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
18 #include <linux/iomap.h>
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_bmap_btree.h"
31 #include "xfs_bmap_util.h"
32 #include "xfs_error.h"
33 #include "xfs_trans.h"
34 #include "xfs_trans_space.h"
35 #include "xfs_iomap.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_quota.h"
39 #include "xfs_dquot_item.h"
40 #include "xfs_dquot.h"
43 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
45 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
51 struct xfs_bmbt_irec *imap)
53 struct xfs_mount *mp = ip->i_mount;
55 if (imap->br_startblock == HOLESTARTBLOCK) {
56 iomap->blkno = IOMAP_NULL_BLOCK;
57 iomap->type = IOMAP_HOLE;
58 } else if (imap->br_startblock == DELAYSTARTBLOCK) {
59 iomap->blkno = IOMAP_NULL_BLOCK;
60 iomap->type = IOMAP_DELALLOC;
62 iomap->blkno = xfs_fsb_to_db(ip, imap->br_startblock);
63 if (imap->br_state == XFS_EXT_UNWRITTEN)
64 iomap->type = IOMAP_UNWRITTEN;
66 iomap->type = IOMAP_MAPPED;
68 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
69 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
70 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
78 struct xfs_mount *mp = ip->i_mount;
79 xfs_extlen_t align = 0;
81 if (!XFS_IS_REALTIME_INODE(ip)) {
83 * Round up the allocation request to a stripe unit
84 * (m_dalign) boundary if the file size is >= stripe unit
85 * size, and we are allocating past the allocation eof.
87 * If mounted with the "-o swalloc" option the alignment is
88 * increased from the strip unit size to the stripe width.
90 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
92 else if (mp->m_dalign)
95 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
100 * Always round up the allocation request to an extent boundary
101 * (when file on a real-time subvolume or has di_extsize hint).
105 align = roundup_64(align, extsize);
114 xfs_iomap_eof_align_last_fsb(
115 struct xfs_inode *ip,
116 xfs_extlen_t extsize,
117 xfs_fileoff_t *last_fsb)
119 xfs_extlen_t align = xfs_eof_alignment(ip, extsize);
122 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
125 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
129 *last_fsb = new_last_fsb;
135 xfs_alert_fsblock_zero(
137 xfs_bmbt_irec_t *imap)
139 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
140 "Access to block zero in inode %llu "
141 "start_block: %llx start_off: %llx "
142 "blkcnt: %llx extent-state: %x",
143 (unsigned long long)ip->i_ino,
144 (unsigned long long)imap->br_startblock,
145 (unsigned long long)imap->br_startoff,
146 (unsigned long long)imap->br_blockcount,
148 return -EFSCORRUPTED;
152 xfs_iomap_write_direct(
156 xfs_bmbt_irec_t *imap,
159 xfs_mount_t *mp = ip->i_mount;
160 xfs_fileoff_t offset_fsb;
161 xfs_fileoff_t last_fsb;
162 xfs_filblks_t count_fsb, resaligned;
163 xfs_fsblock_t firstfsb;
164 xfs_extlen_t extsz, temp;
169 struct xfs_defer_ops dfops;
170 uint qblocks, resblks, resrtextents;
173 int bmapi_flags = XFS_BMAPI_PREALLOC;
176 rt = XFS_IS_REALTIME_INODE(ip);
177 extsz = xfs_get_extsz_hint(ip);
178 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
180 ASSERT(xfs_isilocked(ip, lockmode));
182 offset_fsb = XFS_B_TO_FSBT(mp, offset);
183 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
184 if ((offset + count) > XFS_ISIZE(ip)) {
186 * Assert that the in-core extent list is present since this can
187 * call xfs_iread_extents() and we only have the ilock shared.
