perf tools: Don't clone maps from parent when synthesizing forks
[linux/fpc-iii.git] / fs / xfs / libxfs / xfs_attr_leaf.c
blob6fc5425b1474a52694b4860aa79cefc52d3fd826
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 uint16_t end;
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 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
298 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
299 return __this_address;
300 if (ichdr.freemap[i].base & 0x3)
301 return __this_address;
302 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
303 return __this_address;
304 if (ichdr.freemap[i].size & 0x3)
305 return __this_address;
306 end = ichdr.freemap[i].base + ichdr.freemap[i].size;
307 if (end < ichdr.freemap[i].base)
308 return __this_address;
309 if (end > mp->m_attr_geo->blksize)
310 return __this_address;
313 return NULL;
316 static void
317 xfs_attr3_leaf_write_verify(
318 struct xfs_buf *bp)
320 struct xfs_mount *mp = bp->b_target->bt_mount;
321 struct xfs_buf_log_item *bip = bp->b_log_item;
322 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
323 xfs_failaddr_t fa;
325 fa = xfs_attr3_leaf_verify(bp);
326 if (fa) {
327 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
328 return;
331 if (!xfs_sb_version_hascrc(&mp->m_sb))
332 return;
334 if (bip)
335 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
337 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
341 * leaf/node format detection on trees is sketchy, so a node read can be done on
342 * leaf level blocks when detection identifies the tree as a node format tree
343 * incorrectly. In this case, we need to swap the verifier to match the correct
344 * format of the block being read.
346 static void
347 xfs_attr3_leaf_read_verify(
348 struct xfs_buf *bp)
350 struct xfs_mount *mp = bp->b_target->bt_mount;
351 xfs_failaddr_t fa;
353 if (xfs_sb_version_hascrc(&mp->m_sb) &&
354 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
355 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
356 else {
357 fa = xfs_attr3_leaf_verify(bp);
358 if (fa)
359 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
363 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
364 .name = "xfs_attr3_leaf",
365 .verify_read = xfs_attr3_leaf_read_verify,
366 .verify_write = xfs_attr3_leaf_write_verify,
367 .verify_struct = xfs_attr3_leaf_verify,
371 xfs_attr3_leaf_read(
372 struct xfs_trans *tp,
373 struct xfs_inode *dp,
374 xfs_dablk_t bno,
375 xfs_daddr_t mappedbno,
376 struct xfs_buf **bpp)
378 int err;
380 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
381 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
382 if (!err && tp && *bpp)
383 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
384 return err;
387 /*========================================================================
388 * Namespace helper routines
389 *========================================================================*/
392 * If namespace bits don't match return 0.
393 * If all match then return 1.
395 STATIC int
396 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
398 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
402 /*========================================================================
403 * External routines when attribute fork size < XFS_LITINO(mp).
404 *========================================================================*/
407 * Query whether the requested number of additional bytes of extended
408 * attribute space will be able to fit inline.
410 * Returns zero if not, else the di_forkoff fork offset to be used in the
411 * literal area for attribute data once the new bytes have been added.
413 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
414 * special case for dev/uuid inodes, they have fixed size data forks.
417 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
419 int offset;
420 int minforkoff; /* lower limit on valid forkoff locations */
421 int maxforkoff; /* upper limit on valid forkoff locations */
422 int dsize;
423 xfs_mount_t *mp = dp->i_mount;
425 /* rounded down */
426 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
428 if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
429 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
430 return (offset >= minforkoff) ? minforkoff : 0;
434 * If the requested numbers of bytes is smaller or equal to the
435 * current attribute fork size we can always proceed.
437 * Note that if_bytes in the data fork might actually be larger than
438 * the current data fork size is due to delalloc extents. In that
439 * case either the extent count will go down when they are converted
440 * to real extents, or the delalloc conversion will take care of the
441 * literal area rebalancing.
443 if (bytes <= XFS_IFORK_ASIZE(dp))
444 return dp->i_d.di_forkoff;
447 * For attr2 we can try to move the forkoff if there is space in the
448 * literal area, but for the old format we are done if there is no
449 * space in the fixed attribute fork.
451 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
452 return 0;
454 dsize = dp->i_df.if_bytes;
456 switch (dp->i_d.di_format) {
457 case XFS_DINODE_FMT_EXTENTS:
459 * If there is no attr fork and the data fork is extents,
460 * determine if creating the default attr fork will result
461 * in the extents form migrating to btree. If so, the
462 * minimum offset only needs to be the space required for
463 * the btree root.
465 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
466 xfs_default_attroffset(dp))
467 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
468 break;
469 case XFS_DINODE_FMT_BTREE:
471 * If we have a data btree then keep forkoff if we have one,
472 * otherwise we are adding a new attr, so then we set
473 * minforkoff to where the btree root can finish so we have
474 * plenty of room for attrs
476 if (dp->i_d.di_forkoff) {
477 if (offset < dp->i_d.di_forkoff)
478 return 0;
479 return dp->i_d.di_forkoff;
481 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
482 break;
486 * A data fork btree root must have space for at least
487 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
489 minforkoff = max(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
490 minforkoff = roundup(minforkoff, 8) >> 3;
492 /* attr fork btree root can have at least this many key/ptr pairs */
493 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
494 XFS_BMDR_SPACE_CALC(MINABTPTRS);
495 maxforkoff = maxforkoff >> 3; /* rounded down */
497 if (offset >= maxforkoff)
498 return maxforkoff;
499 if (offset >= minforkoff)
500 return offset;
501 return 0;
505 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
507 STATIC void
508 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
510 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
511 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
512 spin_lock(&mp->m_sb_lock);
513 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
514 xfs_sb_version_addattr2(&mp->m_sb);
515 spin_unlock(&mp->m_sb_lock);
516 xfs_log_sb(tp);
517 } else
518 spin_unlock(&mp->m_sb_lock);
523 * Create the initial contents of a shortform attribute list.
525 void
526 xfs_attr_shortform_create(xfs_da_args_t *args)
528 xfs_attr_sf_hdr_t *hdr;
529 xfs_inode_t *dp;
530 struct xfs_ifork *ifp;
532 trace_xfs_attr_sf_create(args);
534 dp = args->dp;
535 ASSERT(dp != NULL);
536 ifp = dp->i_afp;
537 ASSERT(ifp != NULL);
538 ASSERT(ifp->if_bytes == 0);
539 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
540 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
541 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
542 ifp->if_flags |= XFS_IFINLINE;
543 } else {
544 ASSERT(ifp->if_flags & XFS_IFINLINE);
546 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
547 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
548 hdr->count = 0;
549 hdr->totsize = cpu_to_be16(sizeof(*hdr));
550 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
554 * Add a name/value pair to the shortform attribute list.
555 * Overflow from the inode has already been checked for.
557 void
558 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
560 xfs_attr_shortform_t *sf;
561 xfs_attr_sf_entry_t *sfe;
562 int i, offset, size;
563 xfs_mount_t *mp;
564 xfs_inode_t *dp;
565 struct xfs_ifork *ifp;
567 trace_xfs_attr_sf_add(args);
569 dp = args->dp;
570 mp = dp->i_mount;
571 dp->i_d.di_forkoff = forkoff;
573 ifp = dp->i_afp;
574 ASSERT(ifp->if_flags & XFS_IFINLINE);
575 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
576 sfe = &sf->list[0];
577 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
578 #ifdef DEBUG
579 if (sfe->namelen != args->namelen)
580 continue;
581 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
582 continue;
583 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
584 continue;
585 ASSERT(0);
586 #endif
589 offset = (char *)sfe - (char *)sf;
590 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
591 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
592 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
593 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
595 sfe->namelen = args->namelen;
596 sfe->valuelen = args->valuelen;
597 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
598 memcpy(sfe->nameval, args->name, args->namelen);
599 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
600 sf->hdr.count++;
601 be16_add_cpu(&sf->hdr.totsize, size);
602 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
604 xfs_sbversion_add_attr2(mp, args->trans);
608 * After the last attribute is removed revert to original inode format,
609 * making all literal area available to the data fork once more.
611 void
612 xfs_attr_fork_remove(
613 struct xfs_inode *ip,
614 struct xfs_trans *tp)
616 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
617 ip->i_d.di_forkoff = 0;
618 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
620 ASSERT(ip->i_d.di_anextents == 0);
621 ASSERT(ip->i_afp == NULL);
623 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
627 * Remove an attribute from the shortform attribute list structure.
