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
48 xfs_iomap_eof_align_last_fsb(
52 xfs_fileoff_t *last_fsb)
54 xfs_extlen_t align = 0;
57 if (!XFS_IS_REALTIME_INODE(ip)) {
59 * Round up the allocation request to a stripe unit
60 * (m_dalign) boundary if the file size is >= stripe unit
61 * size, and we are allocating past the allocation eof.
63 * If mounted with the "-o swalloc" option the alignment is
64 * increased from the strip unit size to the stripe width.
66 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
68 else if (mp->m_dalign)
71 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
76 * Always round up the allocation request to an extent boundary
77 * (when file on a real-time subvolume or has di_extsize hint).
81 align = roundup_64(align, extsize);
87 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
88 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
92 *last_fsb = new_last_fsb;
98 xfs_alert_fsblock_zero(
100 xfs_bmbt_irec_t *imap)
102 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
103 "Access to block zero in inode %llu "
104 "start_block: %llx start_off: %llx "
105 "blkcnt: %llx extent-state: %x",
106 (unsigned long long)ip->i_ino,
107 (unsigned long long)imap->br_startblock,
108 (unsigned long long)imap->br_startoff,
109 (unsigned long long)imap->br_blockcount,
111 return -EFSCORRUPTED;
115 xfs_iomap_write_direct(
119 xfs_bmbt_irec_t *imap,
122 xfs_mount_t *mp = ip->i_mount;
123 xfs_fileoff_t offset_fsb;
124 xfs_fileoff_t last_fsb;
125 xfs_filblks_t count_fsb, resaligned;
126 xfs_fsblock_t firstfsb;
127 xfs_extlen_t extsz, temp;
132 struct xfs_defer_ops dfops;
133 uint qblocks, resblks, resrtextents;
136 int bmapi_flags = XFS_BMAPI_PREALLOC;
139 rt = XFS_IS_REALTIME_INODE(ip);
140 extsz = xfs_get_extsz_hint(ip);
141 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
143 ASSERT(xfs_isilocked(ip, lockmode));
145 offset_fsb = XFS_B_TO_FSBT(mp, offset);
146 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
147 if ((offset + count) > XFS_ISIZE(ip)) {
149 * Assert that the in-core extent list is present since this can
150 * call xfs_iread_extents() and we only have the ilock shared.
151 * This should be safe because the lock was held around a bmapi
152 * call in the caller and we only need it to access the in-core
155 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
157 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
161 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
162 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
163 imap->br_blockcount +
166 count_fsb = last_fsb - offset_fsb;
167 ASSERT(count_fsb > 0);
169 resaligned = count_fsb;
170 if (unlikely(extsz)) {
171 if ((temp = do_mod(offset_fsb, extsz)))
173 if ((temp = do_mod(resaligned, extsz)))
174 resaligned += extsz - temp;
178 resrtextents = qblocks = resaligned;
179 resrtextents /= mp->m_sb.sb_rextsize;
180 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
181 quota_flag = XFS_QMOPT_RES_RTBLKS;
184 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
185 quota_flag = XFS_QMOPT_RES_REGBLKS;
189 * Drop the shared lock acquired by the caller, attach the dquot if
190 * necessary and move on to transaction setup.
192 xfs_iunlock(ip, lockmode);
193 error = xfs_qm_dqattach(ip, 0);
198 * For DAX, we do not allocate unwritten extents, but instead we zero
199 * the block before we commit the transaction. Ideally we'd like to do
200 * this outside the transaction context, but if we commit and then crash
201 * we may not have zeroed the blocks and this will be exposed on
202 * recovery of the allocation. Hence we must zero before commit.
204 * Further, if we are mapping unwritten extents here, we need to zero
205 * and convert them to written so that we don't need an unwritten extent
206 * callback for DAX. This also means that we need to be able to dip into
207 * the reserve block pool for bmbt block allocation if there is no space
208 * left but we need to do unwritten extent conversion.
