2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_inode.h"
26 #include "xfs_btree.h"
27 #include "xfs_bmap_btree.h"
29 #include "xfs_bmap_util.h"
30 #include "xfs_error.h"
31 #include "xfs_trans.h"
32 #include "xfs_trans_space.h"
33 #include "xfs_iomap.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_quota.h"
37 #include "xfs_dquot_item.h"
38 #include "xfs_dquot.h"
41 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
43 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
46 xfs_iomap_eof_align_last_fsb(
50 xfs_fileoff_t *last_fsb)
52 xfs_extlen_t align = 0;
55 if (!XFS_IS_REALTIME_INODE(ip)) {
57 * Round up the allocation request to a stripe unit
58 * (m_dalign) boundary if the file size is >= stripe unit
59 * size, and we are allocating past the allocation eof.
61 * If mounted with the "-o swalloc" option the alignment is
62 * increased from the strip unit size to the stripe width.
64 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
66 else if (mp->m_dalign)
69 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
74 * Always round up the allocation request to an extent boundary
75 * (when file on a real-time subvolume or has di_extsize hint).
79 align = roundup_64(align, extsize);
85 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
86 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
90 *last_fsb = new_last_fsb;
96 xfs_alert_fsblock_zero(
98 xfs_bmbt_irec_t *imap)
100 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
101 "Access to block zero in inode %llu "
102 "start_block: %llx start_off: %llx "
103 "blkcnt: %llx extent-state: %x",
104 (unsigned long long)ip->i_ino,
105 (unsigned long long)imap->br_startblock,
106 (unsigned long long)imap->br_startoff,
107 (unsigned long long)imap->br_blockcount,
109 return -EFSCORRUPTED;
113 xfs_iomap_write_direct(
117 xfs_bmbt_irec_t *imap,
120 xfs_mount_t *mp = ip->i_mount;
121 xfs_fileoff_t offset_fsb;
122 xfs_fileoff_t last_fsb;
123 xfs_filblks_t count_fsb, resaligned;
124 xfs_fsblock_t firstfsb;
125 xfs_extlen_t extsz, temp;
130 xfs_bmap_free_t free_list;
131 uint qblocks, resblks, resrtextents;
135 int bmapi_flags = XFS_BMAPI_PREALLOC;
137 rt = XFS_IS_REALTIME_INODE(ip);
138 extsz = xfs_get_extsz_hint(ip);
139 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
141 ASSERT(xfs_isilocked(ip, lockmode));
143 offset_fsb = XFS_B_TO_FSBT(mp, offset);
144 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
145 if ((offset + count) > XFS_ISIZE(ip)) {
147 * Assert that the in-core extent list is present since this can
148 * call xfs_iread_extents() and we only have the ilock shared.
149 * This should be safe because the lock was held around a bmapi
150 * call in the caller and we only need it to access the in-core
153 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
155 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
159 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
160 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
161 imap->br_blockcount +
164 count_fsb = last_fsb - offset_fsb;
165 ASSERT(count_fsb > 0);
167 resaligned = count_fsb;
168 if (unlikely(extsz)) {
169 if ((temp = do_mod(offset_fsb, extsz)))
171 if ((temp = do_mod(resaligned, extsz)))
172 resaligned += extsz - temp;
176 resrtextents = qblocks = resaligned;
177 resrtextents /= mp->m_sb.sb_rextsize;
178 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
179 quota_flag = XFS_QMOPT_RES_RTBLKS;
182 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
183 quota_flag = XFS_QMOPT_RES_REGBLKS;
187 * Drop the shared lock acquired by the caller, attach the dquot if
188 * necessary and move on to transaction setup.
190 xfs_iunlock(ip, lockmode);
191 error = xfs_qm_dqattach(ip, 0);
196 * Allocate and setup the transaction
198 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
201 * For DAX, we do not allocate unwritten extents, but instead we zero
202 * the block before we commit the transaction. Ideally we'd like to do
203 * this outside the transaction context, but if we commit and then crash
204 * we may not have zeroed the blocks and this will be exposed on
205 * recovery of the allocation. Hence we must zero before commit.
