mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / xfs / libxfs / xfs_attr_leaf.c
blobfacb83031ba7782d836bf6fd8ad0c95bae662e06
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_bmap.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_attr_remote.h"
37 #include "xfs_attr.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_error.h"
40 #include "xfs_trace.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_cksum.h"
43 #include "xfs_dir2.h"
44 #include "xfs_log.h"
48 * xfs_attr_leaf.c
50 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
53 /*========================================================================
54 * Function prototypes for the kernel.
55 *========================================================================*/
58 * Routines used for growing the Btree.
60 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
61 xfs_dablk_t which_block, struct xfs_buf **bpp);
62 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
63 struct xfs_attr3_icleaf_hdr *ichdr,
64 struct xfs_da_args *args, int freemap_index);
65 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
66 struct xfs_attr3_icleaf_hdr *ichdr,
67 struct xfs_buf *leaf_buffer);
68 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
69 xfs_da_state_blk_t *blk1,
70 xfs_da_state_blk_t *blk2);
71 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
72 xfs_da_state_blk_t *leaf_blk_1,
73 struct xfs_attr3_icleaf_hdr *ichdr1,
74 xfs_da_state_blk_t *leaf_blk_2,
75 struct xfs_attr3_icleaf_hdr *ichdr2,
76 int *number_entries_in_blk1,
77 int *number_usedbytes_in_blk1);
80 * Utility routines.
82 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
83 struct xfs_attr_leafblock *src_leaf,
84 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
85 struct xfs_attr_leafblock *dst_leaf,
86 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87 int move_count);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
91 * attr3 block 'firstused' conversion helpers.
93 * firstused refers to the offset of the first used byte of the nameval region
94 * of an attr leaf block. The region starts at the tail of the block and expands
95 * backwards towards the middle. As such, firstused is initialized to the block
96 * size for an empty leaf block and is reduced from there.
98 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
99 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
100 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
101 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
102 * the attr block size. The following helpers manage the conversion between the
103 * in-core and on-disk formats.
106 static void
107 xfs_attr3_leaf_firstused_from_disk(
108 struct xfs_da_geometry *geo,
109 struct xfs_attr3_icleaf_hdr *to,
110 struct xfs_attr_leafblock *from)
112 struct xfs_attr3_leaf_hdr *hdr3;
114 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
115 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
116 to->firstused = be16_to_cpu(hdr3->firstused);
117 } else {
118 to->firstused = be16_to_cpu(from->hdr.firstused);
122 * Convert from the magic fsb size value to actual blocksize. This
123 * should only occur for empty blocks when the block size overflows
124 * 16-bits.
126 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
127 ASSERT(!to->count && !to->usedbytes);
128 ASSERT(geo->blksize > USHRT_MAX);
129 to->firstused = geo->blksize;
133 static void
134 xfs_attr3_leaf_firstused_to_disk(
135 struct xfs_da_geometry *geo,
136 struct xfs_attr_leafblock *to,
137 struct xfs_attr3_icleaf_hdr *from)
139 struct xfs_attr3_leaf_hdr *hdr3;
140 uint32_t firstused;
142 /* magic value should only be seen on disk */
143 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
146 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
147 * value. This only overflows at the max supported value of 64k. Use the
148 * magic on-disk value to represent block size in this case.
150 firstused = from->firstused;
151 if (firstused > USHRT_MAX) {
152 ASSERT(from->firstused == geo->blksize);
153 firstused = XFS_ATTR3_LEAF_NULLOFF;
156 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
157 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
158 hdr3->firstused = cpu_to_be16(firstused);
159 } else {
160 to->hdr.firstused = cpu_to_be16(firstused);
164 void
165 xfs_attr3_leaf_hdr_from_disk(
166 struct xfs_da_geometry *geo,
167 struct xfs_attr3_icleaf_hdr *to,
168 struct xfs_attr_leafblock *from)
170 int i;
172 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
173 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
175 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
176 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
178 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
179 to->back = be32_to_cpu(hdr3->info.hdr.back);
180 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
181 to->count = be16_to_cpu(hdr3->count);
182 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
183 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
184 to->holes = hdr3->holes;
186 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
187 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
188 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
190 return;
192 to->forw = be32_to_cpu(from->hdr.info.forw);
193 to->back = be32_to_cpu(from->hdr.info.back);
194 to->magic = be16_to_cpu(from->hdr.info.magic);
195 to->count = be16_to_cpu(from->hdr.count);
196 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
197 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
198 to->holes = from->hdr.holes;
200 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
201 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
202 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
206 void
207 xfs_attr3_leaf_hdr_to_disk(
208 struct xfs_da_geometry *geo,
209 struct xfs_attr_leafblock *to,
210 struct xfs_attr3_icleaf_hdr *from)
212 int i;
214 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
215 from->magic == XFS_ATTR3_LEAF_MAGIC);
217 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
218 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
220 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
221 hdr3->info.hdr.back = cpu_to_be32(from->back);
222 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
223 hdr3->count = cpu_to_be16(from->count);
224 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
225 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
226 hdr3->holes = from->holes;
227 hdr3->pad1 = 0;
229 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
230 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
231 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
233 return;
235 to->hdr.info.forw = cpu_to_be32(from->forw);
236 to->hdr.info.back = cpu_to_be32(from->back);
237 to->hdr.info.magic = cpu_to_be16(from->magic);
238 to->hdr.count = cpu_to_be16(from->count);
239 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
240 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
241 to->hdr.holes = from->holes;
242 to->hdr.pad1 = 0;
244 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
245 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
246 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
250 static bool
251 xfs_attr3_leaf_verify(
252 struct xfs_buf *bp)
254 struct xfs_mount *mp = bp->b_target->bt_mount;
255 struct xfs_attr_leafblock *leaf = bp->b_addr;
256 struct xfs_perag *pag = bp->b_pag;
257 struct xfs_attr3_icleaf_hdr ichdr;
259 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
261 if (xfs_sb_version_hascrc(&mp->m_sb)) {
262 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
264 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
265 return false;
267 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
268 return false;
269 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
270 return false;
271 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
272 return false;
273 } else {
274 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
275 return false;
278 * In recovery there is a transient state where count == 0 is valid
279 * because we may have transitioned an empty shortform attr to a leaf
280 * if the attr didn't fit in shortform.
282 if (pag && pag->pagf_init && ichdr.count == 0)
283 return false;
285 /* XXX: need to range check rest of attr header values */
286 /* XXX: hash order check? */
288 return true;
291 static void
292 xfs_attr3_leaf_write_verify(
293 struct xfs_buf *bp)
295 struct xfs_mount *mp = bp->b_target->bt_mount;
296 struct xfs_buf_log_item *bip = bp->b_fspriv;
297 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
299 if (!xfs_attr3_leaf_verify(bp)) {
300 xfs_buf_ioerror(bp, -EFSCORRUPTED);
301 xfs_verifier_error(bp);
302 return;
305 if (!xfs_sb_version_hascrc(&mp->m_sb))
306 return;
308 if (bip)
309 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
311 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
315 * leaf/node format detection on trees is sketchy, so a node read can be done on
316 * leaf level blocks when detection identifies the tree as a node format tree
317 * incorrectly. In this case, we need to swap the verifier to match the correct
318 * format of the block being read.
320 static void
321 xfs_attr3_leaf_read_verify(
322 struct xfs_buf *bp)
324 struct xfs_mount *mp = bp->b_target->bt_mount;
326 if (xfs_sb_version_hascrc(&mp->m_sb) &&
327 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
328 xfs_buf_ioerror(bp, -EFSBADCRC);
329 else if (!xfs_attr3_leaf_verify(bp))
330 xfs_buf_ioerror(bp, -EFSCORRUPTED);
332 if (bp->b_error)
333 xfs_verifier_error(bp);
336 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
337 .name = "xfs_attr3_leaf",
338 .verify_read = xfs_attr3_leaf_read_verify,
339 .verify_write = xfs_attr3_leaf_write_verify,
343 xfs_attr3_leaf_read(
344 struct xfs_trans *tp,
345 struct xfs_inode *dp,
346 xfs_dablk_t bno,
347 xfs_daddr_t mappedbno,
348 struct xfs_buf **bpp)
350 int err;
352 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
353 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
354 if (!err && tp && *bpp)
355 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
356 return err;
359 /*========================================================================
360 * Namespace helper routines
361 *========================================================================*/
364 * If namespace bits don't match return 0.
365 * If all match then return 1.
367 STATIC int
368 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
370 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
374 /*========================================================================
375 * External routines when attribute fork size < XFS_LITINO(mp).
376 *========================================================================*/
379 * Query whether the requested number of additional bytes of extended
380 * attribute space will be able to fit inline.
382 * Returns zero if not, else the di_forkoff fork offset to be used in the
383 * literal area for attribute data once the new bytes have been added.
385 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
386 * special case for dev/uuid inodes, they have fixed size data forks.
389 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
391 int offset;
392 int minforkoff; /* lower limit on valid forkoff locations */
393 int maxforkoff; /* upper limit on valid forkoff locations */
394 int dsize;
395 xfs_mount_t *mp = dp->i_mount;
397 /* rounded down */
398 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
400 switch (dp->i_d.di_format) {
401 case XFS_DINODE_FMT_DEV:
402 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
403 return (offset >= minforkoff) ? minforkoff : 0;
404 case XFS_DINODE_FMT_UUID:
405 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
406 return (offset >= minforkoff) ? minforkoff : 0;
410 * If the requested numbers of bytes is smaller or equal to the
411 * current attribute fork size we can always proceed.
