1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
16 #include "xfs_dir2_priv.h"
17 #include "xfs_error.h"
18 #include "xfs_trans.h"
19 #include "xfs_buf_item.h"
21 #include "xfs_dir2_priv.h"
23 static xfs_failaddr_t xfs_dir2_data_freefind_verify(
24 struct xfs_dir2_data_hdr *hdr, struct xfs_dir2_data_free *bf,
25 struct xfs_dir2_data_unused *dup,
26 struct xfs_dir2_data_free **bf_ent);
28 struct xfs_dir2_data_free *
29 xfs_dir2_data_bestfree_p(
31 struct xfs_dir2_data_hdr *hdr)
33 if (xfs_sb_version_hascrc(&mp->m_sb))
34 return ((struct xfs_dir3_data_hdr *)hdr)->best_free;
39 * Pointer to an entry's tag word.
42 xfs_dir2_data_entry_tag_p(
44 struct xfs_dir2_data_entry *dep)
46 return (__be16 *)((char *)dep +
47 xfs_dir2_data_entsize(mp, dep->namelen) - sizeof(__be16));
51 xfs_dir2_data_get_ftype(
53 struct xfs_dir2_data_entry *dep)
55 if (xfs_sb_version_hasftype(&mp->m_sb)) {
56 uint8_t ftype = dep->name[dep->namelen];
58 if (likely(ftype < XFS_DIR3_FT_MAX))
62 return XFS_DIR3_FT_UNKNOWN;
66 xfs_dir2_data_put_ftype(
68 struct xfs_dir2_data_entry *dep,
71 ASSERT(ftype < XFS_DIR3_FT_MAX);
72 ASSERT(dep->namelen != 0);
74 if (xfs_sb_version_hasftype(&mp->m_sb))
75 dep->name[dep->namelen] = ftype;
79 * The number of leaf entries is limited by the size of the block and the amount
80 * of space used by the data entries. We don't know how much space is used by
81 * the data entries yet, so just ensure that the count falls somewhere inside
82 * the block right now.
84 static inline unsigned int
85 xfs_dir2_data_max_leaf_entries(
86 struct xfs_da_geometry *geo)
88 return (geo->blksize - sizeof(struct xfs_dir2_block_tail) -
89 geo->data_entry_offset) /
90 sizeof(struct xfs_dir2_leaf_entry);
94 * Check the consistency of the data block.
95 * The input can also be a block-format directory.
96 * Return NULL if the buffer is good, otherwise the address of the error.
99 __xfs_dir3_data_check(
100 struct xfs_inode *dp, /* incore inode pointer */
101 struct xfs_buf *bp) /* data block's buffer */
103 xfs_dir2_dataptr_t addr; /* addr for leaf lookup */
104 xfs_dir2_data_free_t *bf; /* bestfree table */
105 xfs_dir2_block_tail_t *btp=NULL; /* block tail */
106 int count; /* count of entries found */
107 xfs_dir2_data_hdr_t *hdr; /* data block header */
108 xfs_dir2_data_free_t *dfp; /* bestfree entry */
109 int freeseen; /* mask of bestfrees seen */
110 xfs_dahash_t hash; /* hash of current name */
111 int i; /* leaf index */
112 int lastfree; /* last entry was unused */
113 xfs_dir2_leaf_entry_t *lep=NULL; /* block leaf entries */
114 struct xfs_mount *mp = bp->b_mount;
115 int stale; /* count of stale leaves */
116 struct xfs_name name;
119 struct xfs_da_geometry *geo = mp->m_dir_geo;
122 * If this isn't a directory, something is seriously wrong. Bail out.
124 if (dp && !S_ISDIR(VFS_I(dp)->i_mode))
125 return __this_address;
128 offset = geo->data_entry_offset;
130 switch (hdr->magic) {
131 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
132 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
133 btp = xfs_dir2_block_tail_p(geo, hdr);
134 lep = xfs_dir2_block_leaf_p(btp);
136 if (be32_to_cpu(btp->count) >=
137 xfs_dir2_data_max_leaf_entries(geo))
138 return __this_address;
140 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
141 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
144 return __this_address;
146 end = xfs_dir3_data_end_offset(geo, hdr);
148 return __this_address;
151 * Account for zero bestfree entries.