630 xfs_attr_shortform_remove(xfs_da_args_t *args)
632 xfs_attr_shortform_t *sf;
633 xfs_attr_sf_entry_t *sfe;
634 int base, size=0, end, totsize, i;
635 xfs_mount_t *mp;
636 xfs_inode_t *dp;
638 trace_xfs_attr_sf_remove(args);
640 dp = args->dp;
641 mp = dp->i_mount;
642 base = sizeof(xfs_attr_sf_hdr_t);
643 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
644 sfe = &sf->list[0];
645 end = sf->hdr.count;
646 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
647 base += size, i++) {
648 size = XFS_ATTR_SF_ENTSIZE(sfe);
649 if (sfe->namelen != args->namelen)
650 continue;
651 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
652 continue;
653 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
654 continue;
655 break;
657 if (i == end)
658 return -ENOATTR;
661 * Fix up the attribute fork data, covering the hole
663 end = base + size;
664 totsize = be16_to_cpu(sf->hdr.totsize);
665 if (end != totsize)
666 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
667 sf->hdr.count--;
668 be16_add_cpu(&sf->hdr.totsize, -size);
671 * Fix up the start offset of the attribute fork
673 totsize -= size;
674 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
675 (mp->m_flags & XFS_MOUNT_ATTR2) &&
676 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
677 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
678 xfs_attr_fork_remove(dp, args->trans);
679 } else {
680 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
681 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
682 ASSERT(dp->i_d.di_forkoff);
683 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
684 (args->op_flags & XFS_DA_OP_ADDNAME) ||
685 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
686 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
687 xfs_trans_log_inode(args->trans, dp,
688 XFS_ILOG_CORE | XFS_ILOG_ADATA);
691 xfs_sbversion_add_attr2(mp, args->trans);
693 return 0;
697 * Look up a name in a shortform attribute list structure.
699 /*ARGSUSED*/
701 xfs_attr_shortform_lookup(xfs_da_args_t *args)
703 xfs_attr_shortform_t *sf;
704 xfs_attr_sf_entry_t *sfe;
705 int i;
706 struct xfs_ifork *ifp;
708 trace_xfs_attr_sf_lookup(args);
710 ifp = args->dp->i_afp;
711 ASSERT(ifp->if_flags & XFS_IFINLINE);
712 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
713 sfe = &sf->list[0];
714 for (i = 0; i < sf->hdr.count;
715 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
716 if (sfe->namelen != args->namelen)
717 continue;
718 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
719 continue;
720 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
721 continue;
722 return -EEXIST;
724 return -ENOATTR;
728 * Look up a name in a shortform attribute list structure.
730 /*ARGSUSED*/
732 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
734 xfs_attr_shortform_t *sf;
735 xfs_attr_sf_entry_t *sfe;
736 int i;
738 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
739 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
740 sfe = &sf->list[0];
741 for (i = 0; i < sf->hdr.count;
742 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
743 if (sfe->namelen != args->namelen)
744 continue;
745 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
746 continue;
747 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
748 continue;
749 if (args->flags & ATTR_KERNOVAL) {
750 args->valuelen = sfe->valuelen;
751 return -EEXIST;
753 if (args->valuelen < sfe->valuelen) {
754 args->valuelen = sfe->valuelen;
755 return -ERANGE;
757 args->valuelen = sfe->valuelen;
758 memcpy(args->value, &sfe->nameval[args->namelen],
759 args->valuelen);
760 return -EEXIST;
762 return -ENOATTR;
766 * Convert from using the shortform to the leaf. On success, return the
767 * buffer so that we can keep it locked until we're totally done with it.
770 xfs_attr_shortform_to_leaf(
771 struct xfs_da_args *args,
772 struct xfs_buf **leaf_bp)
774 struct xfs_inode *dp;
775 struct xfs_attr_shortform *sf;
776 struct xfs_attr_sf_entry *sfe;
777 struct xfs_da_args nargs;
778 char *tmpbuffer;
779 int error, i, size;
780 xfs_dablk_t blkno;
781 struct xfs_buf *bp;
782 struct xfs_ifork *ifp;
784 trace_xfs_attr_sf_to_leaf(args);
786 dp = args->dp;
787 ifp = dp->i_afp;
788 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
789 size = be16_to_cpu(sf->hdr.totsize);
790 tmpbuffer = kmem_alloc(size, KM_SLEEP);
791 ASSERT(tmpbuffer != NULL);
792 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
793 sf = (xfs_attr_shortform_t *)tmpbuffer;
795 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
796 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
798 bp = NULL;
799 error = xfs_da_grow_inode(args, &blkno);
800 if (error) {
802 * If we hit an IO error middle of the transaction inside
803 * grow_inode(), we may have inconsistent data. Bail out.
805 if (error == -EIO)
806 goto out;
807 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
808 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
809 goto out;
812 ASSERT(blkno == 0);
813 error = xfs_attr3_leaf_create(args, blkno, &bp);
814 if (error) {
815 /* xfs_attr3_leaf_create may not have instantiated a block */
816 if (bp && (xfs_da_shrink_inode(args, 0, bp) != 0))
817 goto out;
818 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
819 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
820 goto out;
823 memset((char *)&nargs, 0, sizeof(nargs));
824 nargs.dp = dp;
825 nargs.geo = args->geo;
826 nargs.total = args->total;
827 nargs.whichfork = XFS_ATTR_FORK;
828 nargs.trans = args->trans;
829 nargs.op_flags = XFS_DA_OP_OKNOENT;
831 sfe = &sf->list[0];
832 for (i = 0; i < sf->hdr.count; i++) {
833 nargs.name = sfe->nameval;
834 nargs.namelen = sfe->namelen;
835 nargs.value = &sfe->nameval[nargs.namelen];
836 nargs.valuelen = sfe->valuelen;
837 nargs.hashval = xfs_da_hashname(sfe->nameval,
838 sfe->namelen);
839 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
840 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
841 ASSERT(error == -ENOATTR);
842 error = xfs_attr3_leaf_add(bp, &nargs);
843 ASSERT(error != -ENOSPC);
844 if (error)
845 goto out;
846 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
848 error = 0;
849 *leaf_bp = bp;
850 out:
851 kmem_free(tmpbuffer);
852 return error;
856 * Check a leaf attribute block to see if all the entries would fit into
857 * a shortform attribute list.
860 xfs_attr_shortform_allfit(
861 struct xfs_buf *bp,
862 struct xfs_inode *dp)
864 struct xfs_attr_leafblock *leaf;
865 struct xfs_attr_leaf_entry *entry;
866 xfs_attr_leaf_name_local_t *name_loc;
867 struct xfs_attr3_icleaf_hdr leafhdr;
868 int bytes;
869 int i;
870 struct xfs_mount *mp = bp->b_target->bt_mount;
872 leaf = bp->b_addr;
873 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
874 entry = xfs_attr3_leaf_entryp(leaf);
876 bytes = sizeof(struct xfs_attr_sf_hdr);
877 for (i = 0; i < leafhdr.count; entry++, i++) {
878 if (entry->flags & XFS_ATTR_INCOMPLETE)
879 continue; /* don't copy partial entries */
880 if (!(entry->flags & XFS_ATTR_LOCAL))
881 return 0;
882 name_loc = xfs_attr3_leaf_name_local(leaf, i);
883 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
884 return 0;
885 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
886 return 0;
887 bytes += sizeof(struct xfs_attr_sf_entry) - 1
888 + name_loc->namelen
889 + be16_to_cpu(name_loc->valuelen);
891 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
892 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
893 (bytes == sizeof(struct xfs_attr_sf_hdr)))
894 return -1;
895 return xfs_attr_shortform_bytesfit(dp, bytes);
898 /* Verify the consistency of an inline attribute fork. */
899 xfs_failaddr_t
900 xfs_attr_shortform_verify(
901 struct xfs_inode *ip)
903 struct xfs_attr_shortform *sfp;
904 struct xfs_attr_sf_entry *sfep;
905 struct xfs_attr_sf_entry *next_sfep;
906 char *endp;
907 struct xfs_ifork *ifp;
908 int i;
909 int size;
911 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL);
912 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
913 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
914 size = ifp->if_bytes;
917 * Give up if the attribute is way too short.
919 if (size < sizeof(struct xfs_attr_sf_hdr))
920 return __this_address;
922 endp = (char *)sfp + size;
924 /* Check all reported entries */
925 sfep = &sfp->list[0];
926 for (i = 0; i < sfp->hdr.count; i++) {
928 * struct xfs_attr_sf_entry has a variable length.
929 * Check the fixed-offset parts of the structure are
930 * within the data buffer.
932 if (((char *)sfep + sizeof(*sfep)) >= endp)
933 return __this_address;
935 /* Don't allow names with known bad length. */
936 if (sfep->namelen == 0)
937 return __this_address;
940 * Check that the variable-length part of the structure is
941 * within the data buffer. The next entry starts after the
942 * name component, so nextentry is an acceptable test.
944 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
945 if ((char *)next_sfep > endp)
946 return __this_address;
949 * Check for unknown flags. Short form doesn't support
950 * the incomplete or local bits, so we can use the namespace
951 * mask here.
