1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2012 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_trans.h"
19 #include "xfs_alloc.h"
21 #include "xfs_bmap_util.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_rtalloc.h"
24 #include "xfs_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_trans_space.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_iomap.h"
30 #include "xfs_reflink.h"
32 /* Kernel only BMAP related definitions and functions */
35 * Convert the given file system block to a disk block. We have to treat it
36 * differently based on whether the file is a real time file or not, because the
40 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
42 if (XFS_IS_REALTIME_INODE(ip))
43 return XFS_FSB_TO_BB(ip->i_mount, fsb);
44 return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
48 * Routine to zero an extent on disk allocated to the specific inode.
50 * The VFS functions take a linearised filesystem block offset, so we have to
51 * convert the sparse xfs fsb to the right format first.
52 * VFS types are real funky, too.
57 xfs_fsblock_t start_fsb,
60 struct xfs_mount *mp = ip->i_mount;
61 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
62 sector_t block = XFS_BB_TO_FSBT(mp, sector);
64 return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)),
65 block << (mp->m_super->s_blocksize_bits - 9),
66 count_fsb << (mp->m_super->s_blocksize_bits - 9),
73 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
75 int error; /* error return value */
76 xfs_mount_t *mp; /* mount point structure */
77 xfs_extlen_t prod = 0; /* product factor for allocators */
78 xfs_extlen_t mod = 0; /* product factor for allocators */
79 xfs_extlen_t ralen = 0; /* realtime allocation length */
80 xfs_extlen_t align; /* minimum allocation alignment */
84 align = xfs_get_extsz_hint(ap->ip);
85 prod = align / mp->m_sb.sb_rextsize;
86 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
88 ap->conv, &ap->offset, &ap->length);
92 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
95 * If the offset & length are not perfectly aligned
96 * then kill prod, it will just get us in trouble.
98 div_u64_rem(ap->offset, align, &mod);
99 if (mod || ap->length % align)
102 * Set ralen to be the actual requested length in rtextents.
104 ralen = ap->length / mp->m_sb.sb_rextsize;
106 * If the old value was close enough to MAXEXTLEN that
107 * we rounded up to it, cut it back so it's valid again.
108 * Note that if it's a really large request (bigger than
109 * MAXEXTLEN), we don't hear about that number, and can't
110 * adjust the starting point to match it.
112 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
113 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
116 * Lock out modifications to both the RT bitmap and summary inodes
118 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
119 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
120 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
121 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
124 * If it's an allocation to an empty file at offset 0,
125 * pick an extent that will space things out in the rt area.
127 if (ap->eof && ap->offset == 0) {
128 xfs_rtblock_t rtx; /* realtime extent no */
130 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
133 ap->blkno = rtx * mp->m_sb.sb_rextsize;
138 xfs_bmap_adjacent(ap);
141 * Realtime allocation, done through xfs_rtallocate_extent.
143 do_div(ap->blkno, mp->m_sb.sb_rextsize);
146 error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
147 &ralen, ap->wasdel, prod, &rtb);
152 if (ap->blkno != NULLFSBLOCK) {
153 ap->blkno *= mp->m_sb.sb_rextsize;
154 ralen *= mp->m_sb.sb_rextsize;
156 ap->ip->i_d.di_nblocks += ralen;
157 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
159 ap->ip->i_delayed_blks -= ralen;
161 * Adjust the disk quota also. This was reserved
164 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
165 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
166 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
168 /* Zero the extent if we were asked to do so */
169 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) {
170 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length);
179 #endif /* CONFIG_XFS_RT */
182 * Check if the endoff is outside the last extent. If so the caller will grow
183 * the allocation to a stripe unit boundary. All offsets are considered outside
184 * the end of file for an empty fork, so 1 is returned in *eof in that case.
188 struct xfs_inode *ip,
189 xfs_fileoff_t endoff,
193 struct xfs_bmbt_irec rec;
196 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
200 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
205 * Extent tree block counting routines.
209 * Count leaf blocks given a range of extent records. Delayed allocation
210 * extents are not counted towards the totals.
213 xfs_bmap_count_leaves(
214 struct xfs_ifork *ifp,
215 xfs_filblks_t *count)
217 struct xfs_iext_cursor icur;
218 struct xfs_bmbt_irec got;
219 xfs_extnum_t numrecs = 0;
221 for_each_xfs_iext(ifp, &icur, &got) {
222 if (!isnullstartblock(got.br_startblock)) {
223 *count += got.br_blockcount;
232 * Count leaf blocks given a range of extent records originally
236 xfs_bmap_disk_count_leaves(
237 struct xfs_mount *mp,
238 struct xfs_btree_block *block,
240 xfs_filblks_t *count)
245 for (b = 1; b <= numrecs; b++) {
246 frp = XFS_BMBT_REC_ADDR(mp, block, b);
247 *count += xfs_bmbt_disk_get_blockcount(frp);
252 * Recursively walks each level of a btree
253 * to count total fsblocks in use.