153 bf = xfs_dir2_data_bestfree_p(mp, hdr);
154 count = lastfree = freeseen = 0;
157 return __this_address;
162 return __this_address;
167 return __this_address;
171 if (be16_to_cpu(bf[0].length) < be16_to_cpu(bf[1].length))
172 return __this_address;
173 if (be16_to_cpu(bf[1].length) < be16_to_cpu(bf[2].length))
174 return __this_address;
176 * Loop over the data/unused entries.
178 while (offset < end) {
179 struct xfs_dir2_data_unused *dup = bp->b_addr + offset;
180 struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
183 * If it's unused, look for the space in the bestfree table.
184 * If we find it, account for that, else make sure it
185 * doesn't need to be there.
187 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
191 return __this_address;
192 if (offset + be16_to_cpu(dup->length) > end)
193 return __this_address;
194 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) !=
196 return __this_address;
197 fa = xfs_dir2_data_freefind_verify(hdr, bf, dup, &dfp);
202 if ((freeseen & (1 << i)) != 0)
203 return __this_address;
206 if (be16_to_cpu(dup->length) >
207 be16_to_cpu(bf[2].length))
208 return __this_address;
210 offset += be16_to_cpu(dup->length);
215 * It's a real entry. Validate the fields.
216 * If this is a block directory then make sure it's
217 * in the leaf section of the block.
218 * The linear search is crude but this is DEBUG code.
220 if (dep->namelen == 0)
221 return __this_address;
222 if (xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)))
223 return __this_address;
224 if (offset + xfs_dir2_data_entsize(mp, dep->namelen) > end)
225 return __this_address;
226 if (be16_to_cpu(*xfs_dir2_data_entry_tag_p(mp, dep)) != offset)
227 return __this_address;
228 if (xfs_dir2_data_get_ftype(mp, dep) >= XFS_DIR3_FT_MAX)
229 return __this_address;
232 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
233 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
234 addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
235 (xfs_dir2_data_aoff_t)
236 ((char *)dep - (char *)hdr));
237 name.name = dep->name;
238 name.len = dep->namelen;
239 hash = xfs_dir2_hashname(mp, &name);
240 for (i = 0; i < be32_to_cpu(btp->count); i++) {
241 if (be32_to_cpu(lep[i].address) == addr &&
242 be32_to_cpu(lep[i].hashval) == hash)
245 if (i >= be32_to_cpu(btp->count))
246 return __this_address;
248 offset += xfs_dir2_data_entsize(mp, dep->namelen);
251 * Need to have seen all the entries and all the bestfree slots.
254 return __this_address;
255 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
256 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
257 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
258 if (lep[i].address ==
259 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
261 if (i > 0 && be32_to_cpu(lep[i].hashval) <
262 be32_to_cpu(lep[i - 1].hashval))
263 return __this_address;
265 if (count != be32_to_cpu(btp->count) - be32_to_cpu(btp->stale))
266 return __this_address;
267 if (stale != be32_to_cpu(btp->stale))
268 return __this_address;
276 struct xfs_inode *dp,
281 fa = __xfs_dir3_data_check(dp, bp);
284 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
285 bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
291 static xfs_failaddr_t
292 xfs_dir3_data_verify(
295 struct xfs_mount *mp = bp->b_mount;
296 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
298 if (!xfs_verify_magic(bp, hdr3->magic))
299 return __this_address;
301 if (xfs_sb_version_hascrc(&mp->m_sb)) {
302 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
303 return __this_address;
304 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
305 return __this_address;
306 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
307 return __this_address;
309 return __xfs_dir3_data_check(NULL, bp);
313 * Readahead of the first block of the directory when it is opened is completely
314 * oblivious to the format of the directory. Hence we can either get a block
315 * format buffer or a data format buffer on readahead.