210 if (IS_DAX(VFS_I(ip))) {
211 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
212 if (ISUNWRITTEN(imap)) {
213 tflags |= XFS_TRANS_RESERVE;
214 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
217 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
222 lockmode = XFS_ILOCK_EXCL;
223 xfs_ilock(ip, lockmode);
225 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
227 goto out_trans_cancel;
229 xfs_trans_ijoin(tp, ip, 0);
232 * From this point onwards we overwrite the imap pointer that the
235 xfs_defer_init(&dfops, &firstfsb);
237 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
238 bmapi_flags, &firstfsb, resblks, imap,
241 goto out_bmap_cancel;
244 * Complete the transaction
246 error = xfs_defer_finish(&tp, &dfops, NULL);
248 goto out_bmap_cancel;
250 error = xfs_trans_commit(tp);
255 * Copy any maps to caller's array and return any error.
262 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
263 error = xfs_alert_fsblock_zero(ip, imap);
266 xfs_iunlock(ip, lockmode);
270 xfs_defer_cancel(&dfops);
271 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
273 xfs_trans_cancel(tp);
278 * If the caller is doing a write at the end of the file, then extend the
279 * allocation out to the file system's write iosize. We clean up any extra
280 * space left over when the file is closed in xfs_inactive().
282 * If we find we already have delalloc preallocation beyond EOF, don't do more
283 * preallocation as it it not needed.
286 xfs_iomap_eof_want_preallocate(
291 xfs_bmbt_irec_t *imap,
295 xfs_fileoff_t start_fsb;
296 xfs_filblks_t count_fsb;
298 int found_delalloc = 0;
301 if (offset + count <= XFS_ISIZE(ip))
305 * If the file is smaller than the minimum prealloc and we are using
306 * dynamic preallocation, don't do any preallocation at all as it is
307 * likely this is the only write to the file that is going to be done.
309 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
310 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
314 * If there are any real blocks past eof, then don't
315 * do any speculative allocation.
317 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
318 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
319 while (count_fsb > 0) {
321 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
325 for (n = 0; n < imaps; n++) {
326 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
327 (imap[n].br_startblock != DELAYSTARTBLOCK))
329 start_fsb += imap[n].br_blockcount;
330 count_fsb -= imap[n].br_blockcount;
332 if (imap[n].br_startblock == DELAYSTARTBLOCK)
342 * Determine the initial size of the preallocation. We are beyond the current
343 * EOF here, but we need to take into account whether this is a sparse write or
344 * an extending write when determining the preallocation size. Hence we need to
345 * look up the extent that ends at the current write offset and use the result
346 * to determine the preallocation size.
348 * If the extent is a hole, then preallocation is essentially disabled.
349 * Otherwise we take the size of the preceeding data extent as the basis for the
350 * preallocation size. If the size of the extent is greater than half the
351 * maximum extent length, then use the current offset as the basis. This ensures
352 * that for large files the preallocation size always extends to MAXEXTLEN
353 * rather than falling short due to things like stripe unit/width alignment of
357 xfs_iomap_eof_prealloc_initial_size(
358 struct xfs_mount *mp,
359 struct xfs_inode *ip,
361 xfs_bmbt_irec_t *imap,
364 xfs_fileoff_t start_fsb;
368 ASSERT(nimaps >= imaps);
370 /* if we are using a specific prealloc size, return now */
371 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
374 /* If the file is small, then use the minimum prealloc */
375 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
379 * As we write multiple pages, the offset will always align to the
380 * start of a page and hence point to a hole at EOF. i.e. if the size is
381 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
382 * will return FSB 1. Hence if there are blocks in the file, we want to
383 * point to the block prior to the EOF block and not the hole that maps
384 * directly at @offset.
