]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blob - libxfs/xfs_attr_leaf.c
xfs: use ->t_dfops for attr set/remove operations
[thirdparty/xfsprogs-dev.git] / libxfs / xfs_attr_leaf.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "libxfs_priv.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_sb.h"
15 #include "xfs_mount.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
19 #include "xfs_trans.h"
20 #include "xfs_bmap_btree.h"
21 #include "xfs_bmap.h"
22 #include "xfs_attr_sf.h"
23 #include "xfs_attr_remote.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_trace.h"
26 #include "xfs_cksum.h"
27 #include "xfs_dir2.h"
28
29
30 /*
31 * xfs_attr_leaf.c
32 *
33 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
34 */
35
36 /*========================================================================
37 * Function prototypes for the kernel.
38 *========================================================================*/
39
40 /*
41 * Routines used for growing the Btree.
42 */
43 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
44 xfs_dablk_t which_block, struct xfs_buf **bpp);
45 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
46 struct xfs_attr3_icleaf_hdr *ichdr,
47 struct xfs_da_args *args, int freemap_index);
48 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
49 struct xfs_attr3_icleaf_hdr *ichdr,
50 struct xfs_buf *leaf_buffer);
51 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
52 xfs_da_state_blk_t *blk1,
53 xfs_da_state_blk_t *blk2);
54 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
55 xfs_da_state_blk_t *leaf_blk_1,
56 struct xfs_attr3_icleaf_hdr *ichdr1,
57 xfs_da_state_blk_t *leaf_blk_2,
58 struct xfs_attr3_icleaf_hdr *ichdr2,
59 int *number_entries_in_blk1,
60 int *number_usedbytes_in_blk1);
61
62 /*
63 * Utility routines.
64 */
65 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
66 struct xfs_attr_leafblock *src_leaf,
67 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
68 struct xfs_attr_leafblock *dst_leaf,
69 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
70 int move_count);
71 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
72
73 /*
74 * attr3 block 'firstused' conversion helpers.
75 *
76 * firstused refers to the offset of the first used byte of the nameval region
77 * of an attr leaf block. The region starts at the tail of the block and expands
78 * backwards towards the middle. As such, firstused is initialized to the block
79 * size for an empty leaf block and is reduced from there.
80 *
81 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
82 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
83 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
84 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
85 * the attr block size. The following helpers manage the conversion between the
86 * in-core and on-disk formats.
87 */
88
89 static void
90 xfs_attr3_leaf_firstused_from_disk(
91 struct xfs_da_geometry *geo,
92 struct xfs_attr3_icleaf_hdr *to,
93 struct xfs_attr_leafblock *from)
94 {
95 struct xfs_attr3_leaf_hdr *hdr3;
96
97 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
98 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
99 to->firstused = be16_to_cpu(hdr3->firstused);
100 } else {
101 to->firstused = be16_to_cpu(from->hdr.firstused);
102 }
103
104 /*
105 * Convert from the magic fsb size value to actual blocksize. This
106 * should only occur for empty blocks when the block size overflows
107 * 16-bits.
108 */
109 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
110 ASSERT(!to->count && !to->usedbytes);
111 ASSERT(geo->blksize > USHRT_MAX);
112 to->firstused = geo->blksize;
113 }
114 }
115
116 static void
117 xfs_attr3_leaf_firstused_to_disk(
118 struct xfs_da_geometry *geo,
119 struct xfs_attr_leafblock *to,
120 struct xfs_attr3_icleaf_hdr *from)
121 {
122 struct xfs_attr3_leaf_hdr *hdr3;
123 uint32_t firstused;
124
125 /* magic value should only be seen on disk */
126 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
127
128 /*
129 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
130 * value. This only overflows at the max supported value of 64k. Use the
131 * magic on-disk value to represent block size in this case.