318 xfs_dir3_data_reada_verify(
321 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
323 switch (hdr->magic) {
324 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
325 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
326 bp->b_ops = &xfs_dir3_block_buf_ops;
327 bp->b_ops->verify_read(bp);
329 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
330 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
331 bp->b_ops = &xfs_dir3_data_buf_ops;
332 bp->b_ops->verify_read(bp);
335 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
341 xfs_dir3_data_read_verify(
344 struct xfs_mount *mp = bp->b_mount;
347 if (xfs_sb_version_hascrc(&mp->m_sb) &&
348 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
349 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
351 fa = xfs_dir3_data_verify(bp);
353 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
358 xfs_dir3_data_write_verify(
361 struct xfs_mount *mp = bp->b_mount;
362 struct xfs_buf_log_item *bip = bp->b_log_item;
363 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
366 fa = xfs_dir3_data_verify(bp);
368 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
372 if (!xfs_sb_version_hascrc(&mp->m_sb))
376 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
378 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
381 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
382 .name = "xfs_dir3_data",
383 .magic = { cpu_to_be32(XFS_DIR2_DATA_MAGIC),
384 cpu_to_be32(XFS_DIR3_DATA_MAGIC) },
385 .verify_read = xfs_dir3_data_read_verify,
386 .verify_write = xfs_dir3_data_write_verify,
387 .verify_struct = xfs_dir3_data_verify,
390 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
391 .name = "xfs_dir3_data_reada",
392 .magic = { cpu_to_be32(XFS_DIR2_DATA_MAGIC),
393 cpu_to_be32(XFS_DIR3_DATA_MAGIC) },
394 .verify_read = xfs_dir3_data_reada_verify,
395 .verify_write = xfs_dir3_data_write_verify,
401 struct xfs_trans *tp,
402 struct xfs_inode *dp,
404 xfs_daddr_t mapped_bno,
405 struct xfs_buf **bpp)
409 err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
410 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
411 if (!err && tp && *bpp)
412 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
417 xfs_dir3_data_readahead(
418 struct xfs_inode *dp,
420 xfs_daddr_t mapped_bno)
422 return xfs_da_reada_buf(dp, bno, mapped_bno,
423 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
427 * Find the bestfree entry that exactly coincides with unused directory space
428 * or a verifier error because the bestfree data are bad.
430 static xfs_failaddr_t
431 xfs_dir2_data_freefind_verify(
432 struct xfs_dir2_data_hdr *hdr,
433 struct xfs_dir2_data_free *bf,
434 struct xfs_dir2_data_unused *dup,
435 struct xfs_dir2_data_free **bf_ent)
437 struct xfs_dir2_data_free *dfp;
438 xfs_dir2_data_aoff_t off;
439 bool matched = false;
440 bool seenzero = false;
443 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
446 * Validate some consistency in the bestfree table.
447 * Check order, non-overlapping entries, and if we find the
448 * one we're looking for it has to be exact.
450 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
453 return __this_address;
458 return __this_address;
459 if (be16_to_cpu(dfp->offset) == off) {
461 if (dfp->length != dup->length)
462 return __this_address;
463 } else if (be16_to_cpu(dfp->offset) > off) {
464 if (off + be16_to_cpu(dup->length) >
465 be16_to_cpu(dfp->offset))
466 return __this_address;
468 if (be16_to_cpu(dfp->offset) +
469 be16_to_cpu(dfp->length) > off)
470 return __this_address;
473 be16_to_cpu(dfp->length) < be16_to_cpu(dup->length))
474 return __this_address;
476 be16_to_cpu(dfp[-1].length) < be16_to_cpu(dfp[0].length))
477 return __this_address;
480 /* Looks ok so far; now try to match up with a bestfree entry. */
481 *bf_ent = xfs_dir2_data_freefind(hdr, bf, dup);
486 * Given a data block and an unused entry from that block,
487 * return the bestfree entry if any that corresponds to it.
489 xfs_dir2_data_free_t *
490 xfs_dir2_data_freefind(
491 struct xfs_dir2_data_hdr *hdr, /* data block header */
492 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
493 struct xfs_dir2_data_unused *dup) /* unused space */
495 xfs_dir2_data_free_t *dfp; /* bestfree entry */
496 xfs_dir2_data_aoff_t off; /* offset value needed */
498 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
501 * If this is smaller than the smallest bestfree entry,
502 * it can't be there since they're sorted.
