1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
16 #include "xfs_bmap_btree.h"
17 #include "xfs_trans.h"
18 #include "xfs_trans_space.h"
19 #include "xfs_icache.h"
20 #include "xfs_rtalloc.h"
22 #include "xfs_rtbitmap.h"
25 * Read and return the summary information for a given extent size,
26 * bitmap block combination.
27 * Keeps track of a current summary block, so we don't keep reading
28 * it from the buffer cache.
32 struct xfs_rtalloc_args *args,
33 int log, /* log2 of extent size */
34 xfs_fileoff_t bbno, /* bitmap block number */
35 xfs_suminfo_t *sum) /* out: summary info for this block */
37 return xfs_rtmodify_summary_int(args, log, bbno, 0, sum);
41 * Return whether there are any free extents in the size range given
42 * by low and high, for the bitmap block bbno.
46 struct xfs_rtalloc_args *args,
47 int low, /* low log2 extent size */
48 int high, /* high log2 extent size */
49 xfs_fileoff_t bbno, /* bitmap block number */
50 int *maxlog) /* out: max log2 extent size free */
52 struct xfs_mount *mp = args->mp;
54 int log; /* loop counter, log2 of ext. size */
55 xfs_suminfo_t sum; /* summary data */
57 /* There are no extents at levels >= m_rsum_cache[bbno]. */
58 if (mp->m_rsum_cache) {
59 high = min(high, mp->m_rsum_cache[bbno] - 1);
67 * Loop over logs of extent sizes.
69 for (log = high; log >= low; log--) {
71 * Get one summary datum.
73 error = xfs_rtget_summary(args, log, bbno, &sum);
78 * If there are any, return success.
86 * Found nothing, return failure.
90 /* There were no extents at levels > log. */
91 if (mp->m_rsum_cache && log + 1 < mp->m_rsum_cache[bbno])
92 mp->m_rsum_cache[bbno] = log + 1;
98 * Copy and transform the summary file, given the old and new
99 * parameters in the mount structures.
103 struct xfs_rtalloc_args *oargs,
104 struct xfs_rtalloc_args *nargs)
106 xfs_fileoff_t bbno; /* bitmap block number */
108 int log; /* summary level number (log length) */
109 xfs_suminfo_t sum; /* summary data */
111 for (log = oargs->mp->m_rsumlevels - 1; log >= 0; log--) {
112 for (bbno = oargs->mp->m_sb.sb_rbmblocks - 1;
113 (xfs_srtblock_t)bbno >= 0;
115 error = xfs_rtget_summary(oargs, log, bbno, &sum);
120 error = xfs_rtmodify_summary(oargs, log, bbno, -sum);
123 error = xfs_rtmodify_summary(nargs, log, bbno, sum);
131 xfs_rtbuf_cache_relse(oargs);
135 * Mark an extent specified by start and len allocated.
136 * Updates all the summary information as well as the bitmap.
139 xfs_rtallocate_range(
140 struct xfs_rtalloc_args *args,
141 xfs_rtxnum_t start, /* start rtext to allocate */
142 xfs_rtxlen_t len) /* in/out: summary block number */
144 struct xfs_mount *mp = args->mp;
145 xfs_rtxnum_t end; /* end of the allocated rtext */
147 xfs_rtxnum_t postblock = 0; /* first rtext allocated > end */
148 xfs_rtxnum_t preblock = 0; /* first rtext allocated < start */
150 end = start + len - 1;
152 * Assume we're allocating out of the middle of a free extent.
153 * We need to find the beginning and end of the extent so we can
154 * properly update the summary.
156 error = xfs_rtfind_back(args, start, 0, &preblock);
161 * Find the next allocated block (end of free extent).
163 error = xfs_rtfind_forw(args, end, mp->m_sb.sb_rextents - 1,
169 * Decrement the summary information corresponding to the entire
172 error = xfs_rtmodify_summary(args,
173 XFS_RTBLOCKLOG(postblock + 1 - preblock),
174 xfs_rtx_to_rbmblock(mp, preblock), -1);
179 * If there are blocks not being allocated at the front of the
180 * old extent, add summary data for them to be free.
182 if (preblock < start) {
183 error = xfs_rtmodify_summary(args,
184 XFS_RTBLOCKLOG(start - preblock),
185 xfs_rtx_to_rbmblock(mp, preblock), 1);
191 * If there are blocks not being allocated at the end of the
192 * old extent, add summary data for them to be free.
194 if (postblock > end) {
195 error = xfs_rtmodify_summary(args,
196 XFS_RTBLOCKLOG(postblock - end),
197 xfs_rtx_to_rbmblock(mp, end + 1), 1);
203 * Modify the bitmap to mark this extent allocated.
