xfs: fix overflow in xfs_attr3_leaf_verify
[linux/fpc-iii.git] / fs / xfs / libxfs / xfs_attr_leaf.c
blob2652d00842d6ba8c6479f816765c87dfc622d1cb
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_fs.h"
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_bit.h"
14 #include "xfs_sb.h"
15 #include "xfs_mount.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
19 #include "xfs_trans.h"
20 #include "xfs_inode_item.h"
21 #include "xfs_bmap_btree.h"
22 #include "xfs_bmap.h"
23 #include "xfs_attr_sf.h"
24 #include "xfs_attr_remote.h"
25 #include "xfs_attr.h"
26 #include "xfs_attr_leaf.h"
27 #include "xfs_error.h"
28 #include "xfs_trace.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_cksum.h"
31 #include "xfs_dir2.h"
32 #include "xfs_log.h"
36 * xfs_attr_leaf.c
38 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
41 /*========================================================================
42 * Function prototypes for the kernel.
43 *========================================================================*/
46 * Routines used for growing the Btree.
48 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
49 xfs_dablk_t which_block, struct xfs_buf **bpp);
50 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
51 struct xfs_attr3_icleaf_hdr *ichdr,
52 struct xfs_da_args *args, int freemap_index);
53 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
54 struct xfs_attr3_icleaf_hdr *ichdr,
55 struct xfs_buf *leaf_buffer);
56 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
57 xfs_da_state_blk_t *blk1,
58 xfs_da_state_blk_t *blk2);
59 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
60 xfs_da_state_blk_t *leaf_blk_1,
61 struct xfs_attr3_icleaf_hdr *ichdr1,
62 xfs_da_state_blk_t *leaf_blk_2,
63 struct xfs_attr3_icleaf_hdr *ichdr2,
64 int *number_entries_in_blk1,
65 int *number_usedbytes_in_blk1);
68 * Utility routines.
70 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
71 struct xfs_attr_leafblock *src_leaf,
72 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
73 struct xfs_attr_leafblock *dst_leaf,
74 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
75 int move_count);
76 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
79 * attr3 block 'firstused' conversion helpers.
81 * firstused refers to the offset of the first used byte of the nameval region
82 * of an attr leaf block. The region starts at the tail of the block and expands
83 * backwards towards the middle. As such, firstused is initialized to the block
84 * size for an empty leaf block and is reduced from there.
86 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
87 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
88 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
89 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
90 * the attr block size. The following helpers manage the conversion between the
91 * in-core and on-disk formats.
94 static void
95 xfs_attr3_leaf_firstused_from_disk(
96 struct xfs_da_geometry *geo,
97 struct xfs_attr3_icleaf_hdr *to,
98 struct xfs_attr_leafblock *from)
100 struct xfs_attr3_leaf_hdr *hdr3;
102 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
103 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
104 to->firstused = be16_to_cpu(hdr3->firstused);
105 } else {
106 to->firstused = be16_to_cpu(from->hdr.firstused);
110 * Convert from the magic fsb size value to actual blocksize. This
111 * should only occur for empty blocks when the block size overflows
112 * 16-bits.
114 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
115 ASSERT(!to->count && !to->usedbytes);
116 ASSERT(geo->blksize > USHRT_MAX);
117 to->firstused = geo->blksize;
121 static void
122 xfs_attr3_leaf_firstused_to_disk(
123 struct xfs_da_geometry *geo,
124 struct xfs_attr_leafblock *to,
125 struct xfs_attr3_icleaf_hdr *from)
127 struct xfs_attr3_leaf_hdr *hdr3;
128 uint32_t firstused;
130 /* magic value should only be seen on disk */
131 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
134 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
135 * value. This only overflows at the max supported value of 64k. Use the
136 * magic on-disk value to represent block size in this case.
138 firstused = from->firstused;
139 if (firstused > USHRT_MAX) {
140 ASSERT(from->firstused == geo->blksize);
141 firstused = XFS_ATTR3_LEAF_NULLOFF;
144 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
145 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
146 hdr3->firstused = cpu_to_be16(firstused);
147 } else {
148 to->hdr.firstused = cpu_to_be16(firstused);
152 void
153 xfs_attr3_leaf_hdr_from_disk(
154 struct xfs_da_geometry *geo,
155 struct xfs_attr3_icleaf_hdr *to,
156 struct xfs_attr_leafblock *from)
158 int i;
160 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
161 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
163 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
164 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
166 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
167 to->back = be32_to_cpu(hdr3->info.hdr.back);
168 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
169 to->count = be16_to_cpu(hdr3->count);
170 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
171 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
172 to->holes = hdr3->holes;
174 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
175 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
176 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
178 return;
180 to->forw = be32_to_cpu(from->hdr.info.forw);
181 to->back = be32_to_cpu(from->hdr.info.back);
182 to->magic = be16_to_cpu(from->hdr.info.magic);
183 to->count = be16_to_cpu(from->hdr.count);
184 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
185 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
186 to->holes = from->hdr.holes;
188 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
189 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
190 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
194 void
195 xfs_attr3_leaf_hdr_to_disk(
196 struct xfs_da_geometry *geo,
197 struct xfs_attr_leafblock *to,
198 struct xfs_attr3_icleaf_hdr *from)
200 int i;
202 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
203 from->magic == XFS_ATTR3_LEAF_MAGIC);
205 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
206 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
208 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
209 hdr3->info.hdr.back = cpu_to_be32(from->back);
210 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
211 hdr3->count = cpu_to_be16(from->count);
212 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
213 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
214 hdr3->holes = from->holes;
215 hdr3->pad1 = 0;
217 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
218 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
219 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
221 return;
223 to->hdr.info.forw = cpu_to_be32(from->forw);
224 to->hdr.info.back = cpu_to_be32(from->back);
225 to->hdr.info.magic = cpu_to_be16(from->magic);
226 to->hdr.count = cpu_to_be16(from->count);
227 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
228 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
229 to->hdr.holes = from->holes;
230 to->hdr.pad1 = 0;
232 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
233 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
234 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
238 static xfs_failaddr_t
239 xfs_attr3_leaf_verify(
240 struct xfs_buf *bp)
242 struct xfs_attr3_icleaf_hdr ichdr;
243 struct xfs_mount *mp = bp->b_target->bt_mount;
244 struct xfs_attr_leafblock *leaf = bp->b_addr;
245 struct xfs_attr_leaf_entry *entries;
246 uint32_t end; /* must be 32bit - see below */
247 int i;
249 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
251 if (xfs_sb_version_hascrc(&mp->m_sb)) {
252 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
254 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
255 return __this_address;
257 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
258 return __this_address;
259 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
260 return __this_address;
261 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
262 return __this_address;
263 } else {
264 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
265 return __this_address;
268 * In recovery there is a transient state where count == 0 is valid
269 * because we may have transitioned an empty shortform attr to a leaf
270 * if the attr didn't fit in shortform.
272 if (!xfs_log_in_recovery(mp) && ichdr.count == 0)
273 return __this_address;
276 * firstused is the block offset of the first name info structure.
277 * Make sure it doesn't go off the block or crash into the header.
279 if (ichdr.firstused > mp->m_attr_geo->blksize)
280 return __this_address;
281 if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
282 return __this_address;
284 /* Make sure the entries array doesn't crash into the name info. */
285 entries = xfs_attr3_leaf_entryp(bp->b_addr);
286 if ((char *)&entries[ichdr.count] >
287 (char *)bp->b_addr + ichdr.firstused)
288 return __this_address;
290 /* XXX: need to range check rest of attr header values */
291 /* XXX: hash order check? */
294 * Quickly check the freemap information. Attribute data has to be
295 * aligned to 4-byte boundaries, and likewise for the free space.
297 * Note that for 64k block size filesystems, the freemap entries cannot
298 * overflow as they are only be16 fields. However, when checking end
299 * pointer of the freemap, we have to be careful to detect overflows and
300 * so use uint32_t for those checks.
302 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
303 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
304 return __this_address;
305 if (ichdr.freemap[i].base & 0x3)
306 return __this_address;
307 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
308 return __this_address;
309 if (ichdr.freemap[i].size & 0x3)
310 return __this_address;
312 /* be care of 16 bit overflows here */
313 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
314 if (end < ichdr.freemap[i].base)
315 return __this_address;
316 if (end > mp->m_attr_geo->blksize)
317 return __this_address;
320 return NULL;
323 static void
324 xfs_attr3_leaf_write_verify(
325 struct xfs_buf *bp)
327 struct xfs_mount *mp = bp->b_target->bt_mount;
328 struct xfs_buf_log_item *bip = bp->b_log_item;
329 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
330 xfs_failaddr_t fa;
332 fa = xfs_attr3_leaf_verify(bp);
333 if (fa) {
334 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
335 return;
338 if (!xfs_sb_version_hascrc(&mp->m_sb))
339 return;
341 if (bip)
342 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
344 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
348 * leaf/node format detection on trees is sketchy, so a node read can be done on
349 * leaf level blocks when detection identifies the tree as a node format tree
350 * incorrectly. In this case, we need to swap the verifier to match the correct
351 * format of the block being read.
353 static void
354 xfs_attr3_leaf_read_verify(
355 struct xfs_buf *bp)
357 struct xfs_mount *mp = bp->b_target->bt_mount;
358 xfs_failaddr_t fa;
360 if (xfs_sb_version_hascrc(&mp->m_sb) &&
361 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
362 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
363 else {
364 fa = xfs_attr3_leaf_verify(bp);
365 if (fa)
366 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
370 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
371 .name = "xfs_attr3_leaf",
372 .verify_read = xfs_attr3_leaf_read_verify,
373 .verify_write = xfs_attr3_leaf_write_verify,
374 .verify_struct = xfs_attr3_leaf_verify,
378 xfs_attr3_leaf_read(
379 struct xfs_trans *tp,
380 struct xfs_inode *dp,
381 xfs_dablk_t bno,
382 xfs_daddr_t mappedbno,
383 struct xfs_buf **bpp)
385 int err;
387 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
388 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
389 if (!err && tp && *bpp)
390 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
391 return err;
394 /*========================================================================
395 * Namespace helper routines
396 *========================================================================*/
399 * If namespace bits don't match return 0.