953 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
954 return __this_address;
957 * Check for invalid namespace combinations. We only allow
958 * one namespace flag per xattr, so we can just count the
959 * bits (i.e. hweight) here.
961 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
962 return __this_address;
964 sfep = next_sfep;
966 if ((void *)sfep != (void *)endp)
967 return __this_address;
969 return NULL;
973 * Convert a leaf attribute list to shortform attribute list
976 xfs_attr3_leaf_to_shortform(
977 struct xfs_buf *bp,
978 struct xfs_da_args *args,
979 int forkoff)
981 struct xfs_attr_leafblock *leaf;
982 struct xfs_attr3_icleaf_hdr ichdr;
983 struct xfs_attr_leaf_entry *entry;
984 struct xfs_attr_leaf_name_local *name_loc;
985 struct xfs_da_args nargs;
986 struct xfs_inode *dp = args->dp;
987 char *tmpbuffer;
988 int error;
989 int i;
991 trace_xfs_attr_leaf_to_sf(args);
993 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
994 if (!tmpbuffer)
995 return -ENOMEM;
997 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
999 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1000 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1001 entry = xfs_attr3_leaf_entryp(leaf);
1003 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1004 memset(bp->b_addr, 0, args->geo->blksize);
1007 * Clean out the prior contents of the attribute list.
1009 error = xfs_da_shrink_inode(args, 0, bp);
1010 if (error)
1011 goto out;
1013 if (forkoff == -1) {
1014 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1015 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1016 xfs_attr_fork_remove(dp, args->trans);
1017 goto out;
1020 xfs_attr_shortform_create(args);
1023 * Copy the attributes
1025 memset((char *)&nargs, 0, sizeof(nargs));
1026 nargs.geo = args->geo;
1027 nargs.dp = dp;
1028 nargs.total = args->total;
1029 nargs.whichfork = XFS_ATTR_FORK;
1030 nargs.trans = args->trans;
1031 nargs.op_flags = XFS_DA_OP_OKNOENT;
1033 for (i = 0; i < ichdr.count; entry++, i++) {
1034 if (entry->flags & XFS_ATTR_INCOMPLETE)
1035 continue; /* don't copy partial entries */
1036 if (!entry->nameidx)
1037 continue;
1038 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1039 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1040 nargs.name = name_loc->nameval;
1041 nargs.namelen = name_loc->namelen;
1042 nargs.value = &name_loc->nameval[nargs.namelen];
1043 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1044 nargs.hashval = be32_to_cpu(entry->hashval);
1045 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1046 xfs_attr_shortform_add(&nargs, forkoff);
1048 error = 0;
1050 out:
1051 kmem_free(tmpbuffer);
1052 return error;
1056 * Convert from using a single leaf to a root node and a leaf.
1059 xfs_attr3_leaf_to_node(
1060 struct xfs_da_args *args)
1062 struct xfs_attr_leafblock *leaf;
1063 struct xfs_attr3_icleaf_hdr icleafhdr;
1064 struct xfs_attr_leaf_entry *entries;
1065 struct xfs_da_node_entry *btree;
1066 struct xfs_da3_icnode_hdr icnodehdr;
1067 struct xfs_da_intnode *node;
1068 struct xfs_inode *dp = args->dp;
1069 struct xfs_mount *mp = dp->i_mount;
1070 struct xfs_buf *bp1 = NULL;
1071 struct xfs_buf *bp2 = NULL;
1072 xfs_dablk_t blkno;
1073 int error;
1075 trace_xfs_attr_leaf_to_node(args);
1077 error = xfs_da_grow_inode(args, &blkno);
1078 if (error)
1079 goto out;
1080 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1081 if (error)
1082 goto out;
1084 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1085 if (error)
1086 goto out;
1088 /* copy leaf to new buffer, update identifiers */
1089 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1090 bp2->b_ops = bp1->b_ops;
1091 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1092 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1093 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1094 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1096 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1099 * Set up the new root node.
1101 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1102 if (error)
1103 goto out;
1104 node = bp1->b_addr;
1105 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1106 btree = dp->d_ops->node_tree_p(node);
1108 leaf = bp2->b_addr;
1109 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1110 entries = xfs_attr3_leaf_entryp(leaf);
1112 /* both on-disk, don't endian-flip twice */
1113 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1114 btree[0].before = cpu_to_be32(blkno);
1115 icnodehdr.count = 1;
1116 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1117 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1118 error = 0;
1119 out:
1120 return error;
1123 /*========================================================================
1124 * Routines used for growing the Btree.
1125 *========================================================================*/
1128 * Create the initial contents of a leaf attribute list
1129 * or a leaf in a node attribute list.
1131 STATIC int
1132 xfs_attr3_leaf_create(
1133 struct xfs_da_args *args,
1134 xfs_dablk_t blkno,
1135 struct xfs_buf **bpp)
1137 struct xfs_attr_leafblock *leaf;
1138 struct xfs_attr3_icleaf_hdr ichdr;
1139 struct xfs_inode *dp = args->dp;
1140 struct xfs_mount *mp = dp->i_mount;
1141 struct xfs_buf *bp;
1142 int error;
1144 trace_xfs_attr_leaf_create(args);
1146 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1147 XFS_ATTR_FORK);
1148 if (error)
1149 return error;
1150 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1151 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1152 leaf = bp->b_addr;
1153 memset(leaf, 0, args->geo->blksize);
1155 memset(&ichdr, 0, sizeof(ichdr));
1156 ichdr.firstused = args->geo->blksize;
1158 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1159 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1161 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1163 hdr3->blkno = cpu_to_be64(bp->b_bn);
1164 hdr3->owner = cpu_to_be64(dp->i_ino);
1165 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1167 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1168 } else {
1169 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1170 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1172 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1174 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1175 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1177 *bpp = bp;
1178 return 0;
1182 * Split the leaf node, rebalance, then add the new entry.
1185 xfs_attr3_leaf_split(
1186 struct xfs_da_state *state,
1187 struct xfs_da_state_blk *oldblk,
1188 struct xfs_da_state_blk *newblk)
1190 xfs_dablk_t blkno;
1191 int error;
1193 trace_xfs_attr_leaf_split(state->args);
1196 * Allocate space for a new leaf node.
1198 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1199 error = xfs_da_grow_inode(state->args, &blkno);
1200 if (error)
1201 return error;
1202 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1203 if (error)
1204 return error;
1205 newblk->blkno = blkno;
1206 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1209 * Rebalance the entries across the two leaves.
1210 * NOTE: rebalance() currently depends on the 2nd block being empty.
1212 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1213 error = xfs_da3_blk_link(state, oldblk, newblk);
1214 if (error)
1215 return error;
1218 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1219 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1220 * "new" attrs info. Will need the "old" info to remove it later.
1222 * Insert the "new" entry in the correct block.
1224 if (state->inleaf) {
1225 trace_xfs_attr_leaf_add_old(state->args);
1226 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1227 } else {
1228 trace_xfs_attr_leaf_add_new(state->args);
1229 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1233 * Update last hashval in each block since we added the name.
1235 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1236 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1237 return error;
1241 * Add a name to the leaf attribute list structure.
1244 xfs_attr3_leaf_add(
1245 struct xfs_buf *bp,
1246 struct xfs_da_args *args)
1248 struct xfs_attr_leafblock *leaf;
1249 struct xfs_attr3_icleaf_hdr ichdr;
1250 int tablesize;
1251 int entsize;
1252 int sum;
1253 int tmp;
1254 int i;
1256 trace_xfs_attr_leaf_add(args);
1258 leaf = bp->b_addr;
1259 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1260 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1261 entsize = xfs_attr_leaf_newentsize(args, NULL);
1264 * Search through freemap for first-fit on new name length.
1265 * (may need to figure in size of entry struct too)
1267 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1268 + xfs_attr3_leaf_hdr_size(leaf);
1269 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1270 if (tablesize > ichdr.firstused) {
1271 sum += ichdr.freemap[i].size;
1272 continue;
1274 if (!ichdr.freemap[i].size)
1275 continue; /* no space in this map */
1276 tmp = entsize;
1277 if (ichdr.freemap[i].base < ichdr.firstused)
1278 tmp += sizeof(xfs_attr_leaf_entry_t);
1279 if (ichdr.freemap[i].size >= tmp) {
1280 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1281 goto out_log_hdr;
1283 sum += ichdr.freemap[i].size;
1287 * If there are no holes in the address space of the block,
1288 * and we don't have enough freespace, then compaction will do us
1289 * no good and we should just give up.
1291 if (!ichdr.holes && sum < entsize)
1292 return -ENOSPC;
1295 * Compact the entries to coalesce free space.
1296 * This may change the hdr->count via dropping INCOMPLETE entries.
1298 xfs_attr3_leaf_compact(args, &ichdr, bp);
1301 * After compaction, the block is guaranteed to have only one
1302 * free region, in freemap[0]. If it is not big enough, give up.