205 error = xfs_rtmodify_range(args, start, len, 0);
210 * Make sure we don't run off the end of the rt volume. Be careful that
211 * adjusting maxlen downwards doesn't cause us to fail the alignment checks.
213 static inline xfs_rtxlen_t
214 xfs_rtallocate_clamp_len(
215 struct xfs_mount *mp,
216 xfs_rtxnum_t startrtx,
222 ret = min(mp->m_sb.sb_rextents, startrtx + rtxlen) - startrtx;
223 return rounddown(ret, prod);
227 * Attempt to allocate an extent minlen<=len<=maxlen starting from
228 * bitmap block bbno. If we don't get maxlen then use prod to trim
229 * the length, if given. Returns error; returns starting block in *rtx.
230 * The lengths are all in rtextents.
233 xfs_rtallocate_extent_block(
234 struct xfs_rtalloc_args *args,
235 xfs_fileoff_t bbno, /* bitmap block number */
236 xfs_rtxlen_t minlen, /* minimum length to allocate */
237 xfs_rtxlen_t maxlen, /* maximum length to allocate */
238 xfs_rtxlen_t *len, /* out: actual length allocated */
239 xfs_rtxnum_t *nextp, /* out: next rtext to try */
240 xfs_rtxlen_t prod, /* extent product factor */
241 xfs_rtxnum_t *rtx) /* out: start rtext allocated */
243 struct xfs_mount *mp = args->mp;
244 xfs_rtxnum_t besti; /* best rtext found so far */
245 xfs_rtxnum_t bestlen;/* best length found so far */
246 xfs_rtxnum_t end; /* last rtext in chunk */
248 xfs_rtxnum_t i; /* current rtext trying */
249 xfs_rtxnum_t next; /* next rtext to try */
250 int stat; /* status from internal calls */
253 * Loop over all the extents starting in this bitmap block,
254 * looking for one that's long enough.
256 for (i = xfs_rbmblock_to_rtx(mp, bbno), besti = -1, bestlen = 0,
257 end = xfs_rbmblock_to_rtx(mp, bbno + 1) - 1;
260 /* Make sure we don't scan off the end of the rt volume. */
261 maxlen = xfs_rtallocate_clamp_len(mp, i, maxlen, prod);
264 * See if there's a free extent of maxlen starting at i.
265 * If it's not so then next will contain the first non-free.
267 error = xfs_rtcheck_range(args, i, maxlen, 1, &next, &stat);
273 * i for maxlen is all free, allocate and return that.
275 error = xfs_rtallocate_range(args, i, maxlen);
284 * In the case where we have a variable-sized allocation
285 * request, figure out how big this free piece is,
286 * and if it's big enough for the minimum, and the best
287 * so far, remember it.
289 if (minlen < maxlen) {
290 xfs_rtxnum_t thislen; /* this extent size */
293 if (thislen >= minlen && thislen > bestlen) {
299 * If not done yet, find the start of the next free space.
302 error = xfs_rtfind_forw(args, next, end, &i);
310 * Searched the whole thing & didn't find a maxlen free extent.
312 if (minlen < maxlen && besti != -1) {
313 xfs_rtxlen_t p; /* amount to trim length by */
316 * If size should be a multiple of prod, make that so.
319 div_u64_rem(bestlen, prod, &p);
325 * Allocate besti for bestlen & return that.
327 error = xfs_rtallocate_range(args, besti, bestlen);
336 * Allocation failed. Set *nextp to the next block to try.
344 * Allocate an extent of length minlen<=len<=maxlen, starting at block
345 * bno. If we don't get maxlen then use prod to trim the length, if given.
346 * Returns error; returns starting block in *rtx.
347 * The lengths are all in rtextents.
350 xfs_rtallocate_extent_exact(
351 struct xfs_rtalloc_args *args,
352 xfs_rtxnum_t start, /* starting rtext number to allocate */
353 xfs_rtxlen_t minlen, /* minimum length to allocate */
354 xfs_rtxlen_t maxlen, /* maximum length to allocate */
355 xfs_rtxlen_t *len, /* out: actual length allocated */
356 xfs_rtxlen_t prod, /* extent product factor */
357 xfs_rtxnum_t *rtx) /* out: start rtext allocated */
360 xfs_rtxlen_t i; /* extent length trimmed due to prod */
361 int isfree; /* extent is free */
362 xfs_rtxnum_t next; /* next rtext to try (dummy) */
364 ASSERT(minlen % prod == 0);
365 ASSERT(maxlen % prod == 0);
367 * Check if the range in question (for maxlen) is free.