413 * Note that if_bytes in the data fork might actually be larger than
414 * the current data fork size is due to delalloc extents. In that
415 * case either the extent count will go down when they are converted
416 * to real extents, or the delalloc conversion will take care of the
417 * literal area rebalancing.
419 if (bytes <= XFS_IFORK_ASIZE(dp))
420 return dp->i_d.di_forkoff;
423 * For attr2 we can try to move the forkoff if there is space in the
424 * literal area, but for the old format we are done if there is no
425 * space in the fixed attribute fork.
427 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
428 return 0;
430 dsize = dp->i_df.if_bytes;
432 switch (dp->i_d.di_format) {
433 case XFS_DINODE_FMT_EXTENTS:
435 * If there is no attr fork and the data fork is extents,
436 * determine if creating the default attr fork will result
437 * in the extents form migrating to btree. If so, the
438 * minimum offset only needs to be the space required for
439 * the btree root.
441 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
442 xfs_default_attroffset(dp))
443 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
444 break;
445 case XFS_DINODE_FMT_BTREE:
447 * If we have a data btree then keep forkoff if we have one,
448 * otherwise we are adding a new attr, so then we set
449 * minforkoff to where the btree root can finish so we have
450 * plenty of room for attrs
452 if (dp->i_d.di_forkoff) {
453 if (offset < dp->i_d.di_forkoff)
454 return 0;
455 return dp->i_d.di_forkoff;
457 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
458 break;
462 * A data fork btree root must have space for at least
463 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
465 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
466 minforkoff = roundup(minforkoff, 8) >> 3;
468 /* attr fork btree root can have at least this many key/ptr pairs */
469 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
470 XFS_BMDR_SPACE_CALC(MINABTPTRS);
471 maxforkoff = maxforkoff >> 3; /* rounded down */
473 if (offset >= maxforkoff)
474 return maxforkoff;
475 if (offset >= minforkoff)
476 return offset;
477 return 0;
481 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
483 STATIC void
484 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
486 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
487 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
488 spin_lock(&mp->m_sb_lock);
489 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
490 xfs_sb_version_addattr2(&mp->m_sb);
491 spin_unlock(&mp->m_sb_lock);
492 xfs_log_sb(tp);
493 } else
494 spin_unlock(&mp->m_sb_lock);
499 * Create the initial contents of a shortform attribute list.
501 void
502 xfs_attr_shortform_create(xfs_da_args_t *args)
504 xfs_attr_sf_hdr_t *hdr;
505 xfs_inode_t *dp;
506 xfs_ifork_t *ifp;
508 trace_xfs_attr_sf_create(args);
510 dp = args->dp;
511 ASSERT(dp != NULL);
512 ifp = dp->i_afp;
513 ASSERT(ifp != NULL);
514 ASSERT(ifp->if_bytes == 0);
515 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
516 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
517 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
518 ifp->if_flags |= XFS_IFINLINE;
519 } else {
520 ASSERT(ifp->if_flags & XFS_IFINLINE);
522 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
523 hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
524 memset(hdr, 0, sizeof(*hdr));
525 hdr->totsize = cpu_to_be16(sizeof(*hdr));
526 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
530 * Add a name/value pair to the shortform attribute list.
531 * Overflow from the inode has already been checked for.
533 void
534 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
536 xfs_attr_shortform_t *sf;
537 xfs_attr_sf_entry_t *sfe;
538 int i, offset, size;
539 xfs_mount_t *mp;
540 xfs_inode_t *dp;
541 xfs_ifork_t *ifp;
543 trace_xfs_attr_sf_add(args);
545 dp = args->dp;
546 mp = dp->i_mount;
547 dp->i_d.di_forkoff = forkoff;
549 ifp = dp->i_afp;
550 ASSERT(ifp->if_flags & XFS_IFINLINE);
551 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
552 sfe = &sf->list[0];
553 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
554 #ifdef DEBUG
555 if (sfe->namelen != args->namelen)
556 continue;
557 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
558 continue;
559 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
560 continue;
561 ASSERT(0);
562 #endif
565 offset = (char *)sfe - (char *)sf;
566 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
567 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
568 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
569 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
571 sfe->namelen = args->namelen;
572 sfe->valuelen = args->valuelen;
573 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
574 memcpy(sfe->nameval, args->name, args->namelen);
575 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
576 sf->hdr.count++;
577 be16_add_cpu(&sf->hdr.totsize, size);
578 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
580 xfs_sbversion_add_attr2(mp, args->trans);
584 * After the last attribute is removed revert to original inode format,
585 * making all literal area available to the data fork once more.
587 void
588 xfs_attr_fork_remove(
589 struct xfs_inode *ip,
590 struct xfs_trans *tp)
592 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
593 ip->i_d.di_forkoff = 0;
594 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
596 ASSERT(ip->i_d.di_anextents == 0);
597 ASSERT(ip->i_afp == NULL);
599 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
603 * Remove an attribute from the shortform attribute list structure.
606 xfs_attr_shortform_remove(xfs_da_args_t *args)
608 xfs_attr_shortform_t *sf;
609 xfs_attr_sf_entry_t *sfe;
610 int base, size=0, end, totsize, i;
611 xfs_mount_t *mp;
612 xfs_inode_t *dp;
614 trace_xfs_attr_sf_remove(args);
616 dp = args->dp;
617 mp = dp->i_mount;
618 base = sizeof(xfs_attr_sf_hdr_t);
619 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
620 sfe = &sf->list[0];
621 end = sf->hdr.count;
622 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
623 base += size, i++) {
624 size = XFS_ATTR_SF_ENTSIZE(sfe);
625 if (sfe->namelen != args->namelen)
626 continue;
627 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
628 continue;
629 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
630 continue;
631 break;
633 if (i == end)
634 return -ENOATTR;
637 * Fix up the attribute fork data, covering the hole
639 end = base + size;
640 totsize = be16_to_cpu(sf->hdr.totsize);
641 if (end != totsize)
642 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
643 sf->hdr.count--;
644 be16_add_cpu(&sf->hdr.totsize, -size);
647 * Fix up the start offset of the attribute fork
649 totsize -= size;
650 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
651 (mp->m_flags & XFS_MOUNT_ATTR2) &&
652 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
653 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
654 xfs_attr_fork_remove(dp, args->trans);
655 } else {
656 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
657 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
658 ASSERT(dp->i_d.di_forkoff);
659 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
660 (args->op_flags & XFS_DA_OP_ADDNAME) ||
661 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
662 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
663 xfs_trans_log_inode(args->trans, dp,
664 XFS_ILOG_CORE | XFS_ILOG_ADATA);
667 xfs_sbversion_add_attr2(mp, args->trans);
669 return 0;
673 * Look up a name in a shortform attribute list structure.
675 /*ARGSUSED*/
677 xfs_attr_shortform_lookup(xfs_da_args_t *args)
679 xfs_attr_shortform_t *sf;
680 xfs_attr_sf_entry_t *sfe;
681 int i;
682 xfs_ifork_t *ifp;
684 trace_xfs_attr_sf_lookup(args);
686 ifp = args->dp->i_afp;
687 ASSERT(ifp->if_flags & XFS_IFINLINE);
688 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
689 sfe = &sf->list[0];
690 for (i = 0; i < sf->hdr.count;
691 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
692 if (sfe->namelen != args->namelen)
693 continue;
694 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
695 continue;
696 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
697 continue;
698 return -EEXIST;
700 return -ENOATTR;
704 * Look up a name in a shortform attribute list structure.
706 /*ARGSUSED*/
708 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
710 xfs_attr_shortform_t *sf;
711 xfs_attr_sf_entry_t *sfe;
712 int i;
714 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
715 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
716 sfe = &sf->list[0];
717 for (i = 0; i < sf->hdr.count;
718 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
719 if (sfe->namelen != args->namelen)
720 continue;
721 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
722 continue;
723 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
724 continue;
725 if (args->flags & ATTR_KERNOVAL) {
726 args->valuelen = sfe->valuelen;
727 return -EEXIST;
729 if (args->valuelen < sfe->valuelen) {
730 args->valuelen = sfe->valuelen;
731 return -ERANGE;
733 args->valuelen = sfe->valuelen;
734 memcpy(args->value, &sfe->nameval[args->namelen],
735 args->valuelen);
736 return -EEXIST;
738 return -ENOATTR;
742 * Convert from using the shortform to the leaf. On success, return the
743 * buffer so that we can keep it locked until we're totally done with it.
746 xfs_attr_shortform_to_leaf(
747 struct xfs_da_args *args,
748 struct xfs_buf **leaf_bp)
750 xfs_inode_t *dp;
751 xfs_attr_shortform_t *sf;
752 xfs_attr_sf_entry_t *sfe;
753 xfs_da_args_t nargs;
754 char *tmpbuffer;
755 int error, i, size;
756 xfs_dablk_t blkno;
757 struct xfs_buf *bp;
758 xfs_ifork_t *ifp;
760 trace_xfs_attr_sf_to_leaf(args);
762 dp = args->dp;
763 ifp = dp->i_afp;
764 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
765 size = be16_to_cpu(sf->hdr.totsize);
766 tmpbuffer = kmem_alloc(size, KM_SLEEP);
767 ASSERT(tmpbuffer != NULL);
768 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
769 sf = (xfs_attr_shortform_t *)tmpbuffer;
771 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
772 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
774 bp = NULL;
775 error = xfs_da_grow_inode(args, &blkno);
776 if (error) {
778 * If we hit an IO error middle of the transaction inside
779 * grow_inode(), we may have inconsistent data. Bail out.
