]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blob - libxfs/xfs_attr_leaf.c
Merge branch 'libxfs-4.15-sync' into for-next
[thirdparty/xfsprogs-dev.git] / libxfs / xfs_attr_leaf.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
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.
9 *
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.
14 *
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
18 */
19 #include "libxfs_priv.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_bmap_btree.h"
33 #include "xfs_bmap.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_attr_remote.h"
36 #include "xfs_attr_leaf.h"
37 #include "xfs_trace.h"
38 #include "xfs_cksum.h"
39 #include "xfs_dir2.h"
40
41
42 /*
43 * xfs_attr_leaf.c
44 *
45 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
46 */
47
48 /*========================================================================
49 * Function prototypes for the kernel.
50 *========================================================================*/
51
52 /*
53 * Routines used for growing the Btree.
54 */
55 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
56 xfs_dablk_t which_block, struct xfs_buf **bpp);
57 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
58 struct xfs_attr3_icleaf_hdr *ichdr,
59 struct xfs_da_args *args, int freemap_index);
60 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
61 struct xfs_attr3_icleaf_hdr *ichdr,
62 struct xfs_buf *leaf_buffer);
63 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
64 xfs_da_state_blk_t *blk1,
65 xfs_da_state_blk_t *blk2);
66 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
67 xfs_da_state_blk_t *leaf_blk_1,
68 struct xfs_attr3_icleaf_hdr *ichdr1,
69 xfs_da_state_blk_t *leaf_blk_2,
70 struct xfs_attr3_icleaf_hdr *ichdr2,
71 int *number_entries_in_blk1,
72 int *number_usedbytes_in_blk1);
73
74 /*
75 * Utility routines.
76 */
77 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
78 struct xfs_attr_leafblock *src_leaf,
79 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
80 struct xfs_attr_leafblock *dst_leaf,
81 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
82 int move_count);
83 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
84
85 /*
86 * attr3 block 'firstused' conversion helpers.
87 *
88 * firstused refers to the offset of the first used byte of the nameval region
89 * of an attr leaf block. The region starts at the tail of the block and expands
90 * backwards towards the middle. As such, firstused is initialized to the block
91 * size for an empty leaf block and is reduced from there.
92 *
93 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
94 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
95 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
96 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
97 * the attr block size. The following helpers manage the conversion between the
98 * in-core and on-disk formats.
99 */
100
101 static void
102 xfs_attr3_leaf_firstused_from_disk(
103 struct xfs_da_geometry *geo,
104 struct xfs_attr3_icleaf_hdr *to,
105 struct xfs_attr_leafblock *from)
106 {
107 struct xfs_attr3_leaf_hdr *hdr3;
108
109 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
110 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
111 to->firstused = be16_to_cpu(hdr3->firstused);
112 } else {
113 to->firstused = be16_to_cpu(from->hdr.firstused);
114 }
115
116 /*
117 * Convert from the magic fsb size value to actual blocksize. This
118 * should only occur for empty blocks when the block size overflows
119 * 16-bits.
120 */
121 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
122 ASSERT(!to->count && !to->usedbytes);
123 ASSERT(geo->blksize > USHRT_MAX);
124 to->firstused = geo->blksize;
125 }
126 }
127
128 static void
129 xfs_attr3_leaf_firstused_to_disk(
130 struct xfs_da_geometry *geo,
131 struct xfs_attr_leafblock *to,
132 struct xfs_attr3_icleaf_hdr *from)
133 {
134 struct xfs_attr3_leaf_hdr *hdr3;
135 uint32_t firstused;
136
137 /* magic value should only be seen on disk */
138 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
139
140 /*
141 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
142 * value. This only overflows at the max supported value of 64k. Use the
143 * magic on-disk value to represent block size in this case.
144 */
145 firstused = from->firstused;
146 if (firstused > USHRT_MAX) {
147 ASSERT(from->firstused == geo->blksize);
148 firstused = XFS_ATTR3_LEAF_NULLOFF;
149 }
150
151 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
152 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
153 hdr3->firstused = cpu_to_be16(firstused);
154 } else {
155 to->hdr.firstused = cpu_to_be16(firstused);
156 }
157 }
158
159 void
160 xfs_attr3_leaf_hdr_from_disk(
161 struct xfs_da_geometry *geo,
162 struct xfs_attr3_icleaf_hdr *to,
163 struct xfs_attr_leafblock *from)
164 {
165 int i;
166
167 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
168 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
169
170 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
171 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
172
173 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
174 to->back = be32_to_cpu(hdr3->info.hdr.back);
175 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
176 to->count = be16_to_cpu(hdr3->count);
177 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
178 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
179 to->holes = hdr3->holes;
180
181 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
182 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
183 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
184 }
185 return;
186 }
187 to->forw = be32_to_cpu(from->hdr.info.forw);
188 to->back = be32_to_cpu(from->hdr.info.back);
189 to->magic = be16_to_cpu(from->hdr.info.magic);
190 to->count = be16_to_cpu(from->hdr.count);
191 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
192 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
193 to->holes = from->hdr.holes;
194
195 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
196 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
197 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
198 }
199 }
200
201 void
202 xfs_attr3_leaf_hdr_to_disk(
203 struct xfs_da_geometry *geo,
204 struct xfs_attr_leafblock *to,
205 struct xfs_attr3_icleaf_hdr *from)
206 {
207 int i;
208
209 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
210 from->magic == XFS_ATTR3_LEAF_MAGIC);
211
212 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
213 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
214
215 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
216 hdr3->info.hdr.back = cpu_to_be32(from->back);
217 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
218 hdr3->count = cpu_to_be16(from->count);
219 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
220 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
221 hdr3->holes = from->holes;
222 hdr3->pad1 = 0;
223
224 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
225 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
226 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
227 }
228 return;
229 }
230 to->hdr.info.forw = cpu_to_be32(from->forw);
231 to->hdr.info.back = cpu_to_be32(from->back);
232 to->hdr.info.magic = cpu_to_be16(from->magic);
233 to->hdr.count = cpu_to_be16(from->count);
234 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
235 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
236 to->hdr.holes = from->holes;
237 to->hdr.pad1 = 0;
238
239 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
240 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
241 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
242 }
243 }
244
245 static bool
246 xfs_attr3_leaf_verify(
247 struct xfs_buf *bp)
248 {
249 struct xfs_mount *mp = bp->b_target->bt_mount;
250 struct xfs_attr_leafblock *leaf = bp->b_addr;
251 struct xfs_perag *pag = bp->b_pag;
252 struct xfs_attr3_icleaf_hdr ichdr;
253
254 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
255
256 if (xfs_sb_version_hascrc(&mp->m_sb)) {
257 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
258
259 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
260 return false;
261
262 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
263 return false;
264 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
265 return false;
266 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
267 return false;
268 } else {
269 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
270 return false;
271 }
272 /*
273 * In recovery there is a transient state where count == 0 is valid
274 * because we may have transitioned an empty shortform attr to a leaf
275 * if the attr didn't fit in shortform.
276 */
277 if (pag && pag->pagf_init && ichdr.count == 0)
278 return false;
279
280 /* XXX: need to range check rest of attr header values */
281 /* XXX: hash order check? */
282
283 return true;
284 }
285
286 static void
287 xfs_attr3_leaf_write_verify(
288 struct xfs_buf *bp)
289 {
290 struct xfs_mount *mp = bp->b_target->bt_mount;
291 struct xfs_buf_log_item *bip = bp->b_fspriv;
292 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
293
294 if (!xfs_attr3_leaf_verify(bp)) {
295 xfs_buf_ioerror(bp, -EFSCORRUPTED);
296 xfs_verifier_error(bp);
297 return;
298 }
299
300 if (!xfs_sb_version_hascrc(&mp->m_sb))
301 return;
302
303 if (bip)
304 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
305
306 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
307 }
308
309 /*
310 * leaf/node format detection on trees is sketchy, so a node read can be done on
311 * leaf level blocks when detection identifies the tree as a node format tree
312 * incorrectly. In this case, we need to swap the verifier to match the correct
313 * format of the block being read.
314 */
315 static void
316 xfs_attr3_leaf_read_verify(
317 struct xfs_buf *bp)
318 {
319 struct xfs_mount *mp = bp->b_target->bt_mount;
320
321 if (xfs_sb_version_hascrc(&mp->m_sb) &&
322 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
323 xfs_buf_ioerror(bp, -EFSBADCRC);
324 else if (!xfs_attr3_leaf_verify(bp))
325 xfs_buf_ioerror(bp, -EFSCORRUPTED);
326
327 if (bp->b_error)
328 xfs_verifier_error(bp);
329 }
330
331 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
332 .name = "xfs_attr3_leaf",
333 .verify_read = xfs_attr3_leaf_read_verify,
334 .verify_write = xfs_attr3_leaf_write_verify,
335 };
336
337 int
338 xfs_attr3_leaf_read(
339 struct xfs_trans *tp,
340 struct xfs_inode *dp,
341 xfs_dablk_t bno,
342 xfs_daddr_t mappedbno,
343 struct xfs_buf **bpp)
344 {
345 int err;
346
347 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
348 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
349 if (!err && tp && *bpp)
350 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
351 return err;
352 }
353
354 /*========================================================================
355 * Namespace helper routines
356 *========================================================================*/
357
358 /*
359 * If namespace bits don't match return 0.
360 * If all match then return 1.
361 */
362 STATIC int
363 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
364 {
365 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
366 }
367
368
369 /*========================================================================
370 * External routines when attribute fork size < XFS_LITINO(mp).
371 *========================================================================*/
372
373 /*
374 * Query whether the requested number of additional bytes of extended
375 * attribute space will be able to fit inline.
376 *
377 * Returns zero if not, else the di_forkoff fork offset to be used in the
378 * literal area for attribute data once the new bytes have been added.
379 *
380 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
381 * special case for dev/uuid inodes, they have fixed size data forks.
382 */
383 int
384 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
385 {
386 int offset;
387 int minforkoff; /* lower limit on valid forkoff locations */
388 int maxforkoff; /* upper limit on valid forkoff locations */
389 int dsize;
390 xfs_mount_t *mp = dp->i_mount;
391
392 /* rounded down */
393 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
394
395 if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
396 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
397 return (offset >= minforkoff) ? minforkoff : 0;
398 }
399
400 /*
401 * If the requested numbers of bytes is smaller or equal to the
402 * current attribute fork size we can always proceed.