132 */
133 firstused = from->firstused;
134 if (firstused > USHRT_MAX) {
135 ASSERT(from->firstused == geo->blksize);
136 firstused = XFS_ATTR3_LEAF_NULLOFF;
137 }
138
139 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
140 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
141 hdr3->firstused = cpu_to_be16(firstused);
142 } else {
143 to->hdr.firstused = cpu_to_be16(firstused);
144 }
145 }
146
147 void
148 xfs_attr3_leaf_hdr_from_disk(
149 struct xfs_da_geometry *geo,
150 struct xfs_attr3_icleaf_hdr *to,
151 struct xfs_attr_leafblock *from)
152 {
153 int i;
154
155 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
156 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
157
158 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
159 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
160
161 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
162 to->back = be32_to_cpu(hdr3->info.hdr.back);
163 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
164 to->count = be16_to_cpu(hdr3->count);
165 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
166 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
167 to->holes = hdr3->holes;
168
169 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
170 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
171 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
172 }
173 return;
174 }
175 to->forw = be32_to_cpu(from->hdr.info.forw);
176 to->back = be32_to_cpu(from->hdr.info.back);
177 to->magic = be16_to_cpu(from->hdr.info.magic);
178 to->count = be16_to_cpu(from->hdr.count);
179 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
180 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
181 to->holes = from->hdr.holes;
182
183 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
184 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
185 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
186 }
187 }
188
189 void
190 xfs_attr3_leaf_hdr_to_disk(
191 struct xfs_da_geometry *geo,
192 struct xfs_attr_leafblock *to,
193 struct xfs_attr3_icleaf_hdr *from)
194 {
195 int i;
196
197 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
198 from->magic == XFS_ATTR3_LEAF_MAGIC);
199
200 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
201 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
202
203 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
204 hdr3->info.hdr.back = cpu_to_be32(from->back);
205 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
206 hdr3->count = cpu_to_be16(from->count);
207 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
208 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
209 hdr3->holes = from->holes;
210 hdr3->pad1 = 0;
211
212 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
213 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
214 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
215 }
216 return;
217 }
218 to->hdr.info.forw = cpu_to_be32(from->forw);
219 to->hdr.info.back = cpu_to_be32(from->back);
220 to->hdr.info.magic = cpu_to_be16(from->magic);
221 to->hdr.count = cpu_to_be16(from->count);
222 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
223 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
224 to->hdr.holes = from->holes;
225 to->hdr.pad1 = 0;
226
227 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
228 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
229 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
230 }
231 }
232
233 static xfs_failaddr_t
234 xfs_attr3_leaf_verify(
235 struct xfs_buf *bp)
236 {
237 struct xfs_attr3_icleaf_hdr ichdr;
238 struct xfs_mount *mp = bp->b_target->bt_mount;
239 struct xfs_attr_leafblock *leaf = bp->b_addr;
240 struct xfs_perag *pag = bp->b_pag;
241 struct xfs_attr_leaf_entry *entries;
242
243 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
244
245 if (xfs_sb_version_hascrc(&mp->m_sb)) {
246 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
247
248 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
249 return __this_address;
250
251 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
252 return __this_address;
253 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
254 return __this_address;
255 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
256 return __this_address;
257 } else {
258 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
259 return __this_address;
260 }
261 /*
262 * In recovery there is a transient state where count == 0 is valid
263 * because we may have transitioned an empty shortform attr to a leaf
264 * if the attr didn't fit in shortform.
265 */
266 if (pag && pag->pagf_init && ichdr.count == 0)
267 return __this_address;
268
269 /*
270 * firstused is the block offset of the first name info structure.
271 * Make sure it doesn't go off the block or crash into the header.