207 * Further, if we are mapping unwritten extents here, we need to zero
208 * and convert them to written so that we don't need an unwritten extent
209 * callback for DAX. This also means that we need to be able to dip into
210 * the reserve block pool for bmbt block allocation if there is no space
211 * left but we need to do unwritten extent conversion.
214 if (IS_DAX(VFS_I(ip))) {
215 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
216 if (ISUNWRITTEN(imap)) {
217 tp->t_flags |= XFS_TRANS_RESERVE;
218 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
221 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
222 resblks, resrtextents);
224 * Check for running out of space, note: need lock to return
227 xfs_trans_cancel(tp);
231 lockmode = XFS_ILOCK_EXCL;
232 xfs_ilock(ip, lockmode);
234 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
236 goto out_trans_cancel;
238 xfs_trans_ijoin(tp, ip, 0);
241 * From this point onwards we overwrite the imap pointer that the
244 xfs_bmap_init(&free_list, &firstfsb);
246 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
247 bmapi_flags, &firstfsb, resblks, imap,
248 &nimaps, &free_list);
250 goto out_bmap_cancel;
253 * Complete the transaction
255 error = xfs_bmap_finish(&tp, &free_list, &committed);
257 goto out_bmap_cancel;
259 error = xfs_trans_commit(tp);
264 * Copy any maps to caller's array and return any error.
271 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
272 error = xfs_alert_fsblock_zero(ip, imap);
275 xfs_iunlock(ip, lockmode);
279 xfs_bmap_cancel(&free_list);
280 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
282 xfs_trans_cancel(tp);
287 * If the caller is doing a write at the end of the file, then extend the
288 * allocation out to the file system's write iosize. We clean up any extra
289 * space left over when the file is closed in xfs_inactive().
291 * If we find we already have delalloc preallocation beyond EOF, don't do more
292 * preallocation as it it not needed.
295 xfs_iomap_eof_want_preallocate(
300 xfs_bmbt_irec_t *imap,
304 xfs_fileoff_t start_fsb;
305 xfs_filblks_t count_fsb;
307 int found_delalloc = 0;
310 if (offset + count <= XFS_ISIZE(ip))
314 * If the file is smaller than the minimum prealloc and we are using
315 * dynamic preallocation, don't do any preallocation at all as it is
316 * likely this is the only write to the file that is going to be done.
318 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
319 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
323 * If there are any real blocks past eof, then don't
324 * do any speculative allocation.
326 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
327 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
328 while (count_fsb > 0) {
330 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
334 for (n = 0; n < imaps; n++) {
335 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
336 (imap[n].br_startblock != DELAYSTARTBLOCK))
338 start_fsb += imap[n].br_blockcount;
339 count_fsb -= imap[n].br_blockcount;
341 if (imap[n].br_startblock == DELAYSTARTBLOCK)
351 * Determine the initial size of the preallocation. We are beyond the current
352 * EOF here, but we need to take into account whether this is a sparse write or
353 * an extending write when determining the preallocation size. Hence we need to
354 * look up the extent that ends at the current write offset and use the result
355 * to determine the preallocation size.
357 * If the extent is a hole, then preallocation is essentially disabled.
358 * Otherwise we take the size of the preceeding data extent as the basis for the
359 * preallocation size. If the size of the extent is greater than half the
360 * maximum extent length, then use the current offset as the basis. This ensures
361 * that for large files the preallocation size always extends to MAXEXTLEN
362 * rather than falling short due to things like stripe unit/width alignment of
366 xfs_iomap_eof_prealloc_initial_size(
367 struct xfs_mount *mp,
368 struct xfs_inode *ip,
370 xfs_bmbt_irec_t *imap,
373 xfs_fileoff_t start_fsb;
377 ASSERT(nimaps >= imaps);
379 /* if we are using a specific prealloc size, return now */
380 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
383 /* If the file is small, then use the minimum prealloc */
384 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
388 * As we write multiple pages, the offset will always align to the
389 * start of a page and hence point to a hole at EOF. i.e. if the size is
390 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
391 * will return FSB 1. Hence if there are blocks in the file, we want to
392 * point to the block prior to the EOF block and not the hole that maps
393 * directly at @offset.