781 if (error == -EIO)
782 goto out;
783 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
784 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
785 goto out;
788 ASSERT(blkno == 0);
789 error = xfs_attr3_leaf_create(args, blkno, &bp);
790 if (error) {
791 /* xfs_attr3_leaf_create may not have instantiated a block */
792 if (bp && (xfs_da_shrink_inode(args, 0, bp) != 0))
793 goto out;
794 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
795 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
796 goto out;
799 memset((char *)&nargs, 0, sizeof(nargs));
800 nargs.dp = dp;
801 nargs.geo = args->geo;
802 nargs.firstblock = args->firstblock;
803 nargs.dfops = args->dfops;
804 nargs.total = args->total;
805 nargs.whichfork = XFS_ATTR_FORK;
806 nargs.trans = args->trans;
807 nargs.op_flags = XFS_DA_OP_OKNOENT;
809 sfe = &sf->list[0];
810 for (i = 0; i < sf->hdr.count; i++) {
811 nargs.name = sfe->nameval;
812 nargs.namelen = sfe->namelen;
813 nargs.value = &sfe->nameval[nargs.namelen];
814 nargs.valuelen = sfe->valuelen;
815 nargs.hashval = xfs_da_hashname(sfe->nameval,
816 sfe->namelen);
817 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
818 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
819 ASSERT(error == -ENOATTR);
820 error = xfs_attr3_leaf_add(bp, &nargs);
821 ASSERT(error != -ENOSPC);
822 if (error)
823 goto out;
824 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
826 error = 0;
827 *leaf_bp = bp;
828 out:
829 kmem_free(tmpbuffer);
830 return error;
834 * Check a leaf attribute block to see if all the entries would fit into
835 * a shortform attribute list.
838 xfs_attr_shortform_allfit(
839 struct xfs_buf *bp,
840 struct xfs_inode *dp)
842 struct xfs_attr_leafblock *leaf;
843 struct xfs_attr_leaf_entry *entry;
844 xfs_attr_leaf_name_local_t *name_loc;
845 struct xfs_attr3_icleaf_hdr leafhdr;
846 int bytes;
847 int i;
848 struct xfs_mount *mp = bp->b_target->bt_mount;
850 leaf = bp->b_addr;
851 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
852 entry = xfs_attr3_leaf_entryp(leaf);
854 bytes = sizeof(struct xfs_attr_sf_hdr);
855 for (i = 0; i < leafhdr.count; entry++, i++) {
856 if (entry->flags & XFS_ATTR_INCOMPLETE)
857 continue; /* don't copy partial entries */
858 if (!(entry->flags & XFS_ATTR_LOCAL))
859 return 0;
860 name_loc = xfs_attr3_leaf_name_local(leaf, i);
861 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
862 return 0;
863 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
864 return 0;
865 bytes += sizeof(struct xfs_attr_sf_entry) - 1
866 + name_loc->namelen
867 + be16_to_cpu(name_loc->valuelen);
869 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
870 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
871 (bytes == sizeof(struct xfs_attr_sf_hdr)))
872 return -1;
873 return xfs_attr_shortform_bytesfit(dp, bytes);
877 * Convert a leaf attribute list to shortform attribute list
880 xfs_attr3_leaf_to_shortform(
881 struct xfs_buf *bp,
882 struct xfs_da_args *args,
883 int forkoff)
885 struct xfs_attr_leafblock *leaf;
886 struct xfs_attr3_icleaf_hdr ichdr;
887 struct xfs_attr_leaf_entry *entry;
888 struct xfs_attr_leaf_name_local *name_loc;
889 struct xfs_da_args nargs;
890 struct xfs_inode *dp = args->dp;
891 char *tmpbuffer;
892 int error;
893 int i;
895 trace_xfs_attr_leaf_to_sf(args);
897 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
898 if (!tmpbuffer)
899 return -ENOMEM;
901 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
903 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
904 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
905 entry = xfs_attr3_leaf_entryp(leaf);
907 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
908 memset(bp->b_addr, 0, args->geo->blksize);
911 * Clean out the prior contents of the attribute list.
913 error = xfs_da_shrink_inode(args, 0, bp);
914 if (error)
915 goto out;
917 if (forkoff == -1) {
918 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
919 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
920 xfs_attr_fork_remove(dp, args->trans);
921 goto out;
924 xfs_attr_shortform_create(args);
927 * Copy the attributes
929 memset((char *)&nargs, 0, sizeof(nargs));
930 nargs.geo = args->geo;
931 nargs.dp = dp;
932 nargs.firstblock = args->firstblock;
933 nargs.dfops = args->dfops;
934 nargs.total = args->total;
935 nargs.whichfork = XFS_ATTR_FORK;
936 nargs.trans = args->trans;
937 nargs.op_flags = XFS_DA_OP_OKNOENT;
939 for (i = 0; i < ichdr.count; entry++, i++) {
940 if (entry->flags & XFS_ATTR_INCOMPLETE)
941 continue; /* don't copy partial entries */
942 if (!entry->nameidx)
943 continue;
944 ASSERT(entry->flags & XFS_ATTR_LOCAL);
945 name_loc = xfs_attr3_leaf_name_local(leaf, i);
946 nargs.name = name_loc->nameval;
947 nargs.namelen = name_loc->namelen;
948 nargs.value = &name_loc->nameval[nargs.namelen];
949 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
950 nargs.hashval = be32_to_cpu(entry->hashval);
951 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
952 xfs_attr_shortform_add(&nargs, forkoff);
954 error = 0;
956 out:
957 kmem_free(tmpbuffer);
958 return error;
962 * Convert from using a single leaf to a root node and a leaf.
965 xfs_attr3_leaf_to_node(
966 struct xfs_da_args *args)
968 struct xfs_attr_leafblock *leaf;
969 struct xfs_attr3_icleaf_hdr icleafhdr;
970 struct xfs_attr_leaf_entry *entries;
971 struct xfs_da_node_entry *btree;
972 struct xfs_da3_icnode_hdr icnodehdr;
973 struct xfs_da_intnode *node;
974 struct xfs_inode *dp = args->dp;
975 struct xfs_mount *mp = dp->i_mount;
976 struct xfs_buf *bp1 = NULL;
977 struct xfs_buf *bp2 = NULL;
978 xfs_dablk_t blkno;
979 int error;
981 trace_xfs_attr_leaf_to_node(args);
983 error = xfs_da_grow_inode(args, &blkno);
984 if (error)
985 goto out;
986 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
987 if (error)
988 goto out;
990 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
991 if (error)
992 goto out;
994 /* copy leaf to new buffer, update identifiers */
995 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
996 bp2->b_ops = bp1->b_ops;
997 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
998 if (xfs_sb_version_hascrc(&mp->m_sb)) {
999 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1000 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1002 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1005 * Set up the new root node.
1007 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1008 if (error)
1009 goto out;
1010 node = bp1->b_addr;
1011 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1012 btree = dp->d_ops->node_tree_p(node);
1014 leaf = bp2->b_addr;
1015 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1016 entries = xfs_attr3_leaf_entryp(leaf);
1018 /* both on-disk, don't endian-flip twice */
1019 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1020 btree[0].before = cpu_to_be32(blkno);
1021 icnodehdr.count = 1;
1022 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1023 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1024 error = 0;
1025 out:
1026 return error;
1029 /*========================================================================
1030 * Routines used for growing the Btree.
1031 *========================================================================*/
1034 * Create the initial contents of a leaf attribute list
1035 * or a leaf in a node attribute list.
1037 STATIC int
1038 xfs_attr3_leaf_create(
1039 struct xfs_da_args *args,
1040 xfs_dablk_t blkno,
1041 struct xfs_buf **bpp)
1043 struct xfs_attr_leafblock *leaf;
1044 struct xfs_attr3_icleaf_hdr ichdr;
1045 struct xfs_inode *dp = args->dp;
1046 struct xfs_mount *mp = dp->i_mount;
1047 struct xfs_buf *bp;
1048 int error;
1050 trace_xfs_attr_leaf_create(args);
1052 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1053 XFS_ATTR_FORK);
1054 if (error)
1055 return error;
1056 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1057 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1058 leaf = bp->b_addr;
1059 memset(leaf, 0, args->geo->blksize);
1061 memset(&ichdr, 0, sizeof(ichdr));
1062 ichdr.firstused = args->geo->blksize;
1064 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1065 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1067 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1069 hdr3->blkno = cpu_to_be64(bp->b_bn);
1070 hdr3->owner = cpu_to_be64(dp->i_ino);
1071 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1073 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1074 } else {
1075 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1076 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1078 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1080 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1081 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1083 *bpp = bp;
1084 return 0;
1088 * Split the leaf node, rebalance, then add the new entry.
1091 xfs_attr3_leaf_split(
1092 struct xfs_da_state *state,
1093 struct xfs_da_state_blk *oldblk,
1094 struct xfs_da_state_blk *newblk)
1096 xfs_dablk_t blkno;
1097 int error;
1099 trace_xfs_attr_leaf_split(state->args);
1102 * Allocate space for a new leaf node.