257 struct xfs_mount *mp,
258 struct xfs_trans *tp,
259 struct xfs_ifork *ifp,
260 xfs_fsblock_t blockno,
262 xfs_extnum_t *nextents,
263 xfs_filblks_t *count)
266 struct xfs_buf *bp, *nbp;
269 xfs_fsblock_t bno = blockno;
270 xfs_fsblock_t nextbno;
271 struct xfs_btree_block *block, *nextblock;
274 error = xfs_btree_read_bufl(mp, tp, bno, &bp, XFS_BMAP_BTREE_REF,
279 block = XFS_BUF_TO_BLOCK(bp);
282 /* Not at node above leaves, count this level of nodes */
283 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
284 while (nextbno != NULLFSBLOCK) {
285 error = xfs_btree_read_bufl(mp, tp, nextbno, &nbp,
291 nextblock = XFS_BUF_TO_BLOCK(nbp);
292 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
293 xfs_trans_brelse(tp, nbp);
296 /* Dive to the next level */
297 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
298 bno = be64_to_cpu(*pp);
299 error = xfs_bmap_count_tree(mp, tp, ifp, bno, level, nextents,
302 xfs_trans_brelse(tp, bp);
303 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
304 XFS_ERRLEVEL_LOW, mp);
305 return -EFSCORRUPTED;
307 xfs_trans_brelse(tp, bp);
309 /* count all level 1 nodes and their leaves */
311 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
312 numrecs = be16_to_cpu(block->bb_numrecs);
313 (*nextents) += numrecs;
314 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
315 xfs_trans_brelse(tp, bp);
316 if (nextbno == NULLFSBLOCK)
319 error = xfs_btree_read_bufl(mp, tp, bno, &bp,
325 block = XFS_BUF_TO_BLOCK(bp);
332 * Count fsblocks of the given fork. Delayed allocation extents are
333 * not counted towards the totals.
336 xfs_bmap_count_blocks(
337 struct xfs_trans *tp,
338 struct xfs_inode *ip,
340 xfs_extnum_t *nextents,
341 xfs_filblks_t *count)
343 struct xfs_mount *mp; /* file system mount structure */
344 __be64 *pp; /* pointer to block address */
345 struct xfs_btree_block *block; /* current btree block */
346 struct xfs_ifork *ifp; /* fork structure */
347 xfs_fsblock_t bno; /* block # of "block" */
348 int level; /* btree level, for checking */
355 ifp = XFS_IFORK_PTR(ip, whichfork);
359 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
360 case XFS_DINODE_FMT_EXTENTS:
361 *nextents = xfs_bmap_count_leaves(ifp, count);
363 case XFS_DINODE_FMT_BTREE:
364 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
365 error = xfs_iread_extents(tp, ip, whichfork);
371 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
373 block = ifp->if_broot;
374 level = be16_to_cpu(block->bb_level);
376 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
377 bno = be64_to_cpu(*pp);
378 ASSERT(bno != NULLFSBLOCK);
379 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
380 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
382 error = xfs_bmap_count_tree(mp, tp, ifp, bno, level,
385 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)",
386 XFS_ERRLEVEL_LOW, mp);
387 return -EFSCORRUPTED;
396 xfs_getbmap_report_one(
397 struct xfs_inode *ip,
398 struct getbmapx *bmv,
399 struct kgetbmap *out,
401 struct xfs_bmbt_irec *got)
403 struct kgetbmap *p = out + bmv->bmv_entries;
407 error = xfs_reflink_trim_around_shared(ip, got, &shared);
411 if (isnullstartblock(got->br_startblock) ||
412 got->br_startblock == DELAYSTARTBLOCK) {
414 * Delalloc extents that start beyond EOF can occur due to
415 * speculative EOF allocation when the delalloc extent is larger
416 * than the largest freespace extent at conversion time. These
417 * extents cannot be converted by data writeback, so can exist
418 * here even if we are not supposed to be finding delalloc
421 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
422 ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
424 p->bmv_oflags |= BMV_OF_DELALLOC;
427 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
430 if (got->br_state == XFS_EXT_UNWRITTEN &&
431 (bmv->bmv_iflags & BMV_IF_PREALLOC))
432 p->bmv_oflags |= BMV_OF_PREALLOC;
435 p->bmv_oflags |= BMV_OF_SHARED;
437 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
438 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
440 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
441 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
447 xfs_getbmap_report_hole(
448 struct xfs_inode *ip,
449 struct getbmapx *bmv,
450 struct kgetbmap *out,
455 struct kgetbmap *p = out + bmv->bmv_entries;
457 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
461 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
462 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
464 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
465 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
471 struct getbmapx *bmv)
473 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
477 xfs_getbmap_next_rec(
478 struct xfs_bmbt_irec *rec,
479 xfs_fileoff_t total_end)
481 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
483 if (end == total_end)
486 rec->br_startoff += rec->br_blockcount;
487 if (!isnullstartblock(rec->br_startblock) &&
488 rec->br_startblock != DELAYSTARTBLOCK)
489 rec->br_startblock += rec->br_blockcount;
490 rec->br_blockcount = total_end - end;
495 * Get inode's extents as described in bmv, and format for output.
496 * Calls formatter to fill the user's buffer until all extents
497 * are mapped, until the passed-in bmv->bmv_count slots have
498 * been filled, or until the formatter short-circuits the loop,
499 * if it is tracking filled-in extents on its own.
