * check for running out of space
1270 * Free the transaction structure.
1272 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1273 xfs_trans_cancel(tp);
1276 xfs_ilock(ip, XFS_ILOCK_EXCL);
1277 error = xfs_trans_reserve_quota(tp, mp,
1278 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1279 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1283 xfs_trans_ijoin(tp, ip, 0);
1286 * issue the bunmapi() call to free the blocks
1288 xfs_bmap_init(&free_list, &firstfsb);
1289 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1290 endoffset_fsb - startoffset_fsb,
1291 0, 2, &firstfsb, &free_list, &done);
1297 * complete the transaction
1299 error = xfs_bmap_finish(&tp, &free_list, &committed);
1304 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1305 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1312 xfs_bmap_cancel(&free_list);
1314 xfs_trans_cancel(tp);
1315 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1320 * Preallocate and zero a range of a file. This mechanism has the allocation
1321 * semantics of fallocate and in addition converts data in the range to zeroes.
1324 xfs_zero_file_space(
1325 struct xfs_inode *ip,
1329 struct xfs_mount *mp = ip->i_mount;
1333 trace_xfs_zero_file_space(ip);
1335 blksize = 1 << mp->m_sb.sb_blocklog;
1338 * Punch a hole and prealloc the range. We use hole punch rather than
1339 * unwritten extent conversion for two reasons:
1341 * 1.) Hole punch handles partial block zeroing for us.
1343 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1344 * by virtue of the hole punch.
1346 error = xfs_free_file_space(ip, offset, len);
1350 error = xfs_alloc_file_space(ip, round_down(offset, blksize),
1351 round_up(offset + len, blksize) -
1352 round_down(offset, blksize),
1353 XFS_BMAPI_PREALLOC);
1360 * @next_fsb will keep track of the extent currently undergoing shift.
1361 * @stop_fsb will keep track of the extent at which we have to stop.
1362 * If we are shifting left, we will start with block (offset + len) and
1363 * shift each extent till last extent.
1364 * If we are shifting right, we will start with last extent inside file space
1365 * and continue until we reach the block corresponding to offset.
1368 xfs_shift_file_space(
1369 struct xfs_inode *ip,
1372 enum shift_direction direction)
1375 struct xfs_mount *mp = ip->i_mount;
1376 struct xfs_trans *tp;
1378 struct xfs_bmap_free free_list;
1379 xfs_fsblock_t first_block;
1381 xfs_fileoff_t stop_fsb;
1382 xfs_fileoff_t next_fsb;
1383 xfs_fileoff_t shift_fsb;
1385 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
1387 if (direction == SHIFT_LEFT) {
1388 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1389 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size);
1392 * If right shift, delegate the work of initialization of
1393 * next_fsb to xfs_bmap_shift_extent as it has ilock held.
1395 next_fsb = NULLFSBLOCK;
1396 stop_fsb = XFS_B_TO_FSB(mp, offset);
1399 shift_fsb = XFS_B_TO_FSB(mp, len);
1402 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1403 * into the accessible region of the file.
1405 if (xfs_can_free_eofblocks(ip, true)) {
1406 error = xfs_free_eofblocks(mp, ip, false);
1412 * Writeback and invalidate cache for the remainder of the file as we're
1413 * about to shift down every extent from offset to EOF.
1415 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1419 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping,
1420 offset >> PAGE_CACHE_SHIFT, -1);
1425 * The extent shiting code works on extent granularity. So, if
1426 * stop_fsb is not the starting block of extent, we need to split
1427 * the extent at stop_fsb.
1429 if (direction == SHIFT_RIGHT) {
1430 error = xfs_bmap_split_extent(ip, stop_fsb);
1435 while (!error && !done) {
1436 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1438 * We would need to reserve permanent block for transaction.
1439 * This will come into picture when after shifting extent into
1440 * hole we found that adjacent extents can be merged which
1441 * may lead to freeing of a block during record update.
1443 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1444 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1446 xfs_trans_cancel(tp);
1450 xfs_ilock(ip, XFS_ILOCK_EXCL);
1451 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1452 ip->i_gdquot, ip->i_pdquot,
1453 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1454 XFS_QMOPT_RES_REGBLKS);
1458 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1460 xfs_bmap_init(&free_list, &first_block);
1463 * We are using the write transaction in which max 2 bmbt
1464 * updates are allowed
1466 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb,
1467 &done, stop_fsb, &first_block, &free_list,
1468 direction, XFS_BMAP_MAX_SHIFT_EXTENTS);
1472 error = xfs_bmap_finish(&tp, &free_list, &committed);
1476 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1482 xfs_trans_cancel(tp);
1487 * xfs_collapse_file_space()
1488 * This routine frees disk space and shift extent for the given file.
1489 * The first thing we do is to free data blocks in the specified range
1490 * by calling xfs_free_file_space(). It would also sync dirty data
1491 * and invalidate page cache over the region on which collapse range
1492 * is working. And Shift extent records to the left to cover a hole.