395 start_fsb = XFS_B_TO_FSB(mp, offset);
398 error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
403 if (imap[0].br_startblock == HOLESTARTBLOCK)
405 if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
406 return imap[0].br_blockcount << 1;
407 return XFS_B_TO_FSB(mp, offset);
411 xfs_quota_need_throttle(
412 struct xfs_inode *ip,
414 xfs_fsblock_t alloc_blocks)
416 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
418 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
421 /* no hi watermark, no throttle */
422 if (!dq->q_prealloc_hi_wmark)
425 /* under the lo watermark, no throttle */
426 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
433 xfs_quota_calc_throttle(
434 struct xfs_inode *ip,
436 xfs_fsblock_t *qblocks,
442 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
444 /* no dq, or over hi wmark, squash the prealloc completely */
445 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
451 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
452 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
454 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
456 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
460 if (freesp < *qfreesp)
463 /* only overwrite the throttle values if we are more aggressive */
464 if ((freesp >> shift) < (*qblocks >> *qshift)) {
471 * If we don't have a user specified preallocation size, dynamically increase
472 * the preallocation size as the size of the file grows. Cap the maximum size
473 * at a single extent or less if the filesystem is near full. The closer the
474 * filesystem is to full, the smaller the maximum prealocation.
477 xfs_iomap_prealloc_size(
478 struct xfs_mount *mp,
479 struct xfs_inode *ip,
481 struct xfs_bmbt_irec *imap,
484 xfs_fsblock_t alloc_blocks = 0;
487 xfs_fsblock_t qblocks;
490 alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
494 qblocks = alloc_blocks;
497 * MAXEXTLEN is not a power of two value but we round the prealloc down
498 * to the nearest power of two value after throttling. To prevent the
499 * round down from unconditionally reducing the maximum supported prealloc
500 * size, we round up first, apply appropriate throttling, round down and
501 * cap the value to MAXEXTLEN.
503 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
506 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
507 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
509 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
511 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
513 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
515 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
520 * Check each quota to cap the prealloc size, provide a shift value to
521 * throttle with and adjust amount of available space.
523 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
524 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
526 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
527 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
529 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
530 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
534 * The final prealloc size is set to the minimum of free space available
535 * in each of the quotas and the overall filesystem.
537 * The shift throttle value is set to the maximum value as determined by
538 * the global low free space values and per-quota low free space values.
540 alloc_blocks = MIN(alloc_blocks, qblocks);
541 shift = MAX(shift, qshift);
544 alloc_blocks >>= shift;
546 * rounddown_pow_of_two() returns an undefined result if we pass in
550 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
551 if (alloc_blocks > MAXEXTLEN)
552 alloc_blocks = MAXEXTLEN;
555 * If we are still trying to allocate more space than is
556 * available, squash the prealloc hard. This can happen if we
557 * have a large file on a small filesystem and the above
558 * lowspace thresholds are smaller than MAXEXTLEN.
560 while (alloc_blocks && alloc_blocks >= freesp)
564 if (alloc_blocks < mp->m_writeio_blocks)
565 alloc_blocks = mp->m_writeio_blocks;
567 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
568 mp->m_writeio_blocks);
574 xfs_iomap_write_delay(
578 xfs_bmbt_irec_t *ret_imap)
580 xfs_mount_t *mp = ip->i_mount;
581 xfs_fileoff_t offset_fsb;
582 xfs_fileoff_t last_fsb;
583 xfs_off_t aligned_offset;
584 xfs_fileoff_t ioalign;
587 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
591 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
594 * Make sure that the dquots are there. This doesn't hold
595 * the ilock across a disk read.