386 start_fsb = XFS_B_TO_FSB(mp, offset);
389 error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
394 if (imap[0].br_startblock == HOLESTARTBLOCK)
396 if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
397 return imap[0].br_blockcount << 1;
398 return XFS_B_TO_FSB(mp, offset);
402 xfs_quota_need_throttle(
403 struct xfs_inode *ip,
405 xfs_fsblock_t alloc_blocks)
407 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
409 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
412 /* no hi watermark, no throttle */
413 if (!dq->q_prealloc_hi_wmark)
416 /* under the lo watermark, no throttle */
417 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
424 xfs_quota_calc_throttle(
425 struct xfs_inode *ip,
427 xfs_fsblock_t *qblocks,
433 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
435 /* no dq, or over hi wmark, squash the prealloc completely */
436 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
442 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
443 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
445 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
447 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
451 if (freesp < *qfreesp)
454 /* only overwrite the throttle values if we are more aggressive */
455 if ((freesp >> shift) < (*qblocks >> *qshift)) {
462 * If we don't have a user specified preallocation size, dynamically increase
463 * the preallocation size as the size of the file grows. Cap the maximum size
464 * at a single extent or less if the filesystem is near full. The closer the
465 * filesystem is to full, the smaller the maximum prealocation.
468 xfs_iomap_prealloc_size(
469 struct xfs_mount *mp,
470 struct xfs_inode *ip,
472 struct xfs_bmbt_irec *imap,
475 xfs_fsblock_t alloc_blocks = 0;
478 xfs_fsblock_t qblocks;
481 alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
485 qblocks = alloc_blocks;
488 * MAXEXTLEN is not a power of two value but we round the prealloc down
489 * to the nearest power of two value after throttling. To prevent the
490 * round down from unconditionally reducing the maximum supported prealloc
491 * size, we round up first, apply appropriate throttling, round down and
492 * cap the value to MAXEXTLEN.
494 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
497 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
498 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
500 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
502 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
504 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
506 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
511 * Check each quota to cap the prealloc size, provide a shift value to
512 * throttle with and adjust amount of available space.
514 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
515 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
517 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
518 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
520 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
521 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
525 * The final prealloc size is set to the minimum of free space available
526 * in each of the quotas and the overall filesystem.
528 * The shift throttle value is set to the maximum value as determined by
529 * the global low free space values and per-quota low free space values.
531 alloc_blocks = MIN(alloc_blocks, qblocks);
532 shift = MAX(shift, qshift);
535 alloc_blocks >>= shift;
537 * rounddown_pow_of_two() returns an undefined result if we pass in
541 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
542 if (alloc_blocks > MAXEXTLEN)
543 alloc_blocks = MAXEXTLEN;
546 * If we are still trying to allocate more space than is
547 * available, squash the prealloc hard. This can happen if we
548 * have a large file on a small filesystem and the above
549 * lowspace thresholds are smaller than MAXEXTLEN.
551 while (alloc_blocks && alloc_blocks >= freesp)
555 if (alloc_blocks < mp->m_writeio_blocks)
556 alloc_blocks = mp->m_writeio_blocks;
558 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
559 mp->m_writeio_blocks);
565 xfs_iomap_write_delay(
569 xfs_bmbt_irec_t *ret_imap)
571 xfs_mount_t *mp = ip->i_mount;
572 xfs_fileoff_t offset_fsb;
573 xfs_fileoff_t last_fsb;
574 xfs_off_t aligned_offset;
575 xfs_fileoff_t ioalign;
578 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
582 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
585 * Make sure that the dquots are there. This doesn't hold
586 * the ilock across a disk read.
588 error = xfs_qm_dqattach_locked(ip, 0);
592 extsz = xfs_get_extsz_hint(ip);
593 offset_fsb = XFS_B_TO_FSBT(mp, offset);
595 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
596 imap, XFS_WRITE_IMAPS, &prealloc);
602 xfs_fsblock_t alloc_blocks;
604 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
607 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
608 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
609 last_fsb = ioalign + alloc_blocks;
611 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
614 if (prealloc || extsz) {
615 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
621 * Make sure preallocation does not create extents beyond the range we
622 * actually support in this filesystem.
624 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
625 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
627 ASSERT(last_fsb > offset_fsb);
629 nimaps = XFS_WRITE_IMAPS;
630 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
631 imap, &nimaps, XFS_BMAPI_ENTIRE);
642 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
643 * without EOF preallocation.