1499 xfs_collapse_file_space(
1500 struct xfs_inode *ip,
1506 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1507 trace_xfs_collapse_file_space(ip);
1509 error = xfs_free_file_space(ip, offset, len);
1513 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT);
1517 * xfs_insert_file_space()
1518 * This routine create hole space by shifting extents for the given file.
1519 * The first thing we do is to sync dirty data and invalidate page cache
1520 * over the region on which insert range is working. And split an extent
1521 * to two extents at given offset by calling xfs_bmap_split_extent.
1522 * And shift all extent records which are laying between [offset,
1523 * last allocated extent] to the right to reserve hole range.
1529 xfs_insert_file_space(
1530 struct xfs_inode *ip,
1534 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1535 trace_xfs_insert_file_space(ip);
1537 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT);
1541 * We need to check that the format of the data fork in the temporary inode is
1542 * valid for the target inode before doing the swap. This is not a problem with
1543 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1544 * data fork depending on the space the attribute fork is taking so we can get
1545 * invalid formats on the target inode.
1547 * E.g. target has space for 7 extents in extent format, temp inode only has
1548 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1549 * btree, but when swapped it needs to be in extent format. Hence we can't just
1550 * blindly swap data forks on attr2 filesystems.
1552 * Note that we check the swap in both directions so that we don't end up with
1553 * a corrupt temporary inode, either.
1555 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1556 * inode will prevent this situation from occurring, so all we do here is
1557 * reject and log the attempt. basically we are putting the responsibility on
1558 * userspace to get this right.
1561 xfs_swap_extents_check_format(
1562 xfs_inode_t *ip, /* target inode */
1563 xfs_inode_t *tip) /* tmp inode */
1566 /* Should never get a local format */
1567 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1568 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1572 * if the target inode has less extents that then temporary inode then
1573 * why did userspace call us?
1575 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1579 * if the target inode is in extent form and the temp inode is in btree
1580 * form then we will end up with the target inode in the wrong format
1581 * as we already know there are less extents in the temp inode.
1583 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1584 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1587 /* Check temp in extent form to max in target */
1588 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1589 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1590 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1593 /* Check target in extent form to max in temp */
1594 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1595 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1596 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1600 * If we are in a btree format, check that the temp root block will fit
1601 * in the target and that it has enough extents to be in btree format
1604 * Note that we have to be careful to allow btree->extent conversions
1605 * (a common defrag case) which will occur when the temp inode is in
1608 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1609 if (XFS_IFORK_BOFF(ip) &&
1610 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1612 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1613 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1617 /* Reciprocal target->temp btree format checks */
1618 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1619 if (XFS_IFORK_BOFF(tip) &&
1620 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1622 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1623 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1631 xfs_swap_extent_flush(
1632 struct xfs_inode *ip)
1636 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1639 truncate_pagecache_range(VFS_I(ip), 0, -1);
1641 /* Verify O_DIRECT for ftmp */
1642 if (VFS_I(ip)->i_mapping->nrpages)
1649 xfs_inode_t *ip, /* target inode */
1650 xfs_inode_t *tip, /* tmp inode */
1653 xfs_mount_t *mp = ip->i_mount;
1655 xfs_bstat_t *sbp = &sxp->sx_stat;
1656 xfs_ifork_t *tempifp, *ifp, *tifp;
1657 int src_log_flags, target_log_flags;
1664 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1671 * Lock the inodes against other IO, page faults and truncate to
1672 * begin with. Then we can ensure the inodes are flushed and have no
1673 * page cache safely. Once we have done this we can take the ilocks and
1674 * do the rest of the checks.
1676 lock_flags = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1677 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1678 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL);
1680 /* Verify that both files have the same format */
1681 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1686 /* Verify both files are either real-time or non-realtime */
1687 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1692 error = xfs_swap_extent_flush(ip);
1695 error = xfs_swap_extent_flush(tip);
1699 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1700 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1702 xfs_trans_cancel(tp);
1707 * Lock and join the inodes to the tansaction so that transaction commit
1708 * or cancel will unlock the inodes from this point onwards.
1710 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1711 lock_flags |= XFS_ILOCK_EXCL;
1712 xfs_trans_ijoin(tp, ip, lock_flags);
1713 xfs_trans_ijoin(tp, tip, lock_flags);
1716 /* Verify all data are being swapped */
1717 if (sxp->sx_offset != 0 ||
1718 sxp->sx_length != ip->i_d.di_size ||
1719 sxp->sx_length != tip->i_d.di_size) {
1721 goto out_trans_cancel;
1724 trace_xfs_swap_extent_before(ip, 0);
1725 trace_xfs_swap_extent_before(tip, 1);
1727 /* check inode formats now that data is flushed */
1728 error = xfs_swap_extents_check_format(ip, tip);
1731 "%s: inode 0x%llx format is incompatible for exchanging.",
1732 __func__, ip->i_ino);
1733 goto out_trans_cancel;
1737 * Compare the current change & modify times with that
1738 * passed in. If they differ, we abort this swap.