1104 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1105 error = xfs_da_grow_inode(state->args, &blkno);
1106 if (error)
1107 return error;
1108 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1109 if (error)
1110 return error;
1111 newblk->blkno = blkno;
1112 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1115 * Rebalance the entries across the two leaves.
1116 * NOTE: rebalance() currently depends on the 2nd block being empty.
1118 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1119 error = xfs_da3_blk_link(state, oldblk, newblk);
1120 if (error)
1121 return error;
1124 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1125 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1126 * "new" attrs info. Will need the "old" info to remove it later.
1128 * Insert the "new" entry in the correct block.
1130 if (state->inleaf) {
1131 trace_xfs_attr_leaf_add_old(state->args);
1132 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1133 } else {
1134 trace_xfs_attr_leaf_add_new(state->args);
1135 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1139 * Update last hashval in each block since we added the name.
1141 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1142 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1143 return error;
1147 * Add a name to the leaf attribute list structure.
1150 xfs_attr3_leaf_add(
1151 struct xfs_buf *bp,
1152 struct xfs_da_args *args)
1154 struct xfs_attr_leafblock *leaf;
1155 struct xfs_attr3_icleaf_hdr ichdr;
1156 int tablesize;
1157 int entsize;
1158 int sum;
1159 int tmp;
1160 int i;
1162 trace_xfs_attr_leaf_add(args);
1164 leaf = bp->b_addr;
1165 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1166 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1167 entsize = xfs_attr_leaf_newentsize(args, NULL);
1170 * Search through freemap for first-fit on new name length.
1171 * (may need to figure in size of entry struct too)
1173 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1174 + xfs_attr3_leaf_hdr_size(leaf);
1175 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1176 if (tablesize > ichdr.firstused) {
1177 sum += ichdr.freemap[i].size;
1178 continue;
1180 if (!ichdr.freemap[i].size)
1181 continue; /* no space in this map */
1182 tmp = entsize;
1183 if (ichdr.freemap[i].base < ichdr.firstused)
1184 tmp += sizeof(xfs_attr_leaf_entry_t);
1185 if (ichdr.freemap[i].size >= tmp) {
1186 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1187 goto out_log_hdr;
1189 sum += ichdr.freemap[i].size;
1193 * If there are no holes in the address space of the block,
1194 * and we don't have enough freespace, then compaction will do us
1195 * no good and we should just give up.
1197 if (!ichdr.holes && sum < entsize)
1198 return -ENOSPC;
1201 * Compact the entries to coalesce free space.
1202 * This may change the hdr->count via dropping INCOMPLETE entries.
1204 xfs_attr3_leaf_compact(args, &ichdr, bp);
1207 * After compaction, the block is guaranteed to have only one
1208 * free region, in freemap[0]. If it is not big enough, give up.
1210 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1211 tmp = -ENOSPC;
1212 goto out_log_hdr;
1215 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1217 out_log_hdr:
1218 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1219 xfs_trans_log_buf(args->trans, bp,
1220 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1221 xfs_attr3_leaf_hdr_size(leaf)));
1222 return tmp;
1226 * Add a name to a leaf attribute list structure.
1228 STATIC int
1229 xfs_attr3_leaf_add_work(
1230 struct xfs_buf *bp,
1231 struct xfs_attr3_icleaf_hdr *ichdr,
1232 struct xfs_da_args *args,
1233 int mapindex)
1235 struct xfs_attr_leafblock *leaf;
1236 struct xfs_attr_leaf_entry *entry;
1237 struct xfs_attr_leaf_name_local *name_loc;
1238 struct xfs_attr_leaf_name_remote *name_rmt;
1239 struct xfs_mount *mp;
1240 int tmp;
1241 int i;
1243 trace_xfs_attr_leaf_add_work(args);
1245 leaf = bp->b_addr;
1246 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1247 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1250 * Force open some space in the entry array and fill it in.
1252 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1253 if (args->index < ichdr->count) {
1254 tmp = ichdr->count - args->index;
1255 tmp *= sizeof(xfs_attr_leaf_entry_t);
1256 memmove(entry + 1, entry, tmp);
1257 xfs_trans_log_buf(args->trans, bp,
1258 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1260 ichdr->count++;
1263 * Allocate space for the new string (at the end of the run).
1265 mp = args->trans->t_mountp;
1266 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1267 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1268 ASSERT(ichdr->freemap[mapindex].size >=
1269 xfs_attr_leaf_newentsize(args, NULL));
1270 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1271 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1273 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1275 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1276 ichdr->freemap[mapindex].size);
1277 entry->hashval = cpu_to_be32(args->hashval);
1278 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1279 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1280 if (args->op_flags & XFS_DA_OP_RENAME) {
1281 entry->flags |= XFS_ATTR_INCOMPLETE;
1282 if ((args->blkno2 == args->blkno) &&
1283 (args->index2 <= args->index)) {
1284 args->index2++;
1287 xfs_trans_log_buf(args->trans, bp,
1288 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1289 ASSERT((args->index == 0) ||
1290 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1291 ASSERT((args->index == ichdr->count - 1) ||
1292 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1295 * For "remote" attribute values, simply note that we need to
1296 * allocate space for the "remote" value. We can't actually
1297 * allocate the extents in this transaction, and we can't decide
1298 * which blocks they should be as we might allocate more blocks
1299 * as part of this transaction (a split operation for example).
1301 if (entry->flags & XFS_ATTR_LOCAL) {
1302 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1303 name_loc->namelen = args->namelen;
1304 name_loc->valuelen = cpu_to_be16(args->valuelen);
1305 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1306 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1307 be16_to_cpu(name_loc->valuelen));
1308 } else {
1309 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1310 name_rmt->namelen = args->namelen;
1311 memcpy((char *)name_rmt->name, args->name, args->namelen);
1312 entry->flags |= XFS_ATTR_INCOMPLETE;
1313 /* just in case */
1314 name_rmt->valuelen = 0;
1315 name_rmt->valueblk = 0;
1316 args->rmtblkno = 1;
1317 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1318 args->rmtvaluelen = args->valuelen;
1320 xfs_trans_log_buf(args->trans, bp,
1321 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1322 xfs_attr_leaf_entsize(leaf, args->index)));
1325 * Update the control info for this leaf node
1327 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1328 ichdr->firstused = be16_to_cpu(entry->nameidx);
1330 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1331 + xfs_attr3_leaf_hdr_size(leaf));
1332 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1333 + xfs_attr3_leaf_hdr_size(leaf);
1335 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1336 if (ichdr->freemap[i].base == tmp) {
1337 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1338 ichdr->freemap[i].size -=
1339 min_t(uint16_t, ichdr->freemap[i].size,
1340 sizeof(xfs_attr_leaf_entry_t));
1343 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1344 return 0;
1348 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1350 STATIC void
1351 xfs_attr3_leaf_compact(
1352 struct xfs_da_args *args,
1353 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1354 struct xfs_buf *bp)
1356 struct xfs_attr_leafblock *leaf_src;
1357 struct xfs_attr_leafblock *leaf_dst;
1358 struct xfs_attr3_icleaf_hdr ichdr_src;
1359 struct xfs_trans *trans = args->trans;
1360 char *tmpbuffer;
1362 trace_xfs_attr_leaf_compact(args);
1364 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1365 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1366 memset(bp->b_addr, 0, args->geo->blksize);
1367 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1368 leaf_dst = bp->b_addr;
1371 * Copy the on-disk header back into the destination buffer to ensure
1372 * all the information in the header that is not part of the incore
1373 * header structure is preserved.
1375 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1377 /* Initialise the incore headers */
1378 ichdr_src = *ichdr_dst; /* struct copy */
1379 ichdr_dst->firstused = args->geo->blksize;
1380 ichdr_dst->usedbytes = 0;
1381 ichdr_dst->count = 0;
1382 ichdr_dst->holes = 0;
1383 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1384 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1385 ichdr_dst->freemap[0].base;
1387 /* write the header back to initialise the underlying buffer */
1388 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1391 * Copy all entry's in the same (sorted) order,
1392 * but allocate name/value pairs packed and in sequence.
1394 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1395 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1397 * this logs the entire buffer, but the caller must write the header
1398 * back to the buffer when it is finished modifying it.
1400 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1402 kmem_free(tmpbuffer);
1406 * Compare two leaf blocks "order".
1407 * Return 0 unless leaf2 should go before leaf1.
1409 static int
1410 xfs_attr3_leaf_order(
1411 struct xfs_buf *leaf1_bp,
1412 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1413 struct xfs_buf *leaf2_bp,
1414 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1416 struct xfs_attr_leaf_entry *entries1;
1417 struct xfs_attr_leaf_entry *entries2;
1419 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1420 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1421 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1422 ((be32_to_cpu(entries2[0].hashval) <
1423 be32_to_cpu(entries1[0].hashval)) ||
1424 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1425 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1426 return 1;
1428 return 0;
1432 xfs_attr_leaf_order(
1433 struct xfs_buf *leaf1_bp,
1434 struct xfs_buf *leaf2_bp)
1436 struct xfs_attr3_icleaf_hdr ichdr1;
1437 struct xfs_attr3_icleaf_hdr ichdr2;
1438 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1440 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1441 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1442 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1446 * Redistribute the attribute list entries between two leaf nodes,
1447 * taking into account the size of the new entry.
