]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blob - libxfs/xfs_attr_leaf.c
Merge branch 'libxfs-commit-script' 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_attr3_icleaf_hdr ichdr;
252
253 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
254
255 if (xfs_sb_version_hascrc(&mp->m_sb)) {
256 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
257
258 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
259 return false;
260
261 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
262 return false;
263 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
264 return false;
265 } else {
266 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
267 return false;
268 }
269 if (ichdr.count == 0)
270 return false;
271
272 /* XXX: need to range check rest of attr header values */
273 /* XXX: hash order check? */
274
275 return true;
276 }
277
278 static void
279 xfs_attr3_leaf_write_verify(
280 struct xfs_buf *bp)
281 {
282 struct xfs_mount *mp = bp->b_target->bt_mount;
283 struct xfs_buf_log_item *bip = bp->b_fspriv;
284 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
285
286 if (!xfs_attr3_leaf_verify(bp)) {
287 xfs_buf_ioerror(bp, -EFSCORRUPTED);
288 xfs_verifier_error(bp);
289 return;
290 }
291
292 if (!xfs_sb_version_hascrc(&mp->m_sb))
293 return;
294
295 if (bip)
296 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
297
298 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
299 }
300
301 /*
302 * leaf/node format detection on trees is sketchy, so a node read can be done on
303 * leaf level blocks when detection identifies the tree as a node format tree
304 * incorrectly. In this case, we need to swap the verifier to match the correct
305 * format of the block being read.
306 */
307 static void
308 xfs_attr3_leaf_read_verify(
309 struct xfs_buf *bp)
310 {
311 struct xfs_mount *mp = bp->b_target->bt_mount;
312
313 if (xfs_sb_version_hascrc(&mp->m_sb) &&
314 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
315 xfs_buf_ioerror(bp, -EFSBADCRC);
316 else if (!xfs_attr3_leaf_verify(bp))
317 xfs_buf_ioerror(bp, -EFSCORRUPTED);
318
319 if (bp->b_error)
320 xfs_verifier_error(bp);
321 }
322
323 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
324 .verify_read = xfs_attr3_leaf_read_verify,
325 .verify_write = xfs_attr3_leaf_write_verify,
326 };
327
328 int
329 xfs_attr3_leaf_read(
330 struct xfs_trans *tp,
331 struct xfs_inode *dp,
332 xfs_dablk_t bno,
333 xfs_daddr_t mappedbno,
334 struct xfs_buf **bpp)
335 {
336 int err;
337
338 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
339 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
340 if (!err && tp)
341 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
342 return err;
343 }
344
345 /*========================================================================
346 * Namespace helper routines
347 *========================================================================*/
348
349 /*
350 * If namespace bits don't match return 0.
351 * If all match then return 1.
352 */
353 STATIC int
354 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
355 {
356 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
357 }
358
359
360 /*========================================================================
361 * External routines when attribute fork size < XFS_LITINO(mp).
362 *========================================================================*/
363
364 /*
365 * Query whether the requested number of additional bytes of extended
366 * attribute space will be able to fit inline.
367 *
368 * Returns zero if not, else the di_forkoff fork offset to be used in the
369 * literal area for attribute data once the new bytes have been added.
370 *
371 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
372 * special case for dev/uuid inodes, they have fixed size data forks.
373 */
374 int
375 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
376 {
377 int offset;
378 int minforkoff; /* lower limit on valid forkoff locations */
379 int maxforkoff; /* upper limit on valid forkoff locations */
380 int dsize;
381 xfs_mount_t *mp = dp->i_mount;
382
383 /* rounded down */
384 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
385
386 switch (dp->i_d.di_format) {
387 case XFS_DINODE_FMT_DEV:
388 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
389 return (offset >= minforkoff) ? minforkoff : 0;
390 case XFS_DINODE_FMT_UUID:
391 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
392 return (offset >= minforkoff) ? minforkoff : 0;
393 }
394
395 /*
396 * If the requested numbers of bytes is smaller or equal to the
397 * current attribute fork size we can always proceed.
398 *
399 * Note that if_bytes in the data fork might actually be larger than
400 * the current data fork size is due to delalloc extents. In that
401 * case either the extent count will go down when they are converted
402 * to real extents, or the delalloc conversion will take care of the
403 * literal area rebalancing.
404 */
405 if (bytes <= XFS_IFORK_ASIZE(dp))
406 return dp->i_d.di_forkoff;
407
408 /*
409 * For attr2 we can try to move the forkoff if there is space in the
410 * literal area, but for the old format we are done if there is no
411 * space in the fixed attribute fork.
412 */
413 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
414 return 0;
415
416 dsize = dp->i_df.if_bytes;
417
418 switch (dp->i_d.di_format) {
419 case XFS_DINODE_FMT_EXTENTS:
420 /*
421 * If there is no attr fork and the data fork is extents,
422 * determine if creating the default attr fork will result
423 * in the extents form migrating to btree. If so, the
424 * minimum offset only needs to be the space required for
425 * the btree root.
426 */
427 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
428 xfs_default_attroffset(dp))
429 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
430 break;
431 case XFS_DINODE_FMT_BTREE:
432 /*
433 * If we have a data btree then keep forkoff if we have one,
434 * otherwise we are adding a new attr, so then we set
435 * minforkoff to where the btree root can finish so we have
436 * plenty of room for attrs
437 */
438 if (dp->i_d.di_forkoff) {
439 if (offset < dp->i_d.di_forkoff)
440 return 0;
441 return dp->i_d.di_forkoff;
442 }
443 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
444 break;
445 }
446
447 /*
448 * A data fork btree root must have space for at least
449 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
450 */
451 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
452 minforkoff = roundup(minforkoff, 8) >> 3;
453
454 /* attr fork btree root can have at least this many key/ptr pairs */
455 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
456 XFS_BMDR_SPACE_CALC(MINABTPTRS);
457 maxforkoff = maxforkoff >> 3; /* rounded down */
458
459 if (offset >= maxforkoff)
460 return maxforkoff;
461 if (offset >= minforkoff)
462 return offset;
463 return 0;
464 }
465
466 /*
467 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
468 */
469 STATIC void
470 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
471 {
472 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
473 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
474 spin_lock(&mp->m_sb_lock);
475 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
476 xfs_sb_version_addattr2(&mp->m_sb);
477 spin_unlock(&mp->m_sb_lock);
478 xfs_log_sb(tp);
479 } else
480 spin_unlock(&mp->m_sb_lock);
481 }
482 }
483
484 /*
485 * Create the initial contents of a shortform attribute list.
486 */
487 void
488 xfs_attr_shortform_create(xfs_da_args_t *args)
489 {
490 xfs_attr_sf_hdr_t *hdr;
491 xfs_inode_t *dp;
492 xfs_ifork_t *ifp;
493
494 trace_xfs_attr_sf_create(args);
495
496 dp = args->dp;
497 ASSERT(dp != NULL);
498 ifp = dp->i_afp;
499 ASSERT(ifp != NULL);
500 ASSERT(ifp->if_bytes == 0);
501 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
502 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
503 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
504 ifp->if_flags |= XFS_IFINLINE;
505 } else {
506 ASSERT(ifp->if_flags & XFS_IFINLINE);
507 }
508 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
509 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
510 hdr->count = 0;
511 hdr->totsize = cpu_to_be16(sizeof(*hdr));
512 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
513 }
514
515 /*
516 * Add a name/value pair to the shortform attribute list.
517 * Overflow from the inode has already been checked for.
518 */
519 void
520 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
521 {
522 xfs_attr_shortform_t *sf;
523 xfs_attr_sf_entry_t *sfe;
524 int i, offset, size;
525 xfs_mount_t *mp;
526 xfs_inode_t *dp;
527 xfs_ifork_t *ifp;
528
529 trace_xfs_attr_sf_add(args);
530
531 dp = args->dp;
532 mp = dp->i_mount;
533 dp->i_d.di_forkoff = forkoff;
534
535 ifp = dp->i_afp;
536 ASSERT(ifp->if_flags & XFS_IFINLINE);
537 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
538 sfe = &sf->list[0];
539 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
540 #ifdef DEBUG
541 if (sfe->namelen != args->namelen)
542 continue;
543 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
544 continue;
545 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
546 continue;
547 ASSERT(0);
548 #endif
549 }
550
551 offset = (char *)sfe - (char *)sf;
552 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
553 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
554 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
555 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
556
557 sfe->namelen = args->namelen;
558 sfe->valuelen = args->valuelen;
559 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
560 memcpy(sfe->nameval, args->name, args->namelen);
561 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
562 sf->hdr.count++;
563 be16_add_cpu(&sf->hdr.totsize, size);
564 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
565
566 xfs_sbversion_add_attr2(mp, args->trans);
567 }
568
569 /*
570 * After the last attribute is removed revert to original inode format,
571 * making all literal area available to the data fork once more.
572 */
573 void
574 xfs_attr_fork_remove(
575 struct xfs_inode *ip,
576 struct xfs_trans *tp)
577 {
578 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
579 ip->i_d.di_forkoff = 0;
580 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
581
582 ASSERT(ip->i_d.di_anextents == 0);
583 ASSERT(ip->i_afp == NULL);
584
585 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
586 }
587
588 /*
589 * Remove an attribute from the shortform attribute list structure.