272 */
273 if (ichdr.firstused > mp->m_attr_geo->blksize)
274 return __this_address;
275 if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
276 return __this_address;
277
278 /* Make sure the entries array doesn't crash into the name info. */
279 entries = xfs_attr3_leaf_entryp(bp->b_addr);
280 if ((char *)&entries[ichdr.count] >
281 (char *)bp->b_addr + ichdr.firstused)
282 return __this_address;
283
284 /* XXX: need to range check rest of attr header values */
285 /* XXX: hash order check? */
286
287 return NULL;
288 }
289
290 static void
291 xfs_attr3_leaf_write_verify(
292 struct xfs_buf *bp)
293 {
294 struct xfs_mount *mp = bp->b_target->bt_mount;
295 struct xfs_buf_log_item *bip = bp->b_log_item;
296 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
297 xfs_failaddr_t fa;
298
299 fa = xfs_attr3_leaf_verify(bp);
300 if (fa) {
301 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
302 return;
303 }
304
305 if (!xfs_sb_version_hascrc(&mp->m_sb))
306 return;
307
308 if (bip)
309 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
310
311 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
312 }
313
314 /*
315 * leaf/node format detection on trees is sketchy, so a node read can be done on
316 * leaf level blocks when detection identifies the tree as a node format tree
317 * incorrectly. In this case, we need to swap the verifier to match the correct
318 * format of the block being read.
319 */
320 static void
321 xfs_attr3_leaf_read_verify(
322 struct xfs_buf *bp)
323 {
324 struct xfs_mount *mp = bp->b_target->bt_mount;
325 xfs_failaddr_t fa;
326
327 if (xfs_sb_version_hascrc(&mp->m_sb) &&
328 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
329 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
330 else {
331 fa = xfs_attr3_leaf_verify(bp);
332 if (fa)
333 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
334 }
335 }
336
337 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
338 .name = "xfs_attr3_leaf",
339 .verify_read = xfs_attr3_leaf_read_verify,
340 .verify_write = xfs_attr3_leaf_write_verify,
341 .verify_struct = xfs_attr3_leaf_verify,
342 };
343
344 int
345 xfs_attr3_leaf_read(
346 struct xfs_trans *tp,
347 struct xfs_inode *dp,
348 xfs_dablk_t bno,
349 xfs_daddr_t mappedbno,
350 struct xfs_buf **bpp)
351 {
352 int err;
353
354 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
355 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
356 if (!err && tp && *bpp)
357 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
358 return err;
359 }
360
361 /*========================================================================
362 * Namespace helper routines
363 *========================================================================*/
364
365 /*
366 * If namespace bits don't match return 0.
367 * If all match then return 1.
368 */
369 STATIC int
370 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
371 {
372 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
373 }
374
375
376 /*========================================================================
377 * External routines when attribute fork size < XFS_LITINO(mp).
378 *========================================================================*/
379
380 /*
381 * Query whether the requested number of additional bytes of extended
382 * attribute space will be able to fit inline.
383 *
384 * Returns zero if not, else the di_forkoff fork offset to be used in the
385 * literal area for attribute data once the new bytes have been added.
386 *
387 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
388 * special case for dev/uuid inodes, they have fixed size data forks.
389 */
390 int
391 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
392 {
393 int offset;
394 int minforkoff; /* lower limit on valid forkoff locations */
395 int maxforkoff; /* upper limit on valid forkoff locations */
396 int dsize;
397 xfs_mount_t *mp = dp->i_mount;
398
399 /* rounded down */
400 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
401
402 if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
403 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
404 return (offset >= minforkoff) ? minforkoff : 0;
405 }
406
407 /*
408 * If the requested numbers of bytes is smaller or equal to the
409 * current attribute fork size we can always proceed.
410 *
411 * Note that if_bytes in the data fork might actually be larger than
412 * the current data fork size is due to delalloc extents. In that
413 * case either the extent count will go down when they are converted
414 * to real extents, or the delalloc conversion will take care of the
415 * literal area rebalancing.
416 */
417 if (bytes <= XFS_IFORK_ASIZE(dp))
418 return dp->i_d.di_forkoff;
419
420 /*
421 * For attr2 we can try to move the forkoff if there is space in the
422 * literal area, but for the old format we are done if there is no
423 * space in the fixed attribute fork.