504 if (be16_to_cpu(dup->length) <
505 be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
508 * Look at the three bestfree entries for our guy.
510 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
513 if (be16_to_cpu(dfp->offset) == off)
517 * Didn't find it. This only happens if there are duplicate lengths.
523 * Insert an unused-space entry into the bestfree table.
525 xfs_dir2_data_free_t * /* entry inserted */
526 xfs_dir2_data_freeinsert(
527 struct xfs_dir2_data_hdr *hdr, /* data block pointer */
528 struct xfs_dir2_data_free *dfp, /* bestfree table pointer */
529 struct xfs_dir2_data_unused *dup, /* unused space */
530 int *loghead) /* log the data header (out) */
532 xfs_dir2_data_free_t new; /* new bestfree entry */
534 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
535 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
536 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
537 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
539 new.length = dup->length;
540 new.offset = cpu_to_be16((char *)dup - (char *)hdr);
543 * Insert at position 0, 1, or 2; or not at all.
545 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
552 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
558 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
567 * Remove a bestfree entry from the table.
570 xfs_dir2_data_freeremove(
571 struct xfs_dir2_data_hdr *hdr, /* data block header */
572 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
573 struct xfs_dir2_data_free *dfp, /* bestfree entry pointer */
574 int *loghead) /* out: log data header */
577 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
578 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
579 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
580 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
583 * It's the first entry, slide the next 2 up.
590 * It's the second entry, slide the 3rd entry up.
592 else if (dfp == &bf[1])
595 * Must be the last entry.
598 ASSERT(dfp == &bf[2]);
600 * Clear the 3rd entry, must be zero now.
608 * Given a data block, reconstruct its bestfree map.
611 xfs_dir2_data_freescan(
612 struct xfs_mount *mp,
613 struct xfs_dir2_data_hdr *hdr,
616 struct xfs_da_geometry *geo = mp->m_dir_geo;
617 struct xfs_dir2_data_free *bf = xfs_dir2_data_bestfree_p(mp, hdr);
619 unsigned int offset = geo->data_entry_offset;
622 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
623 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
624 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
625 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
628 * Start by clearing the table.
630 memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
633 end = xfs_dir3_data_end_offset(geo, addr);
634 while (offset < end) {
635 struct xfs_dir2_data_unused *dup = addr + offset;
636 struct xfs_dir2_data_entry *dep = addr + offset;
639 * If it's a free entry, insert it.
641 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
643 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
644 xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
645 offset += be16_to_cpu(dup->length);
650 * For active entries, check their tags and skip them.
653 be16_to_cpu(*xfs_dir2_data_entry_tag_p(mp, dep)));
654 offset += xfs_dir2_data_entsize(mp, dep->namelen);
659 * Initialize a data block at the given block number in the directory.
660 * Give back the buffer for the created block.
664 struct xfs_da_args *args, /* directory operation args */
665 xfs_dir2_db_t blkno, /* logical dir block number */
666 struct xfs_buf **bpp) /* output block buffer */
668 struct xfs_trans *tp = args->trans;
669 struct xfs_inode *dp = args->dp;
670 struct xfs_mount *mp = dp->i_mount;
671 struct xfs_da_geometry *geo = args->geo;
673 struct xfs_dir2_data_hdr *hdr;
674 struct xfs_dir2_data_unused *dup;
675 struct xfs_dir2_data_free *bf;
680 * Get the buffer set up for the block.
682 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
683 -1, &bp, XFS_DATA_FORK);
686 bp->b_ops = &xfs_dir3_data_buf_ops;
687 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
690 * Initialize the header.