1304 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1305 tmp = -ENOSPC;
1306 goto out_log_hdr;
1309 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1311 out_log_hdr:
1312 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1313 xfs_trans_log_buf(args->trans, bp,
1314 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1315 xfs_attr3_leaf_hdr_size(leaf)));
1316 return tmp;
1320 * Add a name to a leaf attribute list structure.
1322 STATIC int
1323 xfs_attr3_leaf_add_work(
1324 struct xfs_buf *bp,
1325 struct xfs_attr3_icleaf_hdr *ichdr,
1326 struct xfs_da_args *args,
1327 int mapindex)
1329 struct xfs_attr_leafblock *leaf;
1330 struct xfs_attr_leaf_entry *entry;
1331 struct xfs_attr_leaf_name_local *name_loc;
1332 struct xfs_attr_leaf_name_remote *name_rmt;
1333 struct xfs_mount *mp;
1334 int tmp;
1335 int i;
1337 trace_xfs_attr_leaf_add_work(args);
1339 leaf = bp->b_addr;
1340 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1341 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1344 * Force open some space in the entry array and fill it in.
1346 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1347 if (args->index < ichdr->count) {
1348 tmp = ichdr->count - args->index;
1349 tmp *= sizeof(xfs_attr_leaf_entry_t);
1350 memmove(entry + 1, entry, tmp);
1351 xfs_trans_log_buf(args->trans, bp,
1352 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1354 ichdr->count++;
1357 * Allocate space for the new string (at the end of the run).
1359 mp = args->trans->t_mountp;
1360 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1361 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1362 ASSERT(ichdr->freemap[mapindex].size >=
1363 xfs_attr_leaf_newentsize(args, NULL));
1364 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1365 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1367 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1369 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1370 ichdr->freemap[mapindex].size);
1371 entry->hashval = cpu_to_be32(args->hashval);
1372 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1373 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1374 if (args->op_flags & XFS_DA_OP_RENAME) {
1375 entry->flags |= XFS_ATTR_INCOMPLETE;
1376 if ((args->blkno2 == args->blkno) &&
1377 (args->index2 <= args->index)) {
1378 args->index2++;
1381 xfs_trans_log_buf(args->trans, bp,
1382 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1383 ASSERT((args->index == 0) ||
1384 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1385 ASSERT((args->index == ichdr->count - 1) ||
1386 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1389 * For "remote" attribute values, simply note that we need to
1390 * allocate space for the "remote" value. We can't actually
1391 * allocate the extents in this transaction, and we can't decide
1392 * which blocks they should be as we might allocate more blocks
1393 * as part of this transaction (a split operation for example).
1395 if (entry->flags & XFS_ATTR_LOCAL) {
1396 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1397 name_loc->namelen = args->namelen;
1398 name_loc->valuelen = cpu_to_be16(args->valuelen);
1399 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1400 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1401 be16_to_cpu(name_loc->valuelen));
1402 } else {
1403 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1404 name_rmt->namelen = args->namelen;
1405 memcpy((char *)name_rmt->name, args->name, args->namelen);
1406 entry->flags |= XFS_ATTR_INCOMPLETE;
1407 /* just in case */
1408 name_rmt->valuelen = 0;
1409 name_rmt->valueblk = 0;
1410 args->rmtblkno = 1;
1411 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1412 args->rmtvaluelen = args->valuelen;
1414 xfs_trans_log_buf(args->trans, bp,
1415 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1416 xfs_attr_leaf_entsize(leaf, args->index)));
1419 * Update the control info for this leaf node
1421 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1422 ichdr->firstused = be16_to_cpu(entry->nameidx);
1424 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1425 + xfs_attr3_leaf_hdr_size(leaf));
1426 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1427 + xfs_attr3_leaf_hdr_size(leaf);
1429 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1430 if (ichdr->freemap[i].base == tmp) {
1431 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1432 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1435 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1436 return 0;
1440 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1442 STATIC void
1443 xfs_attr3_leaf_compact(
1444 struct xfs_da_args *args,
1445 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1446 struct xfs_buf *bp)
1448 struct xfs_attr_leafblock *leaf_src;
1449 struct xfs_attr_leafblock *leaf_dst;
1450 struct xfs_attr3_icleaf_hdr ichdr_src;
1451 struct xfs_trans *trans = args->trans;
1452 char *tmpbuffer;
1454 trace_xfs_attr_leaf_compact(args);
1456 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1457 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1458 memset(bp->b_addr, 0, args->geo->blksize);
1459 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1460 leaf_dst = bp->b_addr;
1463 * Copy the on-disk header back into the destination buffer to ensure
1464 * all the information in the header that is not part of the incore
1465 * header structure is preserved.
1467 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1469 /* Initialise the incore headers */
1470 ichdr_src = *ichdr_dst; /* struct copy */
1471 ichdr_dst->firstused = args->geo->blksize;
1472 ichdr_dst->usedbytes = 0;
1473 ichdr_dst->count = 0;
1474 ichdr_dst->holes = 0;
1475 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1476 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1477 ichdr_dst->freemap[0].base;
1479 /* write the header back to initialise the underlying buffer */
1480 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1483 * Copy all entry's in the same (sorted) order,
1484 * but allocate name/value pairs packed and in sequence.
1486 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1487 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1489 * this logs the entire buffer, but the caller must write the header
1490 * back to the buffer when it is finished modifying it.
1492 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1494 kmem_free(tmpbuffer);
1498 * Compare two leaf blocks "order".
1499 * Return 0 unless leaf2 should go before leaf1.
1501 static int
1502 xfs_attr3_leaf_order(
1503 struct xfs_buf *leaf1_bp,
1504 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1505 struct xfs_buf *leaf2_bp,
1506 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1508 struct xfs_attr_leaf_entry *entries1;
1509 struct xfs_attr_leaf_entry *entries2;
1511 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1512 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1513 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1514 ((be32_to_cpu(entries2[0].hashval) <
1515 be32_to_cpu(entries1[0].hashval)) ||
1516 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1517 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1518 return 1;
1520 return 0;
1524 xfs_attr_leaf_order(
1525 struct xfs_buf *leaf1_bp,
1526 struct xfs_buf *leaf2_bp)
1528 struct xfs_attr3_icleaf_hdr ichdr1;
1529 struct xfs_attr3_icleaf_hdr ichdr2;
1530 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1532 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1533 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1534 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1538 * Redistribute the attribute list entries between two leaf nodes,
1539 * taking into account the size of the new entry.
1541 * NOTE: if new block is empty, then it will get the upper half of the
1542 * old block. At present, all (one) callers pass in an empty second block.
1544 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1545 * to match what it is doing in splitting the attribute leaf block. Those
1546 * values are used in "atomic rename" operations on attributes. Note that
1547 * the "new" and "old" values can end up in different blocks.
1549 STATIC void
1550 xfs_attr3_leaf_rebalance(
1551 struct xfs_da_state *state,
1552 struct xfs_da_state_blk *blk1,
1553 struct xfs_da_state_blk *blk2)
1555 struct xfs_da_args *args;
1556 struct xfs_attr_leafblock *leaf1;
1557 struct xfs_attr_leafblock *leaf2;
1558 struct xfs_attr3_icleaf_hdr ichdr1;
1559 struct xfs_attr3_icleaf_hdr ichdr2;
1560 struct xfs_attr_leaf_entry *entries1;
1561 struct xfs_attr_leaf_entry *entries2;
1562 int count;
1563 int totallen;
1564 int max;
1565 int space;
1566 int swap;
1569 * Set up environment.
1571 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1572 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1573 leaf1 = blk1->bp->b_addr;
1574 leaf2 = blk2->bp->b_addr;
1575 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1576 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1577 ASSERT(ichdr2.count == 0);
1578 args = state->args;
1580 trace_xfs_attr_leaf_rebalance(args);
1583 * Check ordering of blocks, reverse if it makes things simpler.
1585 * NOTE: Given that all (current) callers pass in an empty
1586 * second block, this code should never set "swap".
1588 swap = 0;
1589 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1590 swap(blk1, blk2);
1592 /* swap structures rather than reconverting them */
1593 swap(ichdr1, ichdr2);
1595 leaf1 = blk1->bp->b_addr;
1596 leaf2 = blk2->bp->b_addr;
1597 swap = 1;
1601 * Examine entries until we reduce the absolute difference in
1602 * byte usage between the two blocks to a minimum. Then get
1603 * the direction to copy and the number of elements to move.
1605 * "inleaf" is true if the new entry should be inserted into blk1.
1606 * If "swap" is also true, then reverse the sense of "inleaf".
1608 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1609 blk2, &ichdr2,
1610 &count, &totallen);
1611 if (swap)
1612 state->inleaf = !state->inleaf;
1615 * Move any entries required from leaf to leaf:
1617 if (count < ichdr1.count) {
1619 * Figure the total bytes to be added to the destination leaf.