1449 * NOTE: if new block is empty, then it will get the upper half of the
1450 * old block. At present, all (one) callers pass in an empty second block.
1452 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1453 * to match what it is doing in splitting the attribute leaf block. Those
1454 * values are used in "atomic rename" operations on attributes. Note that
1455 * the "new" and "old" values can end up in different blocks.
1457 STATIC void
1458 xfs_attr3_leaf_rebalance(
1459 struct xfs_da_state *state,
1460 struct xfs_da_state_blk *blk1,
1461 struct xfs_da_state_blk *blk2)
1463 struct xfs_da_args *args;
1464 struct xfs_attr_leafblock *leaf1;
1465 struct xfs_attr_leafblock *leaf2;
1466 struct xfs_attr3_icleaf_hdr ichdr1;
1467 struct xfs_attr3_icleaf_hdr ichdr2;
1468 struct xfs_attr_leaf_entry *entries1;
1469 struct xfs_attr_leaf_entry *entries2;
1470 int count;
1471 int totallen;
1472 int max;
1473 int space;
1474 int swap;
1477 * Set up environment.
1479 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1480 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1481 leaf1 = blk1->bp->b_addr;
1482 leaf2 = blk2->bp->b_addr;
1483 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1484 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1485 ASSERT(ichdr2.count == 0);
1486 args = state->args;
1488 trace_xfs_attr_leaf_rebalance(args);
1491 * Check ordering of blocks, reverse if it makes things simpler.
1493 * NOTE: Given that all (current) callers pass in an empty
1494 * second block, this code should never set "swap".
1496 swap = 0;
1497 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1498 struct xfs_da_state_blk *tmp_blk;
1499 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1501 tmp_blk = blk1;
1502 blk1 = blk2;
1503 blk2 = tmp_blk;
1505 /* struct copies to swap them rather than reconverting */
1506 tmp_ichdr = ichdr1;
1507 ichdr1 = ichdr2;
1508 ichdr2 = tmp_ichdr;
1510 leaf1 = blk1->bp->b_addr;
1511 leaf2 = blk2->bp->b_addr;
1512 swap = 1;
1516 * Examine entries until we reduce the absolute difference in
1517 * byte usage between the two blocks to a minimum. Then get
1518 * the direction to copy and the number of elements to move.
1520 * "inleaf" is true if the new entry should be inserted into blk1.
1521 * If "swap" is also true, then reverse the sense of "inleaf".
1523 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1524 blk2, &ichdr2,
1525 &count, &totallen);
1526 if (swap)
1527 state->inleaf = !state->inleaf;
1530 * Move any entries required from leaf to leaf:
1532 if (count < ichdr1.count) {
1534 * Figure the total bytes to be added to the destination leaf.
1536 /* number entries being moved */
1537 count = ichdr1.count - count;
1538 space = ichdr1.usedbytes - totallen;
1539 space += count * sizeof(xfs_attr_leaf_entry_t);
1542 * leaf2 is the destination, compact it if it looks tight.
1544 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1545 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1546 if (space > max)
1547 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1550 * Move high entries from leaf1 to low end of leaf2.
1552 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1553 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1555 } else if (count > ichdr1.count) {
1557 * I assert that since all callers pass in an empty
1558 * second buffer, this code should never execute.
1560 ASSERT(0);
1563 * Figure the total bytes to be added to the destination leaf.
1565 /* number entries being moved */
1566 count -= ichdr1.count;
1567 space = totallen - ichdr1.usedbytes;
1568 space += count * sizeof(xfs_attr_leaf_entry_t);
1571 * leaf1 is the destination, compact it if it looks tight.
1573 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1574 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1575 if (space > max)
1576 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1579 * Move low entries from leaf2 to high end of leaf1.
1581 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1582 ichdr1.count, count);
1585 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1586 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1587 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1588 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1591 * Copy out last hashval in each block for B-tree code.
1593 entries1 = xfs_attr3_leaf_entryp(leaf1);
1594 entries2 = xfs_attr3_leaf_entryp(leaf2);
1595 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1596 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1599 * Adjust the expected index for insertion.
1600 * NOTE: this code depends on the (current) situation that the
1601 * second block was originally empty.
1603 * If the insertion point moved to the 2nd block, we must adjust
1604 * the index. We must also track the entry just following the
1605 * new entry for use in an "atomic rename" operation, that entry
1606 * is always the "old" entry and the "new" entry is what we are
1607 * inserting. The index/blkno fields refer to the "old" entry,
1608 * while the index2/blkno2 fields refer to the "new" entry.
1610 if (blk1->index > ichdr1.count) {
1611 ASSERT(state->inleaf == 0);
1612 blk2->index = blk1->index - ichdr1.count;
1613 args->index = args->index2 = blk2->index;
1614 args->blkno = args->blkno2 = blk2->blkno;
1615 } else if (blk1->index == ichdr1.count) {
1616 if (state->inleaf) {
1617 args->index = blk1->index;
1618 args->blkno = blk1->blkno;
1619 args->index2 = 0;
1620 args->blkno2 = blk2->blkno;
1621 } else {
1623 * On a double leaf split, the original attr location
1624 * is already stored in blkno2/index2, so don't
1625 * overwrite it overwise we corrupt the tree.
1627 blk2->index = blk1->index - ichdr1.count;
1628 args->index = blk2->index;
1629 args->blkno = blk2->blkno;
1630 if (!state->extravalid) {
1632 * set the new attr location to match the old
1633 * one and let the higher level split code
1634 * decide where in the leaf to place it.
1636 args->index2 = blk2->index;
1637 args->blkno2 = blk2->blkno;
1640 } else {
1641 ASSERT(state->inleaf == 1);
1642 args->index = args->index2 = blk1->index;
1643 args->blkno = args->blkno2 = blk1->blkno;
1648 * Examine entries until we reduce the absolute difference in
1649 * byte usage between the two blocks to a minimum.
1650 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1651 * GROT: there will always be enough room in either block for a new entry.
1652 * GROT: Do a double-split for this case?
1654 STATIC int
1655 xfs_attr3_leaf_figure_balance(
1656 struct xfs_da_state *state,
1657 struct xfs_da_state_blk *blk1,
1658 struct xfs_attr3_icleaf_hdr *ichdr1,
1659 struct xfs_da_state_blk *blk2,
1660 struct xfs_attr3_icleaf_hdr *ichdr2,
1661 int *countarg,
1662 int *usedbytesarg)
1664 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1665 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1666 struct xfs_attr_leaf_entry *entry;
1667 int count;
1668 int max;
1669 int index;
1670 int totallen = 0;
1671 int half;
1672 int lastdelta;
1673 int foundit = 0;
1674 int tmp;
1677 * Examine entries until we reduce the absolute difference in
1678 * byte usage between the two blocks to a minimum.
1680 max = ichdr1->count + ichdr2->count;
1681 half = (max + 1) * sizeof(*entry);
1682 half += ichdr1->usedbytes + ichdr2->usedbytes +
1683 xfs_attr_leaf_newentsize(state->args, NULL);
1684 half /= 2;
1685 lastdelta = state->args->geo->blksize;
1686 entry = xfs_attr3_leaf_entryp(leaf1);
1687 for (count = index = 0; count < max; entry++, index++, count++) {
1689 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1691 * The new entry is in the first block, account for it.
1693 if (count == blk1->index) {
1694 tmp = totallen + sizeof(*entry) +
1695 xfs_attr_leaf_newentsize(state->args, NULL);
1696 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1697 break;
1698 lastdelta = XFS_ATTR_ABS(half - tmp);
1699 totallen = tmp;
1700 foundit = 1;
1704 * Wrap around into the second block if necessary.
1706 if (count == ichdr1->count) {
1707 leaf1 = leaf2;
1708 entry = xfs_attr3_leaf_entryp(leaf1);
1709 index = 0;
1713 * Figure out if next leaf entry would be too much.
1715 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1716 index);
1717 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1718 break;
1719 lastdelta = XFS_ATTR_ABS(half - tmp);
1720 totallen = tmp;
1721 #undef XFS_ATTR_ABS
1725 * Calculate the number of usedbytes that will end up in lower block.
1726 * If new entry not in lower block, fix up the count.
1728 totallen -= count * sizeof(*entry);
1729 if (foundit) {
1730 totallen -= sizeof(*entry) +
1731 xfs_attr_leaf_newentsize(state->args, NULL);
1734 *countarg = count;
1735 *usedbytesarg = totallen;
1736 return foundit;
1739 /*========================================================================
1740 * Routines used for shrinking the Btree.
1741 *========================================================================*/
1744 * Check a leaf block and its neighbors to see if the block should be
1745 * collapsed into one or the other neighbor. Always keep the block
1746 * with the smaller block number.
1747 * If the current block is over 50% full, don't try to join it, return 0.
1748 * If the block is empty, fill in the state structure and return 2.
1749 * If it can be collapsed, fill in the state structure and return 1.
1750 * If nothing can be done, return 0.
1752 * GROT: allow for INCOMPLETE entries in calculation.
1755 xfs_attr3_leaf_toosmall(
1756 struct xfs_da_state *state,
1757 int *action)
1759 struct xfs_attr_leafblock *leaf;
1760 struct xfs_da_state_blk *blk;
1761 struct xfs_attr3_icleaf_hdr ichdr;
1762 struct xfs_buf *bp;
1763 xfs_dablk_t blkno;
1764 int bytes;
1765 int forward;
1766 int error;
1767 int retval;
1768 int i;
1770 trace_xfs_attr_leaf_toosmall(state->args);
1773 * Check for the degenerate case of the block being over 50% full.