400 * If all match then return 1.
402 STATIC int
403 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
405 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
409 /*========================================================================
410 * External routines when attribute fork size < XFS_LITINO(mp).
411 *========================================================================*/
414 * Query whether the requested number of additional bytes of extended
415 * attribute space will be able to fit inline.
417 * Returns zero if not, else the di_forkoff fork offset to be used in the
418 * literal area for attribute data once the new bytes have been added.
420 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
421 * special case for dev/uuid inodes, they have fixed size data forks.
424 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
426 int offset;
427 int minforkoff; /* lower limit on valid forkoff locations */
428 int maxforkoff; /* upper limit on valid forkoff locations */
429 int dsize;
430 xfs_mount_t *mp = dp->i_mount;
432 /* rounded down */
433 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
435 if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
436 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
437 return (offset >= minforkoff) ? minforkoff : 0;
441 * If the requested numbers of bytes is smaller or equal to the
442 * current attribute fork size we can always proceed.
444 * Note that if_bytes in the data fork might actually be larger than
445 * the current data fork size is due to delalloc extents. In that
446 * case either the extent count will go down when they are converted
447 * to real extents, or the delalloc conversion will take care of the
448 * literal area rebalancing.
450 if (bytes <= XFS_IFORK_ASIZE(dp))
451 return dp->i_d.di_forkoff;
454 * For attr2 we can try to move the forkoff if there is space in the
455 * literal area, but for the old format we are done if there is no
456 * space in the fixed attribute fork.
458 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
459 return 0;
461 dsize = dp->i_df.if_bytes;
463 switch (dp->i_d.di_format) {
464 case XFS_DINODE_FMT_EXTENTS:
466 * If there is no attr fork and the data fork is extents,
467 * determine if creating the default attr fork will result
468 * in the extents form migrating to btree. If so, the
469 * minimum offset only needs to be the space required for
470 * the btree root.
472 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
473 xfs_default_attroffset(dp))
474 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
475 break;
476 case XFS_DINODE_FMT_BTREE:
478 * If we have a data btree then keep forkoff if we have one,
479 * otherwise we are adding a new attr, so then we set
480 * minforkoff to where the btree root can finish so we have
481 * plenty of room for attrs
483 if (dp->i_d.di_forkoff) {
484 if (offset < dp->i_d.di_forkoff)
485 return 0;
486 return dp->i_d.di_forkoff;
488 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
489 break;
493 * A data fork btree root must have space for at least
494 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
496 minforkoff = max(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
497 minforkoff = roundup(minforkoff, 8) >> 3;
499 /* attr fork btree root can have at least this many key/ptr pairs */
500 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
501 XFS_BMDR_SPACE_CALC(MINABTPTRS);
502 maxforkoff = maxforkoff >> 3; /* rounded down */
504 if (offset >= maxforkoff)
505 return maxforkoff;
506 if (offset >= minforkoff)
507 return offset;
508 return 0;
512 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
514 STATIC void
515 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
517 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
518 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
519 spin_lock(&mp->m_sb_lock);
520 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
521 xfs_sb_version_addattr2(&mp->m_sb);
522 spin_unlock(&mp->m_sb_lock);
523 xfs_log_sb(tp);
524 } else
525 spin_unlock(&mp->m_sb_lock);
530 * Create the initial contents of a shortform attribute list.
532 void
533 xfs_attr_shortform_create(xfs_da_args_t *args)
535 xfs_attr_sf_hdr_t *hdr;
536 xfs_inode_t *dp;
537 struct xfs_ifork *ifp;
539 trace_xfs_attr_sf_create(args);
541 dp = args->dp;
542 ASSERT(dp != NULL);
543 ifp = dp->i_afp;
544 ASSERT(ifp != NULL);
545 ASSERT(ifp->if_bytes == 0);
546 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
547 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
548 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
549 ifp->if_flags |= XFS_IFINLINE;
550 } else {
551 ASSERT(ifp->if_flags & XFS_IFINLINE);
553 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
554 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
555 hdr->count = 0;
556 hdr->totsize = cpu_to_be16(sizeof(*hdr));
557 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
561 * Add a name/value pair to the shortform attribute list.
562 * Overflow from the inode has already been checked for.
564 void
565 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
567 xfs_attr_shortform_t *sf;
568 xfs_attr_sf_entry_t *sfe;
569 int i, offset, size;
570 xfs_mount_t *mp;
571 xfs_inode_t *dp;
572 struct xfs_ifork *ifp;
574 trace_xfs_attr_sf_add(args);
576 dp = args->dp;
577 mp = dp->i_mount;
578 dp->i_d.di_forkoff = forkoff;
580 ifp = dp->i_afp;
581 ASSERT(ifp->if_flags & XFS_IFINLINE);
582 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
583 sfe = &sf->list[0];
584 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
585 #ifdef DEBUG
586 if (sfe->namelen != args->namelen)
587 continue;
588 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
589 continue;
590 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
591 continue;
592 ASSERT(0);
593 #endif
596 offset = (char *)sfe - (char *)sf;
597 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
598 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
599 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
600 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
602 sfe->namelen = args->namelen;
603 sfe->valuelen = args->valuelen;
604 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
605 memcpy(sfe->nameval, args->name, args->namelen);
606 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
607 sf->hdr.count++;
608 be16_add_cpu(&sf->hdr.totsize, size);
609 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
611 xfs_sbversion_add_attr2(mp, args->trans);
615 * After the last attribute is removed revert to original inode format,
616 * making all literal area available to the data fork once more.
618 void
619 xfs_attr_fork_remove(
620 struct xfs_inode *ip,
621 struct xfs_trans *tp)
623 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
624 ip->i_d.di_forkoff = 0;
625 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
627 ASSERT(ip->i_d.di_anextents == 0);
628 ASSERT(ip->i_afp == NULL);
630 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
634 * Remove an attribute from the shortform attribute list structure.
637 xfs_attr_shortform_remove(xfs_da_args_t *args)
639 xfs_attr_shortform_t *sf;
640 xfs_attr_sf_entry_t *sfe;
641 int base, size=0, end, totsize, i;
642 xfs_mount_t *mp;
643 xfs_inode_t *dp;
645 trace_xfs_attr_sf_remove(args);
647 dp = args->dp;
648 mp = dp->i_mount;
649 base = sizeof(xfs_attr_sf_hdr_t);
650 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
651 sfe = &sf->list[0];
652 end = sf->hdr.count;
653 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
654 base += size, i++) {
655 size = XFS_ATTR_SF_ENTSIZE(sfe);
656 if (sfe->namelen != args->namelen)
657 continue;
658 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
659 continue;
660 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
661 continue;
662 break;
664 if (i == end)
665 return -ENOATTR;
668 * Fix up the attribute fork data, covering the hole
670 end = base + size;
671 totsize = be16_to_cpu(sf->hdr.totsize);
672 if (end != totsize)
673 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
674 sf->hdr.count--;
675 be16_add_cpu(&sf->hdr.totsize, -size);
678 * Fix up the start offset of the attribute fork
680 totsize -= size;
681 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
682 (mp->m_flags & XFS_MOUNT_ATTR2) &&
683 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
684 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
685 xfs_attr_fork_remove(dp, args->trans);
686 } else {
687 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
688 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
689 ASSERT(dp->i_d.di_forkoff);
690 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
691 (args->op_flags & XFS_DA_OP_ADDNAME) ||
692 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
693 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
694 xfs_trans_log_inode(args->trans, dp,
695 XFS_ILOG_CORE | XFS_ILOG_ADATA);
698 xfs_sbversion_add_attr2(mp, args->trans);
700 return 0;
704 * Look up a name in a shortform attribute list structure.
706 /*ARGSUSED*/
708 xfs_attr_shortform_lookup(xfs_da_args_t *args)
710 xfs_attr_shortform_t *sf;
711 xfs_attr_sf_entry_t *sfe;
712 int i;
713 struct xfs_ifork *ifp;
715 trace_xfs_attr_sf_lookup(args);
717 ifp = args->dp->i_afp;
718 ASSERT(ifp->if_flags & XFS_IFINLINE);
719 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
720 sfe = &sf->list[0];
721 for (i = 0; i < sf->hdr.count;
722 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
723 if (sfe->namelen != args->namelen)
724 continue;
725 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
726 continue;
727 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
728 continue;
729 return -EEXIST;
731 return -ENOATTR;
735 * Look up a name in a shortform attribute list structure.
737 /*ARGSUSED*/
739 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
741 xfs_attr_shortform_t *sf;
742 xfs_attr_sf_entry_t *sfe;
743 int i;
745 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
746 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
747 sfe = &sf->list[0];
748 for (i = 0; i < sf->hdr.count;
749 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
750 if (sfe->namelen != args->namelen)
751 continue;
752 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
753 continue;
754 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
755 continue;
756 if (args->flags & ATTR_KERNOVAL) {
757 args->valuelen = sfe->valuelen;
758 return -EEXIST;
760 if (args->valuelen < sfe->valuelen) {
761 args->valuelen = sfe->valuelen;
762 return -ERANGE;
764 args->valuelen = sfe->valuelen;
765 memcpy(args->value, &sfe->nameval[args->namelen],
766 args->valuelen);
767 return -EEXIST;
769 return -ENOATTR;
773 * Convert from using the shortform to the leaf. On success, return the
774 * buffer so that we can keep it locked until we're totally done with it.