369 error = xfs_rtcheck_range(args, start, maxlen, 1, &next, &isfree);
375 * If it is, allocate it and return success.
377 error = xfs_rtallocate_range(args, start, maxlen);
386 * If not, allocate what there is, if it's at least minlen.
388 maxlen = next - start;
389 if (maxlen < minlen) {
391 * Failed, return failure status.
397 * Trim off tail of extent, if prod is specified.
399 if (prod > 1 && (i = maxlen % prod)) {
401 if (maxlen < minlen) {
403 * Now we can't do it, return failure status.
410 * Allocate what we can and return it.
412 error = xfs_rtallocate_range(args, start, maxlen);
422 * Allocate an extent of length minlen<=len<=maxlen, starting as near
423 * to start as possible. If we don't get maxlen then use prod to trim
424 * the length, if given. The lengths are all in rtextents.
427 xfs_rtallocate_extent_near(
428 struct xfs_rtalloc_args *args,
429 xfs_rtxnum_t start, /* starting rtext number to allocate */
430 xfs_rtxlen_t minlen, /* minimum length to allocate */
431 xfs_rtxlen_t maxlen, /* maximum length to allocate */
432 xfs_rtxlen_t *len, /* out: actual length allocated */
433 xfs_rtxlen_t prod, /* extent product factor */
434 xfs_rtxnum_t *rtx) /* out: start rtext allocated */
436 struct xfs_mount *mp = args->mp;
437 int maxlog; /* max useful extent from summary */
438 xfs_fileoff_t bbno; /* bitmap block number */
440 int i; /* bitmap block offset (loop control) */
441 int j; /* secondary loop control */
442 int log2len; /* log2 of minlen */
443 xfs_rtxnum_t n; /* next rtext to try */
444 xfs_rtxnum_t r; /* result rtext */
446 ASSERT(minlen % prod == 0);
447 ASSERT(maxlen % prod == 0);
450 * If the block number given is off the end, silently set it to
453 if (start >= mp->m_sb.sb_rextents)
454 start = mp->m_sb.sb_rextents - 1;
456 /* Make sure we don't run off the end of the rt volume. */
457 maxlen = xfs_rtallocate_clamp_len(mp, start, maxlen, prod);
458 if (maxlen < minlen) {
464 * Try the exact allocation first.
466 error = xfs_rtallocate_extent_exact(args, start, minlen, maxlen, len,
472 * If the exact allocation worked, return that.
474 if (r != NULLRTEXTNO) {
478 bbno = xfs_rtx_to_rbmblock(mp, start);
482 log2len = xfs_highbit32(minlen);
484 * Loop over all bitmap blocks (bbno + i is current block).
488 * Get summary information of extents of all useful levels
489 * starting in this bitmap block.
491 error = xfs_rtany_summary(args, log2len, mp->m_rsumlevels - 1,
497 * If there are any useful extents starting here, try
501 xfs_extlen_t maxavail =
502 min_t(xfs_rtblock_t, maxlen,
503 (1ULL << (maxlog + 1)) - 1);
505 * On the positive side of the starting location.
509 * Try to allocate an extent starting in
512 error = xfs_rtallocate_extent_block(args,
513 bbno + i, minlen, maxavail, len,
519 * If it worked, return it.
521 if (r != NULLRTEXTNO) {
527 * On the negative side of the starting location.
533 * Loop backwards to find the end of the extent
534 * we found in the realtime summary.
536 * maxblocks is the maximum possible number of
537 * bitmap blocks from the start of the extent
538 * to the end of the extent.
542 else if (maxlog < mp->m_blkbit_log)
545 maxblocks = 2 << (maxlog - mp->m_blkbit_log);
548 * We need to check bbno + i + maxblocks down to
549 * bbno + i. We already checked bbno down to
550 * bbno + j + 1, so we don't need to check those
553 j = min(i + maxblocks, j);
554 for (; j >= i; j--) {
555 error = xfs_rtallocate_extent_block(args,
557 maxavail, len, &n, prod,
563 * If it works, return the extent.
565 if (r != NULLRTEXTNO) {
573 * Loop control. If we were on the positive side, and there's
574 * still more blocks on the negative side, go there.
576 if (i > 0 && (int)bbno - i >= 0)
579 * If positive, and no more negative, but there are more
580 * positive, go there.
582 else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
585 * If negative or 0 (just started), and there are positive
586 * blocks to go, go there. The 0 case moves to block 1.
588 else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
591 * If negative or 0 and there are more negative blocks,
594 else if (i <= 0 && (int)bbno + i > 0)
597 * Must be done. Return failure.