1621 /* number entries being moved */
1622 count = ichdr1.count - count;
1623 space = ichdr1.usedbytes - totallen;
1624 space += count * sizeof(xfs_attr_leaf_entry_t);
1627 * leaf2 is the destination, compact it if it looks tight.
1629 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1630 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1631 if (space > max)
1632 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1635 * Move high entries from leaf1 to low end of leaf2.
1637 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1638 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1640 } else if (count > ichdr1.count) {
1642 * I assert that since all callers pass in an empty
1643 * second buffer, this code should never execute.
1645 ASSERT(0);
1648 * Figure the total bytes to be added to the destination leaf.
1650 /* number entries being moved */
1651 count -= ichdr1.count;
1652 space = totallen - ichdr1.usedbytes;
1653 space += count * sizeof(xfs_attr_leaf_entry_t);
1656 * leaf1 is the destination, compact it if it looks tight.
1658 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1659 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1660 if (space > max)
1661 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1664 * Move low entries from leaf2 to high end of leaf1.
1666 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1667 ichdr1.count, count);
1670 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1671 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1672 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1673 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1676 * Copy out last hashval in each block for B-tree code.
1678 entries1 = xfs_attr3_leaf_entryp(leaf1);
1679 entries2 = xfs_attr3_leaf_entryp(leaf2);
1680 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1681 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1684 * Adjust the expected index for insertion.
1685 * NOTE: this code depends on the (current) situation that the
1686 * second block was originally empty.
1688 * If the insertion point moved to the 2nd block, we must adjust
1689 * the index. We must also track the entry just following the
1690 * new entry for use in an "atomic rename" operation, that entry
1691 * is always the "old" entry and the "new" entry is what we are
1692 * inserting. The index/blkno fields refer to the "old" entry,
1693 * while the index2/blkno2 fields refer to the "new" entry.
1695 if (blk1->index > ichdr1.count) {
1696 ASSERT(state->inleaf == 0);
1697 blk2->index = blk1->index - ichdr1.count;
1698 args->index = args->index2 = blk2->index;
1699 args->blkno = args->blkno2 = blk2->blkno;
1700 } else if (blk1->index == ichdr1.count) {
1701 if (state->inleaf) {
1702 args->index = blk1->index;
1703 args->blkno = blk1->blkno;
1704 args->index2 = 0;
1705 args->blkno2 = blk2->blkno;
1706 } else {
1708 * On a double leaf split, the original attr location
1709 * is already stored in blkno2/index2, so don't
1710 * overwrite it overwise we corrupt the tree.
1712 blk2->index = blk1->index - ichdr1.count;
1713 args->index = blk2->index;
1714 args->blkno = blk2->blkno;
1715 if (!state->extravalid) {
1717 * set the new attr location to match the old
1718 * one and let the higher level split code
1719 * decide where in the leaf to place it.
1721 args->index2 = blk2->index;
1722 args->blkno2 = blk2->blkno;
1725 } else {
1726 ASSERT(state->inleaf == 1);
1727 args->index = args->index2 = blk1->index;
1728 args->blkno = args->blkno2 = blk1->blkno;
1733 * Examine entries until we reduce the absolute difference in
1734 * byte usage between the two blocks to a minimum.
1735 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1736 * GROT: there will always be enough room in either block for a new entry.
1737 * GROT: Do a double-split for this case?
1739 STATIC int
1740 xfs_attr3_leaf_figure_balance(
1741 struct xfs_da_state *state,
1742 struct xfs_da_state_blk *blk1,
1743 struct xfs_attr3_icleaf_hdr *ichdr1,
1744 struct xfs_da_state_blk *blk2,
1745 struct xfs_attr3_icleaf_hdr *ichdr2,
1746 int *countarg,
1747 int *usedbytesarg)
1749 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1750 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1751 struct xfs_attr_leaf_entry *entry;
1752 int count;
1753 int max;
1754 int index;
1755 int totallen = 0;
1756 int half;
1757 int lastdelta;
1758 int foundit = 0;
1759 int tmp;
1762 * Examine entries until we reduce the absolute difference in
1763 * byte usage between the two blocks to a minimum.
1765 max = ichdr1->count + ichdr2->count;
1766 half = (max + 1) * sizeof(*entry);
1767 half += ichdr1->usedbytes + ichdr2->usedbytes +
1768 xfs_attr_leaf_newentsize(state->args, NULL);
1769 half /= 2;
1770 lastdelta = state->args->geo->blksize;
1771 entry = xfs_attr3_leaf_entryp(leaf1);
1772 for (count = index = 0; count < max; entry++, index++, count++) {
1774 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1776 * The new entry is in the first block, account for it.
1778 if (count == blk1->index) {
1779 tmp = totallen + sizeof(*entry) +
1780 xfs_attr_leaf_newentsize(state->args, NULL);
1781 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1782 break;
1783 lastdelta = XFS_ATTR_ABS(half - tmp);
1784 totallen = tmp;
1785 foundit = 1;
1789 * Wrap around into the second block if necessary.
1791 if (count == ichdr1->count) {
1792 leaf1 = leaf2;
1793 entry = xfs_attr3_leaf_entryp(leaf1);
1794 index = 0;
1798 * Figure out if next leaf entry would be too much.
1800 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1801 index);
1802 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1803 break;
1804 lastdelta = XFS_ATTR_ABS(half - tmp);
1805 totallen = tmp;
1806 #undef XFS_ATTR_ABS
1810 * Calculate the number of usedbytes that will end up in lower block.
1811 * If new entry not in lower block, fix up the count.
1813 totallen -= count * sizeof(*entry);
1814 if (foundit) {
1815 totallen -= sizeof(*entry) +
1816 xfs_attr_leaf_newentsize(state->args, NULL);
1819 *countarg = count;
1820 *usedbytesarg = totallen;
1821 return foundit;
1824 /*========================================================================
1825 * Routines used for shrinking the Btree.
1826 *========================================================================*/
1829 * Check a leaf block and its neighbors to see if the block should be
1830 * collapsed into one or the other neighbor. Always keep the block
1831 * with the smaller block number.
1832 * If the current block is over 50% full, don't try to join it, return 0.
1833 * If the block is empty, fill in the state structure and return 2.
1834 * If it can be collapsed, fill in the state structure and return 1.
1835 * If nothing can be done, return 0.
1837 * GROT: allow for INCOMPLETE entries in calculation.
1840 xfs_attr3_leaf_toosmall(
1841 struct xfs_da_state *state,
1842 int *action)
1844 struct xfs_attr_leafblock *leaf;
1845 struct xfs_da_state_blk *blk;
1846 struct xfs_attr3_icleaf_hdr ichdr;
1847 struct xfs_buf *bp;
1848 xfs_dablk_t blkno;
1849 int bytes;
1850 int forward;
1851 int error;
1852 int retval;
1853 int i;
1855 trace_xfs_attr_leaf_toosmall(state->args);
1858 * Check for the degenerate case of the block being over 50% full.
1859 * If so, it's not worth even looking to see if we might be able
1860 * to coalesce with a sibling.
1862 blk = &state->path.blk[ state->path.active-1 ];
1863 leaf = blk->bp->b_addr;
1864 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1865 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1866 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1867 ichdr.usedbytes;
1868 if (bytes > (state->args->geo->blksize >> 1)) {
1869 *action = 0; /* blk over 50%, don't try to join */
1870 return 0;
1874 * Check for the degenerate case of the block being empty.
1875 * If the block is empty, we'll simply delete it, no need to
1876 * coalesce it with a sibling block. We choose (arbitrarily)
1877 * to merge with the forward block unless it is NULL.
1879 if (ichdr.count == 0) {
1881 * Make altpath point to the block we want to keep and
1882 * path point to the block we want to drop (this one).
1884 forward = (ichdr.forw != 0);
1885 memcpy(&state->altpath, &state->path, sizeof(state->path));
1886 error = xfs_da3_path_shift(state, &state->altpath, forward,
1887 0, &retval);
1888 if (error)
1889 return error;
1890 if (retval) {
1891 *action = 0;
1892 } else {
1893 *action = 2;
1895 return 0;
1899 * Examine each sibling block to see if we can coalesce with
1900 * at least 25% free space to spare. We need to figure out
1901 * whether to merge with the forward or the backward block.
1902 * We prefer coalescing with the lower numbered sibling so as
1903 * to shrink an attribute list over time.