597 error = xfs_qm_dqattach_locked(ip, 0);
601 extsz = xfs_get_extsz_hint(ip);
602 offset_fsb = XFS_B_TO_FSBT(mp, offset);
604 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
605 imap, XFS_WRITE_IMAPS, &prealloc);
611 xfs_fsblock_t alloc_blocks;
613 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
616 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
617 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
618 last_fsb = ioalign + alloc_blocks;
620 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
623 if (prealloc || extsz) {
624 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
630 * Make sure preallocation does not create extents beyond the range we
631 * actually support in this filesystem.
633 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
634 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
636 ASSERT(last_fsb > offset_fsb);
638 nimaps = XFS_WRITE_IMAPS;
639 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
640 imap, &nimaps, XFS_BMAPI_ENTIRE);
651 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
652 * without EOF preallocation.
655 trace_xfs_delalloc_enospc(ip, offset, count);
661 return error ? error : -ENOSPC;
664 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
665 return xfs_alert_fsblock_zero(ip, &imap[0]);
668 * Tag the inode as speculatively preallocated so we can reclaim this
669 * space on demand, if necessary.
672 xfs_inode_set_eofblocks_tag(ip);
679 * Pass in a delayed allocate extent, convert it to real extents;
680 * return to the caller the extent we create which maps on top of
681 * the originating callers request.
683 * Called without a lock on the inode.
685 * We no longer bother to look at the incoming map - all we have to
686 * guarantee is that whatever we allocate fills the required range.
689 xfs_iomap_write_allocate(
692 xfs_bmbt_irec_t *imap)
694 xfs_mount_t *mp = ip->i_mount;
695 xfs_fileoff_t offset_fsb, last_block;
696 xfs_fileoff_t end_fsb, map_start_fsb;
697 xfs_fsblock_t first_block;
698 xfs_bmap_free_t free_list;
699 xfs_filblks_t count_fsb;
701 int nimaps, committed;
706 * Make sure that the dquots are there.
708 error = xfs_qm_dqattach(ip, 0);
712 offset_fsb = XFS_B_TO_FSBT(mp, offset);
713 count_fsb = imap->br_blockcount;
714 map_start_fsb = imap->br_startoff;
716 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
718 while (count_fsb != 0) {
720 * Set up a transaction with which to allocate the
721 * backing store for the file. Do allocations in a
722 * loop until we get some space in the range we are
723 * interested in. The other space that might be allocated
724 * is in the delayed allocation extent on which we sit
725 * but before our buffer starts.
729 while (nimaps == 0) {
730 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
731 tp->t_flags |= XFS_TRANS_RESERVE;
732 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
733 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
736 xfs_trans_cancel(tp);
739 xfs_ilock(ip, XFS_ILOCK_EXCL);
740 xfs_trans_ijoin(tp, ip, 0);
742 xfs_bmap_init(&free_list, &first_block);
745 * it is possible that the extents have changed since
746 * we did the read call as we dropped the ilock for a
747 * while. We have to be careful about truncates or hole
748 * punchs here - we are not allowed to allocate
749 * non-delalloc blocks here.
751 * The only protection against truncation is the pages
752 * for the range we are being asked to convert are
753 * locked and hence a truncate will block on them
756 * As a result, if we go beyond the range we really
757 * need and hit an delalloc extent boundary followed by
758 * a hole while we have excess blocks in the map, we
759 * will fill the hole incorrectly and overrun the
760 * transaction reservation.
762 * Using a single map prevents this as we are forced to
763 * check each map we look for overlap with the desired
764 * range and abort as soon as we find it. Also, given
765 * that we only return a single map, having one beyond
766 * what we can return is probably a bit silly.
768 * We also need to check that we don't go beyond EOF;
769 * this is a truncate optimisation as a truncate sets
770 * the new file size before block on the pages we
771 * currently have locked under writeback. Because they
772 * are about to be tossed, we don't need to write them
776 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
777 error = xfs_bmap_last_offset(ip, &last_block,
782 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
783 if ((map_start_fsb + count_fsb) > last_block) {
784 count_fsb = last_block - map_start_fsb;
785 if (count_fsb == 0) {
792 * From this point onwards we overwrite the imap
793 * pointer that the caller gave to us.