777 xfs_attr_shortform_to_leaf(
778 struct xfs_da_args *args,
779 struct xfs_buf **leaf_bp)
781 struct xfs_inode *dp;
782 struct xfs_attr_shortform *sf;
783 struct xfs_attr_sf_entry *sfe;
784 struct xfs_da_args nargs;
785 char *tmpbuffer;
786 int error, i, size;
787 xfs_dablk_t blkno;
788 struct xfs_buf *bp;
789 struct xfs_ifork *ifp;
791 trace_xfs_attr_sf_to_leaf(args);
793 dp = args->dp;
794 ifp = dp->i_afp;
795 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
796 size = be16_to_cpu(sf->hdr.totsize);
797 tmpbuffer = kmem_alloc(size, KM_SLEEP);
798 ASSERT(tmpbuffer != NULL);
799 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
800 sf = (xfs_attr_shortform_t *)tmpbuffer;
802 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
803 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
805 bp = NULL;
806 error = xfs_da_grow_inode(args, &blkno);
807 if (error) {
809 * If we hit an IO error middle of the transaction inside
810 * grow_inode(), we may have inconsistent data. Bail out.
812 if (error == -EIO)
813 goto out;
814 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
815 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
816 goto out;
819 ASSERT(blkno == 0);
820 error = xfs_attr3_leaf_create(args, blkno, &bp);
821 if (error) {
822 /* xfs_attr3_leaf_create may not have instantiated a block */
823 if (bp && (xfs_da_shrink_inode(args, 0, bp) != 0))
824 goto out;
825 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
826 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
827 goto out;
830 memset((char *)&nargs, 0, sizeof(nargs));
831 nargs.dp = dp;
832 nargs.geo = args->geo;
833 nargs.total = args->total;
834 nargs.whichfork = XFS_ATTR_FORK;
835 nargs.trans = args->trans;
836 nargs.op_flags = XFS_DA_OP_OKNOENT;
838 sfe = &sf->list[0];
839 for (i = 0; i < sf->hdr.count; i++) {
840 nargs.name = sfe->nameval;
841 nargs.namelen = sfe->namelen;
842 nargs.value = &sfe->nameval[nargs.namelen];
843 nargs.valuelen = sfe->valuelen;
844 nargs.hashval = xfs_da_hashname(sfe->nameval,
845 sfe->namelen);
846 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
847 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
848 ASSERT(error == -ENOATTR);
849 error = xfs_attr3_leaf_add(bp, &nargs);
850 ASSERT(error != -ENOSPC);
851 if (error)
852 goto out;
853 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
855 error = 0;
856 *leaf_bp = bp;
857 out:
858 kmem_free(tmpbuffer);
859 return error;
863 * Check a leaf attribute block to see if all the entries would fit into
864 * a shortform attribute list.
867 xfs_attr_shortform_allfit(
868 struct xfs_buf *bp,
869 struct xfs_inode *dp)
871 struct xfs_attr_leafblock *leaf;
872 struct xfs_attr_leaf_entry *entry;
873 xfs_attr_leaf_name_local_t *name_loc;
874 struct xfs_attr3_icleaf_hdr leafhdr;
875 int bytes;
876 int i;
877 struct xfs_mount *mp = bp->b_target->bt_mount;
879 leaf = bp->b_addr;
880 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
881 entry = xfs_attr3_leaf_entryp(leaf);
883 bytes = sizeof(struct xfs_attr_sf_hdr);
884 for (i = 0; i < leafhdr.count; entry++, i++) {
885 if (entry->flags & XFS_ATTR_INCOMPLETE)
886 continue; /* don't copy partial entries */
887 if (!(entry->flags & XFS_ATTR_LOCAL))
888 return 0;
889 name_loc = xfs_attr3_leaf_name_local(leaf, i);
890 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
891 return 0;
892 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
893 return 0;
894 bytes += sizeof(struct xfs_attr_sf_entry) - 1
895 + name_loc->namelen
896 + be16_to_cpu(name_loc->valuelen);
898 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
899 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
900 (bytes == sizeof(struct xfs_attr_sf_hdr)))
901 return -1;
902 return xfs_attr_shortform_bytesfit(dp, bytes);
905 /* Verify the consistency of an inline attribute fork. */
906 xfs_failaddr_t
907 xfs_attr_shortform_verify(
908 struct xfs_inode *ip)
910 struct xfs_attr_shortform *sfp;
911 struct xfs_attr_sf_entry *sfep;
912 struct xfs_attr_sf_entry *next_sfep;
913 char *endp;
914 struct xfs_ifork *ifp;
915 int i;
916 int size;
918 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL);
919 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
920 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
921 size = ifp->if_bytes;
924 * Give up if the attribute is way too short.
926 if (size < sizeof(struct xfs_attr_sf_hdr))
927 return __this_address;
929 endp = (char *)sfp + size;
931 /* Check all reported entries */
932 sfep = &sfp->list[0];
933 for (i = 0; i < sfp->hdr.count; i++) {
935 * struct xfs_attr_sf_entry has a variable length.
936 * Check the fixed-offset parts of the structure are
937 * within the data buffer.
939 if (((char *)sfep + sizeof(*sfep)) >= endp)
940 return __this_address;
942 /* Don't allow names with known bad length. */
943 if (sfep->namelen == 0)
944 return __this_address;
947 * Check that the variable-length part of the structure is
948 * within the data buffer. The next entry starts after the
949 * name component, so nextentry is an acceptable test.
951 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
952 if ((char *)next_sfep > endp)
953 return __this_address;
956 * Check for unknown flags. Short form doesn't support
957 * the incomplete or local bits, so we can use the namespace
958 * mask here.
960 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
961 return __this_address;
964 * Check for invalid namespace combinations. We only allow
965 * one namespace flag per xattr, so we can just count the
966 * bits (i.e. hweight) here.
968 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
969 return __this_address;
971 sfep = next_sfep;
973 if ((void *)sfep != (void *)endp)
974 return __this_address;
976 return NULL;
980 * Convert a leaf attribute list to shortform attribute list
983 xfs_attr3_leaf_to_shortform(
984 struct xfs_buf *bp,
985 struct xfs_da_args *args,
986 int forkoff)
988 struct xfs_attr_leafblock *leaf;
989 struct xfs_attr3_icleaf_hdr ichdr;
990 struct xfs_attr_leaf_entry *entry;
991 struct xfs_attr_leaf_name_local *name_loc;
992 struct xfs_da_args nargs;
993 struct xfs_inode *dp = args->dp;
994 char *tmpbuffer;
995 int error;
996 int i;
998 trace_xfs_attr_leaf_to_sf(args);
1000 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1001 if (!tmpbuffer)
1002 return -ENOMEM;
1004 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1006 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1007 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1008 entry = xfs_attr3_leaf_entryp(leaf);
1010 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1011 memset(bp->b_addr, 0, args->geo->blksize);
1014 * Clean out the prior contents of the attribute list.
1016 error = xfs_da_shrink_inode(args, 0, bp);
1017 if (error)
1018 goto out;
1020 if (forkoff == -1) {
1021 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1022 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1023 xfs_attr_fork_remove(dp, args->trans);
1024 goto out;
1027 xfs_attr_shortform_create(args);
1030 * Copy the attributes
1032 memset((char *)&nargs, 0, sizeof(nargs));
1033 nargs.geo = args->geo;
1034 nargs.dp = dp;
1035 nargs.total = args->total;
1036 nargs.whichfork = XFS_ATTR_FORK;
1037 nargs.trans = args->trans;
1038 nargs.op_flags = XFS_DA_OP_OKNOENT;
1040 for (i = 0; i < ichdr.count; entry++, i++) {
1041 if (entry->flags & XFS_ATTR_INCOMPLETE)
1042 continue; /* don't copy partial entries */
1043 if (!entry->nameidx)
1044 continue;
1045 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1046 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1047 nargs.name = name_loc->nameval;
1048 nargs.namelen = name_loc->namelen;
1049 nargs.value = &name_loc->nameval[nargs.namelen];
1050 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1051 nargs.hashval = be32_to_cpu(entry->hashval);
1052 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1053 xfs_attr_shortform_add(&nargs, forkoff);
1055 error = 0;
1057 out:
1058 kmem_free(tmpbuffer);
1059 return error;
1063 * Convert from using a single leaf to a root node and a leaf.
1066 xfs_attr3_leaf_to_node(
1067 struct xfs_da_args *args)
1069 struct xfs_attr_leafblock *leaf;
1070 struct xfs_attr3_icleaf_hdr icleafhdr;
1071 struct xfs_attr_leaf_entry *entries;
1072 struct xfs_da_node_entry *btree;
1073 struct xfs_da3_icnode_hdr icnodehdr;
1074 struct xfs_da_intnode *node;
1075 struct xfs_inode *dp = args->dp;
1076 struct xfs_mount *mp = dp->i_mount;
1077 struct xfs_buf *bp1 = NULL;
1078 struct xfs_buf *bp2 = NULL;
1079 xfs_dablk_t blkno;
1080 int error;
1082 trace_xfs_attr_leaf_to_node(args);
1084 error = xfs_da_grow_inode(args, &blkno);
1085 if (error)
1086 goto out;
1087 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1088 if (error)
1089 goto out;
1091 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1092 if (error)
1093 goto out;
1095 /* copy leaf to new buffer, update identifiers */
1096 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1097 bp2->b_ops = bp1->b_ops;
1098 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1099 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1100 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1101 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1103 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1106 * Set up the new root node.