1905 /* start with smaller blk num */
1906 forward = ichdr.forw < ichdr.back;
1907 for (i = 0; i < 2; forward = !forward, i++) {
1908 struct xfs_attr3_icleaf_hdr ichdr2;
1909 if (forward)
1910 blkno = ichdr.forw;
1911 else
1912 blkno = ichdr.back;
1913 if (blkno == 0)
1914 continue;
1915 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1916 blkno, -1, &bp);
1917 if (error)
1918 return error;
1920 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1922 bytes = state->args->geo->blksize -
1923 (state->args->geo->blksize >> 2) -
1924 ichdr.usedbytes - ichdr2.usedbytes -
1925 ((ichdr.count + ichdr2.count) *
1926 sizeof(xfs_attr_leaf_entry_t)) -
1927 xfs_attr3_leaf_hdr_size(leaf);
1929 xfs_trans_brelse(state->args->trans, bp);
1930 if (bytes >= 0)
1931 break; /* fits with at least 25% to spare */
1933 if (i >= 2) {
1934 *action = 0;
1935 return 0;
1939 * Make altpath point to the block we want to keep (the lower
1940 * numbered block) and path point to the block we want to drop.
1942 memcpy(&state->altpath, &state->path, sizeof(state->path));
1943 if (blkno < blk->blkno) {
1944 error = xfs_da3_path_shift(state, &state->altpath, forward,
1945 0, &retval);
1946 } else {
1947 error = xfs_da3_path_shift(state, &state->path, forward,
1948 0, &retval);
1950 if (error)
1951 return error;
1952 if (retval) {
1953 *action = 0;
1954 } else {
1955 *action = 1;
1957 return 0;
1961 * Remove a name from the leaf attribute list structure.
1963 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1964 * If two leaves are 37% full, when combined they will leave 25% free.
1967 xfs_attr3_leaf_remove(
1968 struct xfs_buf *bp,
1969 struct xfs_da_args *args)
1971 struct xfs_attr_leafblock *leaf;
1972 struct xfs_attr3_icleaf_hdr ichdr;
1973 struct xfs_attr_leaf_entry *entry;
1974 int before;
1975 int after;
1976 int smallest;
1977 int entsize;
1978 int tablesize;
1979 int tmp;
1980 int i;
1982 trace_xfs_attr_leaf_remove(args);
1984 leaf = bp->b_addr;
1985 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1987 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1988 ASSERT(args->index >= 0 && args->index < ichdr.count);
1989 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1990 xfs_attr3_leaf_hdr_size(leaf));
1992 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1994 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1995 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1998 * Scan through free region table:
1999 * check for adjacency of free'd entry with an existing one,
2000 * find smallest free region in case we need to replace it,
2001 * adjust any map that borders the entry table,
2003 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2004 + xfs_attr3_leaf_hdr_size(leaf);
2005 tmp = ichdr.freemap[0].size;
2006 before = after = -1;
2007 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2008 entsize = xfs_attr_leaf_entsize(leaf, args->index);
2009 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2010 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2011 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2012 if (ichdr.freemap[i].base == tablesize) {
2013 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2014 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2017 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2018 be16_to_cpu(entry->nameidx)) {
2019 before = i;
2020 } else if (ichdr.freemap[i].base ==
2021 (be16_to_cpu(entry->nameidx) + entsize)) {
2022 after = i;
2023 } else if (ichdr.freemap[i].size < tmp) {
2024 tmp = ichdr.freemap[i].size;
2025 smallest = i;
2030 * Coalesce adjacent freemap regions,
2031 * or replace the smallest region.
2033 if ((before >= 0) || (after >= 0)) {
2034 if ((before >= 0) && (after >= 0)) {
2035 ichdr.freemap[before].size += entsize;
2036 ichdr.freemap[before].size += ichdr.freemap[after].size;
2037 ichdr.freemap[after].base = 0;
2038 ichdr.freemap[after].size = 0;
2039 } else if (before >= 0) {
2040 ichdr.freemap[before].size += entsize;
2041 } else {
2042 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2043 ichdr.freemap[after].size += entsize;
2045 } else {
2047 * Replace smallest region (if it is smaller than free'd entry)
2049 if (ichdr.freemap[smallest].size < entsize) {
2050 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2051 ichdr.freemap[smallest].size = entsize;
2056 * Did we remove the first entry?
2058 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2059 smallest = 1;
2060 else
2061 smallest = 0;
2064 * Compress the remaining entries and zero out the removed stuff.
2066 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2067 ichdr.usedbytes -= entsize;
2068 xfs_trans_log_buf(args->trans, bp,
2069 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2070 entsize));
2072 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2073 memmove(entry, entry + 1, tmp);
2074 ichdr.count--;
2075 xfs_trans_log_buf(args->trans, bp,
2076 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2078 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2079 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2082 * If we removed the first entry, re-find the first used byte
2083 * in the name area. Note that if the entry was the "firstused",
2084 * then we don't have a "hole" in our block resulting from
2085 * removing the name.
2087 if (smallest) {
2088 tmp = args->geo->blksize;
2089 entry = xfs_attr3_leaf_entryp(leaf);
2090 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2091 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2092 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2094 if (be16_to_cpu(entry->nameidx) < tmp)
2095 tmp = be16_to_cpu(entry->nameidx);
2097 ichdr.firstused = tmp;
2098 ASSERT(ichdr.firstused != 0);
2099 } else {
2100 ichdr.holes = 1; /* mark as needing compaction */
2102 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2103 xfs_trans_log_buf(args->trans, bp,
2104 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2105 xfs_attr3_leaf_hdr_size(leaf)));
2108 * Check if leaf is less than 50% full, caller may want to
2109 * "join" the leaf with a sibling if so.
2111 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2112 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2114 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2118 * Move all the attribute list entries from drop_leaf into save_leaf.
2120 void
2121 xfs_attr3_leaf_unbalance(
2122 struct xfs_da_state *state,
2123 struct xfs_da_state_blk *drop_blk,
2124 struct xfs_da_state_blk *save_blk)
2126 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2127 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2128 struct xfs_attr3_icleaf_hdr drophdr;
2129 struct xfs_attr3_icleaf_hdr savehdr;
2130 struct xfs_attr_leaf_entry *entry;
2132 trace_xfs_attr_leaf_unbalance(state->args);
2134 drop_leaf = drop_blk->bp->b_addr;
2135 save_leaf = save_blk->bp->b_addr;
2136 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2137 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2138 entry = xfs_attr3_leaf_entryp(drop_leaf);
2141 * Save last hashval from dying block for later Btree fixup.
2143 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2146 * Check if we need a temp buffer, or can we do it in place.
2147 * Note that we don't check "leaf" for holes because we will
2148 * always be dropping it, toosmall() decided that for us already.
2150 if (savehdr.holes == 0) {
2152 * dest leaf has no holes, so we add there. May need
2153 * to make some room in the entry array.
2155 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2156 drop_blk->bp, &drophdr)) {
2157 xfs_attr3_leaf_moveents(state->args,
2158 drop_leaf, &drophdr, 0,
2159 save_leaf, &savehdr, 0,
2160 drophdr.count);
2161 } else {
2162 xfs_attr3_leaf_moveents(state->args,
2163 drop_leaf, &drophdr, 0,
2164 save_leaf, &savehdr,
2165 savehdr.count, drophdr.count);
2167 } else {
2169 * Destination has holes, so we make a temporary copy
2170 * of the leaf and add them both to that.
2172 struct xfs_attr_leafblock *tmp_leaf;
2173 struct xfs_attr3_icleaf_hdr tmphdr;
2175 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2178 * Copy the header into the temp leaf so that all the stuff
2179 * not in the incore header is present and gets copied back in
2180 * once we've moved all the entries.
2182 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2184 memset(&tmphdr, 0, sizeof(tmphdr));
2185 tmphdr.magic = savehdr.magic;
2186 tmphdr.forw = savehdr.forw;
2187 tmphdr.back = savehdr.back;
2188 tmphdr.firstused = state->args->geo->blksize;
2190 /* write the header to the temp buffer to initialise it */
2191 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2193 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2194 drop_blk->bp, &drophdr)) {
2195 xfs_attr3_leaf_moveents(state->args,
2196 drop_leaf, &drophdr, 0,
2197 tmp_leaf, &tmphdr, 0,
2198 drophdr.count);
2199 xfs_attr3_leaf_moveents(state->args,
2200 save_leaf, &savehdr, 0,
2201 tmp_leaf, &tmphdr, tmphdr.count,
2202 savehdr.count);
2203 } else {
2204 xfs_attr3_leaf_moveents(state->args,
2205 save_leaf, &savehdr, 0,
2206 tmp_leaf, &tmphdr, 0,
2207 savehdr.count);
2208 xfs_attr3_leaf_moveents(state->args,
2209 drop_leaf, &drophdr, 0,
2210 tmp_leaf, &tmphdr, tmphdr.count,
2211 drophdr.count);
2213 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2214 savehdr = tmphdr; /* struct copy */
2215 kmem_free(tmp_leaf);
2218 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2219 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2220 state->args->geo->blksize - 1);
2223 * Copy out last hashval in each block for B-tree code.