503 struct xfs_inode *ip,
504 struct getbmapx *bmv, /* user bmap structure */
505 struct kgetbmap *out)
507 struct xfs_mount *mp = ip->i_mount;
508 int iflags = bmv->bmv_iflags;
509 int whichfork, lock, error = 0;
510 int64_t bmv_end, max_len;
511 xfs_fileoff_t bno, first_bno;
512 struct xfs_ifork *ifp;
513 struct xfs_bmbt_irec got, rec;
515 struct xfs_iext_cursor icur;
517 if (bmv->bmv_iflags & ~BMV_IF_VALID)
520 /* Only allow CoW fork queries if we're debugging. */
521 if (iflags & BMV_IF_COWFORK)
524 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
527 if (bmv->bmv_length < -1)
529 bmv->bmv_entries = 0;
530 if (bmv->bmv_length == 0)
533 if (iflags & BMV_IF_ATTRFORK)
534 whichfork = XFS_ATTR_FORK;
535 else if (iflags & BMV_IF_COWFORK)
536 whichfork = XFS_COW_FORK;
538 whichfork = XFS_DATA_FORK;
539 ifp = XFS_IFORK_PTR(ip, whichfork);
541 xfs_ilock(ip, XFS_IOLOCK_SHARED);
544 if (!XFS_IFORK_Q(ip))
545 goto out_unlock_iolock;
548 lock = xfs_ilock_attr_map_shared(ip);
551 /* No CoW fork? Just return */
553 goto out_unlock_iolock;
555 if (xfs_get_cowextsz_hint(ip))
556 max_len = mp->m_super->s_maxbytes;
558 max_len = XFS_ISIZE(ip);
560 lock = XFS_ILOCK_SHARED;
564 if (!(iflags & BMV_IF_DELALLOC) &&
565 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
566 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
568 goto out_unlock_iolock;
571 * Even after flushing the inode, there can still be
572 * delalloc blocks on the inode beyond EOF due to
573 * speculative preallocation. These are not removed
574 * until the release function is called or the inode
575 * is inactivated. Hence we cannot assert here that
576 * ip->i_delayed_blks == 0.
580 if (xfs_get_extsz_hint(ip) ||
582 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
583 max_len = mp->m_super->s_maxbytes;
585 max_len = XFS_ISIZE(ip);
587 lock = xfs_ilock_data_map_shared(ip);
591 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
592 case XFS_DINODE_FMT_EXTENTS:
593 case XFS_DINODE_FMT_BTREE:
595 case XFS_DINODE_FMT_LOCAL:
596 /* Local format inode forks report no extents. */
597 goto out_unlock_ilock;
600 goto out_unlock_ilock;
603 if (bmv->bmv_length == -1) {
604 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
605 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
608 bmv_end = bmv->bmv_offset + bmv->bmv_length;
610 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
611 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
613 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
614 error = xfs_iread_extents(NULL, ip, whichfork);
616 goto out_unlock_ilock;
619 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
621 * Report a whole-file hole if the delalloc flag is set to
622 * stay compatible with the old implementation.
624 if (iflags & BMV_IF_DELALLOC)
625 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
626 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
627 goto out_unlock_ilock;
630 while (!xfs_getbmap_full(bmv)) {
631 xfs_trim_extent(&got, first_bno, len);
634 * Report an entry for a hole if this extent doesn't directly
635 * follow the previous one.
637 if (got.br_startoff > bno) {
638 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
640 if (xfs_getbmap_full(bmv))
645 * In order to report shared extents accurately, we report each
646 * distinct shared / unshared part of a single bmbt record with
647 * an individual getbmapx record.
649 bno = got.br_startoff + got.br_blockcount;
652 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
654 if (error || xfs_getbmap_full(bmv))
655 goto out_unlock_ilock;
656 } while (xfs_getbmap_next_rec(&rec, bno));
658 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
659 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
661 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
663 if (whichfork != XFS_ATTR_FORK && bno < end &&
664 !xfs_getbmap_full(bmv)) {
665 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
671 if (bno >= first_bno + len)
676 xfs_iunlock(ip, lock);
678 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
683 * Dead simple method of punching delalyed allocation blocks from a range in
684 * the inode. This will always punch out both the start and end blocks, even
685 * if the ranges only partially overlap them, so it is up to the caller to
686 * ensure that partial blocks are not passed in.
689 xfs_bmap_punch_delalloc_range(
690 struct xfs_inode *ip,
691 xfs_fileoff_t start_fsb,
692 xfs_fileoff_t length)
694 struct xfs_ifork *ifp = &ip->i_df;
695 xfs_fileoff_t end_fsb = start_fsb + length;
696 struct xfs_bmbt_irec got, del;
697 struct xfs_iext_cursor icur;
700 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
702 xfs_ilock(ip, XFS_ILOCK_EXCL);
703 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
706 while (got.br_startoff + got.br_blockcount > start_fsb) {
708 xfs_trim_extent(&del, start_fsb, length);
711 * A delete can push the cursor forward. Step back to the
712 * previous extent on non-delalloc or extents outside the
715 if (!del.br_blockcount ||
716 !isnullstartblock(del.br_startblock)) {
717 if (!xfs_iext_prev_extent(ifp, &icur, &got))
722 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
724 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
729 xfs_iunlock(ip, XFS_ILOCK_EXCL);
734 * Test whether it is appropriate to check an inode for and free post EOF
735 * blocks. The 'force' parameter determines whether we should also consider
736 * regular files that are marked preallocated or append-only.
739 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
741 /* prealloc/delalloc exists only on regular files */
742 if (!S_ISREG(VFS_I(ip)->i_mode))
746 * Zero sized files with no cached pages and delalloc blocks will not
747 * have speculative prealloc/delalloc blocks to remove.