590 */
591 int
592 xfs_attr_shortform_remove(xfs_da_args_t *args)
593 {
594 xfs_attr_shortform_t *sf;
595 xfs_attr_sf_entry_t *sfe;
596 int base, size=0, end, totsize, i;
597 xfs_mount_t *mp;
598 xfs_inode_t *dp;
599
600 trace_xfs_attr_sf_remove(args);
601
602 dp = args->dp;
603 mp = dp->i_mount;
604 base = sizeof(xfs_attr_sf_hdr_t);
605 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
606 sfe = &sf->list[0];
607 end = sf->hdr.count;
608 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
609 base += size, i++) {
610 size = XFS_ATTR_SF_ENTSIZE(sfe);
611 if (sfe->namelen != args->namelen)
612 continue;
613 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
614 continue;
615 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
616 continue;
617 break;
618 }
619 if (i == end)
620 return -ENOATTR;
621
622 /*
623 * Fix up the attribute fork data, covering the hole
624 */
625 end = base + size;
626 totsize = be16_to_cpu(sf->hdr.totsize);
627 if (end != totsize)
628 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
629 sf->hdr.count--;
630 be16_add_cpu(&sf->hdr.totsize, -size);
631
632 /*
633 * Fix up the start offset of the attribute fork
634 */
635 totsize -= size;
636 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
637 (mp->m_flags & XFS_MOUNT_ATTR2) &&
638 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
639 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
640 xfs_attr_fork_remove(dp, args->trans);
641 } else {
642 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
643 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
644 ASSERT(dp->i_d.di_forkoff);
645 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
646 (args->op_flags & XFS_DA_OP_ADDNAME) ||
647 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
648 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
649 xfs_trans_log_inode(args->trans, dp,
650 XFS_ILOG_CORE | XFS_ILOG_ADATA);
651 }
652
653 xfs_sbversion_add_attr2(mp, args->trans);
654
655 return 0;
656 }
657
658 /*
659 * Look up a name in a shortform attribute list structure.
660 */
661 /*ARGSUSED*/
662 int
663 xfs_attr_shortform_lookup(xfs_da_args_t *args)
664 {
665 xfs_attr_shortform_t *sf;
666 xfs_attr_sf_entry_t *sfe;
667 int i;
668 xfs_ifork_t *ifp;
669
670 trace_xfs_attr_sf_lookup(args);
671
672 ifp = args->dp->i_afp;
673 ASSERT(ifp->if_flags & XFS_IFINLINE);
674 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
675 sfe = &sf->list[0];
676 for (i = 0; i < sf->hdr.count;
677 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
678 if (sfe->namelen != args->namelen)
679 continue;
680 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
681 continue;
682 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
683 continue;
684 return -EEXIST;
685 }
686 return -ENOATTR;
687 }
688
689 /*
690 * Look up a name in a shortform attribute list structure.
691 */
692 /*ARGSUSED*/
693 int
694 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
695 {
696 xfs_attr_shortform_t *sf;
697 xfs_attr_sf_entry_t *sfe;
698 int i;
699
700 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
701 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
702 sfe = &sf->list[0];
703 for (i = 0; i < sf->hdr.count;
704 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
705 if (sfe->namelen != args->namelen)
706 continue;
707 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
708 continue;
709 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
710 continue;
711 if (args->flags & ATTR_KERNOVAL) {
712 args->valuelen = sfe->valuelen;
713 return -EEXIST;
714 }
715 if (args->valuelen < sfe->valuelen) {
716 args->valuelen = sfe->valuelen;
717 return -ERANGE;
718 }
719 args->valuelen = sfe->valuelen;
720 memcpy(args->value, &sfe->nameval[args->namelen],
721 args->valuelen);
722 return -EEXIST;
723 }
724 return -ENOATTR;
725 }
726
727 /*
728 * Convert from using the shortform to the leaf.
729 */
730 int
731 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
732 {
733 xfs_inode_t *dp;
734 xfs_attr_shortform_t *sf;
735 xfs_attr_sf_entry_t *sfe;
736 xfs_da_args_t nargs;
737 char *tmpbuffer;
738 int error, i, size;
739 xfs_dablk_t blkno;
740 struct xfs_buf *bp;
741 xfs_ifork_t *ifp;
742
743 trace_xfs_attr_sf_to_leaf(args);
744
745 dp = args->dp;
746 ifp = dp->i_afp;
747 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
748 size = be16_to_cpu(sf->hdr.totsize);
749 tmpbuffer = kmem_alloc(size, KM_SLEEP);
750 ASSERT(tmpbuffer != NULL);
751 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
752 sf = (xfs_attr_shortform_t *)tmpbuffer;
753
754 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
755 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
756
757 bp = NULL;
758 error = xfs_da_grow_inode(args, &blkno);
759 if (error) {
760 /*
761 * If we hit an IO error middle of the transaction inside
762 * grow_inode(), we may have inconsistent data. Bail out.
763 */
764 if (error == -EIO)
765 goto out;
766 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
767 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
768 goto out;
769 }
770
771 ASSERT(blkno == 0);
772 error = xfs_attr3_leaf_create(args, blkno, &bp);
773 if (error) {
774 error = xfs_da_shrink_inode(args, 0, bp);
775 bp = NULL;
776 if (error)
777 goto out;
778 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
779 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
780 goto out;
781 }
782
783 memset((char *)&nargs, 0, sizeof(nargs));
784 nargs.dp = dp;
785 nargs.geo = args->geo;
786 nargs.firstblock = args->firstblock;
787 nargs.flist = args->flist;
788 nargs.total = args->total;
789 nargs.whichfork = XFS_ATTR_FORK;
790 nargs.trans = args->trans;
791 nargs.op_flags = XFS_DA_OP_OKNOENT;
792
793 sfe = &sf->list[0];
794 for (i = 0; i < sf->hdr.count; i++) {
795 nargs.name = sfe->nameval;
796 nargs.namelen = sfe->namelen;
797 nargs.value = &sfe->nameval[nargs.namelen];
798 nargs.valuelen = sfe->valuelen;
799 nargs.hashval = xfs_da_hashname(sfe->nameval,
800 sfe->namelen);
801 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
802 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
803 ASSERT(error == -ENOATTR);
804 error = xfs_attr3_leaf_add(bp, &nargs);
805 ASSERT(error != -ENOSPC);
806 if (error)
807 goto out;
808 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
809 }
810 error = 0;
811
812 out:
813 kmem_free(tmpbuffer);
814 return error;
815 }
816
817 /*
818 * Check a leaf attribute block to see if all the entries would fit into
819 * a shortform attribute list.
820 */
821 int
822 xfs_attr_shortform_allfit(
823 struct xfs_buf *bp,
824 struct xfs_inode *dp)
825 {
826 struct xfs_attr_leafblock *leaf;
827 struct xfs_attr_leaf_entry *entry;
828 xfs_attr_leaf_name_local_t *name_loc;
829 struct xfs_attr3_icleaf_hdr leafhdr;
830 int bytes;
831 int i;
832 struct xfs_mount *mp = bp->b_target->bt_mount;
833
834 leaf = bp->b_addr;
835 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
836 entry = xfs_attr3_leaf_entryp(leaf);
837
838 bytes = sizeof(struct xfs_attr_sf_hdr);
839 for (i = 0; i < leafhdr.count; entry++, i++) {
840 if (entry->flags & XFS_ATTR_INCOMPLETE)
841 continue; /* don't copy partial entries */
842 if (!(entry->flags & XFS_ATTR_LOCAL))
843 return 0;
844 name_loc = xfs_attr3_leaf_name_local(leaf, i);
845 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
846 return 0;
847 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
848 return 0;
849 bytes += sizeof(struct xfs_attr_sf_entry) - 1
850 + name_loc->namelen
851 + be16_to_cpu(name_loc->valuelen);
852 }
853 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
854 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
855 (bytes == sizeof(struct xfs_attr_sf_hdr)))
856 return -1;
857 return xfs_attr_shortform_bytesfit(dp, bytes);
858 }
859
860 /*
861 * Convert a leaf attribute list to shortform attribute list
862 */
863 int
864 xfs_attr3_leaf_to_shortform(
865 struct xfs_buf *bp,
866 struct xfs_da_args *args,
867 int forkoff)
868 {
869 struct xfs_attr_leafblock *leaf;
870 struct xfs_attr3_icleaf_hdr ichdr;
871 struct xfs_attr_leaf_entry *entry;
872 struct xfs_attr_leaf_name_local *name_loc;
873 struct xfs_da_args nargs;
874 struct xfs_inode *dp = args->dp;
875 char *tmpbuffer;
876 int error;
877 int i;
878
879 trace_xfs_attr_leaf_to_sf(args);
880
881 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
882 if (!tmpbuffer)
883 return -ENOMEM;
884
885 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
886
887 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
888 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
889 entry = xfs_attr3_leaf_entryp(leaf);
890
891 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
892 memset(bp->b_addr, 0, args->geo->blksize);
893
894 /*
895 * Clean out the prior contents of the attribute list.