2225 entry = xfs_attr3_leaf_entryp(save_leaf);
2226 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2229 /*========================================================================
2230 * Routines used for finding things in the Btree.
2231 *========================================================================*/
2234 * Look up a name in a leaf attribute list structure.
2235 * This is the internal routine, it uses the caller's buffer.
2237 * Note that duplicate keys are allowed, but only check within the
2238 * current leaf node. The Btree code must check in adjacent leaf nodes.
2240 * Return in args->index the index into the entry[] array of either
2241 * the found entry, or where the entry should have been (insert before
2242 * that entry).
2244 * Don't change the args->value unless we find the attribute.
2247 xfs_attr3_leaf_lookup_int(
2248 struct xfs_buf *bp,
2249 struct xfs_da_args *args)
2251 struct xfs_attr_leafblock *leaf;
2252 struct xfs_attr3_icleaf_hdr ichdr;
2253 struct xfs_attr_leaf_entry *entry;
2254 struct xfs_attr_leaf_entry *entries;
2255 struct xfs_attr_leaf_name_local *name_loc;
2256 struct xfs_attr_leaf_name_remote *name_rmt;
2257 xfs_dahash_t hashval;
2258 int probe;
2259 int span;
2261 trace_xfs_attr_leaf_lookup(args);
2263 leaf = bp->b_addr;
2264 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2265 entries = xfs_attr3_leaf_entryp(leaf);
2266 if (ichdr.count >= args->geo->blksize / 8)
2267 return -EFSCORRUPTED;
2270 * Binary search. (note: small blocks will skip this loop)
2272 hashval = args->hashval;
2273 probe = span = ichdr.count / 2;
2274 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2275 span /= 2;
2276 if (be32_to_cpu(entry->hashval) < hashval)
2277 probe += span;
2278 else if (be32_to_cpu(entry->hashval) > hashval)
2279 probe -= span;
2280 else
2281 break;
2283 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count)))
2284 return -EFSCORRUPTED;
2285 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval))
2286 return -EFSCORRUPTED;
2289 * Since we may have duplicate hashval's, find the first matching
2290 * hashval in the leaf.
2292 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2293 entry--;
2294 probe--;
2296 while (probe < ichdr.count &&
2297 be32_to_cpu(entry->hashval) < hashval) {
2298 entry++;
2299 probe++;
2301 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2302 args->index = probe;
2303 return -ENOATTR;
2307 * Duplicate keys may be present, so search all of them for a match.
2309 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2310 entry++, probe++) {
2312 * GROT: Add code to remove incomplete entries.
2315 * If we are looking for INCOMPLETE entries, show only those.
2316 * If we are looking for complete entries, show only those.
2318 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2319 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2320 continue;
2322 if (entry->flags & XFS_ATTR_LOCAL) {
2323 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2324 if (name_loc->namelen != args->namelen)
2325 continue;
2326 if (memcmp(args->name, name_loc->nameval,
2327 args->namelen) != 0)
2328 continue;
2329 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2330 continue;
2331 args->index = probe;
2332 return -EEXIST;
2333 } else {
2334 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2335 if (name_rmt->namelen != args->namelen)
2336 continue;
2337 if (memcmp(args->name, name_rmt->name,
2338 args->namelen) != 0)
2339 continue;
2340 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2341 continue;
2342 args->index = probe;
2343 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2344 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2345 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2346 args->dp->i_mount,
2347 args->rmtvaluelen);
2348 return -EEXIST;
2351 args->index = probe;
2352 return -ENOATTR;
2356 * Get the value associated with an attribute name from a leaf attribute
2357 * list structure.
2360 xfs_attr3_leaf_getvalue(
2361 struct xfs_buf *bp,
2362 struct xfs_da_args *args)
2364 struct xfs_attr_leafblock *leaf;
2365 struct xfs_attr3_icleaf_hdr ichdr;
2366 struct xfs_attr_leaf_entry *entry;
2367 struct xfs_attr_leaf_name_local *name_loc;
2368 struct xfs_attr_leaf_name_remote *name_rmt;
2369 int valuelen;
2371 leaf = bp->b_addr;
2372 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2373 ASSERT(ichdr.count < args->geo->blksize / 8);
2374 ASSERT(args->index < ichdr.count);
2376 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2377 if (entry->flags & XFS_ATTR_LOCAL) {
2378 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2379 ASSERT(name_loc->namelen == args->namelen);
2380 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2381 valuelen = be16_to_cpu(name_loc->valuelen);
2382 if (args->flags & ATTR_KERNOVAL) {
2383 args->valuelen = valuelen;
2384 return 0;
2386 if (args->valuelen < valuelen) {
2387 args->valuelen = valuelen;
2388 return -ERANGE;
2390 args->valuelen = valuelen;
2391 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2392 } else {
2393 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2394 ASSERT(name_rmt->namelen == args->namelen);
2395 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2396 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2397 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2398 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2399 args->rmtvaluelen);
2400 if (args->flags & ATTR_KERNOVAL) {
2401 args->valuelen = args->rmtvaluelen;
2402 return 0;
2404 if (args->valuelen < args->rmtvaluelen) {
2405 args->valuelen = args->rmtvaluelen;
2406 return -ERANGE;
2408 args->valuelen = args->rmtvaluelen;
2410 return 0;
2413 /*========================================================================
2414 * Utility routines.
2415 *========================================================================*/
2418 * Move the indicated entries from one leaf to another.
2419 * NOTE: this routine modifies both source and destination leaves.
2421 /*ARGSUSED*/
2422 STATIC void
2423 xfs_attr3_leaf_moveents(
2424 struct xfs_da_args *args,
2425 struct xfs_attr_leafblock *leaf_s,
2426 struct xfs_attr3_icleaf_hdr *ichdr_s,
2427 int start_s,
2428 struct xfs_attr_leafblock *leaf_d,
2429 struct xfs_attr3_icleaf_hdr *ichdr_d,
2430 int start_d,
2431 int count)
2433 struct xfs_attr_leaf_entry *entry_s;
2434 struct xfs_attr_leaf_entry *entry_d;
2435 int desti;
2436 int tmp;
2437 int i;
2440 * Check for nothing to do.
2442 if (count == 0)
2443 return;
2446 * Set up environment.
2448 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2449 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2450 ASSERT(ichdr_s->magic == ichdr_d->magic);
2451 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2452 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2453 + xfs_attr3_leaf_hdr_size(leaf_s));
2454 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2455 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2456 + xfs_attr3_leaf_hdr_size(leaf_d));
2458 ASSERT(start_s < ichdr_s->count);
2459 ASSERT(start_d <= ichdr_d->count);
2460 ASSERT(count <= ichdr_s->count);
2464 * Move the entries in the destination leaf up to make a hole?
2466 if (start_d < ichdr_d->count) {
2467 tmp = ichdr_d->count - start_d;
2468 tmp *= sizeof(xfs_attr_leaf_entry_t);
2469 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2470 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2471 memmove(entry_d, entry_s, tmp);
2475 * Copy all entry's in the same (sorted) order,
2476 * but allocate attribute info packed and in sequence.
2478 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2479 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2480 desti = start_d;
2481 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2482 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2483 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2484 #ifdef GROT
2486 * Code to drop INCOMPLETE entries. Difficult to use as we
2487 * may also need to change the insertion index. Code turned
2488 * off for 6.2, should be revisited later.
2490 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2491 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2492 ichdr_s->usedbytes -= tmp;
2493 ichdr_s->count -= 1;
2494 entry_d--; /* to compensate for ++ in loop hdr */
2495 desti--;
2496 if ((start_s + i) < offset)
2497 result++; /* insertion index adjustment */
2498 } else {
2499 #endif /* GROT */
2500 ichdr_d->firstused -= tmp;
2501 /* both on-disk, don't endian flip twice */
2502 entry_d->hashval = entry_s->hashval;
2503 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2504 entry_d->flags = entry_s->flags;
2505 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2506 <= args->geo->blksize);
2507 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2508 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2509 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2510 <= args->geo->blksize);
2511 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2512 ichdr_s->usedbytes -= tmp;
2513 ichdr_d->usedbytes += tmp;
2514 ichdr_s->count -= 1;
2515 ichdr_d->count += 1;
2516 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2517 + xfs_attr3_leaf_hdr_size(leaf_d);
2518 ASSERT(ichdr_d->firstused >= tmp);
2519 #ifdef GROT
2521 #endif /* GROT */
2525 * Zero out the entries we just copied.
2527 if (start_s == ichdr_s->count) {
2528 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2529 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2530 ASSERT(((char *)entry_s + tmp) <=
2531 ((char *)leaf_s + args->geo->blksize));
2532 memset(entry_s, 0, tmp);
2533 } else {
2535 * Move the remaining entries down to fill the hole,
2536 * then zero the entries at the top.