188 * This should be safe because the lock was held around a bmapi
189 * call in the caller and we only need it to access the in-core
192 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
194 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
198 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
199 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
200 imap->br_blockcount +
203 count_fsb = last_fsb - offset_fsb;
204 ASSERT(count_fsb > 0);
206 resaligned = count_fsb;
207 if (unlikely(extsz)) {
208 if ((temp = do_mod(offset_fsb, extsz)))
210 if ((temp = do_mod(resaligned, extsz)))
211 resaligned += extsz - temp;
215 resrtextents = qblocks = resaligned;
216 resrtextents /= mp->m_sb.sb_rextsize;
217 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
218 quota_flag = XFS_QMOPT_RES_RTBLKS;
221 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
222 quota_flag = XFS_QMOPT_RES_REGBLKS;
226 * Drop the shared lock acquired by the caller, attach the dquot if
227 * necessary and move on to transaction setup.
229 xfs_iunlock(ip, lockmode);
230 error = xfs_qm_dqattach(ip, 0);
235 * For DAX, we do not allocate unwritten extents, but instead we zero
236 * the block before we commit the transaction. Ideally we'd like to do
237 * this outside the transaction context, but if we commit and then crash
238 * we may not have zeroed the blocks and this will be exposed on
239 * recovery of the allocation. Hence we must zero before commit.
241 * Further, if we are mapping unwritten extents here, we need to zero
242 * and convert them to written so that we don't need an unwritten extent
243 * callback for DAX. This also means that we need to be able to dip into
244 * the reserve block pool for bmbt block allocation if there is no space
245 * left but we need to do unwritten extent conversion.
247 if (IS_DAX(VFS_I(ip))) {
248 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
249 if (ISUNWRITTEN(imap)) {
250 tflags |= XFS_TRANS_RESERVE;
251 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
254 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
259 lockmode = XFS_ILOCK_EXCL;
260 xfs_ilock(ip, lockmode);
262 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
264 goto out_trans_cancel;
266 xfs_trans_ijoin(tp, ip, 0);
269 * From this point onwards we overwrite the imap pointer that the
272 xfs_defer_init(&dfops, &firstfsb);
274 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
275 bmapi_flags, &firstfsb, resblks, imap,
278 goto out_bmap_cancel;
281 * Complete the transaction
283 error = xfs_defer_finish(&tp, &dfops, NULL);
285 goto out_bmap_cancel;
287 error = xfs_trans_commit(tp);
292 * Copy any maps to caller's array and return any error.
299 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
300 error = xfs_alert_fsblock_zero(ip, imap);
303 xfs_iunlock(ip, lockmode);
307 xfs_defer_cancel(&dfops);
308 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
310 xfs_trans_cancel(tp);
315 * If the caller is doing a write at the end of the file, then extend the
316 * allocation out to the file system's write iosize. We clean up any extra
317 * space left over when the file is closed in xfs_inactive().
319 * If we find we already have delalloc preallocation beyond EOF, don't do more
320 * preallocation as it it not needed.
323 xfs_iomap_eof_want_preallocate(
328 xfs_bmbt_irec_t *imap,
332 xfs_fileoff_t start_fsb;
333 xfs_filblks_t count_fsb;
335 int found_delalloc = 0;
338 if (offset + count <= XFS_ISIZE(ip))
342 * If the file is smaller than the minimum prealloc and we are using
343 * dynamic preallocation, don't do any preallocation at all as it is
344 * likely this is the only write to the file that is going to be done.
346 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
347 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
351 * If there are any real blocks past eof, then don't
352 * do any speculative allocation.
354 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
355 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
356 while (count_fsb > 0) {
358 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
362 for (n = 0; n < imaps; n++) {
363 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
364 (imap[n].br_startblock != DELAYSTARTBLOCK))
366 start_fsb += imap[n].br_blockcount;
367 count_fsb -= imap[n].br_blockcount;
369 if (imap[n].br_startblock == DELAYSTARTBLOCK)
379 * Determine the initial size of the preallocation. We are beyond the current
380 * EOF here, but we need to take into account whether this is a sparse write or
381 * an extending write when determining the preallocation size. Hence we need to
382 * look up the extent that ends at the current write offset and use the result
383 * to determine the preallocation size.
385 * If the extent is a hole, then preallocation is essentially disabled.