1108 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1109 if (error)
1110 goto out;
1111 node = bp1->b_addr;
1112 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1113 btree = dp->d_ops->node_tree_p(node);
1115 leaf = bp2->b_addr;
1116 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1117 entries = xfs_attr3_leaf_entryp(leaf);
1119 /* both on-disk, don't endian-flip twice */
1120 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1121 btree[0].before = cpu_to_be32(blkno);
1122 icnodehdr.count = 1;
1123 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1124 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1125 error = 0;
1126 out:
1127 return error;
1130 /*========================================================================
1131 * Routines used for growing the Btree.
1132 *========================================================================*/
1135 * Create the initial contents of a leaf attribute list
1136 * or a leaf in a node attribute list.
1138 STATIC int
1139 xfs_attr3_leaf_create(
1140 struct xfs_da_args *args,
1141 xfs_dablk_t blkno,
1142 struct xfs_buf **bpp)
1144 struct xfs_attr_leafblock *leaf;
1145 struct xfs_attr3_icleaf_hdr ichdr;
1146 struct xfs_inode *dp = args->dp;
1147 struct xfs_mount *mp = dp->i_mount;
1148 struct xfs_buf *bp;
1149 int error;
1151 trace_xfs_attr_leaf_create(args);
1153 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1154 XFS_ATTR_FORK);
1155 if (error)
1156 return error;
1157 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1158 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1159 leaf = bp->b_addr;
1160 memset(leaf, 0, args->geo->blksize);
1162 memset(&ichdr, 0, sizeof(ichdr));
1163 ichdr.firstused = args->geo->blksize;
1165 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1166 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1168 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1170 hdr3->blkno = cpu_to_be64(bp->b_bn);
1171 hdr3->owner = cpu_to_be64(dp->i_ino);
1172 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1174 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1175 } else {
1176 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1177 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1179 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1181 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1182 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1184 *bpp = bp;
1185 return 0;
1189 * Split the leaf node, rebalance, then add the new entry.
1192 xfs_attr3_leaf_split(
1193 struct xfs_da_state *state,
1194 struct xfs_da_state_blk *oldblk,
1195 struct xfs_da_state_blk *newblk)
1197 xfs_dablk_t blkno;
1198 int error;
1200 trace_xfs_attr_leaf_split(state->args);
1203 * Allocate space for a new leaf node.
1205 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1206 error = xfs_da_grow_inode(state->args, &blkno);
1207 if (error)
1208 return error;
1209 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1210 if (error)
1211 return error;
1212 newblk->blkno = blkno;
1213 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1216 * Rebalance the entries across the two leaves.
1217 * NOTE: rebalance() currently depends on the 2nd block being empty.
1219 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1220 error = xfs_da3_blk_link(state, oldblk, newblk);
1221 if (error)
1222 return error;
1225 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1226 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1227 * "new" attrs info. Will need the "old" info to remove it later.
1229 * Insert the "new" entry in the correct block.
1231 if (state->inleaf) {
1232 trace_xfs_attr_leaf_add_old(state->args);
1233 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1234 } else {
1235 trace_xfs_attr_leaf_add_new(state->args);
1236 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1240 * Update last hashval in each block since we added the name.
1242 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1243 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1244 return error;
1248 * Add a name to the leaf attribute list structure.
1251 xfs_attr3_leaf_add(
1252 struct xfs_buf *bp,
1253 struct xfs_da_args *args)
1255 struct xfs_attr_leafblock *leaf;
1256 struct xfs_attr3_icleaf_hdr ichdr;
1257 int tablesize;
1258 int entsize;
1259 int sum;
1260 int tmp;
1261 int i;
1263 trace_xfs_attr_leaf_add(args);
1265 leaf = bp->b_addr;
1266 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1267 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1268 entsize = xfs_attr_leaf_newentsize(args, NULL);
1271 * Search through freemap for first-fit on new name length.
1272 * (may need to figure in size of entry struct too)
1274 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1275 + xfs_attr3_leaf_hdr_size(leaf);
1276 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1277 if (tablesize > ichdr.firstused) {
1278 sum += ichdr.freemap[i].size;
1279 continue;
1281 if (!ichdr.freemap[i].size)
1282 continue; /* no space in this map */
1283 tmp = entsize;
1284 if (ichdr.freemap[i].base < ichdr.firstused)
1285 tmp += sizeof(xfs_attr_leaf_entry_t);
1286 if (ichdr.freemap[i].size >= tmp) {
1287 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1288 goto out_log_hdr;
1290 sum += ichdr.freemap[i].size;
1294 * If there are no holes in the address space of the block,
1295 * and we don't have enough freespace, then compaction will do us
1296 * no good and we should just give up.
1298 if (!ichdr.holes && sum < entsize)
1299 return -ENOSPC;
1302 * Compact the entries to coalesce free space.
1303 * This may change the hdr->count via dropping INCOMPLETE entries.
1305 xfs_attr3_leaf_compact(args, &ichdr, bp);
1308 * After compaction, the block is guaranteed to have only one
1309 * free region, in freemap[0]. If it is not big enough, give up.
1311 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1312 tmp = -ENOSPC;
1313 goto out_log_hdr;
1316 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1318 out_log_hdr:
1319 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1320 xfs_trans_log_buf(args->trans, bp,
1321 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1322 xfs_attr3_leaf_hdr_size(leaf)));
1323 return tmp;
1327 * Add a name to a leaf attribute list structure.
1329 STATIC int
1330 xfs_attr3_leaf_add_work(
1331 struct xfs_buf *bp,
1332 struct xfs_attr3_icleaf_hdr *ichdr,
1333 struct xfs_da_args *args,
1334 int mapindex)
1336 struct xfs_attr_leafblock *leaf;
1337 struct xfs_attr_leaf_entry *entry;
1338 struct xfs_attr_leaf_name_local *name_loc;
1339 struct xfs_attr_leaf_name_remote *name_rmt;
1340 struct xfs_mount *mp;
1341 int tmp;
1342 int i;
1344 trace_xfs_attr_leaf_add_work(args);
1346 leaf = bp->b_addr;
1347 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1348 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1351 * Force open some space in the entry array and fill it in.
1353 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1354 if (args->index < ichdr->count) {
1355 tmp = ichdr->count - args->index;
1356 tmp *= sizeof(xfs_attr_leaf_entry_t);
1357 memmove(entry + 1, entry, tmp);
1358 xfs_trans_log_buf(args->trans, bp,
1359 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1361 ichdr->count++;
1364 * Allocate space for the new string (at the end of the run).
1366 mp = args->trans->t_mountp;
1367 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1368 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1369 ASSERT(ichdr->freemap[mapindex].size >=
1370 xfs_attr_leaf_newentsize(args, NULL));
1371 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1372 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1374 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1376 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1377 ichdr->freemap[mapindex].size);
1378 entry->hashval = cpu_to_be32(args->hashval);
1379 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1380 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1381 if (args->op_flags & XFS_DA_OP_RENAME) {
1382 entry->flags |= XFS_ATTR_INCOMPLETE;
1383 if ((args->blkno2 == args->blkno) &&
1384 (args->index2 <= args->index)) {
1385 args->index2++;
1388 xfs_trans_log_buf(args->trans, bp,
1389 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1390 ASSERT((args->index == 0) ||
1391 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1392 ASSERT((args->index == ichdr->count - 1) ||
1393 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1396 * For "remote" attribute values, simply note that we need to
1397 * allocate space for the "remote" value. We can't actually
1398 * allocate the extents in this transaction, and we can't decide
1399 * which blocks they should be as we might allocate more blocks
1400 * as part of this transaction (a split operation for example).
1402 if (entry->flags & XFS_ATTR_LOCAL) {
1403 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1404 name_loc->namelen = args->namelen;
1405 name_loc->valuelen = cpu_to_be16(args->valuelen);
1406 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1407 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1408 be16_to_cpu(name_loc->valuelen));
1409 } else {
1410 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1411 name_rmt->namelen = args->namelen;
1412 memcpy((char *)name_rmt->name, args->name, args->namelen);
1413 entry->flags |= XFS_ATTR_INCOMPLETE;
1414 /* just in case */
1415 name_rmt->valuelen = 0;
1416 name_rmt->valueblk = 0;
1417 args->rmtblkno = 1;
1418 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1419 args->rmtvaluelen = args->valuelen;
1421 xfs_trans_log_buf(args->trans, bp,
1422 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1423 xfs_attr_leaf_entsize(leaf, args->index)));
1426 * Update the control info for this leaf node
1428 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1429 ichdr->firstused = be16_to_cpu(entry->nameidx);
1431 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1432 + xfs_attr3_leaf_hdr_size(leaf));
1433 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1434 + xfs_attr3_leaf_hdr_size(leaf);
1436 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1437 if (ichdr->freemap[i].base == tmp) {
1438 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1439 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1442 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1443 return 0;
1447 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1449 STATIC void
1450 xfs_attr3_leaf_compact(
1451 struct xfs_da_args *args,
1452 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1453 struct xfs_buf *bp)
1455 struct xfs_attr_leafblock *leaf_src;
1456 struct xfs_attr_leafblock *leaf_dst;
1457 struct xfs_attr3_icleaf_hdr ichdr_src;
1458 struct xfs_trans *trans = args->trans;
1459 char *tmpbuffer;
1461 trace_xfs_attr_leaf_compact(args);
1463 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1464 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1465 memset(bp->b_addr, 0, args->geo->blksize);
1466 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1467 leaf_dst = bp->b_addr;
1470 * Copy the on-disk header back into the destination buffer to ensure
1471 * all the information in the header that is not part of the incore
1472 * header structure is preserved.
1474 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1476 /* Initialise the incore headers */
1477 ichdr_src = *ichdr_dst; /* struct copy */
1478 ichdr_dst->firstused = args->geo->blksize;
1479 ichdr_dst->usedbytes = 0;
1480 ichdr_dst->count = 0;
1481 ichdr_dst->holes = 0;
1482 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1483 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1484 ichdr_dst->freemap[0].base;
1486 /* write the header back to initialise the underlying buffer */
1487 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1490 * Copy all entry's in the same (sorted) order,
1491 * but allocate name/value pairs packed and in sequence.