646 trace_xfs_delalloc_enospc(ip, offset, count);
652 return error ? error : -ENOSPC;
655 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
656 return xfs_alert_fsblock_zero(ip, &imap[0]);
659 * Tag the inode as speculatively preallocated so we can reclaim this
660 * space on demand, if necessary.
663 xfs_inode_set_eofblocks_tag(ip);
670 * Pass in a delayed allocate extent, convert it to real extents;
671 * return to the caller the extent we create which maps on top of
672 * the originating callers request.
674 * Called without a lock on the inode.
676 * We no longer bother to look at the incoming map - all we have to
677 * guarantee is that whatever we allocate fills the required range.
680 xfs_iomap_write_allocate(
683 xfs_bmbt_irec_t *imap)
685 xfs_mount_t *mp = ip->i_mount;
686 xfs_fileoff_t offset_fsb, last_block;
687 xfs_fileoff_t end_fsb, map_start_fsb;
688 xfs_fsblock_t first_block;
689 struct xfs_defer_ops dfops;
690 xfs_filblks_t count_fsb;
697 * Make sure that the dquots are there.
699 error = xfs_qm_dqattach(ip, 0);
703 offset_fsb = XFS_B_TO_FSBT(mp, offset);
704 count_fsb = imap->br_blockcount;
705 map_start_fsb = imap->br_startoff;
707 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
709 while (count_fsb != 0) {
711 * Set up a transaction with which to allocate the
712 * backing store for the file. Do allocations in a
713 * loop until we get some space in the range we are
714 * interested in. The other space that might be allocated
715 * is in the delayed allocation extent on which we sit
716 * but before our buffer starts.
719 while (nimaps == 0) {
720 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
722 * We have already reserved space for the extent and any
723 * indirect blocks when creating the delalloc extent,
724 * there is no need to reserve space in this transaction
727 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
728 0, XFS_TRANS_RESERVE, &tp);
732 xfs_ilock(ip, XFS_ILOCK_EXCL);
733 xfs_trans_ijoin(tp, ip, 0);
735 xfs_defer_init(&dfops, &first_block);
738 * it is possible that the extents have changed since
739 * we did the read call as we dropped the ilock for a
740 * while. We have to be careful about truncates or hole
741 * punchs here - we are not allowed to allocate
742 * non-delalloc blocks here.
744 * The only protection against truncation is the pages
745 * for the range we are being asked to convert are
746 * locked and hence a truncate will block on them
749 * As a result, if we go beyond the range we really
750 * need and hit an delalloc extent boundary followed by
751 * a hole while we have excess blocks in the map, we
752 * will fill the hole incorrectly and overrun the
753 * transaction reservation.
755 * Using a single map prevents this as we are forced to
756 * check each map we look for overlap with the desired
757 * range and abort as soon as we find it. Also, given
758 * that we only return a single map, having one beyond
759 * what we can return is probably a bit silly.
761 * We also need to check that we don't go beyond EOF;
762 * this is a truncate optimisation as a truncate sets
763 * the new file size before block on the pages we
764 * currently have locked under writeback. Because they
765 * are about to be tossed, we don't need to write them
769 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
770 error = xfs_bmap_last_offset(ip, &last_block,
775 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
776 if ((map_start_fsb + count_fsb) > last_block) {
777 count_fsb = last_block - map_start_fsb;
778 if (count_fsb == 0) {
785 * From this point onwards we overwrite the imap
786 * pointer that the caller gave to us.
788 error = xfs_bmapi_write(tp, ip, map_start_fsb,
789 count_fsb, 0, &first_block,
795 error = xfs_defer_finish(&tp, &dfops, NULL);
799 error = xfs_trans_commit(tp);
803 xfs_iunlock(ip, XFS_ILOCK_EXCL);
807 * See if we were able to allocate an extent that
808 * covers at least part of the callers request
810 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
811 return xfs_alert_fsblock_zero(ip, imap);
813 if ((offset_fsb >= imap->br_startoff) &&
814 (offset_fsb < (imap->br_startoff +
815 imap->br_blockcount))) {
816 XFS_STATS_INC(mp, xs_xstrat_quick);
821 * So far we have not mapped the requested part of the
822 * file, just surrounding data, try again.