607 * Allocate an extent of length minlen<=len<=maxlen, with no position
608 * specified. If we don't get maxlen then use prod to trim
609 * the length, if given. The lengths are all in rtextents.
612 xfs_rtallocate_extent_size(
613 struct xfs_rtalloc_args *args,
614 xfs_rtxlen_t minlen, /* minimum length to allocate */
615 xfs_rtxlen_t maxlen, /* maximum length to allocate */
616 xfs_rtxlen_t *len, /* out: actual length allocated */
617 xfs_rtxlen_t prod, /* extent product factor */
618 xfs_rtxnum_t *rtx) /* out: start rtext allocated */
620 struct xfs_mount *mp = args->mp;
622 xfs_fileoff_t i; /* bitmap block number */
623 int l; /* level number (loop control) */
624 xfs_rtxnum_t n; /* next rtext to be tried */
625 xfs_rtxnum_t r; /* result rtext number */
626 xfs_suminfo_t sum; /* summary information for extents */
628 ASSERT(minlen % prod == 0);
629 ASSERT(maxlen % prod == 0);
633 * Loop over all the levels starting with maxlen.
634 * At each level, look at all the bitmap blocks, to see if there
635 * are extents starting there that are long enough (>= maxlen).
636 * Note, only on the initial level can the allocation fail if
637 * the summary says there's an extent.
639 for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
641 * Loop over all the bitmap blocks.
643 for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
645 * Get the summary for this level/block.
647 error = xfs_rtget_summary(args, l, i, &sum);
652 * Nothing there, on to the next block.
657 * Try allocating the extent.
659 error = xfs_rtallocate_extent_block(args, i, maxlen,
660 maxlen, len, &n, prod, &r);
665 * If it worked, return that.
667 if (r != NULLRTEXTNO) {
672 * If the "next block to try" returned from the
673 * allocator is beyond the next bitmap block,
674 * skip to that bitmap block.
676 if (xfs_rtx_to_rbmblock(mp, n) > i + 1)
677 i = xfs_rtx_to_rbmblock(mp, n) - 1;
681 * Didn't find any maxlen blocks. Try smaller ones, unless
682 * we're asking for a fixed size extent.
684 if (minlen > --maxlen) {
692 * Loop over sizes, from maxlen down to minlen.
693 * This time, when we do the allocations, allow smaller ones
696 for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
698 * Loop over all the bitmap blocks, try an allocation
699 * starting in that block.
701 for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
703 * Get the summary information for this level/block.
705 error = xfs_rtget_summary(args, l, i, &sum);
710 * If nothing there, go on to next.
715 * Try the allocation. Make sure the specified
716 * minlen/maxlen are in the possible range for
717 * this summary level.
719 error = xfs_rtallocate_extent_block(args, i,
720 XFS_RTMAX(minlen, 1 << l),
721 XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
727 * If it worked, return that extent.
729 if (r != NULLRTEXTNO) {
734 * If the "next block to try" returned from the
735 * allocator is beyond the next bitmap block,
736 * skip to that bitmap block.
738 if (xfs_rtx_to_rbmblock(mp, n) > i + 1)
739 i = xfs_rtx_to_rbmblock(mp, n) - 1;
743 * Got nothing, return failure.
750 * Allocate space to the bitmap or summary file, and zero it, for growfs.
754 struct xfs_mount *mp, /* file system mount point */
755 xfs_extlen_t oblocks, /* old count of blocks */
756 xfs_extlen_t nblocks, /* new count of blocks */
757 struct xfs_inode *ip) /* inode (bitmap/summary) */
759 xfs_fileoff_t bno; /* block number in file */
760 struct xfs_buf *bp; /* temporary buffer for zeroing */
761 xfs_daddr_t d; /* disk block address */
762 int error; /* error return value */
763 xfs_fsblock_t fsbno; /* filesystem block for bno */
764 struct xfs_bmbt_irec map; /* block map output */
765 int nmap; /* number of block maps */
766 int resblks; /* space reservation */
767 enum xfs_blft buf_type;
768 struct xfs_trans *tp;
770 if (ip == mp->m_rsumip)
771 buf_type = XFS_BLFT_RTSUMMARY_BUF;
773 buf_type = XFS_BLFT_RTBITMAP_BUF;
776 * Allocate space to the file, as necessary.
778 while (oblocks < nblocks) {
779 resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
781 * Reserve space & log for one extent added to the file.