749 if (VFS_I(ip)->i_size == 0 &&
750 VFS_I(ip)->i_mapping->nrpages == 0 &&
751 ip->i_delayed_blks == 0)
754 /* If we haven't read in the extent list, then don't do it now. */
755 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
759 * Do not free real preallocated or append-only files unless the file
760 * has delalloc blocks and we are forced to remove them.
762 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
763 if (!force || ip->i_delayed_blks == 0)
770 * This is called to free any blocks beyond eof. The caller must hold
771 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
772 * reference to the inode.
776 struct xfs_inode *ip)
778 struct xfs_trans *tp;
780 xfs_fileoff_t end_fsb;
781 xfs_fileoff_t last_fsb;
782 xfs_filblks_t map_len;
784 struct xfs_bmbt_irec imap;
785 struct xfs_mount *mp = ip->i_mount;
788 * Figure out if there are any blocks beyond the end
789 * of the file. If not, then there is nothing to do.
791 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
792 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
793 if (last_fsb <= end_fsb)
795 map_len = last_fsb - end_fsb;
798 xfs_ilock(ip, XFS_ILOCK_SHARED);
799 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
800 xfs_iunlock(ip, XFS_ILOCK_SHARED);
803 * If there are blocks after the end of file, truncate the file to its
804 * current size to free them up.
806 if (!error && (nimaps != 0) &&
807 (imap.br_startblock != HOLESTARTBLOCK ||
808 ip->i_delayed_blks)) {
810 * Attach the dquots to the inode up front.
812 error = xfs_qm_dqattach(ip);
816 /* wait on dio to ensure i_size has settled */
817 inode_dio_wait(VFS_I(ip));
819 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
822 ASSERT(XFS_FORCED_SHUTDOWN(mp));
826 xfs_ilock(ip, XFS_ILOCK_EXCL);
827 xfs_trans_ijoin(tp, ip, 0);
830 * Do not update the on-disk file size. If we update the
831 * on-disk file size and then the system crashes before the
832 * contents of the file are flushed to disk then the files
833 * may be full of holes (ie NULL files bug).
835 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
836 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
839 * If we get an error at this point we simply don't
840 * bother truncating the file.
842 xfs_trans_cancel(tp);
844 error = xfs_trans_commit(tp);
846 xfs_inode_clear_eofblocks_tag(ip);
849 xfs_iunlock(ip, XFS_ILOCK_EXCL);
855 xfs_alloc_file_space(
856 struct xfs_inode *ip,
861 xfs_mount_t *mp = ip->i_mount;
863 xfs_filblks_t allocated_fsb;
864 xfs_filblks_t allocatesize_fsb;
865 xfs_extlen_t extsz, temp;
866 xfs_fileoff_t startoffset_fsb;
867 xfs_fileoff_t endoffset_fsb;
872 xfs_bmbt_irec_t imaps[1], *imapp;
873 uint qblocks, resblks, resrtextents;
876 trace_xfs_alloc_file_space(ip);
878 if (XFS_FORCED_SHUTDOWN(mp))
881 error = xfs_qm_dqattach(ip);
888 rt = XFS_IS_REALTIME_INODE(ip);
889 extsz = xfs_get_extsz_hint(ip);
894 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
895 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
896 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
899 * Allocate file space until done or until there is an error
901 while (allocatesize_fsb && !error) {
905 * Determine space reservations for data/realtime.
907 if (unlikely(extsz)) {
911 e = startoffset_fsb + allocatesize_fsb;
912 div_u64_rem(startoffset_fsb, extsz, &temp);
915 div_u64_rem(e, extsz, &temp);
920 e = allocatesize_fsb;
924 * The transaction reservation is limited to a 32-bit block
925 * count, hence we need to limit the number of blocks we are
926 * trying to reserve to avoid an overflow. We can't allocate
927 * more than @nimaps extents, and an extent is limited on disk
928 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
930 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
932 resrtextents = qblocks = resblks;
933 resrtextents /= mp->m_sb.sb_rextsize;
934 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
935 quota_flag = XFS_QMOPT_RES_RTBLKS;
938 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
939 quota_flag = XFS_QMOPT_RES_REGBLKS;
943 * Allocate and setup the transaction.
945 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks,
946 resrtextents, 0, &tp);
949 * Check for running out of space
953 * Free the transaction structure.