386 * Otherwise we take the size of the preceeding data extent as the basis for the
387 * preallocation size. If the size of the extent is greater than half the
388 * maximum extent length, then use the current offset as the basis. This ensures
389 * that for large files the preallocation size always extends to MAXEXTLEN
390 * rather than falling short due to things like stripe unit/width alignment of
394 xfs_iomap_eof_prealloc_initial_size(
395 struct xfs_mount *mp,
396 struct xfs_inode *ip,
398 xfs_bmbt_irec_t *imap,
401 xfs_fileoff_t start_fsb;
405 ASSERT(nimaps >= imaps);
407 /* if we are using a specific prealloc size, return now */
408 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
411 /* If the file is small, then use the minimum prealloc */
412 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
416 * As we write multiple pages, the offset will always align to the
417 * start of a page and hence point to a hole at EOF. i.e. if the size is
418 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
419 * will return FSB 1. Hence if there are blocks in the file, we want to
420 * point to the block prior to the EOF block and not the hole that maps
421 * directly at @offset.
423 start_fsb = XFS_B_TO_FSB(mp, offset);
426 error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
431 if (imap[0].br_startblock == HOLESTARTBLOCK)
433 if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
434 return imap[0].br_blockcount << 1;
435 return XFS_B_TO_FSB(mp, offset);
439 xfs_quota_need_throttle(
440 struct xfs_inode *ip,
442 xfs_fsblock_t alloc_blocks)
444 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
446 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
449 /* no hi watermark, no throttle */
450 if (!dq->q_prealloc_hi_wmark)
453 /* under the lo watermark, no throttle */
454 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
461 xfs_quota_calc_throttle(
462 struct xfs_inode *ip,
464 xfs_fsblock_t *qblocks,
470 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
472 /* no dq, or over hi wmark, squash the prealloc completely */
473 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
479 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
480 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
482 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
484 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
488 if (freesp < *qfreesp)
491 /* only overwrite the throttle values if we are more aggressive */
492 if ((freesp >> shift) < (*qblocks >> *qshift)) {
499 * If we don't have a user specified preallocation size, dynamically increase
500 * the preallocation size as the size of the file grows. Cap the maximum size
501 * at a single extent or less if the filesystem is near full. The closer the
502 * filesystem is to full, the smaller the maximum prealocation.
505 xfs_iomap_prealloc_size(
506 struct xfs_mount *mp,
507 struct xfs_inode *ip,
509 struct xfs_bmbt_irec *imap,
512 xfs_fsblock_t alloc_blocks = 0;
515 xfs_fsblock_t qblocks;
518 alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
522 qblocks = alloc_blocks;
525 * MAXEXTLEN is not a power of two value but we round the prealloc down
526 * to the nearest power of two value after throttling. To prevent the
527 * round down from unconditionally reducing the maximum supported prealloc
528 * size, we round up first, apply appropriate throttling, round down and
529 * cap the value to MAXEXTLEN.
531 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
534 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
535 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
537 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
539 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
541 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
543 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
548 * Check each quota to cap the prealloc size, provide a shift value to
549 * throttle with and adjust amount of available space.
551 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
552 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
554 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
555 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
557 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
558 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
562 * The final prealloc size is set to the minimum of free space available
563 * in each of the quotas and the overall filesystem.
565 * The shift throttle value is set to the maximum value as determined by
566 * the global low free space values and per-quota low free space values.
568 alloc_blocks = MIN(alloc_blocks, qblocks);
569 shift = MAX(shift, qshift);
572 alloc_blocks >>= shift;
574 * rounddown_pow_of_two() returns an undefined result if we pass in
578 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
579 if (alloc_blocks > MAXEXTLEN)
580 alloc_blocks = MAXEXTLEN;
583 * If we are still trying to allocate more space than is
584 * available, squash the prealloc hard. This can happen if we
585 * have a large file on a small filesystem and the above
586 * lowspace thresholds are smaller than MAXEXTLEN.
588 while (alloc_blocks && alloc_blocks >= freesp)
592 if (alloc_blocks < mp->m_writeio_blocks)
593 alloc_blocks = mp->m_writeio_blocks;
595 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
596 mp->m_writeio_blocks);
602 xfs_iomap_write_delay(
606 xfs_bmbt_irec_t *ret_imap)
608 xfs_mount_t *mp = ip->i_mount;
609 xfs_fileoff_t offset_fsb;
610 xfs_fileoff_t last_fsb;
611 xfs_off_t aligned_offset;
612 xfs_fileoff_t ioalign;
615 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
619 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
622 * Make sure that the dquots are there. This doesn't hold
623 * the ilock across a disk read.