1493 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1494 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1496 * this logs the entire buffer, but the caller must write the header
1497 * back to the buffer when it is finished modifying it.
1499 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1501 kmem_free(tmpbuffer);
1505 * Compare two leaf blocks "order".
1506 * Return 0 unless leaf2 should go before leaf1.
1508 static int
1509 xfs_attr3_leaf_order(
1510 struct xfs_buf *leaf1_bp,
1511 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1512 struct xfs_buf *leaf2_bp,
1513 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1515 struct xfs_attr_leaf_entry *entries1;
1516 struct xfs_attr_leaf_entry *entries2;
1518 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1519 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1520 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1521 ((be32_to_cpu(entries2[0].hashval) <
1522 be32_to_cpu(entries1[0].hashval)) ||
1523 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1524 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1525 return 1;
1527 return 0;
1531 xfs_attr_leaf_order(
1532 struct xfs_buf *leaf1_bp,
1533 struct xfs_buf *leaf2_bp)
1535 struct xfs_attr3_icleaf_hdr ichdr1;
1536 struct xfs_attr3_icleaf_hdr ichdr2;
1537 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1539 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1540 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1541 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1545 * Redistribute the attribute list entries between two leaf nodes,
1546 * taking into account the size of the new entry.
1548 * NOTE: if new block is empty, then it will get the upper half of the
1549 * old block. At present, all (one) callers pass in an empty second block.
1551 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1552 * to match what it is doing in splitting the attribute leaf block. Those
1553 * values are used in "atomic rename" operations on attributes. Note that
1554 * the "new" and "old" values can end up in different blocks.
1556 STATIC void
1557 xfs_attr3_leaf_rebalance(
1558 struct xfs_da_state *state,
1559 struct xfs_da_state_blk *blk1,
1560 struct xfs_da_state_blk *blk2)
1562 struct xfs_da_args *args;
1563 struct xfs_attr_leafblock *leaf1;
1564 struct xfs_attr_leafblock *leaf2;
1565 struct xfs_attr3_icleaf_hdr ichdr1;
1566 struct xfs_attr3_icleaf_hdr ichdr2;
1567 struct xfs_attr_leaf_entry *entries1;
1568 struct xfs_attr_leaf_entry *entries2;
1569 int count;
1570 int totallen;
1571 int max;
1572 int space;
1573 int swap;
1576 * Set up environment.
1578 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1579 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1580 leaf1 = blk1->bp->b_addr;
1581 leaf2 = blk2->bp->b_addr;
1582 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1583 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1584 ASSERT(ichdr2.count == 0);
1585 args = state->args;
1587 trace_xfs_attr_leaf_rebalance(args);
1590 * Check ordering of blocks, reverse if it makes things simpler.
1592 * NOTE: Given that all (current) callers pass in an empty
1593 * second block, this code should never set "swap".
1595 swap = 0;
1596 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1597 swap(blk1, blk2);
1599 /* swap structures rather than reconverting them */
1600 swap(ichdr1, ichdr2);
1602 leaf1 = blk1->bp->b_addr;
1603 leaf2 = blk2->bp->b_addr;
1604 swap = 1;
1608 * Examine entries until we reduce the absolute difference in
1609 * byte usage between the two blocks to a minimum. Then get
1610 * the direction to copy and the number of elements to move.
1612 * "inleaf" is true if the new entry should be inserted into blk1.
1613 * If "swap" is also true, then reverse the sense of "inleaf".
1615 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1616 blk2, &ichdr2,
1617 &count, &totallen);
1618 if (swap)
1619 state->inleaf = !state->inleaf;
1622 * Move any entries required from leaf to leaf:
1624 if (count < ichdr1.count) {
1626 * Figure the total bytes to be added to the destination leaf.
1628 /* number entries being moved */
1629 count = ichdr1.count - count;
1630 space = ichdr1.usedbytes - totallen;
1631 space += count * sizeof(xfs_attr_leaf_entry_t);
1634 * leaf2 is the destination, compact it if it looks tight.
1636 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1637 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1638 if (space > max)
1639 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1642 * Move high entries from leaf1 to low end of leaf2.
1644 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1645 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1647 } else if (count > ichdr1.count) {
1649 * I assert that since all callers pass in an empty
1650 * second buffer, this code should never execute.
1652 ASSERT(0);
1655 * Figure the total bytes to be added to the destination leaf.
1657 /* number entries being moved */
1658 count -= ichdr1.count;
1659 space = totallen - ichdr1.usedbytes;
1660 space += count * sizeof(xfs_attr_leaf_entry_t);
1663 * leaf1 is the destination, compact it if it looks tight.
1665 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1666 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1667 if (space > max)
1668 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1671 * Move low entries from leaf2 to high end of leaf1.
1673 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1674 ichdr1.count, count);
1677 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1678 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1679 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1680 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1683 * Copy out last hashval in each block for B-tree code.
1685 entries1 = xfs_attr3_leaf_entryp(leaf1);
1686 entries2 = xfs_attr3_leaf_entryp(leaf2);
1687 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1688 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1691 * Adjust the expected index for insertion.
1692 * NOTE: this code depends on the (current) situation that the
1693 * second block was originally empty.
1695 * If the insertion point moved to the 2nd block, we must adjust
1696 * the index. We must also track the entry just following the
1697 * new entry for use in an "atomic rename" operation, that entry
1698 * is always the "old" entry and the "new" entry is what we are
1699 * inserting. The index/blkno fields refer to the "old" entry,
1700 * while the index2/blkno2 fields refer to the "new" entry.
1702 if (blk1->index > ichdr1.count) {
1703 ASSERT(state->inleaf == 0);
1704 blk2->index = blk1->index - ichdr1.count;
1705 args->index = args->index2 = blk2->index;
1706 args->blkno = args->blkno2 = blk2->blkno;
1707 } else if (blk1->index == ichdr1.count) {
1708 if (state->inleaf) {
1709 args->index = blk1->index;
1710 args->blkno = blk1->blkno;
1711 args->index2 = 0;
1712 args->blkno2 = blk2->blkno;
1713 } else {
1715 * On a double leaf split, the original attr location
1716 * is already stored in blkno2/index2, so don't
1717 * overwrite it overwise we corrupt the tree.
1719 blk2->index = blk1->index - ichdr1.count;
1720 args->index = blk2->index;
1721 args->blkno = blk2->blkno;
1722 if (!state->extravalid) {
1724 * set the new attr location to match the old
1725 * one and let the higher level split code
1726 * decide where in the leaf to place it.
1728 args->index2 = blk2->index;
1729 args->blkno2 = blk2->blkno;
1732 } else {
1733 ASSERT(state->inleaf == 1);
1734 args->index = args->index2 = blk1->index;
1735 args->blkno = args->blkno2 = blk1->blkno;
1740 * Examine entries until we reduce the absolute difference in
1741 * byte usage between the two blocks to a minimum.
1742 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1743 * GROT: there will always be enough room in either block for a new entry.
1744 * GROT: Do a double-split for this case?
1746 STATIC int
1747 xfs_attr3_leaf_figure_balance(
1748 struct xfs_da_state *state,
1749 struct xfs_da_state_blk *blk1,
1750 struct xfs_attr3_icleaf_hdr *ichdr1,
1751 struct xfs_da_state_blk *blk2,
1752 struct xfs_attr3_icleaf_hdr *ichdr2,
1753 int *countarg,
1754 int *usedbytesarg)
1756 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1757 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1758 struct xfs_attr_leaf_entry *entry;
1759 int count;
1760 int max;
1761 int index;
1762 int totallen = 0;
1763 int half;
1764 int lastdelta;
1765 int foundit = 0;
1766 int tmp;
1769 * Examine entries until we reduce the absolute difference in
1770 * byte usage between the two blocks to a minimum.
1772 max = ichdr1->count + ichdr2->count;
1773 half = (max + 1) * sizeof(*entry);
1774 half += ichdr1->usedbytes + ichdr2->usedbytes +
1775 xfs_attr_leaf_newentsize(state->args, NULL);
1776 half /= 2;
1777 lastdelta = state->args->geo->blksize;
1778 entry = xfs_attr3_leaf_entryp(leaf1);
1779 for (count = index = 0; count < max; entry++, index++, count++) {
1781 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1783 * The new entry is in the first block, account for it.
1785 if (count == blk1->index) {
1786 tmp = totallen + sizeof(*entry) +
1787 xfs_attr_leaf_newentsize(state->args, NULL);
1788 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1789 break;
1790 lastdelta = XFS_ATTR_ABS(half - tmp);
1791 totallen = tmp;
1792 foundit = 1;
1796 * Wrap around into the second block if necessary.
1798 if (count == ichdr1->count) {
1799 leaf1 = leaf2;
1800 entry = xfs_attr3_leaf_entryp(leaf1);
1801 index = 0;
1805 * Figure out if next leaf entry would be too much.
1807 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1808 index);
1809 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1810 break;
1811 lastdelta = XFS_ATTR_ABS(half - tmp);
1812 totallen = tmp;
1813 #undef XFS_ATTR_ABS
1817 * Calculate the number of usedbytes that will end up in lower block.
1818 * If new entry not in lower block, fix up the count.
1820 totallen -= count * sizeof(*entry);
1821 if (foundit) {
1822 totallen -= sizeof(*entry) +
1823 xfs_attr_leaf_newentsize(state->args, NULL);
1826 *countarg = count;
1827 *usedbytesarg = totallen;
1828 return foundit;
1831 /*========================================================================
1832 * Routines used for shrinking the Btree.