424 */
425 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
426 return 0;
427
428 dsize = dp->i_df.if_bytes;
429
430 switch (dp->i_d.di_format) {
431 case XFS_DINODE_FMT_EXTENTS:
432 /*
433 * If there is no attr fork and the data fork is extents,
434 * determine if creating the default attr fork will result
435 * in the extents form migrating to btree. If so, the
436 * minimum offset only needs to be the space required for
437 * the btree root.
438 */
439 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
440 xfs_default_attroffset(dp))
441 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
442 break;
443 case XFS_DINODE_FMT_BTREE:
444 /*
445 * If we have a data btree then keep forkoff if we have one,
446 * otherwise we are adding a new attr, so then we set
447 * minforkoff to where the btree root can finish so we have
448 * plenty of room for attrs
449 */
450 if (dp->i_d.di_forkoff) {
451 if (offset < dp->i_d.di_forkoff)
452 return 0;
453 return dp->i_d.di_forkoff;
454 }
455 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
456 break;
457 }
458
459 /*
460 * A data fork btree root must have space for at least
461 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
462 */
463 minforkoff = max(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
464 minforkoff = roundup(minforkoff, 8) >> 3;
465
466 /* attr fork btree root can have at least this many key/ptr pairs */
467 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
468 XFS_BMDR_SPACE_CALC(MINABTPTRS);
469 maxforkoff = maxforkoff >> 3; /* rounded down */
470
471 if (offset >= maxforkoff)
472 return maxforkoff;
473 if (offset >= minforkoff)
474 return offset;
475 return 0;
476 }
477
478 /*
479 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
480 */
481 STATIC void
482 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
483 {
484 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
485 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
486 spin_lock(&mp->m_sb_lock);
487 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
488 xfs_sb_version_addattr2(&mp->m_sb);
489 spin_unlock(&mp->m_sb_lock);
490 xfs_log_sb(tp);
491 } else
492 spin_unlock(&mp->m_sb_lock);
493 }
494 }
495
496 /*
497 * Create the initial contents of a shortform attribute list.
498 */
499 void
500 xfs_attr_shortform_create(xfs_da_args_t *args)
501 {
502 xfs_attr_sf_hdr_t *hdr;
503 xfs_inode_t *dp;
504 xfs_ifork_t *ifp;
505
506 trace_xfs_attr_sf_create(args);
507
508 dp = args->dp;
509 ASSERT(dp != NULL);
510 ifp = dp->i_afp;
511 ASSERT(ifp != NULL);
512 ASSERT(ifp->if_bytes == 0);
513 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
514 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
515 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
516 ifp->if_flags |= XFS_IFINLINE;
517 } else {
518 ASSERT(ifp->if_flags & XFS_IFINLINE);
519 }
520 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
521 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
522 hdr->count = 0;
523 hdr->totsize = cpu_to_be16(sizeof(*hdr));
524 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
525 }
526
527 /*
528 * Add a name/value pair to the shortform attribute list.
529 * Overflow from the inode has already been checked for.
530 */
531 void
532 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
533 {
534 xfs_attr_shortform_t *sf;
535 xfs_attr_sf_entry_t *sfe;
536 int i, offset, size;
537 xfs_mount_t *mp;
538 xfs_inode_t *dp;
539 xfs_ifork_t *ifp;
540
541 trace_xfs_attr_sf_add(args);
542
543 dp = args->dp;
544 mp = dp->i_mount;
545 dp->i_d.di_forkoff = forkoff;
546
547 ifp = dp->i_afp;
548 ASSERT(ifp->if_flags & XFS_IFINLINE);
549 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
550 sfe = &sf->list[0];
551 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
552 #ifdef DEBUG
553 if (sfe->namelen != args->namelen)
554 continue;
555 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
556 continue;
557 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
558 continue;
559 ASSERT(0);
560 #endif
561 }
562
563 offset = (char *)sfe - (char *)sf;
564 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
565 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
566 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
567 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
568
569 sfe->namelen = args->namelen;
570 sfe->valuelen = args->valuelen;
571 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
572 memcpy(sfe->nameval, args->name, args->namelen);
573 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
574 sf->hdr.count++;
575 be16_add_cpu(&sf->hdr.totsize, size);
576 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
577
578 xfs_sbversion_add_attr2(mp, args->trans);
579 }
580
581 /*
582 * After the last attribute is removed revert to original inode format,
583 * making all literal area available to the data fork once more.