824 count_fsb -= imap->br_blockcount;
825 map_start_fsb = imap->br_startoff + imap->br_blockcount;
829 xfs_defer_cancel(&dfops);
830 xfs_trans_cancel(tp);
832 xfs_iunlock(ip, XFS_ILOCK_EXCL);
837 xfs_iomap_write_unwritten(
842 xfs_mount_t *mp = ip->i_mount;
843 xfs_fileoff_t offset_fsb;
844 xfs_filblks_t count_fsb;
845 xfs_filblks_t numblks_fsb;
846 xfs_fsblock_t firstfsb;
849 xfs_bmbt_irec_t imap;
850 struct xfs_defer_ops dfops;
855 trace_xfs_unwritten_convert(ip, offset, count);
857 offset_fsb = XFS_B_TO_FSBT(mp, offset);
858 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
859 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
862 * Reserve enough blocks in this transaction for two complete extent
863 * btree splits. We may be converting the middle part of an unwritten
864 * extent and in this case we will insert two new extents in the btree
865 * each of which could cause a full split.
867 * This reservation amount will be used in the first call to
868 * xfs_bmbt_split() to select an AG with enough space to satisfy the
869 * rest of the operation.
871 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
875 * Set up a transaction to convert the range of extents
876 * from unwritten to real. Do allocations in a loop until
877 * we have covered the range passed in.
879 * Note that we can't risk to recursing back into the filesystem
880 * here as we might be asked to write out the same inode that we
881 * complete here and might deadlock on the iolock.
883 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
884 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
888 xfs_ilock(ip, XFS_ILOCK_EXCL);
889 xfs_trans_ijoin(tp, ip, 0);
892 * Modify the unwritten extent state of the buffer.
894 xfs_defer_init(&dfops, &firstfsb);
896 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
897 XFS_BMAPI_CONVERT, &firstfsb, resblks,
898 &imap, &nimaps, &dfops);
900 goto error_on_bmapi_transaction;
903 * Log the updated inode size as we go. We have to be careful
904 * to only log it up to the actual write offset if it is
905 * halfway into a block.
907 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
908 if (i_size > offset + count)
909 i_size = offset + count;
911 i_size = xfs_new_eof(ip, i_size);
913 ip->i_d.di_size = i_size;
914 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
917 error = xfs_defer_finish(&tp, &dfops, NULL);
919 goto error_on_bmapi_transaction;
921 error = xfs_trans_commit(tp);
922 xfs_iunlock(ip, XFS_ILOCK_EXCL);
926 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
927 return xfs_alert_fsblock_zero(ip, &imap);
929 if ((numblks_fsb = imap.br_blockcount) == 0) {
931 * The numblks_fsb value should always get
932 * smaller, otherwise the loop is stuck.