1833 *========================================================================*/
1836 * Check a leaf block and its neighbors to see if the block should be
1837 * collapsed into one or the other neighbor. Always keep the block
1838 * with the smaller block number.
1839 * If the current block is over 50% full, don't try to join it, return 0.
1840 * If the block is empty, fill in the state structure and return 2.
1841 * If it can be collapsed, fill in the state structure and return 1.
1842 * If nothing can be done, return 0.
1844 * GROT: allow for INCOMPLETE entries in calculation.
1847 xfs_attr3_leaf_toosmall(
1848 struct xfs_da_state *state,
1849 int *action)
1851 struct xfs_attr_leafblock *leaf;
1852 struct xfs_da_state_blk *blk;
1853 struct xfs_attr3_icleaf_hdr ichdr;
1854 struct xfs_buf *bp;
1855 xfs_dablk_t blkno;
1856 int bytes;
1857 int forward;
1858 int error;
1859 int retval;
1860 int i;
1862 trace_xfs_attr_leaf_toosmall(state->args);
1865 * Check for the degenerate case of the block being over 50% full.
1866 * If so, it's not worth even looking to see if we might be able
1867 * to coalesce with a sibling.
1869 blk = &state->path.blk[ state->path.active-1 ];
1870 leaf = blk->bp->b_addr;
1871 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1872 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1873 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1874 ichdr.usedbytes;
1875 if (bytes > (state->args->geo->blksize >> 1)) {
1876 *action = 0; /* blk over 50%, don't try to join */
1877 return 0;
1881 * Check for the degenerate case of the block being empty.
1882 * If the block is empty, we'll simply delete it, no need to
1883 * coalesce it with a sibling block. We choose (arbitrarily)
1884 * to merge with the forward block unless it is NULL.
1886 if (ichdr.count == 0) {
1888 * Make altpath point to the block we want to keep and
1889 * path point to the block we want to drop (this one).
1891 forward = (ichdr.forw != 0);
1892 memcpy(&state->altpath, &state->path, sizeof(state->path));
1893 error = xfs_da3_path_shift(state, &state->altpath, forward,
1894 0, &retval);
1895 if (error)
1896 return error;
1897 if (retval) {
1898 *action = 0;
1899 } else {
1900 *action = 2;
1902 return 0;
1906 * Examine each sibling block to see if we can coalesce with
1907 * at least 25% free space to spare. We need to figure out
1908 * whether to merge with the forward or the backward block.
1909 * We prefer coalescing with the lower numbered sibling so as
1910 * to shrink an attribute list over time.
1912 /* start with smaller blk num */
1913 forward = ichdr.forw < ichdr.back;
1914 for (i = 0; i < 2; forward = !forward, i++) {
1915 struct xfs_attr3_icleaf_hdr ichdr2;
1916 if (forward)
1917 blkno = ichdr.forw;
1918 else
1919 blkno = ichdr.back;
1920 if (blkno == 0)
1921 continue;
1922 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1923 blkno, -1, &bp);
1924 if (error)
1925 return error;
1927 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1929 bytes = state->args->geo->blksize -
1930 (state->args->geo->blksize >> 2) -
1931 ichdr.usedbytes - ichdr2.usedbytes -
1932 ((ichdr.count + ichdr2.count) *
1933 sizeof(xfs_attr_leaf_entry_t)) -
1934 xfs_attr3_leaf_hdr_size(leaf);
1936 xfs_trans_brelse(state->args->trans, bp);
1937 if (bytes >= 0)
1938 break; /* fits with at least 25% to spare */
1940 if (i >= 2) {
1941 *action = 0;
1942 return 0;
1946 * Make altpath point to the block we want to keep (the lower
1947 * numbered block) and path point to the block we want to drop.
1949 memcpy(&state->altpath, &state->path, sizeof(state->path));
1950 if (blkno < blk->blkno) {
1951 error = xfs_da3_path_shift(state, &state->altpath, forward,
1952 0, &retval);
1953 } else {
1954 error = xfs_da3_path_shift(state, &state->path, forward,
1955 0, &retval);
1957 if (error)
1958 return error;
1959 if (retval) {
1960 *action = 0;
1961 } else {
1962 *action = 1;
1964 return 0;
1968 * Remove a name from the leaf attribute list structure.
1970 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1971 * If two leaves are 37% full, when combined they will leave 25% free.
1974 xfs_attr3_leaf_remove(
1975 struct xfs_buf *bp,
1976 struct xfs_da_args *args)
1978 struct xfs_attr_leafblock *leaf;
1979 struct xfs_attr3_icleaf_hdr ichdr;
1980 struct xfs_attr_leaf_entry *entry;
1981 int before;
1982 int after;
1983 int smallest;
1984 int entsize;
1985 int tablesize;
1986 int tmp;
1987 int i;
1989 trace_xfs_attr_leaf_remove(args);
1991 leaf = bp->b_addr;
1992 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1994 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1995 ASSERT(args->index >= 0 && args->index < ichdr.count);
1996 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1997 xfs_attr3_leaf_hdr_size(leaf));
1999 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2001 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2002 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2005 * Scan through free region table:
2006 * check for adjacency of free'd entry with an existing one,
2007 * find smallest free region in case we need to replace it,
2008 * adjust any map that borders the entry table,
2010 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2011 + xfs_attr3_leaf_hdr_size(leaf);
2012 tmp = ichdr.freemap[0].size;
2013 before = after = -1;
2014 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2015 entsize = xfs_attr_leaf_entsize(leaf, args->index);
2016 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2017 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2018 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2019 if (ichdr.freemap[i].base == tablesize) {
2020 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2021 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2024 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2025 be16_to_cpu(entry->nameidx)) {
2026 before = i;
2027 } else if (ichdr.freemap[i].base ==
2028 (be16_to_cpu(entry->nameidx) + entsize)) {
2029 after = i;
2030 } else if (ichdr.freemap[i].size < tmp) {
2031 tmp = ichdr.freemap[i].size;
2032 smallest = i;
2037 * Coalesce adjacent freemap regions,
2038 * or replace the smallest region.
2040 if ((before >= 0) || (after >= 0)) {
2041 if ((before >= 0) && (after >= 0)) {
2042 ichdr.freemap[before].size += entsize;
2043 ichdr.freemap[before].size += ichdr.freemap[after].size;
2044 ichdr.freemap[after].base = 0;
2045 ichdr.freemap[after].size = 0;
2046 } else if (before >= 0) {
2047 ichdr.freemap[before].size += entsize;
2048 } else {
2049 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2050 ichdr.freemap[after].size += entsize;
2052 } else {
2054 * Replace smallest region (if it is smaller than free'd entry)
2056 if (ichdr.freemap[smallest].size < entsize) {
2057 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2058 ichdr.freemap[smallest].size = entsize;
2063 * Did we remove the first entry?
2065 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2066 smallest = 1;
2067 else
2068 smallest = 0;
2071 * Compress the remaining entries and zero out the removed stuff.
2073 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2074 ichdr.usedbytes -= entsize;
2075 xfs_trans_log_buf(args->trans, bp,
2076 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2077 entsize));
2079 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2080 memmove(entry, entry + 1, tmp);
2081 ichdr.count--;
2082 xfs_trans_log_buf(args->trans, bp,
2083 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2085 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2086 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2089 * If we removed the first entry, re-find the first used byte
2090 * in the name area. Note that if the entry was the "firstused",
2091 * then we don't have a "hole" in our block resulting from
2092 * removing the name.
2094 if (smallest) {
2095 tmp = args->geo->blksize;
2096 entry = xfs_attr3_leaf_entryp(leaf);
2097 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2098 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2099 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2101 if (be16_to_cpu(entry->nameidx) < tmp)
2102 tmp = be16_to_cpu(entry->nameidx);
2104 ichdr.firstused = tmp;
2105 ASSERT(ichdr.firstused != 0);
2106 } else {
2107 ichdr.holes = 1; /* mark as needing compaction */
2109 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2110 xfs_trans_log_buf(args->trans, bp,
2111 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2112 xfs_attr3_leaf_hdr_size(leaf)));
2115 * Check if leaf is less than 50% full, caller may want to
2116 * "join" the leaf with a sibling if so.
2118 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2119 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2121 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2125 * Move all the attribute list entries from drop_leaf into save_leaf.
2127 void
2128 xfs_attr3_leaf_unbalance(
2129 struct xfs_da_state *state,
2130 struct xfs_da_state_blk *drop_blk,
2131 struct xfs_da_state_blk *save_blk)
2133 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2134 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2135 struct xfs_attr3_icleaf_hdr drophdr;
2136 struct xfs_attr3_icleaf_hdr savehdr;
2137 struct xfs_attr_leaf_entry *entry;
2139 trace_xfs_attr_leaf_unbalance(state->args);
2141 drop_leaf = drop_blk->bp->b_addr;
2142 save_leaf = save_blk->bp->b_addr;
2143 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2144 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2145 entry = xfs_attr3_leaf_entryp(drop_leaf);
2148 * Save last hashval from dying block for later Btree fixup.
2150 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2153 * Check if we need a temp buffer, or can we do it in place.
2154 * Note that we don't check "leaf" for holes because we will
2155 * always be dropping it, toosmall() decided that for us already.
2157 if (savehdr.holes == 0) {
2159 * dest leaf has no holes, so we add there. May need
2160 * to make some room in the entry array.
2162 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2163 drop_blk->bp, &drophdr)) {
2164 xfs_attr3_leaf_moveents(state->args,
2165 drop_leaf, &drophdr, 0,
2166 save_leaf, &savehdr, 0,
2167 drophdr.count);
2168 } else {
2169 xfs_attr3_leaf_moveents(state->args,
2170 drop_leaf, &drophdr, 0,
2171 save_leaf, &savehdr,
2172 savehdr.count, drophdr.count);
2174 } else {
2176 * Destination has holes, so we make a temporary copy
2177 * of the leaf and add them both to that.