625 error = xfs_qm_dqattach_locked(ip, 0);
629 extsz = xfs_get_extsz_hint(ip);
630 offset_fsb = XFS_B_TO_FSBT(mp, offset);
632 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
633 imap, XFS_WRITE_IMAPS, &prealloc);
639 xfs_fsblock_t alloc_blocks;
641 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
644 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
645 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
646 last_fsb = ioalign + alloc_blocks;
648 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
651 if (prealloc || extsz) {
652 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
658 * Make sure preallocation does not create extents beyond the range we
659 * actually support in this filesystem.
661 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
662 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
664 ASSERT(last_fsb > offset_fsb);
666 nimaps = XFS_WRITE_IMAPS;
667 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
668 imap, &nimaps, XFS_BMAPI_ENTIRE);
679 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
680 * without EOF preallocation.
683 trace_xfs_delalloc_enospc(ip, offset, count);
689 return error ? error : -ENOSPC;
692 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
693 return xfs_alert_fsblock_zero(ip, &imap[0]);
696 * Tag the inode as speculatively preallocated so we can reclaim this
697 * space on demand, if necessary.
700 xfs_inode_set_eofblocks_tag(ip);
707 * Pass in a delayed allocate extent, convert it to real extents;
708 * return to the caller the extent we create which maps on top of
709 * the originating callers request.
711 * Called without a lock on the inode.
713 * We no longer bother to look at the incoming map - all we have to
714 * guarantee is that whatever we allocate fills the required range.
717 xfs_iomap_write_allocate(
720 xfs_bmbt_irec_t *imap)
722 xfs_mount_t *mp = ip->i_mount;
723 xfs_fileoff_t offset_fsb, last_block;
724 xfs_fileoff_t end_fsb, map_start_fsb;
725 xfs_fsblock_t first_block;
726 struct xfs_defer_ops dfops;
727 xfs_filblks_t count_fsb;
734 * Make sure that the dquots are there.
736 error = xfs_qm_dqattach(ip, 0);
740 offset_fsb = XFS_B_TO_FSBT(mp, offset);
741 count_fsb = imap->br_blockcount;
742 map_start_fsb = imap->br_startoff;
744 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
746 while (count_fsb != 0) {
748 * Set up a transaction with which to allocate the
749 * backing store for the file. Do allocations in a
750 * loop until we get some space in the range we are
751 * interested in. The other space that might be allocated
752 * is in the delayed allocation extent on which we sit
753 * but before our buffer starts.
756 while (nimaps == 0) {
757 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
759 * We have already reserved space for the extent and any
760 * indirect blocks when creating the delalloc extent,
761 * there is no need to reserve space in this transaction
764 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
765 0, XFS_TRANS_RESERVE, &tp);
769 xfs_ilock(ip, XFS_ILOCK_EXCL);
770 xfs_trans_ijoin(tp, ip, 0);
772 xfs_defer_init(&dfops, &first_block);
775 * it is possible that the extents have changed since
776 * we did the read call as we dropped the ilock for a
777 * while. We have to be careful about truncates or hole
778 * punchs here - we are not allowed to allocate
779 * non-delalloc blocks here.
781 * The only protection against truncation is the pages
782 * for the range we are being asked to convert are
783 * locked and hence a truncate will block on them
786 * As a result, if we go beyond the range we really
787 * need and hit an delalloc extent boundary followed by
788 * a hole while we have excess blocks in the map, we
789 * will fill the hole incorrectly and overrun the
790 * transaction reservation.
792 * Using a single map prevents this as we are forced to
793 * check each map we look for overlap with the desired
794 * range and abort as soon as we find it. Also, given
795 * that we only return a single map, having one beyond
796 * what we can return is probably a bit silly.