896 */
897 error = xfs_da_shrink_inode(args, 0, bp);
898 if (error)
899 goto out;
900
901 if (forkoff == -1) {
902 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
903 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
904 xfs_attr_fork_remove(dp, args->trans);
905 goto out;
906 }
907
908 xfs_attr_shortform_create(args);
909
910 /*
911 * Copy the attributes
912 */
913 memset((char *)&nargs, 0, sizeof(nargs));
914 nargs.geo = args->geo;
915 nargs.dp = dp;
916 nargs.firstblock = args->firstblock;
917 nargs.flist = args->flist;
918 nargs.total = args->total;
919 nargs.whichfork = XFS_ATTR_FORK;
920 nargs.trans = args->trans;
921 nargs.op_flags = XFS_DA_OP_OKNOENT;
922
923 for (i = 0; i < ichdr.count; entry++, i++) {
924 if (entry->flags & XFS_ATTR_INCOMPLETE)
925 continue; /* don't copy partial entries */
926 if (!entry->nameidx)
927 continue;
928 ASSERT(entry->flags & XFS_ATTR_LOCAL);
929 name_loc = xfs_attr3_leaf_name_local(leaf, i);
930 nargs.name = name_loc->nameval;
931 nargs.namelen = name_loc->namelen;
932 nargs.value = &name_loc->nameval[nargs.namelen];
933 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
934 nargs.hashval = be32_to_cpu(entry->hashval);
935 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
936 xfs_attr_shortform_add(&nargs, forkoff);
937 }
938 error = 0;
939
940 out:
941 kmem_free(tmpbuffer);
942 return error;
943 }
944
945 /*
946 * Convert from using a single leaf to a root node and a leaf.
947 */
948 int
949 xfs_attr3_leaf_to_node(
950 struct xfs_da_args *args)
951 {
952 struct xfs_attr_leafblock *leaf;
953 struct xfs_attr3_icleaf_hdr icleafhdr;
954 struct xfs_attr_leaf_entry *entries;
955 struct xfs_da_node_entry *btree;
956 struct xfs_da3_icnode_hdr icnodehdr;
957 struct xfs_da_intnode *node;
958 struct xfs_inode *dp = args->dp;
959 struct xfs_mount *mp = dp->i_mount;
960 struct xfs_buf *bp1 = NULL;
961 struct xfs_buf *bp2 = NULL;
962 xfs_dablk_t blkno;
963 int error;
964
965 trace_xfs_attr_leaf_to_node(args);
966
967 error = xfs_da_grow_inode(args, &blkno);
968 if (error)
969 goto out;
970 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
971 if (error)
972 goto out;
973
974 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
975 if (error)
976 goto out;
977
978 /* copy leaf to new buffer, update identifiers */
979 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
980 bp2->b_ops = bp1->b_ops;
981 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
982 if (xfs_sb_version_hascrc(&mp->m_sb)) {
983 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
984 hdr3->blkno = cpu_to_be64(bp2->b_bn);
985 }
986 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
987
988 /*
989 * Set up the new root node.
990 */
991 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
992 if (error)
993 goto out;
994 node = bp1->b_addr;
995 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
996 btree = dp->d_ops->node_tree_p(node);
997
998 leaf = bp2->b_addr;
999 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1000 entries = xfs_attr3_leaf_entryp(leaf);
1001
1002 /* both on-disk, don't endian-flip twice */
1003 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1004 btree[0].before = cpu_to_be32(blkno);
1005 icnodehdr.count = 1;
1006 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1007 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1008 error = 0;
1009 out:
1010 return error;
1011 }
1012
1013 /*========================================================================
1014 * Routines used for growing the Btree.
1015 *========================================================================*/
1016
1017 /*
1018 * Create the initial contents of a leaf attribute list
1019 * or a leaf in a node attribute list.
1020 */
1021 STATIC int
1022 xfs_attr3_leaf_create(
1023 struct xfs_da_args *args,
1024 xfs_dablk_t blkno,
1025 struct xfs_buf **bpp)
1026 {
1027 struct xfs_attr_leafblock *leaf;
1028 struct xfs_attr3_icleaf_hdr ichdr;
1029 struct xfs_inode *dp = args->dp;
1030 struct xfs_mount *mp = dp->i_mount;
1031 struct xfs_buf *bp;
1032 int error;
1033
1034 trace_xfs_attr_leaf_create(args);
1035
1036 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1037 XFS_ATTR_FORK);
1038 if (error)
1039 return error;
1040 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1041 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1042 leaf = bp->b_addr;
1043 memset(leaf, 0, args->geo->blksize);
1044
1045 memset(&ichdr, 0, sizeof(ichdr));
1046 ichdr.firstused = args->geo->blksize;
1047
1048 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1049 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1050
1051 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1052
1053 hdr3->blkno = cpu_to_be64(bp->b_bn);
1054 hdr3->owner = cpu_to_be64(dp->i_ino);
1055 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
1056
1057 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1058 } else {
1059 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1060 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1061 }
1062 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1063
1064 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1065 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1066
1067 *bpp = bp;
1068 return 0;
1069 }
1070
1071 /*
1072 * Split the leaf node, rebalance, then add the new entry.
1073 */
1074 int
1075 xfs_attr3_leaf_split(
1076 struct xfs_da_state *state,
1077 struct xfs_da_state_blk *oldblk,
1078 struct xfs_da_state_blk *newblk)
1079 {
1080 xfs_dablk_t blkno;
1081 int error;
1082
1083 trace_xfs_attr_leaf_split(state->args);
1084
1085 /*
1086 * Allocate space for a new leaf node.
1087 */
1088 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1089 error = xfs_da_grow_inode(state->args, &blkno);
1090 if (error)
1091 return error;
1092 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1093 if (error)
1094 return error;
1095 newblk->blkno = blkno;
1096 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1097
1098 /*
1099 * Rebalance the entries across the two leaves.
1100 * NOTE: rebalance() currently depends on the 2nd block being empty.
1101 */
1102 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1103 error = xfs_da3_blk_link(state, oldblk, newblk);
1104 if (error)
1105 return error;
1106
1107 /*
1108 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1109 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1110 * "new" attrs info. Will need the "old" info to remove it later.
1111 *
1112 * Insert the "new" entry in the correct block.
1113 */
1114 if (state->inleaf) {
1115 trace_xfs_attr_leaf_add_old(state->args);
1116 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1117 } else {
1118 trace_xfs_attr_leaf_add_new(state->args);
1119 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1120 }
1121
1122 /*
1123 * Update last hashval in each block since we added the name.
1124 */
1125 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1126 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1127 return error;
1128 }
1129
1130 /*
1131 * Add a name to the leaf attribute list structure.
1132 */
1133 int
1134 xfs_attr3_leaf_add(
1135 struct xfs_buf *bp,
1136 struct xfs_da_args *args)
1137 {
1138 struct xfs_attr_leafblock *leaf;
1139 struct xfs_attr3_icleaf_hdr ichdr;
1140 int tablesize;
1141 int entsize;
1142 int sum;
1143 int tmp;
1144 int i;
1145
1146 trace_xfs_attr_leaf_add(args);
1147
1148 leaf = bp->b_addr;
1149 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1150 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1151 entsize = xfs_attr_leaf_newentsize(args, NULL);
1152
1153 /*
1154 * Search through freemap for first-fit on new name length.
1155 * (may need to figure in size of entry struct too)
1156 */
1157 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1158 + xfs_attr3_leaf_hdr_size(leaf);
1159 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1160 if (tablesize > ichdr.firstused) {
1161 sum += ichdr.freemap[i].size;
1162 continue;
1163 }
1164 if (!ichdr.freemap[i].size)
1165 continue; /* no space in this map */
1166 tmp = entsize;
1167 if (ichdr.freemap[i].base < ichdr.firstused)
1168 tmp += sizeof(xfs_attr_leaf_entry_t);
1169 if (ichdr.freemap[i].size >= tmp) {
1170 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1171 goto out_log_hdr;
1172 }
1173 sum += ichdr.freemap[i].size;
1174 }
1175
1176 /*
1177 * If there are no holes in the address space of the block,
1178 * and we don't have enough freespace, then compaction will do us
1179 * no good and we should just give up.
1180 */
1181 if (!ichdr.holes && sum < entsize)
1182 return -ENOSPC;
1183
1184 /*
1185 * Compact the entries to coalesce free space.
1186 * This may change the hdr->count via dropping INCOMPLETE entries.
1187 */
1188 xfs_attr3_leaf_compact(args, &ichdr, bp);
1189
1190 /*
1191 * After compaction, the block is guaranteed to have only one
1192 * free region, in freemap[0]. If it is not big enough, give up.
1193 */
1194 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1195 tmp = -ENOSPC;
1196 goto out_log_hdr;
1197 }
1198
1199 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1200
1201 out_log_hdr:
1202 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1203 xfs_trans_log_buf(args->trans, bp,
1204 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1205 xfs_attr3_leaf_hdr_size(leaf)));
1206 return tmp;
1207 }
1208
1209 /*
1210 * Add a name to a leaf attribute list structure.
1211 */
1212 STATIC int
1213 xfs_attr3_leaf_add_work(
1214 struct xfs_buf *bp,
1215 struct xfs_attr3_icleaf_hdr *ichdr,
1216 struct xfs_da_args *args,
1217 int mapindex)
1218 {
1219 struct xfs_attr_leafblock *leaf;
1220 struct xfs_attr_leaf_entry *entry;
1221 struct xfs_attr_leaf_name_local *name_loc;
1222 struct xfs_attr_leaf_name_remote *name_rmt;
1223 struct xfs_mount *mp;
1224 int tmp;
1225 int i;
1226
1227 trace_xfs_attr_leaf_add_work(args);
1228
1229 leaf = bp->b_addr;
1230 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1231 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1232
1233 /*
1234 * Force open some space in the entry array and fill it in.
1235 */
1236 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1237 if (args->index < ichdr->count) {
1238 tmp = ichdr->count - args->index;
1239 tmp *= sizeof(xfs_attr_leaf_entry_t);
1240 memmove(entry + 1, entry, tmp);
1241 xfs_trans_log_buf(args->trans, bp,
1242 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1243 }
1244 ichdr->count++;
1245
1246 /*
1247 * Allocate space for the new string (at the end of the run).