955 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
958 xfs_ilock(ip, XFS_ILOCK_EXCL);
959 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
964 xfs_trans_ijoin(tp, ip, 0);
966 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
967 allocatesize_fsb, alloc_type, resblks,
973 * Complete the transaction
975 error = xfs_trans_commit(tp);
976 xfs_iunlock(ip, XFS_ILOCK_EXCL);
980 allocated_fsb = imapp->br_blockcount;
987 startoffset_fsb += allocated_fsb;
988 allocatesize_fsb -= allocated_fsb;
993 error0: /* unlock inode, unreserve quota blocks, cancel trans */
994 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
996 error1: /* Just cancel transaction */
997 xfs_trans_cancel(tp);
998 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1004 struct xfs_inode *ip,
1005 xfs_fileoff_t startoffset_fsb,
1006 xfs_filblks_t len_fsb,
1009 struct xfs_mount *mp = ip->i_mount;
1010 struct xfs_trans *tp;
1011 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1014 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
1016 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1020 xfs_ilock(ip, XFS_ILOCK_EXCL);
1021 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot,
1022 ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS);
1024 goto out_trans_cancel;
1026 xfs_trans_ijoin(tp, ip, 0);
1028 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
1030 goto out_trans_cancel;
1032 error = xfs_trans_commit(tp);
1034 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1038 xfs_trans_cancel(tp);
1042 /* Caller must first wait for the completion of any pending DIOs if required. */
1044 xfs_flush_unmap_range(
1045 struct xfs_inode *ip,
1049 struct xfs_mount *mp = ip->i_mount;
1050 struct inode *inode = VFS_I(ip);
1051 xfs_off_t rounding, start, end;
1054 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
1055 start = round_down(offset, rounding);
1056 end = round_up(offset + len, rounding) - 1;
1058 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
1061 truncate_pagecache_range(inode, start, end);
1066 xfs_free_file_space(
1067 struct xfs_inode *ip,
1071 struct xfs_mount *mp = ip->i_mount;
1072 xfs_fileoff_t startoffset_fsb;
1073 xfs_fileoff_t endoffset_fsb;
1074 int done = 0, error;
1076 trace_xfs_free_file_space(ip);
1078 error = xfs_qm_dqattach(ip);
1082 if (len <= 0) /* if nothing being freed */
1085 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1086 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1089 * Need to zero the stuff we're not freeing, on disk.
1091 if (endoffset_fsb > startoffset_fsb) {
1093 error = xfs_unmap_extent(ip, startoffset_fsb,
1094 endoffset_fsb - startoffset_fsb, &done);
1101 * Now that we've unmap all full blocks we'll have to zero out any
1102 * partial block at the beginning and/or end. iomap_zero_range is smart
1103 * enough to skip any holes, including those we just created, but we
1104 * must take care not to zero beyond EOF and enlarge i_size.
1106 if (offset >= XFS_ISIZE(ip))
1108 if (offset + len > XFS_ISIZE(ip))
1109 len = XFS_ISIZE(ip) - offset;
1110 error = iomap_zero_range(VFS_I(ip), offset, len, NULL, &xfs_iomap_ops);
1115 * If we zeroed right up to EOF and EOF straddles a page boundary we
1116 * must make sure that the post-EOF area is also zeroed because the
1117 * page could be mmap'd and iomap_zero_range doesn't do that for us.
1118 * Writeback of the eof page will do this, albeit clumsily.
1120 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
1121 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1122 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
1129 * Preallocate and zero a range of a file. This mechanism has the allocation
1130 * semantics of fallocate and in addition converts data in the range to zeroes.
1133 xfs_zero_file_space(
1134 struct xfs_inode *ip,
1138 struct xfs_mount *mp = ip->i_mount;
1142 trace_xfs_zero_file_space(ip);
1144 blksize = 1 << mp->m_sb.sb_blocklog;
1147 * Punch a hole and prealloc the range. We use hole punch rather than
1148 * unwritten extent conversion for two reasons:
1150 * 1.) Hole punch handles partial block zeroing for us.
1152 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued
1153 * by virtue of the hole punch.
1155 error = xfs_free_file_space(ip, offset, len);
1156 if (error || xfs_is_always_cow_inode(ip))
1159 return xfs_alloc_file_space(ip, round_down(offset, blksize),
1160 round_up(offset + len, blksize) -
1161 round_down(offset, blksize),
1162 XFS_BMAPI_PREALLOC);
1167 struct xfs_inode *ip,
1170 struct xfs_mount *mp = ip->i_mount;
1174 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1175 * into the accessible region of the file.
1177 if (xfs_can_free_eofblocks(ip, true)) {
1178 error = xfs_free_eofblocks(ip);
1184 * Shift operations must stabilize the start block offset boundary along
1185 * with the full range of the operation. If we don't, a COW writeback
1186 * completion could race with an insert, front merge with the start
1187 * extent (after split) during the shift and corrupt the file. Start
1188 * with the block just prior to the start to stabilize the boundary.
1190 offset = round_down(offset, 1 << mp->m_sb.sb_blocklog);
1192 offset -= (1 << mp->m_sb.sb_blocklog);
1195 * Writeback and invalidate cache for the remainder of the file as we're
1196 * about to shift down every extent from offset to EOF.
1198 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1203 * Clean out anything hanging around in the cow fork now that
1204 * we've flushed all the dirty data out to disk to avoid having
1205 * CoW extents at the wrong offsets.
1207 if (xfs_inode_has_cow_data(ip)) {
1208 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1218 * xfs_collapse_file_space()
1219 * This routine frees disk space and shift extent for the given file.
1220 * The first thing we do is to free data blocks in the specified range
1221 * by calling xfs_free_file_space(). It would also sync dirty data
1222 * and invalidate page cache over the region on which collapse range
1223 * is working. And Shift extent records to the left to cover a hole.