798 * We also need to check that we don't go beyond EOF;
799 * this is a truncate optimisation as a truncate sets
800 * the new file size before block on the pages we
801 * currently have locked under writeback. Because they
802 * are about to be tossed, we don't need to write them
806 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
807 error = xfs_bmap_last_offset(ip, &last_block,
812 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
813 if ((map_start_fsb + count_fsb) > last_block) {
814 count_fsb = last_block - map_start_fsb;
815 if (count_fsb == 0) {
822 * From this point onwards we overwrite the imap
823 * pointer that the caller gave to us.
825 error = xfs_bmapi_write(tp, ip, map_start_fsb,
826 count_fsb, 0, &first_block,
832 error = xfs_defer_finish(&tp, &dfops, NULL);
836 error = xfs_trans_commit(tp);
840 xfs_iunlock(ip, XFS_ILOCK_EXCL);
844 * See if we were able to allocate an extent that
845 * covers at least part of the callers request
847 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
848 return xfs_alert_fsblock_zero(ip, imap);
850 if ((offset_fsb >= imap->br_startoff) &&
851 (offset_fsb < (imap->br_startoff +
852 imap->br_blockcount))) {
853 XFS_STATS_INC(mp, xs_xstrat_quick);
858 * So far we have not mapped the requested part of the
859 * file, just surrounding data, try again.
861 count_fsb -= imap->br_blockcount;
862 map_start_fsb = imap->br_startoff + imap->br_blockcount;
866 xfs_defer_cancel(&dfops);
867 xfs_trans_cancel(tp);
869 xfs_iunlock(ip, XFS_ILOCK_EXCL);
874 xfs_iomap_write_unwritten(
879 xfs_mount_t *mp = ip->i_mount;
880 xfs_fileoff_t offset_fsb;
881 xfs_filblks_t count_fsb;
882 xfs_filblks_t numblks_fsb;
883 xfs_fsblock_t firstfsb;
886 xfs_bmbt_irec_t imap;
887 struct xfs_defer_ops dfops;
892 trace_xfs_unwritten_convert(ip, offset, count);
894 offset_fsb = XFS_B_TO_FSBT(mp, offset);
895 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
896 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
899 * Reserve enough blocks in this transaction for two complete extent
900 * btree splits. We may be converting the middle part of an unwritten
901 * extent and in this case we will insert two new extents in the btree
902 * each of which could cause a full split.
904 * This reservation amount will be used in the first call to
905 * xfs_bmbt_split() to select an AG with enough space to satisfy the
906 * rest of the operation.
908 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
912 * Set up a transaction to convert the range of extents
913 * from unwritten to real. Do allocations in a loop until
914 * we have covered the range passed in.
916 * Note that we can't risk to recursing back into the filesystem
917 * here as we might be asked to write out the same inode that we
918 * complete here and might deadlock on the iolock.
920 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
921 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
925 xfs_ilock(ip, XFS_ILOCK_EXCL);
926 xfs_trans_ijoin(tp, ip, 0);
929 * Modify the unwritten extent state of the buffer.
931 xfs_defer_init(&dfops, &firstfsb);
933 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
934 XFS_BMAPI_CONVERT, &firstfsb, resblks,
935 &imap, &nimaps, &dfops);
937 goto error_on_bmapi_transaction;
940 * Log the updated inode size as we go. We have to be careful
941 * to only log it up to the actual write offset if it is
942 * halfway into a block.
944 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
945 if (i_size > offset + count)
946 i_size = offset + count;
948 i_size = xfs_new_eof(ip, i_size);
950 ip->i_d.di_size = i_size;
951 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
954 error = xfs_defer_finish(&tp, &dfops, NULL);
956 goto error_on_bmapi_transaction;
958 error = xfs_trans_commit(tp);
959 xfs_iunlock(ip, XFS_ILOCK_EXCL);
963 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
964 return xfs_alert_fsblock_zero(ip, &imap);
966 if ((numblks_fsb = imap.br_blockcount) == 0) {
968 * The numblks_fsb value should always get
969 * smaller, otherwise the loop is stuck.