783 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtalloc, resblks,
790 xfs_ilock(ip, XFS_ILOCK_EXCL);
791 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
793 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
794 XFS_IEXT_ADD_NOSPLIT_CNT);
796 error = xfs_iext_count_upgrade(tp, ip,
797 XFS_IEXT_ADD_NOSPLIT_CNT);
799 goto out_trans_cancel;
802 * Allocate blocks to the bitmap file.
805 error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
806 XFS_BMAPI_METADATA, 0, &map, &nmap);
807 if (!error && nmap < 1)
810 goto out_trans_cancel;
812 * Free any blocks freed up in the transaction, then commit.
814 error = xfs_trans_commit(tp);
818 * Now we need to clear the allocated blocks.
819 * Do this one block per transaction, to keep it simple.
821 for (bno = map.br_startoff, fsbno = map.br_startblock;
822 bno < map.br_startoff + map.br_blockcount;
825 * Reserve log for one block zeroing.
827 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtzero,
832 * Lock the bitmap inode.
834 xfs_ilock(ip, XFS_ILOCK_EXCL);
835 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
837 * Get a buffer for the block.
839 d = XFS_FSB_TO_DADDR(mp, fsbno);
840 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
841 mp->m_bsize, 0, &bp);
843 goto out_trans_cancel;
845 xfs_trans_buf_set_type(tp, bp, buf_type);
846 bp->b_ops = &xfs_rtbuf_ops;
847 memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
848 xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
850 * Commit the transaction.
852 error = xfs_trans_commit(tp);
857 * Go on to the next extent, if any.
859 oblocks = map.br_startoff + map.br_blockcount;
865 xfs_trans_cancel(tp);
870 xfs_alloc_rsum_cache(
871 xfs_mount_t *mp, /* file system mount structure */
872 xfs_extlen_t rbmblocks) /* number of rt bitmap blocks */
875 * The rsum cache is initialized to the maximum value, which is
876 * trivially an upper bound on the maximum level with any free extents.
877 * We can continue without the cache if it couldn't be allocated.
879 mp->m_rsum_cache = kvmalloc(rbmblocks, GFP_KERNEL);
880 if (mp->m_rsum_cache)
881 memset(mp->m_rsum_cache, -1, rbmblocks);
883 xfs_warn(mp, "could not allocate realtime summary cache");
887 * Visible (exported) functions.
891 * Grow the realtime area of the filesystem.
895 xfs_mount_t *mp, /* mount point for filesystem */
896 xfs_growfs_rt_t *in) /* growfs rt input struct */
898 xfs_fileoff_t bmbno; /* bitmap block number */
899 struct xfs_buf *bp; /* temporary buffer */
900 int error; /* error return value */
901 xfs_mount_t *nmp; /* new (fake) mount structure */
902 xfs_rfsblock_t nrblocks; /* new number of realtime blocks */
903 xfs_extlen_t nrbmblocks; /* new number of rt bitmap blocks */
904 xfs_rtxnum_t nrextents; /* new number of realtime extents */
905 uint8_t nrextslog; /* new log2 of sb_rextents */
906 xfs_extlen_t nrsumblocks; /* new number of summary blocks */
907 uint nrsumlevels; /* new rt summary levels */
908 uint nrsumsize; /* new size of rt summary, bytes */
909 xfs_sb_t *nsbp; /* new superblock */
910 xfs_extlen_t rbmblocks; /* current number of rt bitmap blocks */
911 xfs_extlen_t rsumblocks; /* current number of rt summary blks */
912 xfs_sb_t *sbp; /* old superblock */
913 uint8_t *rsum_cache; /* old summary cache */
917 if (!capable(CAP_SYS_ADMIN))
920 /* Needs to have been mounted with an rt device. */
921 if (!XFS_IS_REALTIME_MOUNT(mp))
924 * Mount should fail if the rt bitmap/summary files don't load, but
925 * we'll check anyway.
927 if (!mp->m_rbmip || !mp->m_rsumip)
930 /* Shrink not supported. */
931 if (in->newblocks <= sbp->sb_rblocks)
934 /* Can only change rt extent size when adding rt volume. */
935 if (sbp->sb_rblocks > 0 && in->extsize != sbp->sb_rextsize)
938 /* Range check the extent size. */
939 if (XFS_FSB_TO_B(mp, in->extsize) > XFS_MAX_RTEXTSIZE ||
940 XFS_FSB_TO_B(mp, in->extsize) < XFS_MIN_RTEXTSIZE)
943 /* Unsupported realtime features. */
944 if (xfs_has_rmapbt(mp) || xfs_has_reflink(mp) || xfs_has_quota(mp))
947 nrblocks = in->newblocks;
948 error = xfs_sb_validate_fsb_count(sbp, nrblocks);
952 * Read in the last block of the device, make sure it exists.