403 *
404 * Note that if_bytes in the data fork might actually be larger than
405 * the current data fork size is due to delalloc extents. In that
406 * case either the extent count will go down when they are converted
407 * to real extents, or the delalloc conversion will take care of the
408 * literal area rebalancing.
409 */
410 if (bytes <= XFS_IFORK_ASIZE(dp))
411 return dp->i_d.di_forkoff;
412
413 /*
414 * For attr2 we can try to move the forkoff if there is space in the
415 * literal area, but for the old format we are done if there is no
416 * space in the fixed attribute fork.
417 */
418 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
419 return 0;
420
421 dsize = dp->i_df.if_bytes;
422
423 switch (dp->i_d.di_format) {
424 case XFS_DINODE_FMT_EXTENTS:
425 /*
426 * If there is no attr fork and the data fork is extents,
427 * determine if creating the default attr fork will result
428 * in the extents form migrating to btree. If so, the
429 * minimum offset only needs to be the space required for
430 * the btree root.
431 */
432 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
433 xfs_default_attroffset(dp))
434 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
435 break;
436 case XFS_DINODE_FMT_BTREE:
437 /*
438 * If we have a data btree then keep forkoff if we have one,
439 * otherwise we are adding a new attr, so then we set
440 * minforkoff to where the btree root can finish so we have
441 * plenty of room for attrs
442 */
443 if (dp->i_d.di_forkoff) {
444 if (offset < dp->i_d.di_forkoff)
445 return 0;
446 return dp->i_d.di_forkoff;
447 }
448 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
449 break;
450 }
451
452 /*
453 * A data fork btree root must have space for at least
454 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
455 */
456 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
457 minforkoff = roundup(minforkoff, 8) >> 3;
458
459 /* attr fork btree root can have at least this many key/ptr pairs */
460 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
461 XFS_BMDR_SPACE_CALC(MINABTPTRS);
462 maxforkoff = maxforkoff >> 3; /* rounded down */
463
464 if (offset >= maxforkoff)
465 return maxforkoff;
466 if (offset >= minforkoff)
467 return offset;
468 return 0;
469 }
470
471 /*
472 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
473 */
474 STATIC void
475 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
476 {
477 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
478 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
479 spin_lock(&mp->m_sb_lock);
480 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
481 xfs_sb_version_addattr2(&mp->m_sb);
482 spin_unlock(&mp->m_sb_lock);
483 xfs_log_sb(tp);
484 } else
485 spin_unlock(&mp->m_sb_lock);
486 }
487 }
488
489 /*
490 * Create the initial contents of a shortform attribute list.
491 */
492 void
493 xfs_attr_shortform_create(xfs_da_args_t *args)
494 {
495 xfs_attr_sf_hdr_t *hdr;
496 xfs_inode_t *dp;
497 xfs_ifork_t *ifp;
498
499 trace_xfs_attr_sf_create(args);
500
501 dp = args->dp;
502 ASSERT(dp != NULL);
503 ifp = dp->i_afp;
504 ASSERT(ifp != NULL);
505 ASSERT(ifp->if_bytes == 0);
506 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
507 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
508 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
509 ifp->if_flags |= XFS_IFINLINE;
510 } else {
511 ASSERT(ifp->if_flags & XFS_IFINLINE);
512 }
513 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
514 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
515 hdr->count = 0;
516 hdr->totsize = cpu_to_be16(sizeof(*hdr));
517 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
518 }
519
520 /*
521 * Add a name/value pair to the shortform attribute list.
522 * Overflow from the inode has already been checked for.
523 */
524 void
525 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
526 {
527 xfs_attr_shortform_t *sf;
528 xfs_attr_sf_entry_t *sfe;
529 int i, offset, size;
530 xfs_mount_t *mp;
531 xfs_inode_t *dp;
532 xfs_ifork_t *ifp;
533
534 trace_xfs_attr_sf_add(args);
535
536 dp = args->dp;
537 mp = dp->i_mount;
538 dp->i_d.di_forkoff = forkoff;
539
540 ifp = dp->i_afp;
541 ASSERT(ifp->if_flags & XFS_IFINLINE);
542 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
543 sfe = &sf->list[0];
544 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
545 #ifdef DEBUG
546 if (sfe->namelen != args->namelen)
547 continue;
548 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
549 continue;
550 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
551 continue;
552 ASSERT(0);
553 #endif
554 }
555
556 offset = (char *)sfe - (char *)sf;
557 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
558 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
559 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
560 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
561
562 sfe->namelen = args->namelen;
563 sfe->valuelen = args->valuelen;
564 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
565 memcpy(sfe->nameval, args->name, args->namelen);
566 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
567 sf->hdr.count++;
568 be16_add_cpu(&sf->hdr.totsize, size);
569 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
570
571 xfs_sbversion_add_attr2(mp, args->trans);
572 }
573
574 /*
575 * After the last attribute is removed revert to original inode format,
576 * making all literal area available to the data fork once more.
577 */
578 void
579 xfs_attr_fork_remove(
580 struct xfs_inode *ip,
581 struct xfs_trans *tp)
582 {
583 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
584 ip->i_d.di_forkoff = 0;
585 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
586
587 ASSERT(ip->i_d.di_anextents == 0);
588 ASSERT(ip->i_afp == NULL);
589
590 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
591 }
592
593 /*
594 * Remove an attribute from the shortform attribute list structure.
595 */
596 int
597 xfs_attr_shortform_remove(xfs_da_args_t *args)
598 {
599 xfs_attr_shortform_t *sf;
600 xfs_attr_sf_entry_t *sfe;
601 int base, size=0, end, totsize, i;
602 xfs_mount_t *mp;
603 xfs_inode_t *dp;
604
605 trace_xfs_attr_sf_remove(args);
606
607 dp = args->dp;
608 mp = dp->i_mount;
609 base = sizeof(xfs_attr_sf_hdr_t);
610 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
611 sfe = &sf->list[0];
612 end = sf->hdr.count;
613 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
614 base += size, i++) {
615 size = XFS_ATTR_SF_ENTSIZE(sfe);
616 if (sfe->namelen != args->namelen)
617 continue;
618 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
619 continue;
620 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
621 continue;
622 break;
623 }
624 if (i == end)
625 return -ENOATTR;
626
627 /*
628 * Fix up the attribute fork data, covering the hole
629 */
630 end = base + size;
631 totsize = be16_to_cpu(sf->hdr.totsize);
632 if (end != totsize)
633 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
634 sf->hdr.count--;
635 be16_add_cpu(&sf->hdr.totsize, -size);
636
637 /*
638 * Fix up the start offset of the attribute fork
639 */
640 totsize -= size;
641 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
642 (mp->m_flags & XFS_MOUNT_ATTR2) &&
643 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
644 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
645 xfs_attr_fork_remove(dp, args->trans);
646 } else {
647 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
648 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
649 ASSERT(dp->i_d.di_forkoff);
650 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
651 (args->op_flags & XFS_DA_OP_ADDNAME) ||
652 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
653 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
654 xfs_trans_log_inode(args->trans, dp,
655 XFS_ILOG_CORE | XFS_ILOG_ADATA);
656 }
657
658 xfs_sbversion_add_attr2(mp, args->trans);
659
660 return 0;
661 }
662
663 /*
664 * Look up a name in a shortform attribute list structure.
665 */
666 /*ARGSUSED*/
667 int
668 xfs_attr_shortform_lookup(xfs_da_args_t *args)
669 {
670 xfs_attr_shortform_t *sf;
671 xfs_attr_sf_entry_t *sfe;
672 int i;
673 xfs_ifork_t *ifp;
674
675 trace_xfs_attr_sf_lookup(args);
676
677 ifp = args->dp->i_afp;
678 ASSERT(ifp->if_flags & XFS_IFINLINE);
679 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
680 sfe = &sf->list[0];
681 for (i = 0; i < sf->hdr.count;
682 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
683 if (sfe->namelen != args->namelen)
684 continue;
685 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
686 continue;
687 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
688 continue;
689 return -EEXIST;
690 }
691 return -ENOATTR;
692 }
693
694 /*
695 * Look up a name in a shortform attribute list structure.
696 */
697 /*ARGSUSED*/
698 int
699 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
700 {
701 xfs_attr_shortform_t *sf;
702 xfs_attr_sf_entry_t *sfe;
703 int i;
704
705 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
706 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
707 sfe = &sf->list[0];
708 for (i = 0; i < sf->hdr.count;
709 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
710 if (sfe->namelen != args->namelen)
711 continue;
712 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
713 continue;
714 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
715 continue;
716 if (args->flags & ATTR_KERNOVAL) {
717 args->valuelen = sfe->valuelen;
718 return -EEXIST;
719 }
720 if (args->valuelen < sfe->valuelen) {
721 args->valuelen = sfe->valuelen;
722 return -ERANGE;
723 }
724 args->valuelen = sfe->valuelen;
725 memcpy(args->value, &sfe->nameval[args->namelen],
726 args->valuelen);
727 return -EEXIST;
728 }
729 return -ENOATTR;
730 }
731
732 /*
733 * Convert from using the shortform to the leaf.
734 */
735 int
736 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
737 {
738 xfs_inode_t *dp;
739 xfs_attr_shortform_t *sf;
740 xfs_attr_sf_entry_t *sfe;
741 xfs_da_args_t nargs;
742 char *tmpbuffer;
743 int error, i, size;
744 xfs_dablk_t blkno;
745 struct xfs_buf *bp;
746 xfs_ifork_t *ifp;
747
748 trace_xfs_attr_sf_to_leaf(args);
749
750 dp = args->dp;
751 ifp = dp->i_afp;
752 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
753 size = be16_to_cpu(sf->hdr.totsize);
754 tmpbuffer = kmem_alloc(size, KM_SLEEP);
755 ASSERT(tmpbuffer != NULL);
756 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
757 sf = (xfs_attr_shortform_t *)tmpbuffer;
758
759 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
760 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
761
762 bp = NULL;
763 error = xfs_da_grow_inode(args, &blkno);
764 if (error) {
765 /*
766 * If we hit an IO error middle of the transaction inside
767 * grow_inode(), we may have inconsistent data. Bail out.