971 ASSERT(imap.br_blockcount);
974 offset_fsb += numblks_fsb;
975 count_fsb -= numblks_fsb;
976 } while (count_fsb > 0);
980 error_on_bmapi_transaction:
981 xfs_defer_cancel(&dfops);
982 xfs_trans_cancel(tp);
983 xfs_iunlock(ip, XFS_ILOCK_EXCL);
987 static inline bool imap_needs_alloc(struct xfs_bmbt_irec *imap, int nimaps)
990 imap->br_startblock == HOLESTARTBLOCK ||
991 imap->br_startblock == DELAYSTARTBLOCK;
995 xfs_file_iomap_begin(
1000 struct iomap *iomap)
1002 struct xfs_inode *ip = XFS_I(inode);
1003 struct xfs_mount *mp = ip->i_mount;
1004 struct xfs_bmbt_irec imap;
1005 xfs_fileoff_t offset_fsb, end_fsb;
1006 int nimaps = 1, error = 0;
1008 if (XFS_FORCED_SHUTDOWN(mp))
1011 xfs_ilock(ip, XFS_ILOCK_EXCL);
1013 ASSERT(offset <= mp->m_super->s_maxbytes);
1014 if ((xfs_fsize_t)offset + length > mp->m_super->s_maxbytes)
1015 length = mp->m_super->s_maxbytes - offset;
1016 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1017 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1019 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1020 &nimaps, XFS_BMAPI_ENTIRE);
1022 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1026 if ((flags & IOMAP_WRITE) && imap_needs_alloc(&imap, nimaps)) {
1028 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1029 * pages to keep the chunks of work done where somewhat symmetric
1030 * with the work writeback does. This is a completely arbitrary
1031 * number pulled out of thin air as a best guess for initial
1034 * Note that the values needs to be less than 32-bits wide until
1035 * the lower level functions are updated.
1037 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1038 if (xfs_get_extsz_hint(ip)) {
1040 * xfs_iomap_write_direct() expects the shared lock. It
1041 * is unlocked on return.
1043 xfs_ilock_demote(ip, XFS_ILOCK_EXCL);
1044 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1047 error = xfs_iomap_write_delay(ip, offset, length, &imap);
1048 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1054 trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1058 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1059 trace_xfs_iomap_found(ip, offset, length, 0, &imap);
1062 xfs_bmbt_to_iomap(ip, iomap, &imap);
1067 xfs_file_iomap_end_delalloc(
1068 struct xfs_inode *ip,
1073 struct xfs_mount *mp = ip->i_mount;
1074 xfs_fileoff_t start_fsb;
1075 xfs_fileoff_t end_fsb;
1078 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1079 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1082 * Trim back delalloc blocks if we didn't manage to write the whole
1085 * We don't need to care about racing delalloc as we hold i_mutex
1086 * across the reserve/allocate/unreserve calls. If there are delalloc
1087 * blocks in the range, they are ours.
1089 if (start_fsb < end_fsb) {
1090 xfs_ilock(ip, XFS_ILOCK_EXCL);
1091 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1092 end_fsb - start_fsb);
1093 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1095 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1096 xfs_alert(mp, "%s: unable to clean up ino %lld",
1097 __func__, ip->i_ino);
1107 struct inode *inode,
1112 struct iomap *iomap)
1114 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1115 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1120 struct iomap_ops xfs_iomap_ops = {
1121 .iomap_begin = xfs_file_iomap_begin,
1122 .iomap_end = xfs_file_iomap_end,
1126 xfs_xattr_iomap_begin(
1127 struct inode *inode,
1131 struct iomap *iomap)
1133 struct xfs_inode *ip = XFS_I(inode);
1134 struct xfs_mount *mp = ip->i_mount;
1135 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1136 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1137 struct xfs_bmbt_irec imap;
1138 int nimaps = 1, error = 0;
1141 if (XFS_FORCED_SHUTDOWN(mp))
1144 lockmode = xfs_ilock_data_map_shared(ip);
1146 /* if there are no attribute fork or extents, return ENOENT */
1147 if (XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1152 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1153 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1154 &nimaps, XFS_BMAPI_ENTIRE | XFS_BMAPI_ATTRFORK);
1156 xfs_iunlock(ip, lockmode);
1160 xfs_bmbt_to_iomap(ip, iomap, &imap);
1166 struct iomap_ops xfs_xattr_iomap_ops = {
1167 .iomap_begin = xfs_xattr_iomap_begin,