693 if (xfs_sb_version_hascrc(&mp->m_sb)) {
694 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
696 memset(hdr3, 0, sizeof(*hdr3));
697 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
698 hdr3->blkno = cpu_to_be64(bp->b_bn);
699 hdr3->owner = cpu_to_be64(dp->i_ino);
700 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
703 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
705 bf = xfs_dir2_data_bestfree_p(mp, hdr);
706 bf[0].offset = cpu_to_be16(geo->data_entry_offset);
707 bf[0].length = cpu_to_be16(geo->blksize - geo->data_entry_offset);
708 for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
714 * Set up an unused entry for the block's body.
716 dup = bp->b_addr + geo->data_entry_offset;
717 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
718 dup->length = bf[0].length;
719 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
722 * Log it and return it.
724 xfs_dir2_data_log_header(args, bp);
725 xfs_dir2_data_log_unused(args, bp, dup);
731 * Log an active data entry from the block.
734 xfs_dir2_data_log_entry(
735 struct xfs_da_args *args,
737 xfs_dir2_data_entry_t *dep) /* data entry pointer */
739 struct xfs_mount *mp = bp->b_mount;
740 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
742 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
743 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
744 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
745 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
747 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
748 (uint)((char *)(xfs_dir2_data_entry_tag_p(mp, dep) + 1) -
753 * Log a data block header.
756 xfs_dir2_data_log_header(
757 struct xfs_da_args *args,
761 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
763 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
764 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
765 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
766 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
769 xfs_trans_log_buf(args->trans, bp, 0, args->geo->data_entry_offset - 1);
773 * Log a data unused entry.
776 xfs_dir2_data_log_unused(
777 struct xfs_da_args *args,
779 xfs_dir2_data_unused_t *dup) /* data unused pointer */
781 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
783 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
784 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
785 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
786 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
789 * Log the first part of the unused entry.
791 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
792 (uint)((char *)&dup->length + sizeof(dup->length) -
795 * Log the end (tag) of the unused entry.
797 xfs_trans_log_buf(args->trans, bp,
798 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
799 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
800 sizeof(xfs_dir2_data_off_t) - 1));
804 * Make a byte range in the data block unused.
805 * Its current contents are unimportant.
808 xfs_dir2_data_make_free(
809 struct xfs_da_args *args,
811 xfs_dir2_data_aoff_t offset, /* starting byte offset */
812 xfs_dir2_data_aoff_t len, /* length in bytes */
813 int *needlogp, /* out: log header */
814 int *needscanp) /* out: regen bestfree */
816 xfs_dir2_data_hdr_t *hdr; /* data block pointer */
817 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
818 int needscan; /* need to regen bestfree */
819 xfs_dir2_data_unused_t *newdup; /* new unused entry */
820 xfs_dir2_data_unused_t *postdup; /* unused entry after us */
821 xfs_dir2_data_unused_t *prevdup; /* unused entry before us */
823 struct xfs_dir2_data_free *bf;
828 * Figure out where the end of the data area is.
830 end = xfs_dir3_data_end_offset(args->geo, hdr);
834 * If this isn't the start of the block, then back up to
835 * the previous entry and see if it's free.
837 if (offset > args->geo->data_entry_offset) {
838 __be16 *tagp; /* tag just before us */
840 tagp = (__be16 *)((char *)hdr + offset) - 1;
841 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
842 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
847 * If this isn't the end of the block, see if the entry after
850 if (offset + len < end) {
852 (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
853 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
857 ASSERT(*needscanp == 0);
860 * Previous and following entries are both free,
861 * merge everything into a single free entry.
863 bf = xfs_dir2_data_bestfree_p(args->dp->i_mount, hdr);
864 if (prevdup && postdup) {
865 xfs_dir2_data_free_t *dfp2; /* another bestfree pointer */
868 * See if prevdup and/or postdup are in bestfree table.
870 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
871 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
873 * We need a rescan unless there are exactly 2 free entries
874 * namely our two. Then we know what's happening, otherwise
875 * since the third bestfree is there, there might be more
878 needscan = (bf[2].length != 0);
880 * Fix up the new big freespace.
882 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
883 *xfs_dir2_data_unused_tag_p(prevdup) =
884 cpu_to_be16((char *)prevdup - (char *)hdr);
885 xfs_dir2_data_log_unused(args, bp, prevdup);
888 * Has to be the case that entries 0 and 1 are
889 * dfp and dfp2 (don't know which is which), and
891 * Remove entry 1 first then entry 0.