954 error = xfs_buf_read_uncached(mp->m_rtdev_targp,
955 XFS_FSB_TO_BB(mp, nrblocks - 1),
956 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
962 * Calculate new parameters. These are the final values to be reached.
964 nrextents = nrblocks;
965 do_div(nrextents, in->extsize);
966 nrbmblocks = xfs_rtbitmap_blockcount(mp, nrextents);
967 nrextslog = xfs_highbit32(nrextents);
968 nrsumlevels = nrextslog + 1;
969 nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels, nrbmblocks);
970 nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
972 * New summary size can't be more than half the size of
973 * the log. This prevents us from getting a log overflow,
974 * since we'll log basically the whole summary file at once.
976 if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
979 * Get the old block counts for bitmap and summary inodes.
980 * These can't change since other growfs callers are locked out.
982 rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_disk_size);
983 rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_disk_size);
985 * Allocate space to the bitmap and summary files, as necessary.
987 error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
990 error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
994 rsum_cache = mp->m_rsum_cache;
995 if (nrbmblocks != sbp->sb_rbmblocks)
996 xfs_alloc_rsum_cache(mp, nrbmblocks);
999 * Allocate a new (fake) mount/sb.
1001 nmp = kmem_alloc(sizeof(*nmp), 0);
1003 * Loop over the bitmap blocks.
1004 * We will do everything one bitmap block at a time.
1005 * Skip the current block if it is exactly full.
1006 * This also deals with the case where there were no rtextents before.
1008 for (bmbno = sbp->sb_rbmblocks -
1009 ((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
1012 struct xfs_rtalloc_args args = {
1015 struct xfs_rtalloc_args nargs = {
1018 struct xfs_trans *tp;
1019 xfs_rfsblock_t nrblocks_step;
1024 * Calculate new sb and mount fields for this round.
1026 nsbp->sb_rextsize = in->extsize;
1027 nmp->m_rtxblklog = -1; /* don't use shift or masking */
1028 nsbp->sb_rbmblocks = bmbno + 1;
1029 nrblocks_step = (bmbno + 1) * NBBY * nsbp->sb_blocksize *
1031 nsbp->sb_rblocks = min(nrblocks, nrblocks_step);
1032 nsbp->sb_rextents = xfs_rtb_to_rtx(nmp, nsbp->sb_rblocks);
1033 ASSERT(nsbp->sb_rextents != 0);
1034 nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
1035 nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
1036 nrsumblocks = xfs_rtsummary_blockcount(mp, nrsumlevels,
1037 nsbp->sb_rbmblocks);
1038 nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
1040 * Start a transaction, get the log reservation.
1042 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtfree, 0, 0, 0,
1050 * Lock out other callers by grabbing the bitmap inode lock.
1052 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL | XFS_ILOCK_RTBITMAP);
1053 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
1055 * Update the bitmap inode's size ondisk and incore. We need
1056 * to update the incore size so that inode inactivation won't
1057 * punch what it thinks are "posteof" blocks.
1059 mp->m_rbmip->i_disk_size =
1060 nsbp->sb_rbmblocks * nsbp->sb_blocksize;
1061 i_size_write(VFS_I(mp->m_rbmip), mp->m_rbmip->i_disk_size);
1062 xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
1064 * Get the summary inode into the transaction.
1066 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL | XFS_ILOCK_RTSUM);
1067 xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
1069 * Update the summary inode's size. We need to update the
1070 * incore size so that inode inactivation won't punch what it
1071 * thinks are "posteof" blocks.
1073 mp->m_rsumip->i_disk_size = nmp->m_rsumsize;
1074 i_size_write(VFS_I(mp->m_rsumip), mp->m_rsumip->i_disk_size);
1075 xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
1077 * Copy summary data from old to new sizes.
1078 * Do this when the real size (not block-aligned) changes.
1080 if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
1081 mp->m_rsumlevels != nmp->m_rsumlevels) {
1082 error = xfs_rtcopy_summary(&args, &nargs);
1087 * Update superblock fields.
1089 if (nsbp->sb_rextsize != sbp->sb_rextsize)
1090 xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
1091 nsbp->sb_rextsize - sbp->sb_rextsize);
1092 if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
1093 xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
1094 nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
1095 if (nsbp->sb_rblocks != sbp->sb_rblocks)
1096 xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
1097 nsbp->sb_rblocks - sbp->sb_rblocks);
1098 if (nsbp->sb_rextents != sbp->sb_rextents)
1099 xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
1100 nsbp->sb_rextents - sbp->sb_rextents);
1101 if (nsbp->sb_rextslog != sbp->sb_rextslog)
1102 xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
1103 nsbp->sb_rextslog - sbp->sb_rextslog);
1107 error = xfs_rtfree_range(&nargs, sbp->sb_rextents,
1108 nsbp->sb_rextents - sbp->sb_rextents);
1109 xfs_rtbuf_cache_relse(&nargs);
1112 xfs_trans_cancel(tp);
1116 * Mark more blocks free in the superblock.