1774 * If so, it's not worth even looking to see if we might be able
1775 * to coalesce with a sibling.
1777 blk = &state->path.blk[ state->path.active-1 ];
1778 leaf = blk->bp->b_addr;
1779 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1780 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1781 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1782 ichdr.usedbytes;
1783 if (bytes > (state->args->geo->blksize >> 1)) {
1784 *action = 0; /* blk over 50%, don't try to join */
1785 return 0;
1789 * Check for the degenerate case of the block being empty.
1790 * If the block is empty, we'll simply delete it, no need to
1791 * coalesce it with a sibling block. We choose (arbitrarily)
1792 * to merge with the forward block unless it is NULL.
1794 if (ichdr.count == 0) {
1796 * Make altpath point to the block we want to keep and
1797 * path point to the block we want to drop (this one).
1799 forward = (ichdr.forw != 0);
1800 memcpy(&state->altpath, &state->path, sizeof(state->path));
1801 error = xfs_da3_path_shift(state, &state->altpath, forward,
1802 0, &retval);
1803 if (error)
1804 return error;
1805 if (retval) {
1806 *action = 0;
1807 } else {
1808 *action = 2;
1810 return 0;
1814 * Examine each sibling block to see if we can coalesce with
1815 * at least 25% free space to spare. We need to figure out
1816 * whether to merge with the forward or the backward block.
1817 * We prefer coalescing with the lower numbered sibling so as
1818 * to shrink an attribute list over time.
1820 /* start with smaller blk num */
1821 forward = ichdr.forw < ichdr.back;
1822 for (i = 0; i < 2; forward = !forward, i++) {
1823 struct xfs_attr3_icleaf_hdr ichdr2;
1824 if (forward)
1825 blkno = ichdr.forw;
1826 else
1827 blkno = ichdr.back;
1828 if (blkno == 0)
1829 continue;
1830 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1831 blkno, -1, &bp);
1832 if (error)
1833 return error;
1835 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1837 bytes = state->args->geo->blksize -
1838 (state->args->geo->blksize >> 2) -
1839 ichdr.usedbytes - ichdr2.usedbytes -
1840 ((ichdr.count + ichdr2.count) *
1841 sizeof(xfs_attr_leaf_entry_t)) -
1842 xfs_attr3_leaf_hdr_size(leaf);
1844 xfs_trans_brelse(state->args->trans, bp);
1845 if (bytes >= 0)
1846 break; /* fits with at least 25% to spare */
1848 if (i >= 2) {
1849 *action = 0;
1850 return 0;
1854 * Make altpath point to the block we want to keep (the lower
1855 * numbered block) and path point to the block we want to drop.
1857 memcpy(&state->altpath, &state->path, sizeof(state->path));
1858 if (blkno < blk->blkno) {
1859 error = xfs_da3_path_shift(state, &state->altpath, forward,
1860 0, &retval);
1861 } else {
1862 error = xfs_da3_path_shift(state, &state->path, forward,
1863 0, &retval);
1865 if (error)
1866 return error;
1867 if (retval) {
1868 *action = 0;
1869 } else {
1870 *action = 1;
1872 return 0;
1876 * Remove a name from the leaf attribute list structure.
1878 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1879 * If two leaves are 37% full, when combined they will leave 25% free.
1882 xfs_attr3_leaf_remove(
1883 struct xfs_buf *bp,
1884 struct xfs_da_args *args)
1886 struct xfs_attr_leafblock *leaf;
1887 struct xfs_attr3_icleaf_hdr ichdr;
1888 struct xfs_attr_leaf_entry *entry;
1889 int before;
1890 int after;
1891 int smallest;
1892 int entsize;
1893 int tablesize;
1894 int tmp;
1895 int i;
1897 trace_xfs_attr_leaf_remove(args);
1899 leaf = bp->b_addr;
1900 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1902 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1903 ASSERT(args->index >= 0 && args->index < ichdr.count);
1904 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1905 xfs_attr3_leaf_hdr_size(leaf));
1907 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1909 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1910 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1913 * Scan through free region table:
1914 * check for adjacency of free'd entry with an existing one,
1915 * find smallest free region in case we need to replace it,
1916 * adjust any map that borders the entry table,
1918 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1919 + xfs_attr3_leaf_hdr_size(leaf);
1920 tmp = ichdr.freemap[0].size;
1921 before = after = -1;
1922 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1923 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1924 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1925 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1926 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1927 if (ichdr.freemap[i].base == tablesize) {
1928 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1929 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1932 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1933 be16_to_cpu(entry->nameidx)) {
1934 before = i;
1935 } else if (ichdr.freemap[i].base ==
1936 (be16_to_cpu(entry->nameidx) + entsize)) {
1937 after = i;
1938 } else if (ichdr.freemap[i].size < tmp) {
1939 tmp = ichdr.freemap[i].size;
1940 smallest = i;
1945 * Coalesce adjacent freemap regions,
1946 * or replace the smallest region.
1948 if ((before >= 0) || (after >= 0)) {
1949 if ((before >= 0) && (after >= 0)) {
1950 ichdr.freemap[before].size += entsize;
1951 ichdr.freemap[before].size += ichdr.freemap[after].size;
1952 ichdr.freemap[after].base = 0;
1953 ichdr.freemap[after].size = 0;
1954 } else if (before >= 0) {
1955 ichdr.freemap[before].size += entsize;
1956 } else {
1957 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1958 ichdr.freemap[after].size += entsize;
1960 } else {
1962 * Replace smallest region (if it is smaller than free'd entry)
1964 if (ichdr.freemap[smallest].size < entsize) {
1965 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1966 ichdr.freemap[smallest].size = entsize;
1971 * Did we remove the first entry?
1973 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1974 smallest = 1;
1975 else
1976 smallest = 0;
1979 * Compress the remaining entries and zero out the removed stuff.
1981 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1982 ichdr.usedbytes -= entsize;
1983 xfs_trans_log_buf(args->trans, bp,
1984 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1985 entsize));
1987 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1988 memmove(entry, entry + 1, tmp);
1989 ichdr.count--;
1990 xfs_trans_log_buf(args->trans, bp,
1991 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1993 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1994 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1997 * If we removed the first entry, re-find the first used byte
1998 * in the name area. Note that if the entry was the "firstused",
1999 * then we don't have a "hole" in our block resulting from
2000 * removing the name.
2002 if (smallest) {
2003 tmp = args->geo->blksize;
2004 entry = xfs_attr3_leaf_entryp(leaf);
2005 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2006 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2007 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2009 if (be16_to_cpu(entry->nameidx) < tmp)
2010 tmp = be16_to_cpu(entry->nameidx);
2012 ichdr.firstused = tmp;
2013 ASSERT(ichdr.firstused != 0);
2014 } else {
2015 ichdr.holes = 1; /* mark as needing compaction */
2017 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2018 xfs_trans_log_buf(args->trans, bp,
2019 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2020 xfs_attr3_leaf_hdr_size(leaf)));
2023 * Check if leaf is less than 50% full, caller may want to
2024 * "join" the leaf with a sibling if so.
2026 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2027 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2029 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2033 * Move all the attribute list entries from drop_leaf into save_leaf.
2035 void
2036 xfs_attr3_leaf_unbalance(
2037 struct xfs_da_state *state,
2038 struct xfs_da_state_blk *drop_blk,
2039 struct xfs_da_state_blk *save_blk)
2041 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2042 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2043 struct xfs_attr3_icleaf_hdr drophdr;
2044 struct xfs_attr3_icleaf_hdr savehdr;
2045 struct xfs_attr_leaf_entry *entry;
2047 trace_xfs_attr_leaf_unbalance(state->args);
2049 drop_leaf = drop_blk->bp->b_addr;
2050 save_leaf = save_blk->bp->b_addr;
2051 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2052 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2053 entry = xfs_attr3_leaf_entryp(drop_leaf);
2056 * Save last hashval from dying block for later Btree fixup.
2058 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2061 * Check if we need a temp buffer, or can we do it in place.
2062 * Note that we don't check "leaf" for holes because we will
2063 * always be dropping it, toosmall() decided that for us already.
2065 if (savehdr.holes == 0) {
2067 * dest leaf has no holes, so we add there. May need
2068 * to make some room in the entry array.
2070 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2071 drop_blk->bp, &drophdr)) {
2072 xfs_attr3_leaf_moveents(state->args,
2073 drop_leaf, &drophdr, 0,
2074 save_leaf, &savehdr, 0,
2075 drophdr.count);
2076 } else {
2077 xfs_attr3_leaf_moveents(state->args,
2078 drop_leaf, &drophdr, 0,
2079 save_leaf, &savehdr,
2080 savehdr.count, drophdr.count);
2082 } else {
2084 * Destination has holes, so we make a temporary copy
2085 * of the leaf and add them both to that.
2087 struct xfs_attr_leafblock *tmp_leaf;
2088 struct xfs_attr3_icleaf_hdr tmphdr;
2090 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2093 * Copy the header into the temp leaf so that all the stuff
2094 * not in the incore header is present and gets copied back in
2095 * once we've moved all the entries.