899 xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
900 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
902 * Now insert the new entry.
904 dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
906 ASSERT(dfp == &bf[0]);
907 ASSERT(dfp->length == prevdup->length);
908 ASSERT(!dfp[1].length);
909 ASSERT(!dfp[2].length);
913 * The entry before us is free, merge with it.
916 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
917 be16_add_cpu(&prevdup->length, len);
918 *xfs_dir2_data_unused_tag_p(prevdup) =
919 cpu_to_be16((char *)prevdup - (char *)hdr);
920 xfs_dir2_data_log_unused(args, bp, prevdup);
922 * If the previous entry was in the table, the new entry
923 * is longer, so it will be in the table too. Remove
924 * the old one and add the new one.
927 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
928 xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
931 * Otherwise we need a scan if the new entry is big enough.
934 needscan = be16_to_cpu(prevdup->length) >
935 be16_to_cpu(bf[2].length);
939 * The following entry is free, merge with it.
942 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
943 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
944 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
945 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
946 *xfs_dir2_data_unused_tag_p(newdup) =
947 cpu_to_be16((char *)newdup - (char *)hdr);
948 xfs_dir2_data_log_unused(args, bp, newdup);
950 * If the following entry was in the table, the new entry
951 * is longer, so it will be in the table too. Remove
952 * the old one and add the new one.
955 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
956 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
959 * Otherwise we need a scan if the new entry is big enough.
962 needscan = be16_to_cpu(newdup->length) >
963 be16_to_cpu(bf[2].length);
967 * Neither neighbor is free. Make a new entry.
970 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
971 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
972 newdup->length = cpu_to_be16(len);
973 *xfs_dir2_data_unused_tag_p(newdup) =
974 cpu_to_be16((char *)newdup - (char *)hdr);
975 xfs_dir2_data_log_unused(args, bp, newdup);
976 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
978 *needscanp = needscan;
981 /* Check our free data for obvious signs of corruption. */
982 static inline xfs_failaddr_t
983 xfs_dir2_data_check_free(
984 struct xfs_dir2_data_hdr *hdr,
985 struct xfs_dir2_data_unused *dup,
986 xfs_dir2_data_aoff_t offset,
987 xfs_dir2_data_aoff_t len)
989 if (hdr->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC) &&
990 hdr->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC) &&
991 hdr->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) &&
992 hdr->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
993 return __this_address;
994 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG)
995 return __this_address;
996 if (offset < (char *)dup - (char *)hdr)
997 return __this_address;
998 if (offset + len > (char *)dup + be16_to_cpu(dup->length) - (char *)hdr)
999 return __this_address;
1000 if ((char *)dup - (char *)hdr !=
1001 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)))
1002 return __this_address;
1006 /* Sanity-check a new bestfree entry. */
1007 static inline xfs_failaddr_t
1008 xfs_dir2_data_check_new_free(
1009 struct xfs_dir2_data_hdr *hdr,
1010 struct xfs_dir2_data_free *dfp,
1011 struct xfs_dir2_data_unused *newdup)
1014 return __this_address;
1015 if (dfp->length != newdup->length)
1016 return __this_address;
1017 if (be16_to_cpu(dfp->offset) != (char *)newdup - (char *)hdr)
1018 return __this_address;
1023 * Take a byte range out of an existing unused space and make it un-free.
1026 xfs_dir2_data_use_free(
1027 struct xfs_da_args *args,
1029 xfs_dir2_data_unused_t *dup, /* unused entry */
1030 xfs_dir2_data_aoff_t offset, /* starting offset to use */
1031 xfs_dir2_data_aoff_t len, /* length to use */
1032 int *needlogp, /* out: need to log header */
1033 int *needscanp) /* out: need regen bestfree */
1035 xfs_dir2_data_hdr_t *hdr; /* data block header */
1036 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
1037 xfs_dir2_data_unused_t *newdup; /* new unused entry */
1038 xfs_dir2_data_unused_t *newdup2; /* another new unused entry */
1039 struct xfs_dir2_data_free *bf;
1041 int matchback; /* matches end of freespace */
1042 int matchfront; /* matches start of freespace */
1043 int needscan; /* need to regen bestfree */
1044 int oldlen; /* old unused entry's length */
1047 fa = xfs_dir2_data_check_free(hdr, dup, offset, len);
1051 * Look up the entry in the bestfree table.