1248 */
1249 mp = args->trans->t_mountp;
1250 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1251 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1252 ASSERT(ichdr->freemap[mapindex].size >=
1253 xfs_attr_leaf_newentsize(args, NULL));
1254 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1255 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1256
1257 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1258
1259 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1260 ichdr->freemap[mapindex].size);
1261 entry->hashval = cpu_to_be32(args->hashval);
1262 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1263 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1264 if (args->op_flags & XFS_DA_OP_RENAME) {
1265 entry->flags |= XFS_ATTR_INCOMPLETE;
1266 if ((args->blkno2 == args->blkno) &&
1267 (args->index2 <= args->index)) {
1268 args->index2++;
1269 }
1270 }
1271 xfs_trans_log_buf(args->trans, bp,
1272 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1273 ASSERT((args->index == 0) ||
1274 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1275 ASSERT((args->index == ichdr->count - 1) ||
1276 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1277
1278 /*
1279 * For "remote" attribute values, simply note that we need to
1280 * allocate space for the "remote" value. We can't actually
1281 * allocate the extents in this transaction, and we can't decide
1282 * which blocks they should be as we might allocate more blocks
1283 * as part of this transaction (a split operation for example).
1284 */
1285 if (entry->flags & XFS_ATTR_LOCAL) {
1286 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1287 name_loc->namelen = args->namelen;
1288 name_loc->valuelen = cpu_to_be16(args->valuelen);
1289 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1290 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1291 be16_to_cpu(name_loc->valuelen));
1292 } else {
1293 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1294 name_rmt->namelen = args->namelen;
1295 memcpy((char *)name_rmt->name, args->name, args->namelen);
1296 entry->flags |= XFS_ATTR_INCOMPLETE;
1297 /* just in case */
1298 name_rmt->valuelen = 0;
1299 name_rmt->valueblk = 0;
1300 args->rmtblkno = 1;
1301 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1302 args->rmtvaluelen = args->valuelen;
1303 }
1304 xfs_trans_log_buf(args->trans, bp,
1305 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1306 xfs_attr_leaf_entsize(leaf, args->index)));
1307
1308 /*
1309 * Update the control info for this leaf node
1310 */
1311 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1312 ichdr->firstused = be16_to_cpu(entry->nameidx);
1313
1314 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1315 + xfs_attr3_leaf_hdr_size(leaf));
1316 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1317 + xfs_attr3_leaf_hdr_size(leaf);
1318
1319 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1320 if (ichdr->freemap[i].base == tmp) {
1321 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1322 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1323 }
1324 }
1325 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1326 return 0;
1327 }
1328
1329 /*
1330 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1331 */
1332 STATIC void
1333 xfs_attr3_leaf_compact(
1334 struct xfs_da_args *args,
1335 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1336 struct xfs_buf *bp)
1337 {
1338 struct xfs_attr_leafblock *leaf_src;
1339 struct xfs_attr_leafblock *leaf_dst;
1340 struct xfs_attr3_icleaf_hdr ichdr_src;
1341 struct xfs_trans *trans = args->trans;
1342 char *tmpbuffer;
1343
1344 trace_xfs_attr_leaf_compact(args);
1345
1346 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1347 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1348 memset(bp->b_addr, 0, args->geo->blksize);
1349 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1350 leaf_dst = bp->b_addr;
1351
1352 /*
1353 * Copy the on-disk header back into the destination buffer to ensure
1354 * all the information in the header that is not part of the incore
1355 * header structure is preserved.
1356 */
1357 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1358
1359 /* Initialise the incore headers */
1360 ichdr_src = *ichdr_dst; /* struct copy */
1361 ichdr_dst->firstused = args->geo->blksize;
1362 ichdr_dst->usedbytes = 0;
1363 ichdr_dst->count = 0;
1364 ichdr_dst->holes = 0;
1365 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1366 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1367 ichdr_dst->freemap[0].base;
1368
1369 /* write the header back to initialise the underlying buffer */
1370 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1371
1372 /*
1373 * Copy all entry's in the same (sorted) order,
1374 * but allocate name/value pairs packed and in sequence.
1375 */
1376 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1377 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1378 /*
1379 * this logs the entire buffer, but the caller must write the header
1380 * back to the buffer when it is finished modifying it.
1381 */
1382 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1383
1384 kmem_free(tmpbuffer);
1385 }
1386
1387 /*
1388 * Compare two leaf blocks "order".
1389 * Return 0 unless leaf2 should go before leaf1.
1390 */
1391 static int
1392 xfs_attr3_leaf_order(
1393 struct xfs_buf *leaf1_bp,
1394 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1395 struct xfs_buf *leaf2_bp,
1396 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1397 {
1398 struct xfs_attr_leaf_entry *entries1;
1399 struct xfs_attr_leaf_entry *entries2;
1400
1401 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1402 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1403 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1404 ((be32_to_cpu(entries2[0].hashval) <
1405 be32_to_cpu(entries1[0].hashval)) ||
1406 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1407 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1408 return 1;
1409 }
1410 return 0;
1411 }
1412
1413 int
1414 xfs_attr_leaf_order(
1415 struct xfs_buf *leaf1_bp,
1416 struct xfs_buf *leaf2_bp)
1417 {
1418 struct xfs_attr3_icleaf_hdr ichdr1;
1419 struct xfs_attr3_icleaf_hdr ichdr2;
1420 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1421
1422 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1423 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1424 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1425 }
1426
1427 /*
1428 * Redistribute the attribute list entries between two leaf nodes,
1429 * taking into account the size of the new entry.
1430 *
1431 * NOTE: if new block is empty, then it will get the upper half of the
1432 * old block. At present, all (one) callers pass in an empty second block.
1433 *
1434 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1435 * to match what it is doing in splitting the attribute leaf block. Those
1436 * values are used in "atomic rename" operations on attributes. Note that
1437 * the "new" and "old" values can end up in different blocks.
1438 */
1439 STATIC void
1440 xfs_attr3_leaf_rebalance(
1441 struct xfs_da_state *state,
1442 struct xfs_da_state_blk *blk1,
1443 struct xfs_da_state_blk *blk2)
1444 {
1445 struct xfs_da_args *args;
1446 struct xfs_attr_leafblock *leaf1;
1447 struct xfs_attr_leafblock *leaf2;
1448 struct xfs_attr3_icleaf_hdr ichdr1;
1449 struct xfs_attr3_icleaf_hdr ichdr2;
1450 struct xfs_attr_leaf_entry *entries1;
1451 struct xfs_attr_leaf_entry *entries2;
1452 int count;
1453 int totallen;
1454 int max;
1455 int space;
1456 int swap;
1457
1458 /*
1459 * Set up environment.
1460 */
1461 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1462 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1463 leaf1 = blk1->bp->b_addr;
1464 leaf2 = blk2->bp->b_addr;
1465 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1466 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1467 ASSERT(ichdr2.count == 0);
1468 args = state->args;
1469
1470 trace_xfs_attr_leaf_rebalance(args);
1471
1472 /*
1473 * Check ordering of blocks, reverse if it makes things simpler.
1474 *
1475 * NOTE: Given that all (current) callers pass in an empty
1476 * second block, this code should never set "swap".
1477 */
1478 swap = 0;
1479 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1480 struct xfs_da_state_blk *tmp_blk;
1481 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1482
1483 tmp_blk = blk1;
1484 blk1 = blk2;
1485 blk2 = tmp_blk;
1486
1487 /* struct copies to swap them rather than reconverting */
1488 tmp_ichdr = ichdr1;
1489 ichdr1 = ichdr2;
1490 ichdr2 = tmp_ichdr;
1491
1492 leaf1 = blk1->bp->b_addr;
1493 leaf2 = blk2->bp->b_addr;
1494 swap = 1;
1495 }
1496
1497 /*
1498 * Examine entries until we reduce the absolute difference in
1499 * byte usage between the two blocks to a minimum. Then get
1500 * the direction to copy and the number of elements to move.
1501 *
1502 * "inleaf" is true if the new entry should be inserted into blk1.
1503 * If "swap" is also true, then reverse the sense of "inleaf".
1504 */
1505 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1506 blk2, &ichdr2,
1507 &count, &totallen);
1508 if (swap)
1509 state->inleaf = !state->inleaf;
1510
1511 /*
1512 * Move any entries required from leaf to leaf:
1513 */
1514 if (count < ichdr1.count) {
1515 /*
1516 * Figure the total bytes to be added to the destination leaf.
1517 */
1518 /* number entries being moved */
1519 count = ichdr1.count - count;
1520 space = ichdr1.usedbytes - totallen;
1521 space += count * sizeof(xfs_attr_leaf_entry_t);
1522
1523 /*
1524 * leaf2 is the destination, compact it if it looks tight.
1525 */
1526 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1527 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1528 if (space > max)
1529 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1530
1531 /*
1532 * Move high entries from leaf1 to low end of leaf2.
1533 */
1534 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1535 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1536
1537 } else if (count > ichdr1.count) {
1538 /*
1539 * I assert that since all callers pass in an empty
1540 * second buffer, this code should never execute.
1541 */
1542 ASSERT(0);
1543
1544 /*
1545 * Figure the total bytes to be added to the destination leaf.