1739 * This is the mechanism used to ensure the calling
1740 * process that the file was not changed out from
1743 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1744 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1745 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1746 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1748 goto out_trans_cancel;
1751 * Count the number of extended attribute blocks
1753 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1754 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1755 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1757 goto out_trans_cancel;
1759 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1760 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1761 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1764 goto out_trans_cancel;
1768 * Before we've swapped the forks, lets set the owners of the forks
1769 * appropriately. We have to do this as we are demand paging the btree
1770 * buffers, and so the validation done on read will expect the owner
1771 * field to be correctly set. Once we change the owners, we can swap the
1774 * Note the trickiness in setting the log flags - we set the owner log
1775 * flag on the opposite inode (i.e. the inode we are setting the new
1776 * owner to be) because once we swap the forks and log that, log
1777 * recovery is going to see the fork as owned by the swapped inode,
1778 * not the pre-swapped inodes.
1780 src_log_flags = XFS_ILOG_CORE;
1781 target_log_flags = XFS_ILOG_CORE;
1782 if (ip->i_d.di_version == 3 &&
1783 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1784 target_log_flags |= XFS_ILOG_DOWNER;
1785 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1788 goto out_trans_cancel;
1791 if (tip->i_d.di_version == 3 &&
1792 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1793 src_log_flags |= XFS_ILOG_DOWNER;
1794 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1797 goto out_trans_cancel;
1801 * Swap the data forks of the inodes
1805 *tempifp = *ifp; /* struct copy */
1806 *ifp = *tifp; /* struct copy */
1807 *tifp = *tempifp; /* struct copy */
1810 * Fix the on-disk inode values
1812 tmp = (__uint64_t)ip->i_d.di_nblocks;
1813 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1814 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1816 tmp = (__uint64_t) ip->i_d.di_nextents;
1817 ip->i_d.di_nextents = tip->i_d.di_nextents;
1818 tip->i_d.di_nextents = tmp;
1820 tmp = (__uint64_t) ip->i_d.di_format;
1821 ip->i_d.di_format = tip->i_d.di_format;
1822 tip->i_d.di_format = tmp;
1825 * The extents in the source inode could still contain speculative
1826 * preallocation beyond EOF (e.g. the file is open but not modified
1827 * while defrag is in progress). In that case, we need to copy over the
1828 * number of delalloc blocks the data fork in the source inode is
1829 * tracking beyond EOF so that when the fork is truncated away when the
1830 * temporary inode is unlinked we don't underrun the i_delayed_blks
1831 * counter on that inode.
1833 ASSERT(tip->i_delayed_blks == 0);
1834 tip->i_delayed_blks = ip->i_delayed_blks;
1835 ip->i_delayed_blks = 0;
1837 switch (ip->i_d.di_format) {
1838 case XFS_DINODE_FMT_EXTENTS:
1839 /* If the extents fit in the inode, fix the
1840 * pointer. Otherwise it's already NULL or
1841 * pointing to the extent.
1843 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1844 ifp->if_u1.if_extents =
1845 ifp->if_u2.if_inline_ext;
1847 src_log_flags |= XFS_ILOG_DEXT;
1849 case XFS_DINODE_FMT_BTREE:
1850 ASSERT(ip->i_d.di_version < 3 ||
1851 (src_log_flags & XFS_ILOG_DOWNER));
1852 src_log_flags |= XFS_ILOG_DBROOT;
1856 switch (tip->i_d.di_format) {
1857 case XFS_DINODE_FMT_EXTENTS:
1858 /* If the extents fit in the inode, fix the
1859 * pointer. Otherwise it's already NULL or
1860 * pointing to the extent.
1862 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1863 tifp->if_u1.if_extents =
1864 tifp->if_u2.if_inline_ext;
1866 target_log_flags |= XFS_ILOG_DEXT;
1868 case XFS_DINODE_FMT_BTREE:
1869 target_log_flags |= XFS_ILOG_DBROOT;
1870 ASSERT(tip->i_d.di_version < 3 ||
1871 (target_log_flags & XFS_ILOG_DOWNER));
1875 xfs_trans_log_inode(tp, ip, src_log_flags);
1876 xfs_trans_log_inode(tp, tip, target_log_flags);
1879 * If this is a synchronous mount, make sure that the
1880 * transaction goes to disk before returning to the user.
1882 if (mp->m_flags & XFS_MOUNT_WSYNC)
1883 xfs_trans_set_sync(tp);
1885 error = xfs_trans_commit(tp, 0);
1887 trace_xfs_swap_extent_after(ip, 0);
1888 trace_xfs_swap_extent_after(tip, 1);
1894 xfs_iunlock(ip, lock_flags);
1895 xfs_iunlock(tip, lock_flags);
1899 xfs_trans_cancel(tp);