1118 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
1119 nsbp->sb_rextents - sbp->sb_rextents);
1121 * Update mp values into the real mp structure.
1123 mp->m_rsumlevels = nrsumlevels;
1124 mp->m_rsumsize = nrsumsize;
1126 error = xfs_trans_commit(tp);
1130 /* Ensure the mount RT feature flag is now set. */
1131 mp->m_features |= XFS_FEAT_REALTIME;
1136 /* Update secondary superblocks now the physical grow has completed */
1137 error = xfs_update_secondary_sbs(mp);
1141 * Free the fake mp structure.
1146 * If we had to allocate a new rsum_cache, we either need to free the
1147 * old one (if we succeeded) or free the new one and restore the old one
1148 * (if there was an error).
1150 if (rsum_cache != mp->m_rsum_cache) {
1152 kmem_free(mp->m_rsum_cache);
1153 mp->m_rsum_cache = rsum_cache;
1155 kmem_free(rsum_cache);
1163 * Allocate an extent in the realtime subvolume, with the usual allocation
1164 * parameters. The length units are all in realtime extents, as is the
1165 * result block number.
1168 xfs_rtallocate_extent(
1169 struct xfs_trans *tp,
1170 xfs_rtxnum_t start, /* starting rtext number to allocate */
1171 xfs_rtxlen_t minlen, /* minimum length to allocate */
1172 xfs_rtxlen_t maxlen, /* maximum length to allocate */
1173 xfs_rtxlen_t *len, /* out: actual length allocated */
1174 int wasdel, /* was a delayed allocation extent */
1175 xfs_rtxlen_t prod, /* extent product factor */
1176 xfs_rtxnum_t *rtblock) /* out: start rtext allocated */
1178 struct xfs_rtalloc_args args = {
1182 int error; /* error value */
1183 xfs_rtxnum_t r; /* result allocated rtext */
1185 ASSERT(xfs_isilocked(args.mp->m_rbmip, XFS_ILOCK_EXCL));
1186 ASSERT(minlen > 0 && minlen <= maxlen);
1189 * If prod is set then figure out what to do to minlen and maxlen.
1194 if ((i = maxlen % prod))
1196 if ((i = minlen % prod))
1198 if (maxlen < minlen) {
1199 *rtblock = NULLRTEXTNO;
1206 error = xfs_rtallocate_extent_size(&args, minlen,
1207 maxlen, len, prod, &r);
1209 error = xfs_rtallocate_extent_near(&args, start, minlen,
1210 maxlen, len, prod, &r);
1213 xfs_rtbuf_cache_relse(&args);
1218 * If it worked, update the superblock.
1220 if (r != NULLRTEXTNO) {
1221 long slen = (long)*len;
1223 ASSERT(*len >= minlen && *len <= maxlen);
1225 xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
1227 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
1228 } else if (prod > 1) {
1238 * Initialize realtime fields in the mount structure.
1242 struct xfs_mount *mp) /* file system mount structure */
1244 struct xfs_buf *bp; /* buffer for last block of subvolume */
1245 struct xfs_sb *sbp; /* filesystem superblock copy in mount */
1246 xfs_daddr_t d; /* address of last block of subvolume */
1247 unsigned int rsumblocks;
1251 if (sbp->sb_rblocks == 0)
1253 if (mp->m_rtdev_targp == NULL) {
1255 "Filesystem has a realtime volume, use rtdev=device option");
1258 mp->m_rsumlevels = sbp->sb_rextslog + 1;
1259 rsumblocks = xfs_rtsummary_blockcount(mp, mp->m_rsumlevels,
1260 mp->m_sb.sb_rbmblocks);
1261 mp->m_rsumsize = XFS_FSB_TO_B(mp, rsumblocks);
1262 mp->m_rbmip = mp->m_rsumip = NULL;
1264 * Check that the realtime section is an ok size.