2097 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2099 memset(&tmphdr, 0, sizeof(tmphdr));
2100 tmphdr.magic = savehdr.magic;
2101 tmphdr.forw = savehdr.forw;
2102 tmphdr.back = savehdr.back;
2103 tmphdr.firstused = state->args->geo->blksize;
2105 /* write the header to the temp buffer to initialise it */
2106 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2108 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2109 drop_blk->bp, &drophdr)) {
2110 xfs_attr3_leaf_moveents(state->args,
2111 drop_leaf, &drophdr, 0,
2112 tmp_leaf, &tmphdr, 0,
2113 drophdr.count);
2114 xfs_attr3_leaf_moveents(state->args,
2115 save_leaf, &savehdr, 0,
2116 tmp_leaf, &tmphdr, tmphdr.count,
2117 savehdr.count);
2118 } else {
2119 xfs_attr3_leaf_moveents(state->args,
2120 save_leaf, &savehdr, 0,
2121 tmp_leaf, &tmphdr, 0,
2122 savehdr.count);
2123 xfs_attr3_leaf_moveents(state->args,
2124 drop_leaf, &drophdr, 0,
2125 tmp_leaf, &tmphdr, tmphdr.count,
2126 drophdr.count);
2128 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2129 savehdr = tmphdr; /* struct copy */
2130 kmem_free(tmp_leaf);
2133 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2134 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2135 state->args->geo->blksize - 1);
2138 * Copy out last hashval in each block for B-tree code.
2140 entry = xfs_attr3_leaf_entryp(save_leaf);
2141 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2144 /*========================================================================
2145 * Routines used for finding things in the Btree.
2146 *========================================================================*/
2149 * Look up a name in a leaf attribute list structure.
2150 * This is the internal routine, it uses the caller's buffer.
2152 * Note that duplicate keys are allowed, but only check within the
2153 * current leaf node. The Btree code must check in adjacent leaf nodes.
2155 * Return in args->index the index into the entry[] array of either
2156 * the found entry, or where the entry should have been (insert before
2157 * that entry).
2159 * Don't change the args->value unless we find the attribute.
2162 xfs_attr3_leaf_lookup_int(
2163 struct xfs_buf *bp,
2164 struct xfs_da_args *args)
2166 struct xfs_attr_leafblock *leaf;
2167 struct xfs_attr3_icleaf_hdr ichdr;
2168 struct xfs_attr_leaf_entry *entry;
2169 struct xfs_attr_leaf_entry *entries;
2170 struct xfs_attr_leaf_name_local *name_loc;
2171 struct xfs_attr_leaf_name_remote *name_rmt;
2172 xfs_dahash_t hashval;
2173 int probe;
2174 int span;
2176 trace_xfs_attr_leaf_lookup(args);
2178 leaf = bp->b_addr;
2179 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2180 entries = xfs_attr3_leaf_entryp(leaf);
2181 ASSERT(ichdr.count < args->geo->blksize / 8);
2184 * Binary search. (note: small blocks will skip this loop)
2186 hashval = args->hashval;
2187 probe = span = ichdr.count / 2;
2188 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2189 span /= 2;
2190 if (be32_to_cpu(entry->hashval) < hashval)
2191 probe += span;
2192 else if (be32_to_cpu(entry->hashval) > hashval)
2193 probe -= span;
2194 else
2195 break;
2197 ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2198 ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2201 * Since we may have duplicate hashval's, find the first matching
2202 * hashval in the leaf.
2204 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2205 entry--;
2206 probe--;
2208 while (probe < ichdr.count &&
2209 be32_to_cpu(entry->hashval) < hashval) {
2210 entry++;
2211 probe++;
2213 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2214 args->index = probe;
2215 return -ENOATTR;
2219 * Duplicate keys may be present, so search all of them for a match.
2221 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2222 entry++, probe++) {
2224 * GROT: Add code to remove incomplete entries.
2227 * If we are looking for INCOMPLETE entries, show only those.
2228 * If we are looking for complete entries, show only those.
2230 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2231 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2232 continue;
2234 if (entry->flags & XFS_ATTR_LOCAL) {
2235 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2236 if (name_loc->namelen != args->namelen)
2237 continue;
2238 if (memcmp(args->name, name_loc->nameval,
2239 args->namelen) != 0)
2240 continue;
2241 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2242 continue;
2243 args->index = probe;
2244 return -EEXIST;
2245 } else {
2246 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2247 if (name_rmt->namelen != args->namelen)
2248 continue;
2249 if (memcmp(args->name, name_rmt->name,
2250 args->namelen) != 0)
2251 continue;
2252 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2253 continue;
2254 args->index = probe;
2255 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2256 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2257 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2258 args->dp->i_mount,
2259 args->rmtvaluelen);
2260 return -EEXIST;
2263 args->index = probe;
2264 return -ENOATTR;
2268 * Get the value associated with an attribute name from a leaf attribute
2269 * list structure.
2272 xfs_attr3_leaf_getvalue(
2273 struct xfs_buf *bp,
2274 struct xfs_da_args *args)
2276 struct xfs_attr_leafblock *leaf;
2277 struct xfs_attr3_icleaf_hdr ichdr;
2278 struct xfs_attr_leaf_entry *entry;
2279 struct xfs_attr_leaf_name_local *name_loc;
2280 struct xfs_attr_leaf_name_remote *name_rmt;
2281 int valuelen;
2283 leaf = bp->b_addr;
2284 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2285 ASSERT(ichdr.count < args->geo->blksize / 8);
2286 ASSERT(args->index < ichdr.count);
2288 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2289 if (entry->flags & XFS_ATTR_LOCAL) {
2290 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2291 ASSERT(name_loc->namelen == args->namelen);
2292 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2293 valuelen = be16_to_cpu(name_loc->valuelen);
2294 if (args->flags & ATTR_KERNOVAL) {
2295 args->valuelen = valuelen;
2296 return 0;
2298 if (args->valuelen < valuelen) {
2299 args->valuelen = valuelen;
2300 return -ERANGE;
2302 args->valuelen = valuelen;
2303 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2304 } else {
2305 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2306 ASSERT(name_rmt->namelen == args->namelen);
2307 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2308 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2309 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2310 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2311 args->rmtvaluelen);
2312 if (args->flags & ATTR_KERNOVAL) {
2313 args->valuelen = args->rmtvaluelen;
2314 return 0;
2316 if (args->valuelen < args->rmtvaluelen) {
2317 args->valuelen = args->rmtvaluelen;
2318 return -ERANGE;
2320 args->valuelen = args->rmtvaluelen;
2322 return 0;
2325 /*========================================================================
2326 * Utility routines.
2327 *========================================================================*/
2330 * Move the indicated entries from one leaf to another.
2331 * NOTE: this routine modifies both source and destination leaves.
2333 /*ARGSUSED*/
2334 STATIC void
2335 xfs_attr3_leaf_moveents(
2336 struct xfs_da_args *args,
2337 struct xfs_attr_leafblock *leaf_s,
2338 struct xfs_attr3_icleaf_hdr *ichdr_s,
2339 int start_s,
2340 struct xfs_attr_leafblock *leaf_d,
2341 struct xfs_attr3_icleaf_hdr *ichdr_d,
2342 int start_d,
2343 int count)
2345 struct xfs_attr_leaf_entry *entry_s;
2346 struct xfs_attr_leaf_entry *entry_d;
2347 int desti;
2348 int tmp;
2349 int i;
2352 * Check for nothing to do.
2354 if (count == 0)
2355 return;
2358 * Set up environment.
2360 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2361 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2362 ASSERT(ichdr_s->magic == ichdr_d->magic);
2363 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2364 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2365 + xfs_attr3_leaf_hdr_size(leaf_s));
2366 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2367 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2368 + xfs_attr3_leaf_hdr_size(leaf_d));
2370 ASSERT(start_s < ichdr_s->count);
2371 ASSERT(start_d <= ichdr_d->count);
2372 ASSERT(count <= ichdr_s->count);
2376 * Move the entries in the destination leaf up to make a hole?
2378 if (start_d < ichdr_d->count) {
2379 tmp = ichdr_d->count - start_d;
2380 tmp *= sizeof(xfs_attr_leaf_entry_t);
2381 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2382 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2383 memmove(entry_d, entry_s, tmp);
2387 * Copy all entry's in the same (sorted) order,
2388 * but allocate attribute info packed and in sequence.
2390 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2391 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2392 desti = start_d;
2393 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2394 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2395 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2396 #ifdef GROT
2398 * Code to drop INCOMPLETE entries. Difficult to use as we
2399 * may also need to change the insertion index. Code turned
2400 * off for 6.2, should be revisited later.
2402 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2403 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2404 ichdr_s->usedbytes -= tmp;
2405 ichdr_s->count -= 1;
2406 entry_d--; /* to compensate for ++ in loop hdr */
2407 desti--;
2408 if ((start_s + i) < offset)
2409 result++; /* insertion index adjustment */
2410 } else {
2411 #endif /* GROT */
2412 ichdr_d->firstused -= tmp;
2413 /* both on-disk, don't endian flip twice */
2414 entry_d->hashval = entry_s->hashval;
2415 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2416 entry_d->flags = entry_s->flags;
2417 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2418 <= args->geo->blksize);
2419 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2420 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2421 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2422 <= args->geo->blksize);
2423 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2424 ichdr_s->usedbytes -= tmp;
2425 ichdr_d->usedbytes += tmp;
2426 ichdr_s->count -= 1;
2427 ichdr_d->count += 1;
2428 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2429 + xfs_attr3_leaf_hdr_size(leaf_d);
2430 ASSERT(ichdr_d->firstused >= tmp);
2431 #ifdef GROT
2433 #endif /* GROT */
2437 * Zero out the entries we just copied.