768 */
769 if (error == -EIO)
770 goto out;
771 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
772 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
773 goto out;
774 }
775
776 ASSERT(blkno == 0);
777 error = xfs_attr3_leaf_create(args, blkno, &bp);
778 if (error) {
779 error = xfs_da_shrink_inode(args, 0, bp);
780 bp = NULL;
781 if (error)
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;
786 }
787
788 memset((char *)&nargs, 0, sizeof(nargs));
789 nargs.dp = dp;
790 nargs.geo = args->geo;
791 nargs.firstblock = args->firstblock;
792 nargs.dfops = args->dfops;
793 nargs.total = args->total;
794 nargs.whichfork = XFS_ATTR_FORK;
795 nargs.trans = args->trans;
796 nargs.op_flags = XFS_DA_OP_OKNOENT;
797
798 sfe = &sf->list[0];
799 for (i = 0; i < sf->hdr.count; i++) {
800 nargs.name = sfe->nameval;
801 nargs.namelen = sfe->namelen;
802 nargs.value = &sfe->nameval[nargs.namelen];
803 nargs.valuelen = sfe->valuelen;
804 nargs.hashval = xfs_da_hashname(sfe->nameval,
805 sfe->namelen);
806 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
807 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
808 ASSERT(error == -ENOATTR);
809 error = xfs_attr3_leaf_add(bp, &nargs);
810 ASSERT(error != -ENOSPC);
811 if (error)
812 goto out;
813 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
814 }
815 error = 0;
816
817 out:
818 kmem_free(tmpbuffer);
819 return error;
820 }
821
822 /*
823 * Check a leaf attribute block to see if all the entries would fit into
824 * a shortform attribute list.
825 */
826 int
827 xfs_attr_shortform_allfit(
828 struct xfs_buf *bp,
829 struct xfs_inode *dp)
830 {
831 struct xfs_attr_leafblock *leaf;
832 struct xfs_attr_leaf_entry *entry;
833 xfs_attr_leaf_name_local_t *name_loc;
834 struct xfs_attr3_icleaf_hdr leafhdr;
835 int bytes;
836 int i;
837 struct xfs_mount *mp = bp->b_target->bt_mount;
838
839 leaf = bp->b_addr;
840 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
841 entry = xfs_attr3_leaf_entryp(leaf);
842
843 bytes = sizeof(struct xfs_attr_sf_hdr);
844 for (i = 0; i < leafhdr.count; entry++, i++) {
845 if (entry->flags & XFS_ATTR_INCOMPLETE)
846 continue; /* don't copy partial entries */
847 if (!(entry->flags & XFS_ATTR_LOCAL))
848 return 0;
849 name_loc = xfs_attr3_leaf_name_local(leaf, i);
850 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
851 return 0;
852 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
853 return 0;
854 bytes += sizeof(struct xfs_attr_sf_entry) - 1
855 + name_loc->namelen
856 + be16_to_cpu(name_loc->valuelen);
857 }
858 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
859 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
860 (bytes == sizeof(struct xfs_attr_sf_hdr)))
861 return -1;
862 return xfs_attr_shortform_bytesfit(dp, bytes);
863 }
864
865 /*
866 * Convert a leaf attribute list to shortform attribute list
867 */
868 int
869 xfs_attr3_leaf_to_shortform(
870 struct xfs_buf *bp,
871 struct xfs_da_args *args,
872 int forkoff)
873 {
874 struct xfs_attr_leafblock *leaf;
875 struct xfs_attr3_icleaf_hdr ichdr;
876 struct xfs_attr_leaf_entry *entry;
877 struct xfs_attr_leaf_name_local *name_loc;
878 struct xfs_da_args nargs;
879 struct xfs_inode *dp = args->dp;
880 char *tmpbuffer;
881 int error;
882 int i;
883
884 trace_xfs_attr_leaf_to_sf(args);
885
886 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
887 if (!tmpbuffer)
888 return -ENOMEM;
889
890 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
891
892 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
893 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
894 entry = xfs_attr3_leaf_entryp(leaf);
895
896 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
897 memset(bp->b_addr, 0, args->geo->blksize);
898
899 /*
900 * Clean out the prior contents of the attribute list.
901 */
902 error = xfs_da_shrink_inode(args, 0, bp);
903 if (error)
904 goto out;
905
906 if (forkoff == -1) {
907 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
908 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
909 xfs_attr_fork_remove(dp, args->trans);
910 goto out;
911 }
912
913 xfs_attr_shortform_create(args);
914
915 /*
916 * Copy the attributes
917 */
918 memset((char *)&nargs, 0, sizeof(nargs));
919 nargs.geo = args->geo;
920 nargs.dp = dp;
921 nargs.firstblock = args->firstblock;
922 nargs.dfops = args->dfops;
923 nargs.total = args->total;
924 nargs.whichfork = XFS_ATTR_FORK;
925 nargs.trans = args->trans;
926 nargs.op_flags = XFS_DA_OP_OKNOENT;
927
928 for (i = 0; i < ichdr.count; entry++, i++) {
929 if (entry->flags & XFS_ATTR_INCOMPLETE)
930 continue; /* don't copy partial entries */
931 if (!entry->nameidx)
932 continue;
933 ASSERT(entry->flags & XFS_ATTR_LOCAL);
934 name_loc = xfs_attr3_leaf_name_local(leaf, i);
935 nargs.name = name_loc->nameval;
936 nargs.namelen = name_loc->namelen;
937 nargs.value = &name_loc->nameval[nargs.namelen];
938 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
939 nargs.hashval = be32_to_cpu(entry->hashval);
940 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
941 xfs_attr_shortform_add(&nargs, forkoff);
942 }
943 error = 0;
944
945 out:
946 kmem_free(tmpbuffer);
947 return error;
948 }
949
950 /*
951 * Convert from using a single leaf to a root node and a leaf.
952 */
953 int
954 xfs_attr3_leaf_to_node(
955 struct xfs_da_args *args)
956 {
957 struct xfs_attr_leafblock *leaf;
958 struct xfs_attr3_icleaf_hdr icleafhdr;
959 struct xfs_attr_leaf_entry *entries;
960 struct xfs_da_node_entry *btree;
961 struct xfs_da3_icnode_hdr icnodehdr;
962 struct xfs_da_intnode *node;
963 struct xfs_inode *dp = args->dp;
964 struct xfs_mount *mp = dp->i_mount;
965 struct xfs_buf *bp1 = NULL;
966 struct xfs_buf *bp2 = NULL;
967 xfs_dablk_t blkno;
968 int error;
969
970 trace_xfs_attr_leaf_to_node(args);
971
972 error = xfs_da_grow_inode(args, &blkno);
973 if (error)
974 goto out;
975 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
976 if (error)
977 goto out;
978
979 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
980 if (error)
981 goto out;
982
983 /* copy leaf to new buffer, update identifiers */
984 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
985 bp2->b_ops = bp1->b_ops;
986 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
987 if (xfs_sb_version_hascrc(&mp->m_sb)) {
988 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
989 hdr3->blkno = cpu_to_be64(bp2->b_bn);
990 }
991 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
992
993 /*
994 * Set up the new root node.
995 */
996 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
997 if (error)
998 goto out;
999 node = bp1->b_addr;
1000 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1001 btree = dp->d_ops->node_tree_p(node);
1002
1003 leaf = bp2->b_addr;
1004 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1005 entries = xfs_attr3_leaf_entryp(leaf);
1006
1007 /* both on-disk, don't endian-flip twice */
1008 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1009 btree[0].before = cpu_to_be32(blkno);
1010 icnodehdr.count = 1;
1011 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1012 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1013 error = 0;
1014 out:
1015 return error;
1016 }
1017
1018 /*========================================================================
1019 * Routines used for growing the Btree.
1020 *========================================================================*/
1021
1022 /*
1023 * Create the initial contents of a leaf attribute list
1024 * or a leaf in a node attribute list.
1025 */
1026 STATIC int
1027 xfs_attr3_leaf_create(
1028 struct xfs_da_args *args,
1029 xfs_dablk_t blkno,
1030 struct xfs_buf **bpp)
1031 {
1032 struct xfs_attr_leafblock *leaf;
1033 struct xfs_attr3_icleaf_hdr ichdr;
1034 struct xfs_inode *dp = args->dp;
1035 struct xfs_mount *mp = dp->i_mount;
1036 struct xfs_buf *bp;
1037 int error;
1038
1039 trace_xfs_attr_leaf_create(args);
1040
1041 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1042 XFS_ATTR_FORK);
1043 if (error)
1044 return error;
1045 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1046 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1047 leaf = bp->b_addr;
1048 memset(leaf, 0, args->geo->blksize);
1049
1050 memset(&ichdr, 0, sizeof(ichdr));
1051 ichdr.firstused = args->geo->blksize;
1052
1053 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1054 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1055
1056 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1057
1058 hdr3->blkno = cpu_to_be64(bp->b_bn);
1059 hdr3->owner = cpu_to_be64(dp->i_ino);
1060 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1061
1062 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1063 } else {
1064 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1065 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1066 }
1067 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1068
1069 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1070 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1071
1072 *bpp = bp;
1073 return 0;
1074 }
1075
1076 /*
1077 * Split the leaf node, rebalance, then add the new entry.
1078 */
1079 int
1080 xfs_attr3_leaf_split(
1081 struct xfs_da_state *state,
1082 struct xfs_da_state_blk *oldblk,
1083 struct xfs_da_state_blk *newblk)
1084 {
1085 xfs_dablk_t blkno;
1086 int error;
1087
1088 trace_xfs_attr_leaf_split(state->args);
1089
1090 /*
1091 * Allocate space for a new leaf node.
1092 */
1093 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1094 error = xfs_da_grow_inode(state->args, &blkno);
1095 if (error)
1096 return error;
1097 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1098 if (error)
1099 return error;
1100 newblk->blkno = blkno;
1101 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1102
1103 /*
1104 * Rebalance the entries across the two leaves.
1105 * NOTE: rebalance() currently depends on the 2nd block being empty.
1106 */
1107 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1108 error = xfs_da3_blk_link(state, oldblk, newblk);
1109 if (error)
1110 return error;
1111
1112 /*
1113 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1114 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1115 * "new" attrs info. Will need the "old" info to remove it later.