1546 */
1547 /* number entries being moved */
1548 count -= ichdr1.count;
1549 space = totallen - ichdr1.usedbytes;
1550 space += count * sizeof(xfs_attr_leaf_entry_t);
1551
1552 /*
1553 * leaf1 is the destination, compact it if it looks tight.
1554 */
1555 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1556 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1557 if (space > max)
1558 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1559
1560 /*
1561 * Move low entries from leaf2 to high end of leaf1.
1562 */
1563 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1564 ichdr1.count, count);
1565 }
1566
1567 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1568 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1569 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1570 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1571
1572 /*
1573 * Copy out last hashval in each block for B-tree code.
1574 */
1575 entries1 = xfs_attr3_leaf_entryp(leaf1);
1576 entries2 = xfs_attr3_leaf_entryp(leaf2);
1577 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1578 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1579
1580 /*
1581 * Adjust the expected index for insertion.
1582 * NOTE: this code depends on the (current) situation that the
1583 * second block was originally empty.
1584 *
1585 * If the insertion point moved to the 2nd block, we must adjust
1586 * the index. We must also track the entry just following the
1587 * new entry for use in an "atomic rename" operation, that entry
1588 * is always the "old" entry and the "new" entry is what we are
1589 * inserting. The index/blkno fields refer to the "old" entry,
1590 * while the index2/blkno2 fields refer to the "new" entry.
1591 */
1592 if (blk1->index > ichdr1.count) {
1593 ASSERT(state->inleaf == 0);
1594 blk2->index = blk1->index - ichdr1.count;
1595 args->index = args->index2 = blk2->index;
1596 args->blkno = args->blkno2 = blk2->blkno;
1597 } else if (blk1->index == ichdr1.count) {
1598 if (state->inleaf) {
1599 args->index = blk1->index;
1600 args->blkno = blk1->blkno;
1601 args->index2 = 0;
1602 args->blkno2 = blk2->blkno;
1603 } else {
1604 /*
1605 * On a double leaf split, the original attr location
1606 * is already stored in blkno2/index2, so don't
1607 * overwrite it overwise we corrupt the tree.
1608 */
1609 blk2->index = blk1->index - ichdr1.count;
1610 args->index = blk2->index;
1611 args->blkno = blk2->blkno;
1612 if (!state->extravalid) {
1613 /*
1614 * set the new attr location to match the old
1615 * one and let the higher level split code
1616 * decide where in the leaf to place it.
1617 */
1618 args->index2 = blk2->index;
1619 args->blkno2 = blk2->blkno;
1620 }
1621 }
1622 } else {
1623 ASSERT(state->inleaf == 1);
1624 args->index = args->index2 = blk1->index;
1625 args->blkno = args->blkno2 = blk1->blkno;
1626 }
1627 }
1628
1629 /*
1630 * Examine entries until we reduce the absolute difference in
1631 * byte usage between the two blocks to a minimum.
1632 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1633 * GROT: there will always be enough room in either block for a new entry.
1634 * GROT: Do a double-split for this case?
1635 */
1636 STATIC int
1637 xfs_attr3_leaf_figure_balance(
1638 struct xfs_da_state *state,
1639 struct xfs_da_state_blk *blk1,
1640 struct xfs_attr3_icleaf_hdr *ichdr1,
1641 struct xfs_da_state_blk *blk2,
1642 struct xfs_attr3_icleaf_hdr *ichdr2,
1643 int *countarg,
1644 int *usedbytesarg)
1645 {
1646 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1647 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1648 struct xfs_attr_leaf_entry *entry;
1649 int count;
1650 int max;
1651 int index;
1652 int totallen = 0;
1653 int half;
1654 int lastdelta;
1655 int foundit = 0;
1656 int tmp;
1657
1658 /*
1659 * Examine entries until we reduce the absolute difference in
1660 * byte usage between the two blocks to a minimum.
1661 */
1662 max = ichdr1->count + ichdr2->count;
1663 half = (max + 1) * sizeof(*entry);
1664 half += ichdr1->usedbytes + ichdr2->usedbytes +
1665 xfs_attr_leaf_newentsize(state->args, NULL);
1666 half /= 2;
1667 lastdelta = state->args->geo->blksize;
1668 entry = xfs_attr3_leaf_entryp(leaf1);
1669 for (count = index = 0; count < max; entry++, index++, count++) {
1670
1671 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1672 /*
1673 * The new entry is in the first block, account for it.
1674 */
1675 if (count == blk1->index) {
1676 tmp = totallen + sizeof(*entry) +
1677 xfs_attr_leaf_newentsize(state->args, NULL);
1678 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1679 break;
1680 lastdelta = XFS_ATTR_ABS(half - tmp);
1681 totallen = tmp;
1682 foundit = 1;
1683 }
1684
1685 /*
1686 * Wrap around into the second block if necessary.
1687 */
1688 if (count == ichdr1->count) {
1689 leaf1 = leaf2;
1690 entry = xfs_attr3_leaf_entryp(leaf1);
1691 index = 0;
1692 }
1693
1694 /*
1695 * Figure out if next leaf entry would be too much.
1696 */
1697 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1698 index);
1699 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1700 break;
1701 lastdelta = XFS_ATTR_ABS(half - tmp);
1702 totallen = tmp;
1703 #undef XFS_ATTR_ABS
1704 }
1705
1706 /*
1707 * Calculate the number of usedbytes that will end up in lower block.
1708 * If new entry not in lower block, fix up the count.
1709 */
1710 totallen -= count * sizeof(*entry);
1711 if (foundit) {
1712 totallen -= sizeof(*entry) +
1713 xfs_attr_leaf_newentsize(state->args, NULL);
1714 }
1715
1716 *countarg = count;
1717 *usedbytesarg = totallen;
1718 return foundit;
1719 }
1720
1721 /*========================================================================
1722 * Routines used for shrinking the Btree.
1723 *========================================================================*/
1724
1725 /*
1726 * Check a leaf block and its neighbors to see if the block should be
1727 * collapsed into one or the other neighbor. Always keep the block
1728 * with the smaller block number.
1729 * If the current block is over 50% full, don't try to join it, return 0.
1730 * If the block is empty, fill in the state structure and return 2.
1731 * If it can be collapsed, fill in the state structure and return 1.
1732 * If nothing can be done, return 0.
1733 *
1734 * GROT: allow for INCOMPLETE entries in calculation.
1735 */
1736 int
1737 xfs_attr3_leaf_toosmall(
1738 struct xfs_da_state *state,
1739 int *action)
1740 {
1741 struct xfs_attr_leafblock *leaf;
1742 struct xfs_da_state_blk *blk;
1743 struct xfs_attr3_icleaf_hdr ichdr;
1744 struct xfs_buf *bp;
1745 xfs_dablk_t blkno;
1746 int bytes;
1747 int forward;
1748 int error;
1749 int retval;
1750 int i;
1751
1752 trace_xfs_attr_leaf_toosmall(state->args);
1753
1754 /*
1755 * Check for the degenerate case of the block being over 50% full.
1756 * If so, it's not worth even looking to see if we might be able
1757 * to coalesce with a sibling.
1758 */
1759 blk = &state->path.blk[ state->path.active-1 ];
1760 leaf = blk->bp->b_addr;
1761 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1762 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1763 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1764 ichdr.usedbytes;
1765 if (bytes > (state->args->geo->blksize >> 1)) {
1766 *action = 0; /* blk over 50%, don't try to join */
1767 return 0;
1768 }
1769
1770 /*
1771 * Check for the degenerate case of the block being empty.
1772 * If the block is empty, we'll simply delete it, no need to
1773 * coalesce it with a sibling block. We choose (arbitrarily)
1774 * to merge with the forward block unless it is NULL.
1775 */
1776 if (ichdr.count == 0) {
1777 /*
1778 * Make altpath point to the block we want to keep and
1779 * path point to the block we want to drop (this one).
1780 */
1781 forward = (ichdr.forw != 0);
1782 memcpy(&state->altpath, &state->path, sizeof(state->path));
1783 error = xfs_da3_path_shift(state, &state->altpath, forward,
1784 0, &retval);
1785 if (error)
1786 return error;
1787 if (retval) {
1788 *action = 0;
1789 } else {
1790 *action = 2;
1791 }
1792 return 0;
1793 }
1794
1795 /*
1796 * Examine each sibling block to see if we can coalesce with
1797 * at least 25% free space to spare. We need to figure out
1798 * whether to merge with the forward or the backward block.
1799 * We prefer coalescing with the lower numbered sibling so as
1800 * to shrink an attribute list over time.
1801 */
1802 /* start with smaller blk num */
1803 forward = ichdr.forw < ichdr.back;
1804 for (i = 0; i < 2; forward = !forward, i++) {
1805 struct xfs_attr3_icleaf_hdr ichdr2;
1806 if (forward)
1807 blkno = ichdr.forw;
1808 else
1809 blkno = ichdr.back;
1810 if (blkno == 0)
1811 continue;
1812 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1813 blkno, -1, &bp);
1814 if (error)
1815 return error;
1816
1817 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1818
1819 bytes = state->args->geo->blksize -
1820 (state->args->geo->blksize >> 2) -
1821 ichdr.usedbytes - ichdr2.usedbytes -
1822 ((ichdr.count + ichdr2.count) *
1823 sizeof(xfs_attr_leaf_entry_t)) -
1824 xfs_attr3_leaf_hdr_size(leaf);
1825
1826 xfs_trans_brelse(state->args->trans, bp);
1827 if (bytes >= 0)
1828 break; /* fits with at least 25% to spare */
1829 }
1830 if (i >= 2) {
1831 *action = 0;
1832 return 0;
1833 }
1834
1835 /*
1836 * Make altpath point to the block we want to keep (the lower
1837 * numbered block) and path point to the block we want to drop.