1230 xfs_collapse_file_space(
1231 struct xfs_inode *ip,
1235 struct xfs_mount *mp = ip->i_mount;
1236 struct xfs_trans *tp;
1238 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
1239 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1242 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1243 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1245 trace_xfs_collapse_file_space(ip);
1247 error = xfs_free_file_space(ip, offset, len);
1251 error = xfs_prepare_shift(ip, offset);
1255 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1259 xfs_ilock(ip, XFS_ILOCK_EXCL);
1260 xfs_trans_ijoin(tp, ip, 0);
1263 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1266 goto out_trans_cancel;
1270 /* finish any deferred frees and roll the transaction */
1271 error = xfs_defer_finish(&tp);
1273 goto out_trans_cancel;
1276 error = xfs_trans_commit(tp);
1277 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1281 xfs_trans_cancel(tp);
1282 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1287 * xfs_insert_file_space()
1288 * This routine create hole space by shifting extents for the given file.
1289 * The first thing we do is to sync dirty data and invalidate page cache
1290 * over the region on which insert range is working. And split an extent
1291 * to two extents at given offset by calling xfs_bmap_split_extent.
1292 * And shift all extent records which are laying between [offset,
1293 * last allocated extent] to the right to reserve hole range.
1299 xfs_insert_file_space(
1300 struct xfs_inode *ip,
1304 struct xfs_mount *mp = ip->i_mount;
1305 struct xfs_trans *tp;
1307 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1308 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1309 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1312 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1313 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1315 trace_xfs_insert_file_space(ip);
1317 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1321 error = xfs_prepare_shift(ip, offset);
1325 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1326 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1330 xfs_ilock(ip, XFS_ILOCK_EXCL);
1331 xfs_trans_ijoin(tp, ip, 0);
1334 * The extent shifting code works on extent granularity. So, if stop_fsb
1335 * is not the starting block of extent, we need to split the extent at
1338 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1340 goto out_trans_cancel;
1343 error = xfs_defer_finish(&tp);
1345 goto out_trans_cancel;
1347 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1350 goto out_trans_cancel;
1353 error = xfs_trans_commit(tp);
1354 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1358 xfs_trans_cancel(tp);
1359 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1364 * We need to check that the format of the data fork in the temporary inode is
1365 * valid for the target inode before doing the swap. This is not a problem with
1366 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1367 * data fork depending on the space the attribute fork is taking so we can get
1368 * invalid formats on the target inode.
1370 * E.g. target has space for 7 extents in extent format, temp inode only has
1371 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1372 * btree, but when swapped it needs to be in extent format. Hence we can't just
1373 * blindly swap data forks on attr2 filesystems.
1375 * Note that we check the swap in both directions so that we don't end up with
1376 * a corrupt temporary inode, either.
1378 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1379 * inode will prevent this situation from occurring, so all we do here is
1380 * reject and log the attempt. basically we are putting the responsibility on
1381 * userspace to get this right.
1384 xfs_swap_extents_check_format(
1385 struct xfs_inode *ip, /* target inode */
1386 struct xfs_inode *tip) /* tmp inode */
1389 /* Should never get a local format */
1390 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1391 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1395 * if the target inode has less extents that then temporary inode then
1396 * why did userspace call us?
1398 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1402 * If we have to use the (expensive) rmap swap method, we can
1403 * handle any number of extents and any format.
1405 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1409 * if the target inode is in extent form and the temp inode is in btree
1410 * form then we will end up with the target inode in the wrong format
1411 * as we already know there are less extents in the temp inode.
1413 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1414 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1417 /* Check temp in extent form to max in target */
1418 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1419 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1420 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1423 /* Check target in extent form to max in temp */
1424 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1425 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1426 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1430 * If we are in a btree format, check that the temp root block will fit
1431 * in the target and that it has enough extents to be in btree format
1434 * Note that we have to be careful to allow btree->extent conversions
1435 * (a common defrag case) which will occur when the temp inode is in
1438 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1439 if (XFS_IFORK_Q(ip) &&
1440 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1442 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1443 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1447 /* Reciprocal target->temp btree format checks */
1448 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1449 if (XFS_IFORK_Q(tip) &&
1450 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1452 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1453 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1461 xfs_swap_extent_flush(
1462 struct xfs_inode *ip)
1466 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1469 truncate_pagecache_range(VFS_I(ip), 0, -1);
1471 /* Verify O_DIRECT for ftmp */
1472 if (VFS_I(ip)->i_mapping->nrpages)
1478 * Move extents from one file to another, when rmap is enabled.
1481 xfs_swap_extent_rmap(
1482 struct xfs_trans **tpp,
1483 struct xfs_inode *ip,
1484 struct xfs_inode *tip)
1486 struct xfs_trans *tp = *tpp;
1487 struct xfs_bmbt_irec irec;
1488 struct xfs_bmbt_irec uirec;
1489 struct xfs_bmbt_irec tirec;
1490 xfs_fileoff_t offset_fsb;
1491 xfs_fileoff_t end_fsb;
1492 xfs_filblks_t count_fsb;
1497 uint64_t tip_flags2;
1500 * If the source file has shared blocks, we must flag the donor
1501 * file as having shared blocks so that we get the shared-block
1502 * rmap functions when we go to fix up the rmaps. The flags
1503 * will be switch for reals later.