934 ASSERT(imap.br_blockcount);
937 offset_fsb += numblks_fsb;
938 count_fsb -= numblks_fsb;
939 } while (count_fsb > 0);
943 error_on_bmapi_transaction:
944 xfs_defer_cancel(&dfops);
945 xfs_trans_cancel(tp);
946 xfs_iunlock(ip, XFS_ILOCK_EXCL);
952 struct xfs_inode *ip,
954 struct xfs_bmbt_irec *imap)
956 struct xfs_mount *mp = ip->i_mount;
958 if (imap->br_startblock == HOLESTARTBLOCK) {
959 iomap->blkno = IOMAP_NULL_BLOCK;
960 iomap->type = IOMAP_HOLE;
961 } else if (imap->br_startblock == DELAYSTARTBLOCK) {
962 iomap->blkno = IOMAP_NULL_BLOCK;
963 iomap->type = IOMAP_DELALLOC;
965 iomap->blkno = xfs_fsb_to_db(ip, imap->br_startblock);
966 if (imap->br_state == XFS_EXT_UNWRITTEN)
967 iomap->type = IOMAP_UNWRITTEN;
969 iomap->type = IOMAP_MAPPED;
971 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
972 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
973 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
976 static inline bool imap_needs_alloc(struct xfs_bmbt_irec *imap, int nimaps)
979 imap->br_startblock == HOLESTARTBLOCK ||
980 imap->br_startblock == DELAYSTARTBLOCK;
984 xfs_file_iomap_begin(
991 struct xfs_inode *ip = XFS_I(inode);
992 struct xfs_mount *mp = ip->i_mount;
993 struct xfs_bmbt_irec imap;
994 xfs_fileoff_t offset_fsb, end_fsb;
995 int nimaps = 1, error = 0;
997 if (XFS_FORCED_SHUTDOWN(mp))
1000 xfs_ilock(ip, XFS_ILOCK_EXCL);
1002 ASSERT(offset <= mp->m_super->s_maxbytes);
1003 if ((xfs_fsize_t)offset + length > mp->m_super->s_maxbytes)
1004 length = mp->m_super->s_maxbytes - offset;
1005 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1006 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1008 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1009 &nimaps, XFS_BMAPI_ENTIRE);
1011 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1015 if ((flags & IOMAP_WRITE) && imap_needs_alloc(&imap, nimaps)) {
1017 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1018 * pages to keep the chunks of work done where somewhat symmetric
1019 * with the work writeback does. This is a completely arbitrary
1020 * number pulled out of thin air as a best guess for initial
1023 * Note that the values needs to be less than 32-bits wide until
1024 * the lower level functions are updated.
1026 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1027 if (xfs_get_extsz_hint(ip)) {
1029 * xfs_iomap_write_direct() expects the shared lock. It
1030 * is unlocked on return.
1032 xfs_ilock_demote(ip, XFS_ILOCK_EXCL);
1033 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1036 error = xfs_iomap_write_delay(ip, offset, length, &imap);
1037 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1043 trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1047 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1048 trace_xfs_iomap_found(ip, offset, length, 0, &imap);
1051 xfs_bmbt_to_iomap(ip, iomap, &imap);
1056 xfs_file_iomap_end_delalloc(
1057 struct xfs_inode *ip,
1062 struct xfs_mount *mp = ip->i_mount;
1063 xfs_fileoff_t start_fsb;
1064 xfs_fileoff_t end_fsb;
1067 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1068 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1071 * Trim back delalloc blocks if we didn't manage to write the whole
1074 * We don't need to care about racing delalloc as we hold i_mutex
1075 * across the reserve/allocate/unreserve calls. If there are delalloc
1076 * blocks in the range, they are ours.
1078 if (start_fsb < end_fsb) {
1079 xfs_ilock(ip, XFS_ILOCK_EXCL);
1080 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1081 end_fsb - start_fsb);
1082 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1084 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1085 xfs_alert(mp, "%s: unable to clean up ino %lld",
1086 __func__, ip->i_ino);
1096 struct inode *inode,
1101 struct iomap *iomap)
1103 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1104 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1109 struct iomap_ops xfs_iomap_ops = {
1110 .iomap_begin = xfs_file_iomap_begin,
1111 .iomap_end = xfs_file_iomap_end,
1115 xfs_xattr_iomap_begin(
1116 struct inode *inode,
1120 struct iomap *iomap)
1122 struct xfs_inode *ip = XFS_I(inode);
1123 struct xfs_mount *mp = ip->i_mount;
1124 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1125 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1126 struct xfs_bmbt_irec imap;
1127 int nimaps = 1, error = 0;
1130 if (XFS_FORCED_SHUTDOWN(mp))
1133 lockmode = xfs_ilock_data_map_shared(ip);
1135 /* if there are no attribute fork or extents, return ENOENT */
1136 if (XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1141 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1142 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1143 &nimaps, XFS_BMAPI_ENTIRE | XFS_BMAPI_ATTRFORK);
1145 xfs_iunlock(ip, lockmode);
1149 xfs_bmbt_to_iomap(ip, iomap, &imap);
1155 struct iomap_ops xfs_xattr_iomap_ops = {
1156 .iomap_begin = xfs_xattr_iomap_begin,