1838 */
1839 memcpy(&state->altpath, &state->path, sizeof(state->path));
1840 if (blkno < blk->blkno) {
1841 error = xfs_da3_path_shift(state, &state->altpath, forward,
1842 0, &retval);
1843 } else {
1844 error = xfs_da3_path_shift(state, &state->path, forward,
1845 0, &retval);
1846 }
1847 if (error)
1848 return error;
1849 if (retval) {
1850 *action = 0;
1851 } else {
1852 *action = 1;
1853 }
1854 return 0;
1855 }
1856
1857 /*
1858 * Remove a name from the leaf attribute list structure.
1859 *
1860 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1861 * If two leaves are 37% full, when combined they will leave 25% free.
1862 */
1863 int
1864 xfs_attr3_leaf_remove(
1865 struct xfs_buf *bp,
1866 struct xfs_da_args *args)
1867 {
1868 struct xfs_attr_leafblock *leaf;
1869 struct xfs_attr3_icleaf_hdr ichdr;
1870 struct xfs_attr_leaf_entry *entry;
1871 int before;
1872 int after;
1873 int smallest;
1874 int entsize;
1875 int tablesize;
1876 int tmp;
1877 int i;
1878
1879 trace_xfs_attr_leaf_remove(args);
1880
1881 leaf = bp->b_addr;
1882 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1883
1884 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1885 ASSERT(args->index >= 0 && args->index < ichdr.count);
1886 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1887 xfs_attr3_leaf_hdr_size(leaf));
1888
1889 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1890
1891 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1892 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1893
1894 /*
1895 * Scan through free region table:
1896 * check for adjacency of free'd entry with an existing one,
1897 * find smallest free region in case we need to replace it,
1898 * adjust any map that borders the entry table,
1899 */
1900 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1901 + xfs_attr3_leaf_hdr_size(leaf);
1902 tmp = ichdr.freemap[0].size;
1903 before = after = -1;
1904 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1905 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1906 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1907 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1908 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1909 if (ichdr.freemap[i].base == tablesize) {
1910 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1911 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1912 }
1913
1914 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1915 be16_to_cpu(entry->nameidx)) {
1916 before = i;
1917 } else if (ichdr.freemap[i].base ==
1918 (be16_to_cpu(entry->nameidx) + entsize)) {
1919 after = i;
1920 } else if (ichdr.freemap[i].size < tmp) {
1921 tmp = ichdr.freemap[i].size;
1922 smallest = i;
1923 }
1924 }
1925
1926 /*
1927 * Coalesce adjacent freemap regions,
1928 * or replace the smallest region.
1929 */
1930 if ((before >= 0) || (after >= 0)) {
1931 if ((before >= 0) && (after >= 0)) {
1932 ichdr.freemap[before].size += entsize;
1933 ichdr.freemap[before].size += ichdr.freemap[after].size;
1934 ichdr.freemap[after].base = 0;
1935 ichdr.freemap[after].size = 0;
1936 } else if (before >= 0) {
1937 ichdr.freemap[before].size += entsize;
1938 } else {
1939 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1940 ichdr.freemap[after].size += entsize;
1941 }
1942 } else {
1943 /*
1944 * Replace smallest region (if it is smaller than free'd entry)
1945 */
1946 if (ichdr.freemap[smallest].size < entsize) {
1947 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1948 ichdr.freemap[smallest].size = entsize;
1949 }
1950 }
1951
1952 /*
1953 * Did we remove the first entry?
1954 */
1955 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1956 smallest = 1;
1957 else
1958 smallest = 0;
1959
1960 /*
1961 * Compress the remaining entries and zero out the removed stuff.
1962 */
1963 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1964 ichdr.usedbytes -= entsize;
1965 xfs_trans_log_buf(args->trans, bp,
1966 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1967 entsize));
1968
1969 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1970 memmove(entry, entry + 1, tmp);
1971 ichdr.count--;
1972 xfs_trans_log_buf(args->trans, bp,
1973 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1974
1975 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1976 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1977
1978 /*
1979 * If we removed the first entry, re-find the first used byte
1980 * in the name area. Note that if the entry was the "firstused",
1981 * then we don't have a "hole" in our block resulting from
1982 * removing the name.
1983 */
1984 if (smallest) {
1985 tmp = args->geo->blksize;
1986 entry = xfs_attr3_leaf_entryp(leaf);
1987 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1988 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1989 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1990
1991 if (be16_to_cpu(entry->nameidx) < tmp)
1992 tmp = be16_to_cpu(entry->nameidx);
1993 }
1994 ichdr.firstused = tmp;
1995 ASSERT(ichdr.firstused != 0);
1996 } else {
1997 ichdr.holes = 1; /* mark as needing compaction */
1998 }
1999 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2000 xfs_trans_log_buf(args->trans, bp,
2001 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2002 xfs_attr3_leaf_hdr_size(leaf)));
2003
2004 /*
2005 * Check if leaf is less than 50% full, caller may want to
2006 * "join" the leaf with a sibling if so.
2007 */
2008 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2009 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2010
2011 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2012 }
2013
2014 /*
2015 * Move all the attribute list entries from drop_leaf into save_leaf.
2016 */
2017 void
2018 xfs_attr3_leaf_unbalance(
2019 struct xfs_da_state *state,
2020 struct xfs_da_state_blk *drop_blk,
2021 struct xfs_da_state_blk *save_blk)
2022 {
2023 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2024 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2025 struct xfs_attr3_icleaf_hdr drophdr;
2026 struct xfs_attr3_icleaf_hdr savehdr;
2027 struct xfs_attr_leaf_entry *entry;
2028
2029 trace_xfs_attr_leaf_unbalance(state->args);
2030
2031 drop_leaf = drop_blk->bp->b_addr;
2032 save_leaf = save_blk->bp->b_addr;
2033 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2034 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2035 entry = xfs_attr3_leaf_entryp(drop_leaf);
2036
2037 /*
2038 * Save last hashval from dying block for later Btree fixup.
2039 */
2040 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2041
2042 /*
2043 * Check if we need a temp buffer, or can we do it in place.
2044 * Note that we don't check "leaf" for holes because we will
2045 * always be dropping it, toosmall() decided that for us already.
2046 */
2047 if (savehdr.holes == 0) {
2048 /*
2049 * dest leaf has no holes, so we add there. May need
2050 * to make some room in the entry array.
2051 */
2052 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2053 drop_blk->bp, &drophdr)) {
2054 xfs_attr3_leaf_moveents(state->args,
2055 drop_leaf, &drophdr, 0,
2056 save_leaf, &savehdr, 0,
2057 drophdr.count);
2058 } else {
2059 xfs_attr3_leaf_moveents(state->args,
2060 drop_leaf, &drophdr, 0,
2061 save_leaf, &savehdr,
2062 savehdr.count, drophdr.count);
2063 }
2064 } else {
2065 /*
2066 * Destination has holes, so we make a temporary copy
2067 * of the leaf and add them both to that.
2068 */
2069 struct xfs_attr_leafblock *tmp_leaf;
2070 struct xfs_attr3_icleaf_hdr tmphdr;
2071
2072 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2073
2074 /*
2075 * Copy the header into the temp leaf so that all the stuff
2076 * not in the incore header is present and gets copied back in
2077 * once we've moved all the entries.
2078 */
2079 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2080
2081 memset(&tmphdr, 0, sizeof(tmphdr));
2082 tmphdr.magic = savehdr.magic;
2083 tmphdr.forw = savehdr.forw;
2084 tmphdr.back = savehdr.back;
2085 tmphdr.firstused = state->args->geo->blksize;
2086
2087 /* write the header to the temp buffer to initialise it */
2088 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2089
2090 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2091 drop_blk->bp, &drophdr)) {
2092 xfs_attr3_leaf_moveents(state->args,
2093 drop_leaf, &drophdr, 0,
2094 tmp_leaf, &tmphdr, 0,
2095 drophdr.count);
2096 xfs_attr3_leaf_moveents(state->args,
2097 save_leaf, &savehdr, 0,
2098 tmp_leaf, &tmphdr, tmphdr.count,
2099 savehdr.count);
2100 } else {
2101 xfs_attr3_leaf_moveents(state->args,
2102 save_leaf, &savehdr, 0,
2103 tmp_leaf, &tmphdr, 0,
2104 savehdr.count);
2105 xfs_attr3_leaf_moveents(state->args,
2106 drop_leaf, &drophdr, 0,
2107 tmp_leaf, &tmphdr, tmphdr.count,
2108 drophdr.count);
2109 }
2110 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2111 savehdr = tmphdr; /* struct copy */
2112 kmem_free(tmp_leaf);
2113 }
2114
2115 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2116 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2117 state->args->geo->blksize - 1);
2118
2119 /*
2120 * Copy out last hashval in each block for B-tree code.
2121 */
2122 entry = xfs_attr3_leaf_entryp(save_leaf);
2123 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2124 }
2125
2126 /*========================================================================
2127 * Routines used for finding things in the Btree.
2128 *========================================================================*/
2129
2130 /*
2131 * Look up a name in a leaf attribute list structure.
2132 * This is the internal routine, it uses the caller's buffer.