1116 *
1117 * Insert the "new" entry in the correct block.
1118 */
1119 if (state->inleaf) {
1120 trace_xfs_attr_leaf_add_old(state->args);
1121 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1122 } else {
1123 trace_xfs_attr_leaf_add_new(state->args);
1124 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1125 }
1126
1127 /*
1128 * Update last hashval in each block since we added the name.
1129 */
1130 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1131 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1132 return error;
1133 }
1134
1135 /*
1136 * Add a name to the leaf attribute list structure.
1137 */
1138 int
1139 xfs_attr3_leaf_add(
1140 struct xfs_buf *bp,
1141 struct xfs_da_args *args)
1142 {
1143 struct xfs_attr_leafblock *leaf;
1144 struct xfs_attr3_icleaf_hdr ichdr;
1145 int tablesize;
1146 int entsize;
1147 int sum;
1148 int tmp;
1149 int i;
1150
1151 trace_xfs_attr_leaf_add(args);
1152
1153 leaf = bp->b_addr;
1154 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1155 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1156 entsize = xfs_attr_leaf_newentsize(args, NULL);
1157
1158 /*
1159 * Search through freemap for first-fit on new name length.
1160 * (may need to figure in size of entry struct too)
1161 */
1162 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1163 + xfs_attr3_leaf_hdr_size(leaf);
1164 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1165 if (tablesize > ichdr.firstused) {
1166 sum += ichdr.freemap[i].size;
1167 continue;
1168 }
1169 if (!ichdr.freemap[i].size)
1170 continue; /* no space in this map */
1171 tmp = entsize;
1172 if (ichdr.freemap[i].base < ichdr.firstused)
1173 tmp += sizeof(xfs_attr_leaf_entry_t);
1174 if (ichdr.freemap[i].size >= tmp) {
1175 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1176 goto out_log_hdr;
1177 }
1178 sum += ichdr.freemap[i].size;
1179 }
1180
1181 /*
1182 * If there are no holes in the address space of the block,
1183 * and we don't have enough freespace, then compaction will do us
1184 * no good and we should just give up.
1185 */
1186 if (!ichdr.holes && sum < entsize)
1187 return -ENOSPC;
1188
1189 /*
1190 * Compact the entries to coalesce free space.
1191 * This may change the hdr->count via dropping INCOMPLETE entries.
1192 */
1193 xfs_attr3_leaf_compact(args, &ichdr, bp);
1194
1195 /*
1196 * After compaction, the block is guaranteed to have only one
1197 * free region, in freemap[0]. If it is not big enough, give up.
1198 */
1199 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1200 tmp = -ENOSPC;
1201 goto out_log_hdr;
1202 }
1203
1204 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1205
1206 out_log_hdr:
1207 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1208 xfs_trans_log_buf(args->trans, bp,
1209 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1210 xfs_attr3_leaf_hdr_size(leaf)));
1211 return tmp;
1212 }
1213
1214 /*
1215 * Add a name to a leaf attribute list structure.
1216 */
1217 STATIC int
1218 xfs_attr3_leaf_add_work(
1219 struct xfs_buf *bp,
1220 struct xfs_attr3_icleaf_hdr *ichdr,
1221 struct xfs_da_args *args,
1222 int mapindex)
1223 {
1224 struct xfs_attr_leafblock *leaf;
1225 struct xfs_attr_leaf_entry *entry;
1226 struct xfs_attr_leaf_name_local *name_loc;
1227 struct xfs_attr_leaf_name_remote *name_rmt;
1228 struct xfs_mount *mp;
1229 int tmp;
1230 int i;
1231
1232 trace_xfs_attr_leaf_add_work(args);
1233
1234 leaf = bp->b_addr;
1235 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1236 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1237
1238 /*
1239 * Force open some space in the entry array and fill it in.
1240 */
1241 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1242 if (args->index < ichdr->count) {
1243 tmp = ichdr->count - args->index;
1244 tmp *= sizeof(xfs_attr_leaf_entry_t);
1245 memmove(entry + 1, entry, tmp);
1246 xfs_trans_log_buf(args->trans, bp,
1247 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1248 }
1249 ichdr->count++;
1250
1251 /*
1252 * Allocate space for the new string (at the end of the run).
1253 */
1254 mp = args->trans->t_mountp;
1255 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1256 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1257 ASSERT(ichdr->freemap[mapindex].size >=
1258 xfs_attr_leaf_newentsize(args, NULL));
1259 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1260 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1261
1262 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1263
1264 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1265 ichdr->freemap[mapindex].size);
1266 entry->hashval = cpu_to_be32(args->hashval);
1267 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1268 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1269 if (args->op_flags & XFS_DA_OP_RENAME) {
1270 entry->flags |= XFS_ATTR_INCOMPLETE;
1271 if ((args->blkno2 == args->blkno) &&
1272 (args->index2 <= args->index)) {
1273 args->index2++;
1274 }
1275 }
1276 xfs_trans_log_buf(args->trans, bp,
1277 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1278 ASSERT((args->index == 0) ||
1279 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1280 ASSERT((args->index == ichdr->count - 1) ||
1281 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1282
1283 /*
1284 * For "remote" attribute values, simply note that we need to
1285 * allocate space for the "remote" value. We can't actually
1286 * allocate the extents in this transaction, and we can't decide
1287 * which blocks they should be as we might allocate more blocks
1288 * as part of this transaction (a split operation for example).
1289 */
1290 if (entry->flags & XFS_ATTR_LOCAL) {
1291 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1292 name_loc->namelen = args->namelen;
1293 name_loc->valuelen = cpu_to_be16(args->valuelen);
1294 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1295 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1296 be16_to_cpu(name_loc->valuelen));
1297 } else {
1298 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1299 name_rmt->namelen = args->namelen;
1300 memcpy((char *)name_rmt->name, args->name, args->namelen);
1301 entry->flags |= XFS_ATTR_INCOMPLETE;
1302 /* just in case */
1303 name_rmt->valuelen = 0;
1304 name_rmt->valueblk = 0;
1305 args->rmtblkno = 1;
1306 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1307 args->rmtvaluelen = args->valuelen;
1308 }
1309 xfs_trans_log_buf(args->trans, bp,
1310 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1311 xfs_attr_leaf_entsize(leaf, args->index)));
1312
1313 /*
1314 * Update the control info for this leaf node
1315 */
1316 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1317 ichdr->firstused = be16_to_cpu(entry->nameidx);
1318
1319 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1320 + xfs_attr3_leaf_hdr_size(leaf));
1321 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1322 + xfs_attr3_leaf_hdr_size(leaf);
1323
1324 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1325 if (ichdr->freemap[i].base == tmp) {
1326 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1327 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1328 }
1329 }
1330 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1331 return 0;
1332 }
1333
1334 /*
1335 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1336 */
1337 STATIC void
1338 xfs_attr3_leaf_compact(
1339 struct xfs_da_args *args,
1340 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1341 struct xfs_buf *bp)
1342 {
1343 struct xfs_attr_leafblock *leaf_src;
1344 struct xfs_attr_leafblock *leaf_dst;
1345 struct xfs_attr3_icleaf_hdr ichdr_src;
1346 struct xfs_trans *trans = args->trans;
1347 char *tmpbuffer;
1348
1349 trace_xfs_attr_leaf_compact(args);
1350
1351 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1352 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1353 memset(bp->b_addr, 0, args->geo->blksize);
1354 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1355 leaf_dst = bp->b_addr;
1356
1357 /*
1358 * Copy the on-disk header back into the destination buffer to ensure
1359 * all the information in the header that is not part of the incore
1360 * header structure is preserved.
1361 */
1362 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1363
1364 /* Initialise the incore headers */
1365 ichdr_src = *ichdr_dst; /* struct copy */
1366 ichdr_dst->firstused = args->geo->blksize;
1367 ichdr_dst->usedbytes = 0;
1368 ichdr_dst->count = 0;
1369 ichdr_dst->holes = 0;
1370 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1371 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1372 ichdr_dst->freemap[0].base;
1373
1374 /* write the header back to initialise the underlying buffer */
1375 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1376
1377 /*
1378 * Copy all entry's in the same (sorted) order,
1379 * but allocate name/value pairs packed and in sequence.
1380 */
1381 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1382 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1383 /*
1384 * this logs the entire buffer, but the caller must write the header
1385 * back to the buffer when it is finished modifying it.
1386 */
1387 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1388
1389 kmem_free(tmpbuffer);
1390 }
1391
1392 /*
1393 * Compare two leaf blocks "order".
1394 * Return 0 unless leaf2 should go before leaf1.
1395 */
1396 static int
1397 xfs_attr3_leaf_order(
1398 struct xfs_buf *leaf1_bp,
1399 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1400 struct xfs_buf *leaf2_bp,
1401 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1402 {
1403 struct xfs_attr_leaf_entry *entries1;
1404 struct xfs_attr_leaf_entry *entries2;
1405
1406 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1407 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1408 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1409 ((be32_to_cpu(entries2[0].hashval) <
1410 be32_to_cpu(entries1[0].hashval)) ||
1411 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1412 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1413 return 1;
1414 }
1415 return 0;
1416 }
1417
1418 int
1419 xfs_attr_leaf_order(
1420 struct xfs_buf *leaf1_bp,
1421 struct xfs_buf *leaf2_bp)
1422 {
1423 struct xfs_attr3_icleaf_hdr ichdr1;
1424 struct xfs_attr3_icleaf_hdr ichdr2;
1425 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1426
1427 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1428 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1429 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1430 }
1431
1432 /*
1433 * Redistribute the attribute list entries between two leaf nodes,
1434 * taking into account the size of the new entry.
1435 *
1436 * NOTE: if new block is empty, then it will get the upper half of the
1437 * old block. At present, all (one) callers pass in an empty second block.
1438 *
1439 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1440 * to match what it is doing in splitting the attribute leaf block. Those
1441 * values are used in "atomic rename" operations on attributes. Note that
1442 * the "new" and "old" values can end up in different blocks.