1505 tip_flags2 = tip->i_d.di_flags2;
1506 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1507 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1510 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1511 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1514 /* Read extent from the donor file */
1516 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1520 ASSERT(nimaps == 1);
1521 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1523 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1524 ilen = tirec.br_blockcount;
1526 /* Unmap the old blocks in the source file. */
1527 while (tirec.br_blockcount) {
1528 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1529 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1531 /* Read extent from the source file */
1533 error = xfs_bmapi_read(ip, tirec.br_startoff,
1534 tirec.br_blockcount, &irec,
1538 ASSERT(nimaps == 1);
1539 ASSERT(tirec.br_startoff == irec.br_startoff);
1540 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1542 /* Trim the extent. */
1544 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1545 tirec.br_blockcount,
1546 irec.br_blockcount);
1547 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1549 /* Remove the mapping from the donor file. */
1550 xfs_bmap_unmap_extent(tp, tip, &uirec);
1552 /* Remove the mapping from the source file. */
1553 xfs_bmap_unmap_extent(tp, ip, &irec);
1555 /* Map the donor file's blocks into the source file. */
1556 xfs_bmap_map_extent(tp, ip, &uirec);
1558 /* Map the source file's blocks into the donor file. */
1559 xfs_bmap_map_extent(tp, tip, &irec);
1561 error = xfs_defer_finish(tpp);
1566 tirec.br_startoff += rlen;
1567 if (tirec.br_startblock != HOLESTARTBLOCK &&
1568 tirec.br_startblock != DELAYSTARTBLOCK)
1569 tirec.br_startblock += rlen;
1570 tirec.br_blockcount -= rlen;
1578 tip->i_d.di_flags2 = tip_flags2;
1582 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1583 tip->i_d.di_flags2 = tip_flags2;
1587 /* Swap the extents of two files by swapping data forks. */
1589 xfs_swap_extent_forks(
1590 struct xfs_trans *tp,
1591 struct xfs_inode *ip,
1592 struct xfs_inode *tip,
1594 int *target_log_flags)
1596 xfs_filblks_t aforkblks = 0;
1597 xfs_filblks_t taforkblks = 0;
1603 * Count the number of extended attribute blocks
1605 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1606 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1607 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1612 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1613 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1614 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1621 * Btree format (v3) inodes have the inode number stamped in the bmbt
1622 * block headers. We can't start changing the bmbt blocks until the
1623 * inode owner change is logged so recovery does the right thing in the
1624 * event of a crash. Set the owner change log flags now and leave the
1625 * bmbt scan as the last step.
1627 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
1628 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1629 (*target_log_flags) |= XFS_ILOG_DOWNER;
1630 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1631 (*src_log_flags) |= XFS_ILOG_DOWNER;
1635 * Swap the data forks of the inodes
1637 swap(ip->i_df, tip->i_df);
1640 * Fix the on-disk inode values
1642 tmp = (uint64_t)ip->i_d.di_nblocks;
1643 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1644 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1646 swap(ip->i_d.di_nextents, tip->i_d.di_nextents);
1647 swap(ip->i_d.di_format, tip->i_d.di_format);
1650 * The extents in the source inode could still contain speculative
1651 * preallocation beyond EOF (e.g. the file is open but not modified
1652 * while defrag is in progress). In that case, we need to copy over the
1653 * number of delalloc blocks the data fork in the source inode is
1654 * tracking beyond EOF so that when the fork is truncated away when the
1655 * temporary inode is unlinked we don't underrun the i_delayed_blks
1656 * counter on that inode.
1658 ASSERT(tip->i_delayed_blks == 0);
1659 tip->i_delayed_blks = ip->i_delayed_blks;
1660 ip->i_delayed_blks = 0;
1662 switch (ip->i_d.di_format) {
1663 case XFS_DINODE_FMT_EXTENTS:
1664 (*src_log_flags) |= XFS_ILOG_DEXT;
1666 case XFS_DINODE_FMT_BTREE:
1667 ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1668 (*src_log_flags & XFS_ILOG_DOWNER));
1669 (*src_log_flags) |= XFS_ILOG_DBROOT;
1673 switch (tip->i_d.di_format) {
1674 case XFS_DINODE_FMT_EXTENTS:
1675 (*target_log_flags) |= XFS_ILOG_DEXT;
1677 case XFS_DINODE_FMT_BTREE:
1678 (*target_log_flags) |= XFS_ILOG_DBROOT;
1679 ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1680 (*target_log_flags & XFS_ILOG_DOWNER));
1688 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1689 * change owner scan attempts to order all modified buffers in the current
1690 * transaction. In the event of ordered buffer failure, the offending buffer is
1691 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1692 * the transaction in this case to replenish the fallback log reservation and
1693 * restart the scan. This process repeats until the scan completes.
1696 xfs_swap_change_owner(
1697 struct xfs_trans **tpp,
1698 struct xfs_inode *ip,
1699 struct xfs_inode *tmpip)
1702 struct xfs_trans *tp = *tpp;
1705 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1707 /* success or fatal error */
1708 if (error != -EAGAIN)
1711 error = xfs_trans_roll(tpp);
1717 * Redirty both inodes so they can relog and keep the log tail
1720 xfs_trans_ijoin(tp, ip, 0);
1721 xfs_trans_ijoin(tp, tmpip, 0);
1722 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1723 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1731 struct xfs_inode *ip, /* target inode */
1732 struct xfs_inode *tip, /* tmp inode */
1733 struct xfs_swapext *sxp)
1735 struct xfs_mount *mp = ip->i_mount;
1736 struct xfs_trans *tp;
1737 struct xfs_bstat *sbp = &sxp->sx_stat;
1738 int src_log_flags, target_log_flags;
1743 unsigned int flags = 0;
1746 * Lock the inodes against other IO, page faults and truncate to
1747 * begin with. Then we can ensure the inodes are flushed and have no
1748 * page cache safely. Once we have done this we can take the ilocks and
1749 * do the rest of the checks.