2439 if (start_s == ichdr_s->count) {
2440 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2441 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2442 ASSERT(((char *)entry_s + tmp) <=
2443 ((char *)leaf_s + args->geo->blksize));
2444 memset(entry_s, 0, tmp);
2445 } else {
2447 * Move the remaining entries down to fill the hole,
2448 * then zero the entries at the top.
2450 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2451 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2452 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2453 memmove(entry_d, entry_s, tmp);
2455 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2456 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2457 ASSERT(((char *)entry_s + tmp) <=
2458 ((char *)leaf_s + args->geo->blksize));
2459 memset(entry_s, 0, tmp);
2463 * Fill in the freemap information
2465 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2466 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2467 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2468 ichdr_d->freemap[1].base = 0;
2469 ichdr_d->freemap[2].base = 0;
2470 ichdr_d->freemap[1].size = 0;
2471 ichdr_d->freemap[2].size = 0;
2472 ichdr_s->holes = 1; /* leaf may not be compact */
2476 * Pick up the last hashvalue from a leaf block.
2478 xfs_dahash_t
2479 xfs_attr_leaf_lasthash(
2480 struct xfs_buf *bp,
2481 int *count)
2483 struct xfs_attr3_icleaf_hdr ichdr;
2484 struct xfs_attr_leaf_entry *entries;
2485 struct xfs_mount *mp = bp->b_target->bt_mount;
2487 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2488 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2489 if (count)
2490 *count = ichdr.count;
2491 if (!ichdr.count)
2492 return 0;
2493 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2497 * Calculate the number of bytes used to store the indicated attribute
2498 * (whether local or remote only calculate bytes in this block).
2500 STATIC int
2501 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2503 struct xfs_attr_leaf_entry *entries;
2504 xfs_attr_leaf_name_local_t *name_loc;
2505 xfs_attr_leaf_name_remote_t *name_rmt;
2506 int size;
2508 entries = xfs_attr3_leaf_entryp(leaf);
2509 if (entries[index].flags & XFS_ATTR_LOCAL) {
2510 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2511 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2512 be16_to_cpu(name_loc->valuelen));
2513 } else {
2514 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2515 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2517 return size;
2521 * Calculate the number of bytes that would be required to store the new
2522 * attribute (whether local or remote only calculate bytes in this block).
2523 * This routine decides as a side effect whether the attribute will be
2524 * a "local" or a "remote" attribute.
2527 xfs_attr_leaf_newentsize(
2528 struct xfs_da_args *args,
2529 int *local)
2531 int size;
2533 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2534 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2535 if (local)
2536 *local = 1;
2537 return size;
2539 if (local)
2540 *local = 0;
2541 return xfs_attr_leaf_entsize_remote(args->namelen);
2545 /*========================================================================
2546 * Manage the INCOMPLETE flag in a leaf entry
2547 *========================================================================*/
2550 * Clear the INCOMPLETE flag on an entry in a leaf block.
2553 xfs_attr3_leaf_clearflag(
2554 struct xfs_da_args *args)
2556 struct xfs_attr_leafblock *leaf;
2557 struct xfs_attr_leaf_entry *entry;
2558 struct xfs_attr_leaf_name_remote *name_rmt;
2559 struct xfs_buf *bp;
2560 int error;
2561 #ifdef DEBUG
2562 struct xfs_attr3_icleaf_hdr ichdr;
2563 xfs_attr_leaf_name_local_t *name_loc;
2564 int namelen;
2565 char *name;
2566 #endif /* DEBUG */
2568 trace_xfs_attr_leaf_clearflag(args);
2570 * Set up the operation.
2572 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2573 if (error)
2574 return error;
2576 leaf = bp->b_addr;
2577 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2578 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2580 #ifdef DEBUG
2581 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2582 ASSERT(args->index < ichdr.count);
2583 ASSERT(args->index >= 0);
2585 if (entry->flags & XFS_ATTR_LOCAL) {
2586 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2587 namelen = name_loc->namelen;
2588 name = (char *)name_loc->nameval;
2589 } else {
2590 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2591 namelen = name_rmt->namelen;
2592 name = (char *)name_rmt->name;
2594 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2595 ASSERT(namelen == args->namelen);
2596 ASSERT(memcmp(name, args->name, namelen) == 0);
2597 #endif /* DEBUG */
2599 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2600 xfs_trans_log_buf(args->trans, bp,
2601 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2603 if (args->rmtblkno) {
2604 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2605 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2606 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2607 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2608 xfs_trans_log_buf(args->trans, bp,
2609 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2613 * Commit the flag value change and start the next trans in series.
2615 return xfs_trans_roll_inode(&args->trans, args->dp);
2619 * Set the INCOMPLETE flag on an entry in a leaf block.
2622 xfs_attr3_leaf_setflag(
2623 struct xfs_da_args *args)
2625 struct xfs_attr_leafblock *leaf;
2626 struct xfs_attr_leaf_entry *entry;
2627 struct xfs_attr_leaf_name_remote *name_rmt;
2628 struct xfs_buf *bp;
2629 int error;
2630 #ifdef DEBUG
2631 struct xfs_attr3_icleaf_hdr ichdr;
2632 #endif
2634 trace_xfs_attr_leaf_setflag(args);
2637 * Set up the operation.
2639 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2640 if (error)
2641 return error;
2643 leaf = bp->b_addr;
2644 #ifdef DEBUG
2645 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2646 ASSERT(args->index < ichdr.count);
2647 ASSERT(args->index >= 0);
2648 #endif
2649 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2651 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2652 entry->flags |= XFS_ATTR_INCOMPLETE;
2653 xfs_trans_log_buf(args->trans, bp,
2654 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2655 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2656 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2657 name_rmt->valueblk = 0;
2658 name_rmt->valuelen = 0;
2659 xfs_trans_log_buf(args->trans, bp,
2660 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2664 * Commit the flag value change and start the next trans in series.
2666 return xfs_trans_roll_inode(&args->trans, args->dp);
2670 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2671 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2672 * entry given by args->blkno2/index2.
2674 * Note that they could be in different blocks, or in the same block.
2677 xfs_attr3_leaf_flipflags(
2678 struct xfs_da_args *args)
2680 struct xfs_attr_leafblock *leaf1;
2681 struct xfs_attr_leafblock *leaf2;
2682 struct xfs_attr_leaf_entry *entry1;
2683 struct xfs_attr_leaf_entry *entry2;
2684 struct xfs_attr_leaf_name_remote *name_rmt;
2685 struct xfs_buf *bp1;
2686 struct xfs_buf *bp2;
2687 int error;
2688 #ifdef DEBUG
2689 struct xfs_attr3_icleaf_hdr ichdr1;
2690 struct xfs_attr3_icleaf_hdr ichdr2;
2691 xfs_attr_leaf_name_local_t *name_loc;
2692 int namelen1, namelen2;
2693 char *name1, *name2;
2694 #endif /* DEBUG */
2696 trace_xfs_attr_leaf_flipflags(args);
2699 * Read the block containing the "old" attr
2701 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2702 if (error)
2703 return error;
2706 * Read the block containing the "new" attr, if it is different
2708 if (args->blkno2 != args->blkno) {
2709 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2710 -1, &bp2);
2711 if (error)
2712 return error;
2713 } else {
2714 bp2 = bp1;
2717 leaf1 = bp1->b_addr;
2718 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2720 leaf2 = bp2->b_addr;
2721 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2723 #ifdef DEBUG
2724 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2725 ASSERT(args->index < ichdr1.count);
2726 ASSERT(args->index >= 0);
2728 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2729 ASSERT(args->index2 < ichdr2.count);
2730 ASSERT(args->index2 >= 0);
2732 if (entry1->flags & XFS_ATTR_LOCAL) {
2733 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2734 namelen1 = name_loc->namelen;
2735 name1 = (char *)name_loc->nameval;
2736 } else {
2737 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2738 namelen1 = name_rmt->namelen;
2739 name1 = (char *)name_rmt->name;
2741 if (entry2->flags & XFS_ATTR_LOCAL) {
2742 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2743 namelen2 = name_loc->namelen;
2744 name2 = (char *)name_loc->nameval;
2745 } else {
2746 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2747 namelen2 = name_rmt->namelen;
2748 name2 = (char *)name_rmt->name;
2750 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2751 ASSERT(namelen1 == namelen2);
2752 ASSERT(memcmp(name1, name2, namelen1) == 0);
2753 #endif /* DEBUG */
2755 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2756 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2758 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2759 xfs_trans_log_buf(args->trans, bp1,
2760 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2761 if (args->rmtblkno) {
2762 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2763 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2764 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2765 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2766 xfs_trans_log_buf(args->trans, bp1,
2767 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2770 entry2->flags |= XFS_ATTR_INCOMPLETE;
2771 xfs_trans_log_buf(args->trans, bp2,
2772 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2773 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2774 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2775 name_rmt->valueblk = 0;
2776 name_rmt->valuelen = 0;
2777 xfs_trans_log_buf(args->trans, bp2,
2778 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2782 * Commit the flag value change and start the next trans in series.
2784 error = xfs_trans_roll_inode(&args->trans, args->dp);
2786 return error;