1443 */
1444 STATIC void
1445 xfs_attr3_leaf_rebalance(
1446 struct xfs_da_state *state,
1447 struct xfs_da_state_blk *blk1,
1448 struct xfs_da_state_blk *blk2)
1449 {
1450 struct xfs_da_args *args;
1451 struct xfs_attr_leafblock *leaf1;
1452 struct xfs_attr_leafblock *leaf2;
1453 struct xfs_attr3_icleaf_hdr ichdr1;
1454 struct xfs_attr3_icleaf_hdr ichdr2;
1455 struct xfs_attr_leaf_entry *entries1;
1456 struct xfs_attr_leaf_entry *entries2;
1457 int count;
1458 int totallen;
1459 int max;
1460 int space;
1461 int swap;
1462
1463 /*
1464 * Set up environment.
1465 */
1466 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1467 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1468 leaf1 = blk1->bp->b_addr;
1469 leaf2 = blk2->bp->b_addr;
1470 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1471 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1472 ASSERT(ichdr2.count == 0);
1473 args = state->args;
1474
1475 trace_xfs_attr_leaf_rebalance(args);
1476
1477 /*
1478 * Check ordering of blocks, reverse if it makes things simpler.
1479 *
1480 * NOTE: Given that all (current) callers pass in an empty
1481 * second block, this code should never set "swap".
1482 */
1483 swap = 0;
1484 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1485 struct xfs_da_state_blk *tmp_blk;
1486 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1487
1488 tmp_blk = blk1;
1489 blk1 = blk2;
1490 blk2 = tmp_blk;
1491
1492 /* struct copies to swap them rather than reconverting */
1493 tmp_ichdr = ichdr1;
1494 ichdr1 = ichdr2;
1495 ichdr2 = tmp_ichdr;
1496
1497 leaf1 = blk1->bp->b_addr;
1498 leaf2 = blk2->bp->b_addr;
1499 swap = 1;
1500 }
1501
1502 /*
1503 * Examine entries until we reduce the absolute difference in
1504 * byte usage between the two blocks to a minimum. Then get
1505 * the direction to copy and the number of elements to move.
1506 *
1507 * "inleaf" is true if the new entry should be inserted into blk1.
1508 * If "swap" is also true, then reverse the sense of "inleaf".
1509 */
1510 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1511 blk2, &ichdr2,
1512 &count, &totallen);
1513 if (swap)
1514 state->inleaf = !state->inleaf;
1515
1516 /*
1517 * Move any entries required from leaf to leaf:
1518 */
1519 if (count < ichdr1.count) {
1520 /*
1521 * Figure the total bytes to be added to the destination leaf.
1522 */
1523 /* number entries being moved */
1524 count = ichdr1.count - count;
1525 space = ichdr1.usedbytes - totallen;
1526 space += count * sizeof(xfs_attr_leaf_entry_t);
1527
1528 /*
1529 * leaf2 is the destination, compact it if it looks tight.
1530 */
1531 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1532 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1533 if (space > max)
1534 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1535
1536 /*
1537 * Move high entries from leaf1 to low end of leaf2.
1538 */
1539 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1540 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1541
1542 } else if (count > ichdr1.count) {
1543 /*
1544 * I assert that since all callers pass in an empty
1545 * second buffer, this code should never execute.
1546 */
1547 ASSERT(0);
1548
1549 /*
1550 * Figure the total bytes to be added to the destination leaf.
1551 */
1552 /* number entries being moved */
1553 count -= ichdr1.count;
1554 space = totallen - ichdr1.usedbytes;
1555 space += count * sizeof(xfs_attr_leaf_entry_t);
1556
1557 /*
1558 * leaf1 is the destination, compact it if it looks tight.
1559 */
1560 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1561 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1562 if (space > max)
1563 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1564
1565 /*
1566 * Move low entries from leaf2 to high end of leaf1.
1567 */
1568 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1569 ichdr1.count, count);
1570 }
1571
1572 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1573 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1574 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1575 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1576
1577 /*
1578 * Copy out last hashval in each block for B-tree code.
1579 */
1580 entries1 = xfs_attr3_leaf_entryp(leaf1);
1581 entries2 = xfs_attr3_leaf_entryp(leaf2);
1582 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1583 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1584
1585 /*
1586 * Adjust the expected index for insertion.
1587 * NOTE: this code depends on the (current) situation that the
1588 * second block was originally empty.
1589 *
1590 * If the insertion point moved to the 2nd block, we must adjust
1591 * the index. We must also track the entry just following the
1592 * new entry for use in an "atomic rename" operation, that entry
1593 * is always the "old" entry and the "new" entry is what we are
1594 * inserting. The index/blkno fields refer to the "old" entry,
1595 * while the index2/blkno2 fields refer to the "new" entry.
1596 */
1597 if (blk1->index > ichdr1.count) {
1598 ASSERT(state->inleaf == 0);
1599 blk2->index = blk1->index - ichdr1.count;
1600 args->index = args->index2 = blk2->index;
1601 args->blkno = args->blkno2 = blk2->blkno;
1602 } else if (blk1->index == ichdr1.count) {
1603 if (state->inleaf) {
1604 args->index = blk1->index;
1605 args->blkno = blk1->blkno;
1606 args->index2 = 0;
1607 args->blkno2 = blk2->blkno;
1608 } else {
1609 /*
1610 * On a double leaf split, the original attr location
1611 * is already stored in blkno2/index2, so don't
1612 * overwrite it overwise we corrupt the tree.
1613 */
1614 blk2->index = blk1->index - ichdr1.count;
1615 args->index = blk2->index;
1616 args->blkno = blk2->blkno;
1617 if (!state->extravalid) {
1618 /*
1619 * set the new attr location to match the old
1620 * one and let the higher level split code
1621 * decide where in the leaf to place it.
1622 */
1623 args->index2 = blk2->index;
1624 args->blkno2 = blk2->blkno;
1625 }
1626 }
1627 } else {
1628 ASSERT(state->inleaf == 1);
1629 args->index = args->index2 = blk1->index;
1630 args->blkno = args->blkno2 = blk1->blkno;
1631 }
1632 }
1633
1634 /*
1635 * Examine entries until we reduce the absolute difference in
1636 * byte usage between the two blocks to a minimum.
1637 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1638 * GROT: there will always be enough room in either block for a new entry.
1639 * GROT: Do a double-split for this case?
1640 */
1641 STATIC int
1642 xfs_attr3_leaf_figure_balance(
1643 struct xfs_da_state *state,
1644 struct xfs_da_state_blk *blk1,
1645 struct xfs_attr3_icleaf_hdr *ichdr1,
1646 struct xfs_da_state_blk *blk2,
1647 struct xfs_attr3_icleaf_hdr *ichdr2,
1648 int *countarg,
1649 int *usedbytesarg)
1650 {
1651 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1652 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1653 struct xfs_attr_leaf_entry *entry;
1654 int count;
1655 int max;
1656 int index;
1657 int totallen = 0;
1658 int half;
1659 int lastdelta;
1660 int foundit = 0;
1661 int tmp;
1662
1663 /*
1664 * Examine entries until we reduce the absolute difference in
1665 * byte usage between the two blocks to a minimum.
1666 */
1667 max = ichdr1->count + ichdr2->count;
1668 half = (max + 1) * sizeof(*entry);
1669 half += ichdr1->usedbytes + ichdr2->usedbytes +
1670 xfs_attr_leaf_newentsize(state->args, NULL);
1671 half /= 2;
1672 lastdelta = state->args->geo->blksize;
1673 entry = xfs_attr3_leaf_entryp(leaf1);
1674 for (count = index = 0; count < max; entry++, index++, count++) {
1675
1676 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1677 /*
1678 * The new entry is in the first block, account for it.
1679 */
1680 if (count == blk1->index) {
1681 tmp = totallen + sizeof(*entry) +
1682 xfs_attr_leaf_newentsize(state->args, NULL);
1683 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1684 break;
1685 lastdelta = XFS_ATTR_ABS(half - tmp);
1686 totallen = tmp;
1687 foundit = 1;
1688 }
1689
1690 /*
1691 * Wrap around into the second block if necessary.
1692 */
1693 if (count == ichdr1->count) {
1694 leaf1 = leaf2;
1695 entry = xfs_attr3_leaf_entryp(leaf1);
1696 index = 0;
1697 }
1698
1699 /*
1700 * Figure out if next leaf entry would be too much.
1701 */
1702 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1703 index);
1704 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1705 break;
1706 lastdelta = XFS_ATTR_ABS(half - tmp);
1707 totallen = tmp;
1708 #undef XFS_ATTR_ABS
1709 }
1710
1711 /*
1712 * Calculate the number of usedbytes that will end up in lower block.
1713 * If new entry not in lower block, fix up the count.
1714 */
1715 totallen -= count * sizeof(*entry);
1716 if (foundit) {
1717 totallen -= sizeof(*entry) +
1718 xfs_attr_leaf_newentsize(state->args, NULL);
1719 }
1720
1721 *countarg = count;
1722 *usedbytesarg = totallen;
1723 return foundit;
1724 }
1725
1726 /*========================================================================
1727 * Routines used for shrinking the Btree.
1728 *========================================================================*/
1729
1730 /*
1731 * Check a leaf block and its neighbors to see if the block should be
1732 * collapsed into one or the other neighbor. Always keep the block
1733 * with the smaller block number.
1734 * If the current block is over 50% full, don't try to join it, return 0.
1735 * If the block is empty, fill in the state structure and return 2.
1736 * If it can be collapsed, fill in the state structure and return 1.
1737 * If nothing can be done, return 0.
1738 *
1739 * GROT: allow for INCOMPLETE entries in calculation.
1740 */
1741 int
1742 xfs_attr3_leaf_toosmall(
1743 struct xfs_da_state *state,
1744 int *action)
1745 {
1746 struct xfs_attr_leafblock *leaf;
1747 struct xfs_da_state_blk *blk;
1748 struct xfs_attr3_icleaf_hdr ichdr;
1749 struct xfs_buf *bp;
1750 xfs_dablk_t blkno;
1751 int bytes;
1752 int forward;
1753 int error;
1754 int retval;
1755 int i;
1756
1757 trace_xfs_attr_leaf_toosmall(state->args);
1758
1759 /*
1760 * Check for the degenerate case of the block being over 50% full.
1761 * If so, it's not worth even looking to see if we might be able
1762 * to coalesce with a sibling.