2133 *
2134 * Note that duplicate keys are allowed, but only check within the
2135 * current leaf node. The Btree code must check in adjacent leaf nodes.
2136 *
2137 * Return in args->index the index into the entry[] array of either
2138 * the found entry, or where the entry should have been (insert before
2139 * that entry).
2140 *
2141 * Don't change the args->value unless we find the attribute.
2142 */
2143 int
2144 xfs_attr3_leaf_lookup_int(
2145 struct xfs_buf *bp,
2146 struct xfs_da_args *args)
2147 {
2148 struct xfs_attr_leafblock *leaf;
2149 struct xfs_attr3_icleaf_hdr ichdr;
2150 struct xfs_attr_leaf_entry *entry;
2151 struct xfs_attr_leaf_entry *entries;
2152 struct xfs_attr_leaf_name_local *name_loc;
2153 struct xfs_attr_leaf_name_remote *name_rmt;
2154 xfs_dahash_t hashval;
2155 int probe;
2156 int span;
2157
2158 trace_xfs_attr_leaf_lookup(args);
2159
2160 leaf = bp->b_addr;
2161 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2162 entries = xfs_attr3_leaf_entryp(leaf);
2163 ASSERT(ichdr.count < args->geo->blksize / 8);
2164
2165 /*
2166 * Binary search. (note: small blocks will skip this loop)
2167 */
2168 hashval = args->hashval;
2169 probe = span = ichdr.count / 2;
2170 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2171 span /= 2;
2172 if (be32_to_cpu(entry->hashval) < hashval)
2173 probe += span;
2174 else if (be32_to_cpu(entry->hashval) > hashval)
2175 probe -= span;
2176 else
2177 break;
2178 }
2179 ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2180 ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2181
2182 /*
2183 * Since we may have duplicate hashval's, find the first matching
2184 * hashval in the leaf.
2185 */
2186 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2187 entry--;
2188 probe--;
2189 }
2190 while (probe < ichdr.count &&
2191 be32_to_cpu(entry->hashval) < hashval) {
2192 entry++;
2193 probe++;
2194 }
2195 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2196 args->index = probe;
2197 return -ENOATTR;
2198 }
2199
2200 /*
2201 * Duplicate keys may be present, so search all of them for a match.
2202 */
2203 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2204 entry++, probe++) {
2205 /*
2206 * GROT: Add code to remove incomplete entries.
2207 */
2208 /*
2209 * If we are looking for INCOMPLETE entries, show only those.
2210 * If we are looking for complete entries, show only those.
2211 */
2212 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2213 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2214 continue;
2215 }
2216 if (entry->flags & XFS_ATTR_LOCAL) {
2217 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2218 if (name_loc->namelen != args->namelen)
2219 continue;
2220 if (memcmp(args->name, name_loc->nameval,
2221 args->namelen) != 0)
2222 continue;
2223 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2224 continue;
2225 args->index = probe;
2226 return -EEXIST;
2227 } else {
2228 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2229 if (name_rmt->namelen != args->namelen)
2230 continue;
2231 if (memcmp(args->name, name_rmt->name,
2232 args->namelen) != 0)
2233 continue;
2234 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2235 continue;
2236 args->index = probe;
2237 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2238 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2239 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2240 args->dp->i_mount,
2241 args->rmtvaluelen);
2242 return -EEXIST;
2243 }
2244 }
2245 args->index = probe;
2246 return -ENOATTR;
2247 }
2248
2249 /*
2250 * Get the value associated with an attribute name from a leaf attribute
2251 * list structure.
2252 */
2253 int
2254 xfs_attr3_leaf_getvalue(
2255 struct xfs_buf *bp,
2256 struct xfs_da_args *args)
2257 {
2258 struct xfs_attr_leafblock *leaf;
2259 struct xfs_attr3_icleaf_hdr ichdr;
2260 struct xfs_attr_leaf_entry *entry;
2261 struct xfs_attr_leaf_name_local *name_loc;
2262 struct xfs_attr_leaf_name_remote *name_rmt;
2263 int valuelen;
2264
2265 leaf = bp->b_addr;
2266 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2267 ASSERT(ichdr.count < args->geo->blksize / 8);
2268 ASSERT(args->index < ichdr.count);
2269
2270 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2271 if (entry->flags & XFS_ATTR_LOCAL) {
2272 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2273 ASSERT(name_loc->namelen == args->namelen);
2274 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2275 valuelen = be16_to_cpu(name_loc->valuelen);
2276 if (args->flags & ATTR_KERNOVAL) {
2277 args->valuelen = valuelen;
2278 return 0;
2279 }
2280 if (args->valuelen < valuelen) {
2281 args->valuelen = valuelen;
2282 return -ERANGE;
2283 }
2284 args->valuelen = valuelen;
2285 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2286 } else {
2287 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2288 ASSERT(name_rmt->namelen == args->namelen);
2289 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2290 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2291 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2292 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2293 args->rmtvaluelen);
2294 if (args->flags & ATTR_KERNOVAL) {
2295 args->valuelen = args->rmtvaluelen;
2296 return 0;
2297 }
2298 if (args->valuelen < args->rmtvaluelen) {
2299 args->valuelen = args->rmtvaluelen;
2300 return -ERANGE;
2301 }
2302 args->valuelen = args->rmtvaluelen;
2303 }
2304 return 0;
2305 }
2306
2307 /*========================================================================
2308 * Utility routines.
2309 *========================================================================*/
2310
2311 /*
2312 * Move the indicated entries from one leaf to another.
2313 * NOTE: this routine modifies both source and destination leaves.
2314 */
2315 /*ARGSUSED*/
2316 STATIC void
2317 xfs_attr3_leaf_moveents(
2318 struct xfs_da_args *args,
2319 struct xfs_attr_leafblock *leaf_s,
2320 struct xfs_attr3_icleaf_hdr *ichdr_s,
2321 int start_s,
2322 struct xfs_attr_leafblock *leaf_d,
2323 struct xfs_attr3_icleaf_hdr *ichdr_d,
2324 int start_d,
2325 int count)
2326 {
2327 struct xfs_attr_leaf_entry *entry_s;
2328 struct xfs_attr_leaf_entry *entry_d;
2329 int desti;
2330 int tmp;
2331 int i;
2332
2333 /*
2334 * Check for nothing to do.
2335 */
2336 if (count == 0)
2337 return;
2338
2339 /*
2340 * Set up environment.
2341 */
2342 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2343 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2344 ASSERT(ichdr_s->magic == ichdr_d->magic);
2345 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2346 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2347 + xfs_attr3_leaf_hdr_size(leaf_s));
2348 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2349 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2350 + xfs_attr3_leaf_hdr_size(leaf_d));
2351
2352 ASSERT(start_s < ichdr_s->count);
2353 ASSERT(start_d <= ichdr_d->count);
2354 ASSERT(count <= ichdr_s->count);
2355
2356
2357 /*
2358 * Move the entries in the destination leaf up to make a hole?
2359 */
2360 if (start_d < ichdr_d->count) {
2361 tmp = ichdr_d->count - start_d;
2362 tmp *= sizeof(xfs_attr_leaf_entry_t);
2363 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2364 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2365 memmove(entry_d, entry_s, tmp);
2366 }
2367
2368 /*
2369 * Copy all entry's in the same (sorted) order,
2370 * but allocate attribute info packed and in sequence.
2371 */
2372 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2373 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2374 desti = start_d;
2375 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2376 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2377 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2378 #ifdef GROT
2379 /*
2380 * Code to drop INCOMPLETE entries. Difficult to use as we
2381 * may also need to change the insertion index. Code turned
2382 * off for 6.2, should be revisited later.
2383 */
2384 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2385 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2386 ichdr_s->usedbytes -= tmp;
2387 ichdr_s->count -= 1;
2388 entry_d--; /* to compensate for ++ in loop hdr */
2389 desti--;
2390 if ((start_s + i) < offset)
2391 result++; /* insertion index adjustment */
2392 } else {
2393 #endif /* GROT */
2394 ichdr_d->firstused -= tmp;
2395 /* both on-disk, don't endian flip twice */
2396 entry_d->hashval = entry_s->hashval;
2397 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2398 entry_d->flags = entry_s->flags;
2399 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2400 <= args->geo->blksize);
2401 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2402 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2403 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2404 <= args->geo->blksize);
2405 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2406 ichdr_s->usedbytes -= tmp;
2407 ichdr_d->usedbytes += tmp;
2408 ichdr_s->count -= 1;
2409 ichdr_d->count += 1;
2410 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2411 + xfs_attr3_leaf_hdr_size(leaf_d);
2412 ASSERT(ichdr_d->firstused >= tmp);
2413 #ifdef GROT
2414 }
2415 #endif /* GROT */
2416 }
2417
2418 /*
2419 * Zero out the entries we just copied.
2420 */
2421 if (start_s == ichdr_s->count) {
2422 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2423 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2424 ASSERT(((char *)entry_s + tmp) <=
2425 ((char *)leaf_s + args->geo->blksize));
2426 memset(entry_s, 0, tmp);
2427 } else {
2428 /*
2429 * Move the remaining entries down to fill the hole,
2430 * then zero the entries at the top.