1053 oldlen = be16_to_cpu(dup->length);
1054 bf = xfs_dir2_data_bestfree_p(args->dp->i_mount, hdr);
1055 dfp = xfs_dir2_data_freefind(hdr, bf, dup);
1056 ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
1058 * Check for alignment with front and back of the entry.
1060 matchfront = (char *)dup - (char *)hdr == offset;
1061 matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
1062 ASSERT(*needscanp == 0);
1065 * If we matched it exactly we just need to get rid of it from
1066 * the bestfree table.
1068 if (matchfront && matchback) {
1070 needscan = (bf[2].offset != 0);
1072 xfs_dir2_data_freeremove(hdr, bf, dfp,
1077 * We match the first part of the entry.
1078 * Make a new entry with the remaining freespace.
1080 else if (matchfront) {
1081 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1082 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1083 newdup->length = cpu_to_be16(oldlen - len);
1084 *xfs_dir2_data_unused_tag_p(newdup) =
1085 cpu_to_be16((char *)newdup - (char *)hdr);
1086 xfs_dir2_data_log_unused(args, bp, newdup);
1088 * If it was in the table, remove it and add the new one.
1091 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1092 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1094 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup);
1098 * If we got inserted at the last slot,
1099 * that means we don't know if there was a better
1100 * choice for the last slot, or not. Rescan.
1102 needscan = dfp == &bf[2];
1106 * We match the last part of the entry.
1107 * Trim the allocated space off the tail of the entry.
1109 else if (matchback) {
1111 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1112 *xfs_dir2_data_unused_tag_p(newdup) =
1113 cpu_to_be16((char *)newdup - (char *)hdr);
1114 xfs_dir2_data_log_unused(args, bp, newdup);
1116 * If it was in the table, remove it and add the new one.
1119 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1120 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1122 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup);
1126 * If we got inserted at the last slot,
1127 * that means we don't know if there was a better
1128 * choice for the last slot, or not. Rescan.
1130 needscan = dfp == &bf[2];
1134 * Poking out the middle of an entry.
1135 * Make two new entries.
1139 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1140 *xfs_dir2_data_unused_tag_p(newdup) =
1141 cpu_to_be16((char *)newdup - (char *)hdr);
1142 xfs_dir2_data_log_unused(args, bp, newdup);
1143 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1144 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1145 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1146 *xfs_dir2_data_unused_tag_p(newdup2) =
1147 cpu_to_be16((char *)newdup2 - (char *)hdr);
1148 xfs_dir2_data_log_unused(args, bp, newdup2);
1150 * If the old entry was in the table, we need to scan
1151 * if the 3rd entry was valid, since these entries
1152 * are smaller than the old one.
1153 * If we don't need to scan that means there were 1 or 2
1154 * entries in the table, and removing the old and adding
1155 * the 2 new will work.
1158 needscan = (bf[2].length != 0);
1160 xfs_dir2_data_freeremove(hdr, bf, dfp,
1162 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1164 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1169 *needscanp = needscan;
1172 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, args->dp->i_mount,
1173 hdr, sizeof(*hdr), __FILE__, __LINE__, fa);
1174 return -EFSCORRUPTED;
1177 /* Find the end of the entry data in a data/block format dir block. */
1179 xfs_dir3_data_end_offset(
1180 struct xfs_da_geometry *geo,
1181 struct xfs_dir2_data_hdr *hdr)
1185 switch (hdr->magic) {
1186 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
1187 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
1188 p = xfs_dir2_block_leaf_p(xfs_dir2_block_tail_p(geo, hdr));
1189 return p - (void *)hdr;
1190 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
1191 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
1192 return geo->blksize;