1763 */
1764 blk = &state->path.blk[ state->path.active-1 ];
1765 leaf = blk->bp->b_addr;
1766 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1767 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1768 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1769 ichdr.usedbytes;
1770 if (bytes > (state->args->geo->blksize >> 1)) {
1771 *action = 0; /* blk over 50%, don't try to join */
1772 return 0;
1773 }
1774
1775 /*
1776 * Check for the degenerate case of the block being empty.
1777 * If the block is empty, we'll simply delete it, no need to
1778 * coalesce it with a sibling block. We choose (arbitrarily)
1779 * to merge with the forward block unless it is NULL.
1780 */
1781 if (ichdr.count == 0) {
1782 /*
1783 * Make altpath point to the block we want to keep and
1784 * path point to the block we want to drop (this one).
1785 */
1786 forward = (ichdr.forw != 0);
1787 memcpy(&state->altpath, &state->path, sizeof(state->path));
1788 error = xfs_da3_path_shift(state, &state->altpath, forward,
1789 0, &retval);
1790 if (error)
1791 return error;
1792 if (retval) {
1793 *action = 0;
1794 } else {
1795 *action = 2;
1796 }
1797 return 0;
1798 }
1799
1800 /*
1801 * Examine each sibling block to see if we can coalesce with
1802 * at least 25% free space to spare. We need to figure out
1803 * whether to merge with the forward or the backward block.
1804 * We prefer coalescing with the lower numbered sibling so as
1805 * to shrink an attribute list over time.
1806 */
1807 /* start with smaller blk num */
1808 forward = ichdr.forw < ichdr.back;
1809 for (i = 0; i < 2; forward = !forward, i++) {
1810 struct xfs_attr3_icleaf_hdr ichdr2;
1811 if (forward)
1812 blkno = ichdr.forw;
1813 else
1814 blkno = ichdr.back;
1815 if (blkno == 0)
1816 continue;
1817 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1818 blkno, -1, &bp);
1819 if (error)
1820 return error;
1821
1822 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1823
1824 bytes = state->args->geo->blksize -
1825 (state->args->geo->blksize >> 2) -
1826 ichdr.usedbytes - ichdr2.usedbytes -
1827 ((ichdr.count + ichdr2.count) *
1828 sizeof(xfs_attr_leaf_entry_t)) -
1829 xfs_attr3_leaf_hdr_size(leaf);
1830
1831 xfs_trans_brelse(state->args->trans, bp);
1832 if (bytes >= 0)
1833 break; /* fits with at least 25% to spare */
1834 }
1835 if (i >= 2) {
1836 *action = 0;
1837 return 0;
1838 }
1839
1840 /*
1841 * Make altpath point to the block we want to keep (the lower
1842 * numbered block) and path point to the block we want to drop.
1843 */
1844 memcpy(&state->altpath, &state->path, sizeof(state->path));
1845 if (blkno < blk->blkno) {
1846 error = xfs_da3_path_shift(state, &state->altpath, forward,
1847 0, &retval);
1848 } else {
1849 error = xfs_da3_path_shift(state, &state->path, forward,
1850 0, &retval);
1851 }
1852 if (error)
1853 return error;
1854 if (retval) {
1855 *action = 0;
1856 } else {
1857 *action = 1;
1858 }
1859 return 0;
1860 }
1861
1862 /*
1863 * Remove a name from the leaf attribute list structure.
1864 *
1865 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1866 * If two leaves are 37% full, when combined they will leave 25% free.
1867 */
1868 int
1869 xfs_attr3_leaf_remove(
1870 struct xfs_buf *bp,
1871 struct xfs_da_args *args)
1872 {
1873 struct xfs_attr_leafblock *leaf;
1874 struct xfs_attr3_icleaf_hdr ichdr;
1875 struct xfs_attr_leaf_entry *entry;
1876 int before;
1877 int after;
1878 int smallest;
1879 int entsize;
1880 int tablesize;
1881 int tmp;
1882 int i;
1883
1884 trace_xfs_attr_leaf_remove(args);
1885
1886 leaf = bp->b_addr;
1887 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1888
1889 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1890 ASSERT(args->index >= 0 && args->index < ichdr.count);
1891 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1892 xfs_attr3_leaf_hdr_size(leaf));
1893
1894 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1895
1896 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1897 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1898
1899 /*
1900 * Scan through free region table:
1901 * check for adjacency of free'd entry with an existing one,
1902 * find smallest free region in case we need to replace it,
1903 * adjust any map that borders the entry table,
1904 */
1905 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1906 + xfs_attr3_leaf_hdr_size(leaf);
1907 tmp = ichdr.freemap[0].size;
1908 before = after = -1;
1909 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1910 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1911 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1912 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1913 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1914 if (ichdr.freemap[i].base == tablesize) {
1915 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1916 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1917 }
1918
1919 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1920 be16_to_cpu(entry->nameidx)) {
1921 before = i;
1922 } else if (ichdr.freemap[i].base ==
1923 (be16_to_cpu(entry->nameidx) + entsize)) {
1924 after = i;
1925 } else if (ichdr.freemap[i].size < tmp) {
1926 tmp = ichdr.freemap[i].size;
1927 smallest = i;
1928 }
1929 }
1930
1931 /*
1932 * Coalesce adjacent freemap regions,
1933 * or replace the smallest region.
1934 */
1935 if ((before >= 0) || (after >= 0)) {
1936 if ((before >= 0) && (after >= 0)) {
1937 ichdr.freemap[before].size += entsize;
1938 ichdr.freemap[before].size += ichdr.freemap[after].size;
1939 ichdr.freemap[after].base = 0;
1940 ichdr.freemap[after].size = 0;
1941 } else if (before >= 0) {
1942 ichdr.freemap[before].size += entsize;
1943 } else {
1944 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1945 ichdr.freemap[after].size += entsize;
1946 }
1947 } else {
1948 /*
1949 * Replace smallest region (if it is smaller than free'd entry)
1950 */
1951 if (ichdr.freemap[smallest].size < entsize) {
1952 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1953 ichdr.freemap[smallest].size = entsize;
1954 }
1955 }
1956
1957 /*
1958 * Did we remove the first entry?
1959 */
1960 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1961 smallest = 1;
1962 else
1963 smallest = 0;
1964
1965 /*
1966 * Compress the remaining entries and zero out the removed stuff.
1967 */
1968 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1969 ichdr.usedbytes -= entsize;
1970 xfs_trans_log_buf(args->trans, bp,
1971 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1972 entsize));
1973
1974 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1975 memmove(entry, entry + 1, tmp);
1976 ichdr.count--;
1977 xfs_trans_log_buf(args->trans, bp,
1978 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1979
1980 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1981 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1982
1983 /*
1984 * If we removed the first entry, re-find the first used byte
1985 * in the name area. Note that if the entry was the "firstused",
1986 * then we don't have a "hole" in our block resulting from
1987 * removing the name.
1988 */
1989 if (smallest) {
1990 tmp = args->geo->blksize;
1991 entry = xfs_attr3_leaf_entryp(leaf);
1992 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1993 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1994 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1995
1996 if (be16_to_cpu(entry->nameidx) < tmp)
1997 tmp = be16_to_cpu(entry->nameidx);
1998 }
1999 ichdr.firstused = tmp;
2000 ASSERT(ichdr.firstused != 0);
2001 } else {
2002 ichdr.holes = 1; /* mark as needing compaction */
2003 }
2004 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2005 xfs_trans_log_buf(args->trans, bp,
2006 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2007 xfs_attr3_leaf_hdr_size(leaf)));
2008
2009 /*
2010 * Check if leaf is less than 50% full, caller may want to
2011 * "join" the leaf with a sibling if so.
2012 */
2013 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2014 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2015
2016 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2017 }
2018
2019 /*
2020 * Move all the attribute list entries from drop_leaf into save_leaf.
2021 */
2022 void
2023 xfs_attr3_leaf_unbalance(
2024 struct xfs_da_state *state,
2025 struct xfs_da_state_blk *drop_blk,
2026 struct xfs_da_state_blk *save_blk)
2027 {
2028 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2029 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2030 struct xfs_attr3_icleaf_hdr drophdr;
2031 struct xfs_attr3_icleaf_hdr savehdr;
2032 struct xfs_attr_leaf_entry *entry;
2033
2034 trace_xfs_attr_leaf_unbalance(state->args);
2035
2036 drop_leaf = drop_blk->bp->b_addr;
2037 save_leaf = save_blk->bp->b_addr;
2038 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2039 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2040 entry = xfs_attr3_leaf_entryp(drop_leaf);
2041
2042 /*
2043 * Save last hashval from dying block for later Btree fixup.
2044 */
2045 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2046
2047 /*
2048 * Check if we need a temp buffer, or can we do it in place.
2049 * Note that we don't check "leaf" for holes because we will
2050 * always be dropping it, toosmall() decided that for us already.
2051 */
2052 if (savehdr.holes == 0) {
2053 /*
2054 * dest leaf has no holes, so we add there. May need
2055 * to make some room in the entry array.
2056 */
2057 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2058 drop_blk->bp, &drophdr)) {
2059 xfs_attr3_leaf_moveents(state->args,
2060 drop_leaf, &drophdr, 0,
2061 save_leaf, &savehdr, 0,
2062 drophdr.count);
2063 } else {
2064 xfs_attr3_leaf_moveents(state->args,
2065 drop_leaf, &drophdr, 0,
2066 save_leaf, &savehdr,
2067 savehdr.count, drophdr.count);
2068 }
2069 } else {
2070 /*
2071 * Destination has holes, so we make a temporary copy
2072 * of the leaf and add them both to that.
2073 */
2074 struct xfs_attr_leafblock *tmp_leaf;
2075 struct xfs_attr3_icleaf_hdr tmphdr;
2076
2077 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2078
2079 /*
2080 * Copy the header into the temp leaf so that all the stuff
2081 * not in the incore header is present and gets copied back in
2082 * once we've moved all the entries.