2538 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2539 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2540 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2541 memmove(entry_d, entry_s, tmp);
2543 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2544 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2545 ASSERT(((char *)entry_s + tmp) <=
2546 ((char *)leaf_s + args->geo->blksize));
2547 memset(entry_s, 0, tmp);
2551 * Fill in the freemap information
2553 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2554 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2555 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2556 ichdr_d->freemap[1].base = 0;
2557 ichdr_d->freemap[2].base = 0;
2558 ichdr_d->freemap[1].size = 0;
2559 ichdr_d->freemap[2].size = 0;
2560 ichdr_s->holes = 1; /* leaf may not be compact */
2564 * Pick up the last hashvalue from a leaf block.
2566 xfs_dahash_t
2567 xfs_attr_leaf_lasthash(
2568 struct xfs_buf *bp,
2569 int *count)
2571 struct xfs_attr3_icleaf_hdr ichdr;
2572 struct xfs_attr_leaf_entry *entries;
2573 struct xfs_mount *mp = bp->b_target->bt_mount;
2575 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2576 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2577 if (count)
2578 *count = ichdr.count;
2579 if (!ichdr.count)
2580 return 0;
2581 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2585 * Calculate the number of bytes used to store the indicated attribute
2586 * (whether local or remote only calculate bytes in this block).
2588 STATIC int
2589 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2591 struct xfs_attr_leaf_entry *entries;
2592 xfs_attr_leaf_name_local_t *name_loc;
2593 xfs_attr_leaf_name_remote_t *name_rmt;
2594 int size;
2596 entries = xfs_attr3_leaf_entryp(leaf);
2597 if (entries[index].flags & XFS_ATTR_LOCAL) {
2598 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2599 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2600 be16_to_cpu(name_loc->valuelen));
2601 } else {
2602 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2603 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2605 return size;
2609 * Calculate the number of bytes that would be required to store the new
2610 * attribute (whether local or remote only calculate bytes in this block).
2611 * This routine decides as a side effect whether the attribute will be
2612 * a "local" or a "remote" attribute.
2615 xfs_attr_leaf_newentsize(
2616 struct xfs_da_args *args,
2617 int *local)
2619 int size;
2621 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2622 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2623 if (local)
2624 *local = 1;
2625 return size;
2627 if (local)
2628 *local = 0;
2629 return xfs_attr_leaf_entsize_remote(args->namelen);
2633 /*========================================================================
2634 * Manage the INCOMPLETE flag in a leaf entry
2635 *========================================================================*/
2638 * Clear the INCOMPLETE flag on an entry in a leaf block.
2641 xfs_attr3_leaf_clearflag(
2642 struct xfs_da_args *args)
2644 struct xfs_attr_leafblock *leaf;
2645 struct xfs_attr_leaf_entry *entry;
2646 struct xfs_attr_leaf_name_remote *name_rmt;
2647 struct xfs_buf *bp;
2648 int error;
2649 #ifdef DEBUG
2650 struct xfs_attr3_icleaf_hdr ichdr;
2651 xfs_attr_leaf_name_local_t *name_loc;
2652 int namelen;
2653 char *name;
2654 #endif /* DEBUG */
2656 trace_xfs_attr_leaf_clearflag(args);
2658 * Set up the operation.
2660 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2661 if (error)
2662 return error;
2664 leaf = bp->b_addr;
2665 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2666 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2668 #ifdef DEBUG
2669 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2670 ASSERT(args->index < ichdr.count);
2671 ASSERT(args->index >= 0);
2673 if (entry->flags & XFS_ATTR_LOCAL) {
2674 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2675 namelen = name_loc->namelen;
2676 name = (char *)name_loc->nameval;
2677 } else {
2678 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2679 namelen = name_rmt->namelen;
2680 name = (char *)name_rmt->name;
2682 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2683 ASSERT(namelen == args->namelen);
2684 ASSERT(memcmp(name, args->name, namelen) == 0);
2685 #endif /* DEBUG */
2687 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2688 xfs_trans_log_buf(args->trans, bp,
2689 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2691 if (args->rmtblkno) {
2692 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2693 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2694 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2695 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2696 xfs_trans_log_buf(args->trans, bp,
2697 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2701 * Commit the flag value change and start the next trans in series.
2703 return xfs_trans_roll_inode(&args->trans, args->dp);
2707 * Set the INCOMPLETE flag on an entry in a leaf block.
2710 xfs_attr3_leaf_setflag(
2711 struct xfs_da_args *args)
2713 struct xfs_attr_leafblock *leaf;
2714 struct xfs_attr_leaf_entry *entry;
2715 struct xfs_attr_leaf_name_remote *name_rmt;
2716 struct xfs_buf *bp;
2717 int error;
2718 #ifdef DEBUG
2719 struct xfs_attr3_icleaf_hdr ichdr;
2720 #endif
2722 trace_xfs_attr_leaf_setflag(args);
2725 * Set up the operation.
2727 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2728 if (error)
2729 return error;
2731 leaf = bp->b_addr;
2732 #ifdef DEBUG
2733 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2734 ASSERT(args->index < ichdr.count);
2735 ASSERT(args->index >= 0);
2736 #endif
2737 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2739 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2740 entry->flags |= XFS_ATTR_INCOMPLETE;
2741 xfs_trans_log_buf(args->trans, bp,
2742 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2743 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2744 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2745 name_rmt->valueblk = 0;
2746 name_rmt->valuelen = 0;
2747 xfs_trans_log_buf(args->trans, bp,
2748 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2752 * Commit the flag value change and start the next trans in series.
2754 return xfs_trans_roll_inode(&args->trans, args->dp);
2758 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2759 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2760 * entry given by args->blkno2/index2.
2762 * Note that they could be in different blocks, or in the same block.
2765 xfs_attr3_leaf_flipflags(
2766 struct xfs_da_args *args)
2768 struct xfs_attr_leafblock *leaf1;
2769 struct xfs_attr_leafblock *leaf2;
2770 struct xfs_attr_leaf_entry *entry1;
2771 struct xfs_attr_leaf_entry *entry2;
2772 struct xfs_attr_leaf_name_remote *name_rmt;
2773 struct xfs_buf *bp1;
2774 struct xfs_buf *bp2;
2775 int error;
2776 #ifdef DEBUG
2777 struct xfs_attr3_icleaf_hdr ichdr1;
2778 struct xfs_attr3_icleaf_hdr ichdr2;
2779 xfs_attr_leaf_name_local_t *name_loc;
2780 int namelen1, namelen2;
2781 char *name1, *name2;
2782 #endif /* DEBUG */
2784 trace_xfs_attr_leaf_flipflags(args);
2787 * Read the block containing the "old" attr
2789 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2790 if (error)
2791 return error;
2794 * Read the block containing the "new" attr, if it is different
2796 if (args->blkno2 != args->blkno) {
2797 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2798 -1, &bp2);
2799 if (error)
2800 return error;
2801 } else {
2802 bp2 = bp1;
2805 leaf1 = bp1->b_addr;
2806 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2808 leaf2 = bp2->b_addr;
2809 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2811 #ifdef DEBUG
2812 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2813 ASSERT(args->index < ichdr1.count);
2814 ASSERT(args->index >= 0);
2816 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2817 ASSERT(args->index2 < ichdr2.count);
2818 ASSERT(args->index2 >= 0);
2820 if (entry1->flags & XFS_ATTR_LOCAL) {
2821 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2822 namelen1 = name_loc->namelen;
2823 name1 = (char *)name_loc->nameval;
2824 } else {
2825 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2826 namelen1 = name_rmt->namelen;
2827 name1 = (char *)name_rmt->name;
2829 if (entry2->flags & XFS_ATTR_LOCAL) {
2830 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2831 namelen2 = name_loc->namelen;
2832 name2 = (char *)name_loc->nameval;
2833 } else {
2834 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2835 namelen2 = name_rmt->namelen;
2836 name2 = (char *)name_rmt->name;
2838 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2839 ASSERT(namelen1 == namelen2);
2840 ASSERT(memcmp(name1, name2, namelen1) == 0);
2841 #endif /* DEBUG */
2843 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2844 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2846 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2847 xfs_trans_log_buf(args->trans, bp1,
2848 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2849 if (args->rmtblkno) {
2850 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2851 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2852 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2853 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2854 xfs_trans_log_buf(args->trans, bp1,
2855 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2858 entry2->flags |= XFS_ATTR_INCOMPLETE;
2859 xfs_trans_log_buf(args->trans, bp2,
2860 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2861 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2862 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2863 name_rmt->valueblk = 0;
2864 name_rmt->valuelen = 0;
2865 xfs_trans_log_buf(args->trans, bp2,
2866 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2870 * Commit the flag value change and start the next trans in series.
2872 error = xfs_trans_roll_inode(&args->trans, args->dp);
2874 return error;