2179 struct xfs_attr_leafblock *tmp_leaf;
2180 struct xfs_attr3_icleaf_hdr tmphdr;
2182 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2185 * Copy the header into the temp leaf so that all the stuff
2186 * not in the incore header is present and gets copied back in
2187 * once we've moved all the entries.
2189 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2191 memset(&tmphdr, 0, sizeof(tmphdr));
2192 tmphdr.magic = savehdr.magic;
2193 tmphdr.forw = savehdr.forw;
2194 tmphdr.back = savehdr.back;
2195 tmphdr.firstused = state->args->geo->blksize;
2197 /* write the header to the temp buffer to initialise it */
2198 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2200 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2201 drop_blk->bp, &drophdr)) {
2202 xfs_attr3_leaf_moveents(state->args,
2203 drop_leaf, &drophdr, 0,
2204 tmp_leaf, &tmphdr, 0,
2205 drophdr.count);
2206 xfs_attr3_leaf_moveents(state->args,
2207 save_leaf, &savehdr, 0,
2208 tmp_leaf, &tmphdr, tmphdr.count,
2209 savehdr.count);
2210 } else {
2211 xfs_attr3_leaf_moveents(state->args,
2212 save_leaf, &savehdr, 0,
2213 tmp_leaf, &tmphdr, 0,
2214 savehdr.count);
2215 xfs_attr3_leaf_moveents(state->args,
2216 drop_leaf, &drophdr, 0,
2217 tmp_leaf, &tmphdr, tmphdr.count,
2218 drophdr.count);
2220 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2221 savehdr = tmphdr; /* struct copy */
2222 kmem_free(tmp_leaf);
2225 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2226 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2227 state->args->geo->blksize - 1);
2230 * Copy out last hashval in each block for B-tree code.
2232 entry = xfs_attr3_leaf_entryp(save_leaf);
2233 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2236 /*========================================================================
2237 * Routines used for finding things in the Btree.
2238 *========================================================================*/
2241 * Look up a name in a leaf attribute list structure.
2242 * This is the internal routine, it uses the caller's buffer.
2244 * Note that duplicate keys are allowed, but only check within the
2245 * current leaf node. The Btree code must check in adjacent leaf nodes.
2247 * Return in args->index the index into the entry[] array of either
2248 * the found entry, or where the entry should have been (insert before
2249 * that entry).
2251 * Don't change the args->value unless we find the attribute.
2254 xfs_attr3_leaf_lookup_int(
2255 struct xfs_buf *bp,
2256 struct xfs_da_args *args)
2258 struct xfs_attr_leafblock *leaf;
2259 struct xfs_attr3_icleaf_hdr ichdr;
2260 struct xfs_attr_leaf_entry *entry;
2261 struct xfs_attr_leaf_entry *entries;
2262 struct xfs_attr_leaf_name_local *name_loc;
2263 struct xfs_attr_leaf_name_remote *name_rmt;
2264 xfs_dahash_t hashval;
2265 int probe;
2266 int span;
2268 trace_xfs_attr_leaf_lookup(args);
2270 leaf = bp->b_addr;
2271 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2272 entries = xfs_attr3_leaf_entryp(leaf);
2273 if (ichdr.count >= args->geo->blksize / 8)
2274 return -EFSCORRUPTED;
2277 * Binary search. (note: small blocks will skip this loop)
2279 hashval = args->hashval;
2280 probe = span = ichdr.count / 2;
2281 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2282 span /= 2;
2283 if (be32_to_cpu(entry->hashval) < hashval)
2284 probe += span;
2285 else if (be32_to_cpu(entry->hashval) > hashval)
2286 probe -= span;
2287 else
2288 break;
2290 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count)))
2291 return -EFSCORRUPTED;
2292 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval))
2293 return -EFSCORRUPTED;
2296 * Since we may have duplicate hashval's, find the first matching
2297 * hashval in the leaf.
2299 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2300 entry--;
2301 probe--;
2303 while (probe < ichdr.count &&
2304 be32_to_cpu(entry->hashval) < hashval) {
2305 entry++;
2306 probe++;
2308 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2309 args->index = probe;
2310 return -ENOATTR;
2314 * Duplicate keys may be present, so search all of them for a match.
2316 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2317 entry++, probe++) {
2319 * GROT: Add code to remove incomplete entries.
2322 * If we are looking for INCOMPLETE entries, show only those.
2323 * If we are looking for complete entries, show only those.
2325 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2326 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2327 continue;
2329 if (entry->flags & XFS_ATTR_LOCAL) {
2330 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2331 if (name_loc->namelen != args->namelen)
2332 continue;
2333 if (memcmp(args->name, name_loc->nameval,
2334 args->namelen) != 0)
2335 continue;
2336 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2337 continue;
2338 args->index = probe;
2339 return -EEXIST;
2340 } else {
2341 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2342 if (name_rmt->namelen != args->namelen)
2343 continue;
2344 if (memcmp(args->name, name_rmt->name,
2345 args->namelen) != 0)
2346 continue;
2347 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2348 continue;
2349 args->index = probe;
2350 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2351 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2352 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2353 args->dp->i_mount,
2354 args->rmtvaluelen);
2355 return -EEXIST;
2358 args->index = probe;
2359 return -ENOATTR;
2363 * Get the value associated with an attribute name from a leaf attribute
2364 * list structure.
2367 xfs_attr3_leaf_getvalue(
2368 struct xfs_buf *bp,
2369 struct xfs_da_args *args)
2371 struct xfs_attr_leafblock *leaf;
2372 struct xfs_attr3_icleaf_hdr ichdr;
2373 struct xfs_attr_leaf_entry *entry;
2374 struct xfs_attr_leaf_name_local *name_loc;
2375 struct xfs_attr_leaf_name_remote *name_rmt;
2376 int valuelen;
2378 leaf = bp->b_addr;
2379 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2380 ASSERT(ichdr.count < args->geo->blksize / 8);
2381 ASSERT(args->index < ichdr.count);
2383 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2384 if (entry->flags & XFS_ATTR_LOCAL) {
2385 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2386 ASSERT(name_loc->namelen == args->namelen);
2387 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2388 valuelen = be16_to_cpu(name_loc->valuelen);
2389 if (args->flags & ATTR_KERNOVAL) {
2390 args->valuelen = valuelen;
2391 return 0;
2393 if (args->valuelen < valuelen) {
2394 args->valuelen = valuelen;
2395 return -ERANGE;
2397 args->valuelen = valuelen;
2398 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2399 } else {
2400 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2401 ASSERT(name_rmt->namelen == args->namelen);
2402 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2403 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2404 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2405 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2406 args->rmtvaluelen);
2407 if (args->flags & ATTR_KERNOVAL) {
2408 args->valuelen = args->rmtvaluelen;
2409 return 0;
2411 if (args->valuelen < args->rmtvaluelen) {
2412 args->valuelen = args->rmtvaluelen;
2413 return -ERANGE;
2415 args->valuelen = args->rmtvaluelen;
2417 return 0;
2420 /*========================================================================
2421 * Utility routines.
2422 *========================================================================*/
2425 * Move the indicated entries from one leaf to another.
2426 * NOTE: this routine modifies both source and destination leaves.
2428 /*ARGSUSED*/
2429 STATIC void
2430 xfs_attr3_leaf_moveents(
2431 struct xfs_da_args *args,
2432 struct xfs_attr_leafblock *leaf_s,
2433 struct xfs_attr3_icleaf_hdr *ichdr_s,
2434 int start_s,
2435 struct xfs_attr_leafblock *leaf_d,
2436 struct xfs_attr3_icleaf_hdr *ichdr_d,
2437 int start_d,
2438 int count)
2440 struct xfs_attr_leaf_entry *entry_s;
2441 struct xfs_attr_leaf_entry *entry_d;
2442 int desti;
2443 int tmp;
2444 int i;
2447 * Check for nothing to do.
2449 if (count == 0)
2450 return;
2453 * Set up environment.
2455 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2456 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2457 ASSERT(ichdr_s->magic == ichdr_d->magic);
2458 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2459 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2460 + xfs_attr3_leaf_hdr_size(leaf_s));
2461 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2462 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2463 + xfs_attr3_leaf_hdr_size(leaf_d));
2465 ASSERT(start_s < ichdr_s->count);
2466 ASSERT(start_d <= ichdr_d->count);
2467 ASSERT(count <= ichdr_s->count);
2471 * Move the entries in the destination leaf up to make a hole?
2473 if (start_d < ichdr_d->count) {
2474 tmp = ichdr_d->count - start_d;
2475 tmp *= sizeof(xfs_attr_leaf_entry_t);
2476 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2477 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2478 memmove(entry_d, entry_s, tmp);
2482 * Copy all entry's in the same (sorted) order,
2483 * but allocate attribute info packed and in sequence.
2485 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2486 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2487 desti = start_d;
2488 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2489 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2490 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2491 #ifdef GROT
2493 * Code to drop INCOMPLETE entries. Difficult to use as we
2494 * may also need to change the insertion index. Code turned
2495 * off for 6.2, should be revisited later.