1751 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1752 lock_flags = XFS_MMAPLOCK_EXCL;
1753 xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1755 /* Verify that both files have the same format */
1756 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1761 /* Verify both files are either real-time or non-realtime */
1762 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1767 error = xfs_swap_extent_flush(ip);
1770 error = xfs_swap_extent_flush(tip);
1774 if (xfs_inode_has_cow_data(tip)) {
1775 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1781 * Extent "swapping" with rmap requires a permanent reservation and
1782 * a block reservation because it's really just a remap operation
1783 * performed with log redo items!
1785 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1786 int w = XFS_DATA_FORK;
1787 uint32_t ipnext = XFS_IFORK_NEXTENTS(ip, w);
1788 uint32_t tipnext = XFS_IFORK_NEXTENTS(tip, w);
1791 * Conceptually this shouldn't affect the shape of either bmbt,
1792 * but since we atomically move extents one by one, we reserve
1793 * enough space to rebuild both trees.
1795 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1796 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1799 * If either inode straddles a bmapbt block allocation boundary,
1800 * the rmapbt algorithm triggers repeated allocs and frees as
1801 * extents are remapped. This can exhaust the block reservation
1802 * prematurely and cause shutdown. Return freed blocks to the
1803 * transaction reservation to counter this behavior.
1805 flags |= XFS_TRANS_RES_FDBLKS;
1807 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1813 * Lock and join the inodes to the tansaction so that transaction commit
1814 * or cancel will unlock the inodes from this point onwards.
1816 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1817 lock_flags |= XFS_ILOCK_EXCL;
1818 xfs_trans_ijoin(tp, ip, 0);
1819 xfs_trans_ijoin(tp, tip, 0);
1822 /* Verify all data are being swapped */
1823 if (sxp->sx_offset != 0 ||
1824 sxp->sx_length != ip->i_d.di_size ||
1825 sxp->sx_length != tip->i_d.di_size) {
1827 goto out_trans_cancel;
1830 trace_xfs_swap_extent_before(ip, 0);
1831 trace_xfs_swap_extent_before(tip, 1);
1833 /* check inode formats now that data is flushed */
1834 error = xfs_swap_extents_check_format(ip, tip);
1837 "%s: inode 0x%llx format is incompatible for exchanging.",
1838 __func__, ip->i_ino);
1839 goto out_trans_cancel;
1843 * Compare the current change & modify times with that
1844 * passed in. If they differ, we abort this swap.
1845 * This is the mechanism used to ensure the calling
1846 * process that the file was not changed out from
1849 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1850 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1851 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1852 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1854 goto out_trans_cancel;
1858 * Note the trickiness in setting the log flags - we set the owner log
1859 * flag on the opposite inode (i.e. the inode we are setting the new
1860 * owner to be) because once we swap the forks and log that, log
1861 * recovery is going to see the fork as owned by the swapped inode,
1862 * not the pre-swapped inodes.
1864 src_log_flags = XFS_ILOG_CORE;
1865 target_log_flags = XFS_ILOG_CORE;
1867 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1868 error = xfs_swap_extent_rmap(&tp, ip, tip);
1870 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1873 goto out_trans_cancel;
1875 /* Do we have to swap reflink flags? */
1876 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1877 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1878 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1879 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1880 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1881 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1882 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1885 /* Swap the cow forks. */
1886 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1887 ASSERT(ip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1888 ASSERT(tip->i_cformat == XFS_DINODE_FMT_EXTENTS);
1890 swap(ip->i_cnextents, tip->i_cnextents);
1891 swap(ip->i_cowfp, tip->i_cowfp);
1893 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1894 xfs_inode_set_cowblocks_tag(ip);
1896 xfs_inode_clear_cowblocks_tag(ip);
1897 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1898 xfs_inode_set_cowblocks_tag(tip);
1900 xfs_inode_clear_cowblocks_tag(tip);
1903 xfs_trans_log_inode(tp, ip, src_log_flags);
1904 xfs_trans_log_inode(tp, tip, target_log_flags);
1907 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1908 * have inode number owner values in the bmbt blocks that still refer to
1909 * the old inode. Scan each bmbt to fix up the owner values with the
1910 * inode number of the current inode.
1912 if (src_log_flags & XFS_ILOG_DOWNER) {
1913 error = xfs_swap_change_owner(&tp, ip, tip);
1915 goto out_trans_cancel;
1917 if (target_log_flags & XFS_ILOG_DOWNER) {
1918 error = xfs_swap_change_owner(&tp, tip, ip);
1920 goto out_trans_cancel;
1924 * If this is a synchronous mount, make sure that the
1925 * transaction goes to disk before returning to the user.
1927 if (mp->m_flags & XFS_MOUNT_WSYNC)
1928 xfs_trans_set_sync(tp);
1930 error = xfs_trans_commit(tp);
1932 trace_xfs_swap_extent_after(ip, 0);
1933 trace_xfs_swap_extent_after(tip, 1);
1936 xfs_iunlock(ip, lock_flags);
1937 xfs_iunlock(tip, lock_flags);
1938 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1942 xfs_trans_cancel(tp);