1266 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
1267 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
1268 xfs_warn(mp, "realtime mount -- %llu != %llu",
1269 (unsigned long long) XFS_BB_TO_FSB(mp, d),
1270 (unsigned long long) mp->m_sb.sb_rblocks);
1273 error = xfs_buf_read_uncached(mp->m_rtdev_targp,
1274 d - XFS_FSB_TO_BB(mp, 1),
1275 XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
1277 xfs_warn(mp, "realtime device size check failed");
1285 xfs_rtalloc_count_frextent(
1286 struct xfs_mount *mp,
1287 struct xfs_trans *tp,
1288 const struct xfs_rtalloc_rec *rec,
1291 uint64_t *valp = priv;
1293 *valp += rec->ar_extcount;
1298 * Reinitialize the number of free realtime extents from the realtime bitmap.
1299 * Callers must ensure that there is no other activity in the filesystem.
1302 xfs_rtalloc_reinit_frextents(
1303 struct xfs_mount *mp)
1308 xfs_ilock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
1309 error = xfs_rtalloc_query_all(mp, NULL, xfs_rtalloc_count_frextent,
1311 xfs_iunlock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
1315 spin_lock(&mp->m_sb_lock);
1316 mp->m_sb.sb_frextents = val;
1317 spin_unlock(&mp->m_sb_lock);
1318 percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1323 * Read in the bmbt of an rt metadata inode so that we never have to load them
1324 * at runtime. This enables the use of shared ILOCKs for rtbitmap scans. Use
1325 * an empty transaction to avoid deadlocking on loops in the bmbt.
1328 xfs_rtmount_iread_extents(
1329 struct xfs_inode *ip,
1330 unsigned int lock_class)
1332 struct xfs_trans *tp;
1335 error = xfs_trans_alloc_empty(ip->i_mount, &tp);
1339 xfs_ilock(ip, XFS_ILOCK_EXCL | lock_class);
1341 error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
1345 if (xfs_inode_has_attr_fork(ip)) {
1346 error = xfs_iread_extents(tp, ip, XFS_ATTR_FORK);
1352 xfs_iunlock(ip, XFS_ILOCK_EXCL | lock_class);
1353 xfs_trans_cancel(tp);
1358 * Get the bitmap and summary inodes and the summary cache into the mount
1359 * structure at mount time.
1363 xfs_mount_t *mp) /* file system mount structure */
1365 int error; /* error return value */
1369 error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
1372 ASSERT(mp->m_rbmip != NULL);
1374 error = xfs_rtmount_iread_extents(mp->m_rbmip, XFS_ILOCK_RTBITMAP);
1376 goto out_rele_bitmap;
1378 error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
1380 goto out_rele_bitmap;
1381 ASSERT(mp->m_rsumip != NULL);
1383 error = xfs_rtmount_iread_extents(mp->m_rsumip, XFS_ILOCK_RTSUM);
1385 goto out_rele_summary;
1387 xfs_alloc_rsum_cache(mp, sbp->sb_rbmblocks);
1391 xfs_irele(mp->m_rsumip);
1393 xfs_irele(mp->m_rbmip);
1398 xfs_rtunmount_inodes(
1399 struct xfs_mount *mp)
1401 kmem_free(mp->m_rsum_cache);
1403 xfs_irele(mp->m_rbmip);
1405 xfs_irele(mp->m_rsumip);
1409 * Pick an extent for allocation at the start of a new realtime file.
1410 * Use the sequence number stored in the atime field of the bitmap inode.
1411 * Translate this to a fraction of the rtextents, and return the product
1412 * of rtextents and the fraction.
1413 * The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
1417 xfs_mount_t *mp, /* file system mount point */
1418 xfs_trans_t *tp, /* transaction pointer */
1419 xfs_rtxlen_t len, /* allocation length (rtextents) */
1420 xfs_rtxnum_t *pick) /* result rt extent */
1422 xfs_rtxnum_t b; /* result rtext */
1423 int log2; /* log of sequence number */
1424 uint64_t resid; /* residual after log removed */
1425 uint64_t seq; /* sequence number of file creation */
1426 struct timespec64 ts; /* timespec in inode */
1428 ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
1430 ts = inode_get_atime(VFS_I(mp->m_rbmip));
1431 if (!(mp->m_rbmip->i_diflags & XFS_DIFLAG_NEWRTBM)) {
1432 mp->m_rbmip->i_diflags |= XFS_DIFLAG_NEWRTBM;
1437 if ((log2 = xfs_highbit64(seq)) == -1)
1440 resid = seq - (1ULL << log2);
1441 b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
1443 if (b >= mp->m_sb.sb_rextents)
1444 div64_u64_rem(b, mp->m_sb.sb_rextents, &b);
1445 if (b + len > mp->m_sb.sb_rextents)
1446 b = mp->m_sb.sb_rextents - len;
1448 ts.tv_sec = seq + 1;
1449 inode_set_atime_to_ts(VFS_I(mp->m_rbmip), ts);
1450 xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);