795 error = xfs_bmapi_write(tp, ip, map_start_fsb,
796 count_fsb, 0, &first_block,
802 error = xfs_bmap_finish(&tp, &free_list, &committed);
806 error = xfs_trans_commit(tp);
810 xfs_iunlock(ip, XFS_ILOCK_EXCL);
814 * See if we were able to allocate an extent that
815 * covers at least part of the callers request
817 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
818 return xfs_alert_fsblock_zero(ip, imap);
820 if ((offset_fsb >= imap->br_startoff) &&
821 (offset_fsb < (imap->br_startoff +
822 imap->br_blockcount))) {
823 XFS_STATS_INC(mp, xs_xstrat_quick);
828 * So far we have not mapped the requested part of the
829 * file, just surrounding data, try again.
831 count_fsb -= imap->br_blockcount;
832 map_start_fsb = imap->br_startoff + imap->br_blockcount;
836 xfs_bmap_cancel(&free_list);
837 xfs_trans_cancel(tp);
839 xfs_iunlock(ip, XFS_ILOCK_EXCL);
844 xfs_iomap_write_unwritten(
849 xfs_mount_t *mp = ip->i_mount;
850 xfs_fileoff_t offset_fsb;
851 xfs_filblks_t count_fsb;
852 xfs_filblks_t numblks_fsb;
853 xfs_fsblock_t firstfsb;
856 xfs_bmbt_irec_t imap;
857 xfs_bmap_free_t free_list;
863 trace_xfs_unwritten_convert(ip, offset, count);
865 offset_fsb = XFS_B_TO_FSBT(mp, offset);
866 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
867 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
870 * Reserve enough blocks in this transaction for two complete extent
871 * btree splits. We may be converting the middle part of an unwritten
872 * extent and in this case we will insert two new extents in the btree
873 * each of which could cause a full split.
875 * This reservation amount will be used in the first call to
876 * xfs_bmbt_split() to select an AG with enough space to satisfy the
877 * rest of the operation.
879 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
883 * set up a transaction to convert the range of extents
884 * from unwritten to real. Do allocations in a loop until
885 * we have covered the range passed in.
887 * Note that we open code the transaction allocation here
888 * to pass KM_NOFS--we can't risk to recursing back into
889 * the filesystem here as we might be asked to write out
890 * the same inode that we complete here and might deadlock
893 sb_start_intwrite(mp->m_super);
894 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
895 tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
896 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
899 xfs_trans_cancel(tp);
903 xfs_ilock(ip, XFS_ILOCK_EXCL);
904 xfs_trans_ijoin(tp, ip, 0);
907 * Modify the unwritten extent state of the buffer.
909 xfs_bmap_init(&free_list, &firstfsb);
911 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
912 XFS_BMAPI_CONVERT, &firstfsb, resblks,
913 &imap, &nimaps, &free_list);
915 goto error_on_bmapi_transaction;
918 * Log the updated inode size as we go. We have to be careful
919 * to only log it up to the actual write offset if it is
920 * halfway into a block.
922 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
923 if (i_size > offset + count)
924 i_size = offset + count;
926 i_size = xfs_new_eof(ip, i_size);
928 ip->i_d.di_size = i_size;
929 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
932 error = xfs_bmap_finish(&tp, &free_list, &committed);
934 goto error_on_bmapi_transaction;
936 error = xfs_trans_commit(tp);
937 xfs_iunlock(ip, XFS_ILOCK_EXCL);
941 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
942 return xfs_alert_fsblock_zero(ip, &imap);
944 if ((numblks_fsb = imap.br_blockcount) == 0) {
946 * The numblks_fsb value should always get
947 * smaller, otherwise the loop is stuck.
949 ASSERT(imap.br_blockcount);
952 offset_fsb += numblks_fsb;
953 count_fsb -= numblks_fsb;
954 } while (count_fsb > 0);
958 error_on_bmapi_transaction:
959 xfs_bmap_cancel(&free_list);
960 xfs_trans_cancel(tp);
961 xfs_iunlock(ip, XFS_ILOCK_EXCL);