2497 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2498 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2499 ichdr_s->usedbytes -= tmp;
2500 ichdr_s->count -= 1;
2501 entry_d--; /* to compensate for ++ in loop hdr */
2502 desti--;
2503 if ((start_s + i) < offset)
2504 result++; /* insertion index adjustment */
2505 } else {
2506 #endif /* GROT */
2507 ichdr_d->firstused -= tmp;
2508 /* both on-disk, don't endian flip twice */
2509 entry_d->hashval = entry_s->hashval;
2510 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2511 entry_d->flags = entry_s->flags;
2512 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2513 <= args->geo->blksize);
2514 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2515 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2516 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2517 <= args->geo->blksize);
2518 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2519 ichdr_s->usedbytes -= tmp;
2520 ichdr_d->usedbytes += tmp;
2521 ichdr_s->count -= 1;
2522 ichdr_d->count += 1;
2523 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2524 + xfs_attr3_leaf_hdr_size(leaf_d);
2525 ASSERT(ichdr_d->firstused >= tmp);
2526 #ifdef GROT
2528 #endif /* GROT */
2532 * Zero out the entries we just copied.
2534 if (start_s == ichdr_s->count) {
2535 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2536 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2537 ASSERT(((char *)entry_s + tmp) <=
2538 ((char *)leaf_s + args->geo->blksize));
2539 memset(entry_s, 0, tmp);
2540 } else {
2542 * Move the remaining entries down to fill the hole,
2543 * then zero the entries at the top.
2545 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2546 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2547 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2548 memmove(entry_d, entry_s, tmp);
2550 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2551 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2552 ASSERT(((char *)entry_s + tmp) <=
2553 ((char *)leaf_s + args->geo->blksize));
2554 memset(entry_s, 0, tmp);
2558 * Fill in the freemap information
2560 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2561 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2562 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2563 ichdr_d->freemap[1].base = 0;
2564 ichdr_d->freemap[2].base = 0;
2565 ichdr_d->freemap[1].size = 0;
2566 ichdr_d->freemap[2].size = 0;
2567 ichdr_s->holes = 1; /* leaf may not be compact */
2571 * Pick up the last hashvalue from a leaf block.
2573 xfs_dahash_t
2574 xfs_attr_leaf_lasthash(
2575 struct xfs_buf *bp,
2576 int *count)
2578 struct xfs_attr3_icleaf_hdr ichdr;
2579 struct xfs_attr_leaf_entry *entries;
2580 struct xfs_mount *mp = bp->b_target->bt_mount;
2582 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2583 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2584 if (count)
2585 *count = ichdr.count;
2586 if (!ichdr.count)
2587 return 0;
2588 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2592 * Calculate the number of bytes used to store the indicated attribute
2593 * (whether local or remote only calculate bytes in this block).
2595 STATIC int
2596 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2598 struct xfs_attr_leaf_entry *entries;
2599 xfs_attr_leaf_name_local_t *name_loc;
2600 xfs_attr_leaf_name_remote_t *name_rmt;
2601 int size;
2603 entries = xfs_attr3_leaf_entryp(leaf);
2604 if (entries[index].flags & XFS_ATTR_LOCAL) {
2605 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2606 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2607 be16_to_cpu(name_loc->valuelen));
2608 } else {
2609 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2610 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2612 return size;
2616 * Calculate the number of bytes that would be required to store the new
2617 * attribute (whether local or remote only calculate bytes in this block).
2618 * This routine decides as a side effect whether the attribute will be
2619 * a "local" or a "remote" attribute.
2622 xfs_attr_leaf_newentsize(
2623 struct xfs_da_args *args,
2624 int *local)
2626 int size;
2628 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2629 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2630 if (local)
2631 *local = 1;
2632 return size;
2634 if (local)
2635 *local = 0;
2636 return xfs_attr_leaf_entsize_remote(args->namelen);
2640 /*========================================================================
2641 * Manage the INCOMPLETE flag in a leaf entry
2642 *========================================================================*/
2645 * Clear the INCOMPLETE flag on an entry in a leaf block.
2648 xfs_attr3_leaf_clearflag(
2649 struct xfs_da_args *args)
2651 struct xfs_attr_leafblock *leaf;
2652 struct xfs_attr_leaf_entry *entry;
2653 struct xfs_attr_leaf_name_remote *name_rmt;
2654 struct xfs_buf *bp;
2655 int error;
2656 #ifdef DEBUG
2657 struct xfs_attr3_icleaf_hdr ichdr;
2658 xfs_attr_leaf_name_local_t *name_loc;
2659 int namelen;
2660 char *name;
2661 #endif /* DEBUG */
2663 trace_xfs_attr_leaf_clearflag(args);
2665 * Set up the operation.
2667 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2668 if (error)
2669 return error;
2671 leaf = bp->b_addr;
2672 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2673 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2675 #ifdef DEBUG
2676 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2677 ASSERT(args->index < ichdr.count);
2678 ASSERT(args->index >= 0);
2680 if (entry->flags & XFS_ATTR_LOCAL) {
2681 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2682 namelen = name_loc->namelen;
2683 name = (char *)name_loc->nameval;
2684 } else {
2685 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2686 namelen = name_rmt->namelen;
2687 name = (char *)name_rmt->name;
2689 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2690 ASSERT(namelen == args->namelen);
2691 ASSERT(memcmp(name, args->name, namelen) == 0);
2692 #endif /* DEBUG */
2694 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2695 xfs_trans_log_buf(args->trans, bp,
2696 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2698 if (args->rmtblkno) {
2699 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2700 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2701 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2702 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2703 xfs_trans_log_buf(args->trans, bp,
2704 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2708 * Commit the flag value change and start the next trans in series.
2710 return xfs_trans_roll_inode(&args->trans, args->dp);
2714 * Set the INCOMPLETE flag on an entry in a leaf block.
2717 xfs_attr3_leaf_setflag(
2718 struct xfs_da_args *args)
2720 struct xfs_attr_leafblock *leaf;
2721 struct xfs_attr_leaf_entry *entry;
2722 struct xfs_attr_leaf_name_remote *name_rmt;
2723 struct xfs_buf *bp;
2724 int error;
2725 #ifdef DEBUG
2726 struct xfs_attr3_icleaf_hdr ichdr;
2727 #endif
2729 trace_xfs_attr_leaf_setflag(args);
2732 * Set up the operation.
2734 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2735 if (error)
2736 return error;
2738 leaf = bp->b_addr;
2739 #ifdef DEBUG
2740 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2741 ASSERT(args->index < ichdr.count);
2742 ASSERT(args->index >= 0);
2743 #endif
2744 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2746 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2747 entry->flags |= XFS_ATTR_INCOMPLETE;
2748 xfs_trans_log_buf(args->trans, bp,
2749 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2750 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2751 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2752 name_rmt->valueblk = 0;
2753 name_rmt->valuelen = 0;
2754 xfs_trans_log_buf(args->trans, bp,
2755 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2759 * Commit the flag value change and start the next trans in series.
2761 return xfs_trans_roll_inode(&args->trans, args->dp);
2765 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2766 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2767 * entry given by args->blkno2/index2.
2769 * Note that they could be in different blocks, or in the same block.
2772 xfs_attr3_leaf_flipflags(
2773 struct xfs_da_args *args)
2775 struct xfs_attr_leafblock *leaf1;
2776 struct xfs_attr_leafblock *leaf2;
2777 struct xfs_attr_leaf_entry *entry1;
2778 struct xfs_attr_leaf_entry *entry2;
2779 struct xfs_attr_leaf_name_remote *name_rmt;
2780 struct xfs_buf *bp1;
2781 struct xfs_buf *bp2;
2782 int error;
2783 #ifdef DEBUG
2784 struct xfs_attr3_icleaf_hdr ichdr1;
2785 struct xfs_attr3_icleaf_hdr ichdr2;
2786 xfs_attr_leaf_name_local_t *name_loc;
2787 int namelen1, namelen2;
2788 char *name1, *name2;
2789 #endif /* DEBUG */
2791 trace_xfs_attr_leaf_flipflags(args);
2794 * Read the block containing the "old" attr
2796 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2797 if (error)
2798 return error;
2801 * Read the block containing the "new" attr, if it is different
2803 if (args->blkno2 != args->blkno) {
2804 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2805 -1, &bp2);
2806 if (error)
2807 return error;
2808 } else {
2809 bp2 = bp1;
2812 leaf1 = bp1->b_addr;
2813 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2815 leaf2 = bp2->b_addr;
2816 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2818 #ifdef DEBUG
2819 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2820 ASSERT(args->index < ichdr1.count);
2821 ASSERT(args->index >= 0);
2823 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2824 ASSERT(args->index2 < ichdr2.count);
2825 ASSERT(args->index2 >= 0);
2827 if (entry1->flags & XFS_ATTR_LOCAL) {
2828 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2829 namelen1 = name_loc->namelen;
2830 name1 = (char *)name_loc->nameval;
2831 } else {
2832 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2833 namelen1 = name_rmt->namelen;
2834 name1 = (char *)name_rmt->name;
2836 if (entry2->flags & XFS_ATTR_LOCAL) {
2837 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2838 namelen2 = name_loc->namelen;
2839 name2 = (char *)name_loc->nameval;
2840 } else {
2841 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2842 namelen2 = name_rmt->namelen;
2843 name2 = (char *)name_rmt->name;
2845 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2846 ASSERT(namelen1 == namelen2);
2847 ASSERT(memcmp(name1, name2, namelen1) == 0);
2848 #endif /* DEBUG */
2850 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2851 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2853 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2854 xfs_trans_log_buf(args->trans, bp1,
2855 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2856 if (args->rmtblkno) {
2857 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2858 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2859 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2860 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2861 xfs_trans_log_buf(args->trans, bp1,
2862 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2865 entry2->flags |= XFS_ATTR_INCOMPLETE;
2866 xfs_trans_log_buf(args->trans, bp2,
2867 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2868 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2869 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2870 name_rmt->valueblk = 0;
2871 name_rmt->valuelen = 0;
2872 xfs_trans_log_buf(args->trans, bp2,
2873 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2877 * Commit the flag value change and start the next trans in series.
2879 error = xfs_trans_roll_inode(&args->trans, args->dp);
2881 return error;