2083 */
2084 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2085
2086 memset(&tmphdr, 0, sizeof(tmphdr));
2087 tmphdr.magic = savehdr.magic;
2088 tmphdr.forw = savehdr.forw;
2089 tmphdr.back = savehdr.back;
2090 tmphdr.firstused = state->args->geo->blksize;
2091
2092 /* write the header to the temp buffer to initialise it */
2093 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2094
2095 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2096 drop_blk->bp, &drophdr)) {
2097 xfs_attr3_leaf_moveents(state->args,
2098 drop_leaf, &drophdr, 0,
2099 tmp_leaf, &tmphdr, 0,
2100 drophdr.count);
2101 xfs_attr3_leaf_moveents(state->args,
2102 save_leaf, &savehdr, 0,
2103 tmp_leaf, &tmphdr, tmphdr.count,
2104 savehdr.count);
2105 } else {
2106 xfs_attr3_leaf_moveents(state->args,
2107 save_leaf, &savehdr, 0,
2108 tmp_leaf, &tmphdr, 0,
2109 savehdr.count);
2110 xfs_attr3_leaf_moveents(state->args,
2111 drop_leaf, &drophdr, 0,
2112 tmp_leaf, &tmphdr, tmphdr.count,
2113 drophdr.count);
2114 }
2115 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2116 savehdr = tmphdr; /* struct copy */
2117 kmem_free(tmp_leaf);
2118 }
2119
2120 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2121 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2122 state->args->geo->blksize - 1);
2123
2124 /*
2125 * Copy out last hashval in each block for B-tree code.
2126 */
2127 entry = xfs_attr3_leaf_entryp(save_leaf);
2128 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2129 }
2130
2131 /*========================================================================
2132 * Routines used for finding things in the Btree.
2133 *========================================================================*/
2134
2135 /*
2136 * Look up a name in a leaf attribute list structure.
2137 * This is the internal routine, it uses the caller's buffer.
2138 *
2139 * Note that duplicate keys are allowed, but only check within the
2140 * current leaf node. The Btree code must check in adjacent leaf nodes.
2141 *
2142 * Return in args->index the index into the entry[] array of either
2143 * the found entry, or where the entry should have been (insert before
2144 * that entry).
2145 *
2146 * Don't change the args->value unless we find the attribute.
2147 */
2148 int
2149 xfs_attr3_leaf_lookup_int(
2150 struct xfs_buf *bp,
2151 struct xfs_da_args *args)
2152 {
2153 struct xfs_attr_leafblock *leaf;
2154 struct xfs_attr3_icleaf_hdr ichdr;
2155 struct xfs_attr_leaf_entry *entry;
2156 struct xfs_attr_leaf_entry *entries;
2157 struct xfs_attr_leaf_name_local *name_loc;
2158 struct xfs_attr_leaf_name_remote *name_rmt;
2159 xfs_dahash_t hashval;
2160 int probe;
2161 int span;
2162
2163 trace_xfs_attr_leaf_lookup(args);
2164
2165 leaf = bp->b_addr;
2166 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2167 entries = xfs_attr3_leaf_entryp(leaf);
2168 ASSERT(ichdr.count < args->geo->blksize / 8);
2169
2170 /*
2171 * Binary search. (note: small blocks will skip this loop)
2172 */
2173 hashval = args->hashval;
2174 probe = span = ichdr.count / 2;
2175 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2176 span /= 2;
2177 if (be32_to_cpu(entry->hashval) < hashval)
2178 probe += span;
2179 else if (be32_to_cpu(entry->hashval) > hashval)
2180 probe -= span;
2181 else
2182 break;
2183 }
2184 ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2185 ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2186
2187 /*
2188 * Since we may have duplicate hashval's, find the first matching
2189 * hashval in the leaf.
2190 */
2191 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2192 entry--;
2193 probe--;
2194 }
2195 while (probe < ichdr.count &&
2196 be32_to_cpu(entry->hashval) < hashval) {
2197 entry++;
2198 probe++;
2199 }
2200 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2201 args->index = probe;
2202 return -ENOATTR;
2203 }
2204
2205 /*
2206 * Duplicate keys may be present, so search all of them for a match.
2207 */
2208 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2209 entry++, probe++) {
2210 /*
2211 * GROT: Add code to remove incomplete entries.
2212 */
2213 /*
2214 * If we are looking for INCOMPLETE entries, show only those.
2215 * If we are looking for complete entries, show only those.
2216 */
2217 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2218 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2219 continue;
2220 }
2221 if (entry->flags & XFS_ATTR_LOCAL) {
2222 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2223 if (name_loc->namelen != args->namelen)
2224 continue;
2225 if (memcmp(args->name, name_loc->nameval,
2226 args->namelen) != 0)
2227 continue;
2228 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2229 continue;
2230 args->index = probe;
2231 return -EEXIST;
2232 } else {
2233 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2234 if (name_rmt->namelen != args->namelen)
2235 continue;
2236 if (memcmp(args->name, name_rmt->name,
2237 args->namelen) != 0)
2238 continue;
2239 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2240 continue;
2241 args->index = probe;
2242 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2243 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2244 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2245 args->dp->i_mount,
2246 args->rmtvaluelen);
2247 return -EEXIST;
2248 }
2249 }
2250 args->index = probe;
2251 return -ENOATTR;
2252 }
2253
2254 /*
2255 * Get the value associated with an attribute name from a leaf attribute
2256 * list structure.
2257 */
2258 int
2259 xfs_attr3_leaf_getvalue(
2260 struct xfs_buf *bp,
2261 struct xfs_da_args *args)
2262 {
2263 struct xfs_attr_leafblock *leaf;
2264 struct xfs_attr3_icleaf_hdr ichdr;
2265 struct xfs_attr_leaf_entry *entry;
2266 struct xfs_attr_leaf_name_local *name_loc;
2267 struct xfs_attr_leaf_name_remote *name_rmt;
2268 int valuelen;
2269
2270 leaf = bp->b_addr;
2271 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2272 ASSERT(ichdr.count < args->geo->blksize / 8);
2273 ASSERT(args->index < ichdr.count);
2274
2275 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2276 if (entry->flags & XFS_ATTR_LOCAL) {
2277 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2278 ASSERT(name_loc->namelen == args->namelen);
2279 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2280 valuelen = be16_to_cpu(name_loc->valuelen);
2281 if (args->flags & ATTR_KERNOVAL) {
2282 args->valuelen = valuelen;
2283 return 0;
2284 }
2285 if (args->valuelen < valuelen) {
2286 args->valuelen = valuelen;
2287 return -ERANGE;
2288 }
2289 args->valuelen = valuelen;
2290 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2291 } else {
2292 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2293 ASSERT(name_rmt->namelen == args->namelen);
2294 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2295 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2296 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2297 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2298 args->rmtvaluelen);
2299 if (args->flags & ATTR_KERNOVAL) {
2300 args->valuelen = args->rmtvaluelen;
2301 return 0;
2302 }
2303 if (args->valuelen < args->rmtvaluelen) {
2304 args->valuelen = args->rmtvaluelen;
2305 return -ERANGE;
2306 }
2307 args->valuelen = args->rmtvaluelen;
2308 }
2309 return 0;
2310 }
2311
2312 /*========================================================================
2313 * Utility routines.
2314 *========================================================================*/
2315
2316 /*
2317 * Move the indicated entries from one leaf to another.
2318 * NOTE: this routine modifies both source and destination leaves.
2319 */
2320 /*ARGSUSED*/
2321 STATIC void
2322 xfs_attr3_leaf_moveents(
2323 struct xfs_da_args *args,
2324 struct xfs_attr_leafblock *leaf_s,
2325 struct xfs_attr3_icleaf_hdr *ichdr_s,
2326 int start_s,
2327 struct xfs_attr_leafblock *leaf_d,
2328 struct xfs_attr3_icleaf_hdr *ichdr_d,
2329 int start_d,
2330 int count)
2331 {
2332 struct xfs_attr_leaf_entry *entry_s;
2333 struct xfs_attr_leaf_entry *entry_d;
2334 int desti;
2335 int tmp;
2336 int i;
2337
2338 /*
2339 * Check for nothing to do.
2340 */
2341 if (count == 0)
2342 return;
2343
2344 /*
2345 * Set up environment.
2346 */
2347 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2348 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2349 ASSERT(ichdr_s->magic == ichdr_d->magic);
2350 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2351 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2352 + xfs_attr3_leaf_hdr_size(leaf_s));
2353 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2354 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2355 + xfs_attr3_leaf_hdr_size(leaf_d));
2356
2357 ASSERT(start_s < ichdr_s->count);
2358 ASSERT(start_d <= ichdr_d->count);
2359 ASSERT(count <= ichdr_s->count);
2360
2361
2362 /*
2363 * Move the entries in the destination leaf up to make a hole?
2364 */
2365 if (start_d < ichdr_d->count) {
2366 tmp = ichdr_d->count - start_d;
2367 tmp *= sizeof(xfs_attr_leaf_entry_t);
2368 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2369 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2370 memmove(entry_d, entry_s, tmp);
2371 }
2372
2373 /*
2374 * Copy all entry's in the same (sorted) order,
2375 * but allocate attribute info packed and in sequence.
2376 */
2377 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2378 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2379 desti = start_d;
2380 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2381 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2382 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2383 #ifdef GROT
2384 /*
2385 * Code to drop INCOMPLETE entries. Difficult to use as we
2386 * may also need to change the insertion index. Code turned
2387 * off for 6.2, should be revisited later.
2388 */
2389 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2390 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2391 ichdr_s->usedbytes -= tmp;
2392 ichdr_s->count -= 1;
2393 entry_d--; /* to compensate for ++ in loop hdr */
2394 desti--;
2395 if ((start_s + i) < offset)
2396 result++; /* insertion index adjustment */
2397 } else {
2398 #endif /* GROT */
2399 ichdr_d->firstused -= tmp;
2400 /* both on-disk, don't endian flip twice */
2401 entry_d->hashval = entry_s->hashval;
2402 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2403 entry_d->flags = entry_s->flags;
2404 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2405 <= args->geo->blksize);
2406 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2407 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2408 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2409 <= args->geo->blksize);
2410 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2411 ichdr_s->usedbytes -= tmp;
2412 ichdr_d->usedbytes += tmp;
2413 ichdr_s->count -= 1;
2414 ichdr_d->count += 1;
2415 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2416 + xfs_attr3_leaf_hdr_size(leaf_d);
2417 ASSERT(ichdr_d->firstused >= tmp);
2418 #ifdef GROT
2419 }
2420 #endif /* GROT */
2421 }
2422
2423 /*
2424 * Zero out the entries we just copied.