2431 */
2432 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2433 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2434 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2435 memmove(entry_d, entry_s, tmp);
2436
2437 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2438 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2439 ASSERT(((char *)entry_s + tmp) <=
2440 ((char *)leaf_s + args->geo->blksize));
2441 memset(entry_s, 0, tmp);
2442 }
2443
2444 /*
2445 * Fill in the freemap information
2446 */
2447 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2448 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2449 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2450 ichdr_d->freemap[1].base = 0;
2451 ichdr_d->freemap[2].base = 0;
2452 ichdr_d->freemap[1].size = 0;
2453 ichdr_d->freemap[2].size = 0;
2454 ichdr_s->holes = 1; /* leaf may not be compact */
2455 }
2456
2457 /*
2458 * Pick up the last hashvalue from a leaf block.
2459 */
2460 xfs_dahash_t
2461 xfs_attr_leaf_lasthash(
2462 struct xfs_buf *bp,
2463 int *count)
2464 {
2465 struct xfs_attr3_icleaf_hdr ichdr;
2466 struct xfs_attr_leaf_entry *entries;
2467 struct xfs_mount *mp = bp->b_target->bt_mount;
2468
2469 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2470 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2471 if (count)
2472 *count = ichdr.count;
2473 if (!ichdr.count)
2474 return 0;
2475 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2476 }
2477
2478 /*
2479 * Calculate the number of bytes used to store the indicated attribute
2480 * (whether local or remote only calculate bytes in this block).
2481 */
2482 STATIC int
2483 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2484 {
2485 struct xfs_attr_leaf_entry *entries;
2486 xfs_attr_leaf_name_local_t *name_loc;
2487 xfs_attr_leaf_name_remote_t *name_rmt;
2488 int size;
2489
2490 entries = xfs_attr3_leaf_entryp(leaf);
2491 if (entries[index].flags & XFS_ATTR_LOCAL) {
2492 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2493 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2494 be16_to_cpu(name_loc->valuelen));
2495 } else {
2496 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2497 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2498 }
2499 return size;
2500 }
2501
2502 /*
2503 * Calculate the number of bytes that would be required to store the new
2504 * attribute (whether local or remote only calculate bytes in this block).
2505 * This routine decides as a side effect whether the attribute will be
2506 * a "local" or a "remote" attribute.
2507 */
2508 int
2509 xfs_attr_leaf_newentsize(
2510 struct xfs_da_args *args,
2511 int *local)
2512 {
2513 int size;
2514
2515 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2516 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2517 if (local)
2518 *local = 1;
2519 return size;
2520 }
2521 if (local)
2522 *local = 0;
2523 return xfs_attr_leaf_entsize_remote(args->namelen);
2524 }
2525
2526
2527 /*========================================================================
2528 * Manage the INCOMPLETE flag in a leaf entry
2529 *========================================================================*/
2530
2531 /*
2532 * Clear the INCOMPLETE flag on an entry in a leaf block.
2533 */
2534 int
2535 xfs_attr3_leaf_clearflag(
2536 struct xfs_da_args *args)
2537 {
2538 struct xfs_attr_leafblock *leaf;
2539 struct xfs_attr_leaf_entry *entry;
2540 struct xfs_attr_leaf_name_remote *name_rmt;
2541 struct xfs_buf *bp;
2542 int error;
2543 #ifdef DEBUG
2544 struct xfs_attr3_icleaf_hdr ichdr;
2545 xfs_attr_leaf_name_local_t *name_loc;
2546 int namelen;
2547 char *name;
2548 #endif /* DEBUG */
2549
2550 trace_xfs_attr_leaf_clearflag(args);
2551 /*
2552 * Set up the operation.
2553 */
2554 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2555 if (error)
2556 return error;
2557
2558 leaf = bp->b_addr;
2559 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2560 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2561
2562 #ifdef DEBUG
2563 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2564 ASSERT(args->index < ichdr.count);
2565 ASSERT(args->index >= 0);
2566
2567 if (entry->flags & XFS_ATTR_LOCAL) {
2568 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2569 namelen = name_loc->namelen;
2570 name = (char *)name_loc->nameval;
2571 } else {
2572 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2573 namelen = name_rmt->namelen;
2574 name = (char *)name_rmt->name;
2575 }
2576 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2577 ASSERT(namelen == args->namelen);
2578 ASSERT(memcmp(name, args->name, namelen) == 0);
2579 #endif /* DEBUG */
2580
2581 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2582 xfs_trans_log_buf(args->trans, bp,
2583 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2584
2585 if (args->rmtblkno) {
2586 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2587 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2588 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2589 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2590 xfs_trans_log_buf(args->trans, bp,
2591 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2592 }
2593
2594 /*
2595 * Commit the flag value change and start the next trans in series.
2596 */
2597 return xfs_trans_roll(&args->trans, args->dp);
2598 }
2599
2600 /*
2601 * Set the INCOMPLETE flag on an entry in a leaf block.
2602 */
2603 int
2604 xfs_attr3_leaf_setflag(
2605 struct xfs_da_args *args)
2606 {
2607 struct xfs_attr_leafblock *leaf;
2608 struct xfs_attr_leaf_entry *entry;
2609 struct xfs_attr_leaf_name_remote *name_rmt;
2610 struct xfs_buf *bp;
2611 int error;
2612 #ifdef DEBUG
2613 struct xfs_attr3_icleaf_hdr ichdr;
2614 #endif
2615
2616 trace_xfs_attr_leaf_setflag(args);
2617
2618 /*
2619 * Set up the operation.
2620 */
2621 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2622 if (error)
2623 return error;
2624
2625 leaf = bp->b_addr;
2626 #ifdef DEBUG
2627 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2628 ASSERT(args->index < ichdr.count);
2629 ASSERT(args->index >= 0);
2630 #endif
2631 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2632
2633 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2634 entry->flags |= XFS_ATTR_INCOMPLETE;
2635 xfs_trans_log_buf(args->trans, bp,
2636 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2637 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2638 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2639 name_rmt->valueblk = 0;
2640 name_rmt->valuelen = 0;
2641 xfs_trans_log_buf(args->trans, bp,
2642 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2643 }
2644
2645 /*
2646 * Commit the flag value change and start the next trans in series.
2647 */
2648 return xfs_trans_roll(&args->trans, args->dp);
2649 }
2650
2651 /*
2652 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2653 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2654 * entry given by args->blkno2/index2.
2655 *
2656 * Note that they could be in different blocks, or in the same block.
2657 */
2658 int
2659 xfs_attr3_leaf_flipflags(
2660 struct xfs_da_args *args)
2661 {
2662 struct xfs_attr_leafblock *leaf1;
2663 struct xfs_attr_leafblock *leaf2;
2664 struct xfs_attr_leaf_entry *entry1;
2665 struct xfs_attr_leaf_entry *entry2;
2666 struct xfs_attr_leaf_name_remote *name_rmt;
2667 struct xfs_buf *bp1;
2668 struct xfs_buf *bp2;
2669 int error;
2670 #ifdef DEBUG
2671 struct xfs_attr3_icleaf_hdr ichdr1;
2672 struct xfs_attr3_icleaf_hdr ichdr2;
2673 xfs_attr_leaf_name_local_t *name_loc;
2674 int namelen1, namelen2;
2675 char *name1, *name2;
2676 #endif /* DEBUG */
2677
2678 trace_xfs_attr_leaf_flipflags(args);
2679
2680 /*
2681 * Read the block containing the "old" attr
2682 */
2683 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2684 if (error)
2685 return error;
2686
2687 /*
2688 * Read the block containing the "new" attr, if it is different
2689 */
2690 if (args->blkno2 != args->blkno) {
2691 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2692 -1, &bp2);
2693 if (error)
2694 return error;
2695 } else {
2696 bp2 = bp1;
2697 }
2698
2699 leaf1 = bp1->b_addr;
2700 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2701
2702 leaf2 = bp2->b_addr;
2703 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2704
2705 #ifdef DEBUG
2706 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2707 ASSERT(args->index < ichdr1.count);
2708 ASSERT(args->index >= 0);
2709
2710 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2711 ASSERT(args->index2 < ichdr2.count);
2712 ASSERT(args->index2 >= 0);
2713
2714 if (entry1->flags & XFS_ATTR_LOCAL) {
2715 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2716 namelen1 = name_loc->namelen;
2717 name1 = (char *)name_loc->nameval;
2718 } else {
2719 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2720 namelen1 = name_rmt->namelen;
2721 name1 = (char *)name_rmt->name;
2722 }
2723 if (entry2->flags & XFS_ATTR_LOCAL) {
2724 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2725 namelen2 = name_loc->namelen;
2726 name2 = (char *)name_loc->nameval;
2727 } else {
2728 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2729 namelen2 = name_rmt->namelen;
2730 name2 = (char *)name_rmt->name;
2731 }
2732 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2733 ASSERT(namelen1 == namelen2);
2734 ASSERT(memcmp(name1, name2, namelen1) == 0);
2735 #endif /* DEBUG */
2736
2737 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2738 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2739
2740 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2741 xfs_trans_log_buf(args->trans, bp1,
2742 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2743 if (args->rmtblkno) {
2744 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2745 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2746 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2747 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2748 xfs_trans_log_buf(args->trans, bp1,
2749 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2750 }
2751
2752 entry2->flags |= XFS_ATTR_INCOMPLETE;
2753 xfs_trans_log_buf(args->trans, bp2,
2754 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2755 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2756 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2757 name_rmt->valueblk = 0;
2758 name_rmt->valuelen = 0;
2759 xfs_trans_log_buf(args->trans, bp2,
2760 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2761 }
2762
2763 /*
2764 * Commit the flag value change and start the next trans in series.
2765 */
2766 error = xfs_trans_roll(&args->trans, args->dp);
2767
2768 return error;
2769 }