2425 */
2426 if (start_s == ichdr_s->count) {
2427 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2428 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2429 ASSERT(((char *)entry_s + tmp) <=
2430 ((char *)leaf_s + args->geo->blksize));
2431 memset(entry_s, 0, tmp);
2432 } else {
2433 /*
2434 * Move the remaining entries down to fill the hole,
2435 * then zero the entries at the top.
2436 */
2437 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2438 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2439 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2440 memmove(entry_d, entry_s, tmp);
2441
2442 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2443 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2444 ASSERT(((char *)entry_s + tmp) <=
2445 ((char *)leaf_s + args->geo->blksize));
2446 memset(entry_s, 0, tmp);
2447 }
2448
2449 /*
2450 * Fill in the freemap information
2451 */
2452 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2453 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2454 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2455 ichdr_d->freemap[1].base = 0;
2456 ichdr_d->freemap[2].base = 0;
2457 ichdr_d->freemap[1].size = 0;
2458 ichdr_d->freemap[2].size = 0;
2459 ichdr_s->holes = 1; /* leaf may not be compact */
2460 }
2461
2462 /*
2463 * Pick up the last hashvalue from a leaf block.
2464 */
2465 xfs_dahash_t
2466 xfs_attr_leaf_lasthash(
2467 struct xfs_buf *bp,
2468 int *count)
2469 {
2470 struct xfs_attr3_icleaf_hdr ichdr;
2471 struct xfs_attr_leaf_entry *entries;
2472 struct xfs_mount *mp = bp->b_target->bt_mount;
2473
2474 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2475 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2476 if (count)
2477 *count = ichdr.count;
2478 if (!ichdr.count)
2479 return 0;
2480 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2481 }
2482
2483 /*
2484 * Calculate the number of bytes used to store the indicated attribute
2485 * (whether local or remote only calculate bytes in this block).
2486 */
2487 STATIC int
2488 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2489 {
2490 struct xfs_attr_leaf_entry *entries;
2491 xfs_attr_leaf_name_local_t *name_loc;
2492 xfs_attr_leaf_name_remote_t *name_rmt;
2493 int size;
2494
2495 entries = xfs_attr3_leaf_entryp(leaf);
2496 if (entries[index].flags & XFS_ATTR_LOCAL) {
2497 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2498 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2499 be16_to_cpu(name_loc->valuelen));
2500 } else {
2501 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2502 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2503 }
2504 return size;
2505 }
2506
2507 /*
2508 * Calculate the number of bytes that would be required to store the new
2509 * attribute (whether local or remote only calculate bytes in this block).
2510 * This routine decides as a side effect whether the attribute will be
2511 * a "local" or a "remote" attribute.
2512 */
2513 int
2514 xfs_attr_leaf_newentsize(
2515 struct xfs_da_args *args,
2516 int *local)
2517 {
2518 int size;
2519
2520 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2521 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2522 if (local)
2523 *local = 1;
2524 return size;
2525 }
2526 if (local)
2527 *local = 0;
2528 return xfs_attr_leaf_entsize_remote(args->namelen);
2529 }
2530
2531
2532 /*========================================================================
2533 * Manage the INCOMPLETE flag in a leaf entry
2534 *========================================================================*/
2535
2536 /*
2537 * Clear the INCOMPLETE flag on an entry in a leaf block.
2538 */
2539 int
2540 xfs_attr3_leaf_clearflag(
2541 struct xfs_da_args *args)
2542 {
2543 struct xfs_attr_leafblock *leaf;
2544 struct xfs_attr_leaf_entry *entry;
2545 struct xfs_attr_leaf_name_remote *name_rmt;
2546 struct xfs_buf *bp;
2547 int error;
2548 #ifdef DEBUG
2549 struct xfs_attr3_icleaf_hdr ichdr;
2550 xfs_attr_leaf_name_local_t *name_loc;
2551 int namelen;
2552 char *name;
2553 #endif /* DEBUG */
2554
2555 trace_xfs_attr_leaf_clearflag(args);
2556 /*
2557 * Set up the operation.
2558 */
2559 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2560 if (error)
2561 return error;
2562
2563 leaf = bp->b_addr;
2564 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2565 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2566
2567 #ifdef DEBUG
2568 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2569 ASSERT(args->index < ichdr.count);
2570 ASSERT(args->index >= 0);
2571
2572 if (entry->flags & XFS_ATTR_LOCAL) {
2573 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2574 namelen = name_loc->namelen;
2575 name = (char *)name_loc->nameval;
2576 } else {
2577 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2578 namelen = name_rmt->namelen;
2579 name = (char *)name_rmt->name;
2580 }
2581 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2582 ASSERT(namelen == args->namelen);
2583 ASSERT(memcmp(name, args->name, namelen) == 0);
2584 #endif /* DEBUG */
2585
2586 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2587 xfs_trans_log_buf(args->trans, bp,
2588 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2589
2590 if (args->rmtblkno) {
2591 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2592 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2593 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2594 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2595 xfs_trans_log_buf(args->trans, bp,
2596 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2597 }
2598
2599 /*
2600 * Commit the flag value change and start the next trans in series.
2601 */
2602 return xfs_trans_roll_inode(&args->trans, args->dp);
2603 }
2604
2605 /*
2606 * Set the INCOMPLETE flag on an entry in a leaf block.
2607 */
2608 int
2609 xfs_attr3_leaf_setflag(
2610 struct xfs_da_args *args)
2611 {
2612 struct xfs_attr_leafblock *leaf;
2613 struct xfs_attr_leaf_entry *entry;
2614 struct xfs_attr_leaf_name_remote *name_rmt;
2615 struct xfs_buf *bp;
2616 int error;
2617 #ifdef DEBUG
2618 struct xfs_attr3_icleaf_hdr ichdr;
2619 #endif
2620
2621 trace_xfs_attr_leaf_setflag(args);
2622
2623 /*
2624 * Set up the operation.
2625 */
2626 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2627 if (error)
2628 return error;
2629
2630 leaf = bp->b_addr;
2631 #ifdef DEBUG
2632 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2633 ASSERT(args->index < ichdr.count);
2634 ASSERT(args->index >= 0);
2635 #endif
2636 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2637
2638 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2639 entry->flags |= XFS_ATTR_INCOMPLETE;
2640 xfs_trans_log_buf(args->trans, bp,
2641 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2642 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2643 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2644 name_rmt->valueblk = 0;
2645 name_rmt->valuelen = 0;
2646 xfs_trans_log_buf(args->trans, bp,
2647 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2648 }
2649
2650 /*
2651 * Commit the flag value change and start the next trans in series.
2652 */
2653 return xfs_trans_roll_inode(&args->trans, args->dp);
2654 }
2655
2656 /*
2657 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2658 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2659 * entry given by args->blkno2/index2.
2660 *
2661 * Note that they could be in different blocks, or in the same block.
2662 */
2663 int
2664 xfs_attr3_leaf_flipflags(
2665 struct xfs_da_args *args)
2666 {
2667 struct xfs_attr_leafblock *leaf1;
2668 struct xfs_attr_leafblock *leaf2;
2669 struct xfs_attr_leaf_entry *entry1;
2670 struct xfs_attr_leaf_entry *entry2;
2671 struct xfs_attr_leaf_name_remote *name_rmt;
2672 struct xfs_buf *bp1;
2673 struct xfs_buf *bp2;
2674 int error;
2675 #ifdef DEBUG
2676 struct xfs_attr3_icleaf_hdr ichdr1;
2677 struct xfs_attr3_icleaf_hdr ichdr2;
2678 xfs_attr_leaf_name_local_t *name_loc;
2679 int namelen1, namelen2;
2680 char *name1, *name2;
2681 #endif /* DEBUG */
2682
2683 trace_xfs_attr_leaf_flipflags(args);
2684
2685 /*
2686 * Read the block containing the "old" attr
2687 */
2688 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2689 if (error)
2690 return error;
2691
2692 /*
2693 * Read the block containing the "new" attr, if it is different
2694 */
2695 if (args->blkno2 != args->blkno) {
2696 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2697 -1, &bp2);
2698 if (error)
2699 return error;
2700 } else {
2701 bp2 = bp1;
2702 }
2703
2704 leaf1 = bp1->b_addr;
2705 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2706
2707 leaf2 = bp2->b_addr;
2708 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2709
2710 #ifdef DEBUG
2711 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2712 ASSERT(args->index < ichdr1.count);
2713 ASSERT(args->index >= 0);
2714
2715 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2716 ASSERT(args->index2 < ichdr2.count);
2717 ASSERT(args->index2 >= 0);
2718
2719 if (entry1->flags & XFS_ATTR_LOCAL) {
2720 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2721 namelen1 = name_loc->namelen;
2722 name1 = (char *)name_loc->nameval;
2723 } else {
2724 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2725 namelen1 = name_rmt->namelen;
2726 name1 = (char *)name_rmt->name;
2727 }
2728 if (entry2->flags & XFS_ATTR_LOCAL) {
2729 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2730 namelen2 = name_loc->namelen;
2731 name2 = (char *)name_loc->nameval;
2732 } else {
2733 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2734 namelen2 = name_rmt->namelen;
2735 name2 = (char *)name_rmt->name;
2736 }
2737 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2738 ASSERT(namelen1 == namelen2);
2739 ASSERT(memcmp(name1, name2, namelen1) == 0);
2740 #endif /* DEBUG */
2741
2742 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2743 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2744
2745 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2746 xfs_trans_log_buf(args->trans, bp1,
2747 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2748 if (args->rmtblkno) {
2749 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2750 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2751 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2752 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2753 xfs_trans_log_buf(args->trans, bp1,
2754 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2755 }
2756
2757 entry2->flags |= XFS_ATTR_INCOMPLETE;
2758 xfs_trans_log_buf(args->trans, bp2,
2759 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2760 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2761 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2762 name_rmt->valueblk = 0;
2763 name_rmt->valuelen = 0;
2764 xfs_trans_log_buf(args->trans, bp2,
2765 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2766 }
2767
2768 /*
2769 * Commit the flag value change and start the next trans in series.
2770 */
2771 error = xfs_trans_roll_inode(&args->trans, args->dp);
2772
2773 return error;
2774 }