584 */
585 void
586 xfs_attr_fork_remove(
587 struct xfs_inode *ip,
588 struct xfs_trans *tp)
589 {
590 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
591 ip->i_d.di_forkoff = 0;
592 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
593
594 ASSERT(ip->i_d.di_anextents == 0);
595 ASSERT(ip->i_afp == NULL);
596
597 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
598 }
599
600 /*
601 * Remove an attribute from the shortform attribute list structure.
602 */
603 int
604 xfs_attr_shortform_remove(xfs_da_args_t *args)
605 {
606 xfs_attr_shortform_t *sf;
607 xfs_attr_sf_entry_t *sfe;
608 int base, size=0, end, totsize, i;
609 xfs_mount_t *mp;
610 xfs_inode_t *dp;
611
612 trace_xfs_attr_sf_remove(args);
613
614 dp = args->dp;
615 mp = dp->i_mount;
616 base = sizeof(xfs_attr_sf_hdr_t);
617 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
618 sfe = &sf->list[0];
619 end = sf->hdr.count;
620 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
621 base += size, i++) {
622 size = XFS_ATTR_SF_ENTSIZE(sfe);
623 if (sfe->namelen != args->namelen)
624 continue;
625 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
626 continue;
627 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
628 continue;
629 break;
630 }
631 if (i == end)
632 return -ENOATTR;
633
634 /*
635 * Fix up the attribute fork data, covering the hole
636 */
637 end = base + size;
638 totsize = be16_to_cpu(sf->hdr.totsize);
639 if (end != totsize)
640 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
641 sf->hdr.count--;
642 be16_add_cpu(&sf->hdr.totsize, -size);
643
644 /*
645 * Fix up the start offset of the attribute fork
646 */
647 totsize -= size;
648 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
649 (mp->m_flags & XFS_MOUNT_ATTR2) &&
650 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
651 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
652 xfs_attr_fork_remove(dp, args->trans);
653 } else {
654 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
655 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
656 ASSERT(dp->i_d.di_forkoff);
657 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
658 (args->op_flags & XFS_DA_OP_ADDNAME) ||
659 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
660 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
661 xfs_trans_log_inode(args->trans, dp,
662 XFS_ILOG_CORE | XFS_ILOG_ADATA);
663 }
664
665 xfs_sbversion_add_attr2(mp, args->trans);
666
667 return 0;
668 }
669
670 /*
671 * Look up a name in a shortform attribute list structure.
672 */
673 /*ARGSUSED*/
674 int
675 xfs_attr_shortform_lookup(xfs_da_args_t *args)
676 {
677 xfs_attr_shortform_t *sf;
678 xfs_attr_sf_entry_t *sfe;
679 int i;
680 xfs_ifork_t *ifp;
681
682 trace_xfs_attr_sf_lookup(args);
683
684 ifp = args->dp->i_afp;
685 ASSERT(ifp->if_flags & XFS_IFINLINE);
686 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
687 sfe = &sf->list[0];
688 for (i = 0; i < sf->hdr.count;
689 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
690 if (sfe->namelen != args->namelen)
691 continue;
692 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
693 continue;
694 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
695 continue;
696 return -EEXIST;
697 }
698 return -ENOATTR;
699 }
700
701 /*
702 * Look up a name in a shortform attribute list structure.
703 */
704 /*ARGSUSED*/
705 int
706 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
707 {
708 xfs_attr_shortform_t *sf;
709 xfs_attr_sf_entry_t *sfe;
710 int i;
711
712 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
713 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
714 sfe = &sf->list[0];
715 for (i = 0; i < sf->hdr.count;
716 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
717 if (sfe->namelen != args->namelen)
718 continue;
719 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
720 continue;
721 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
722 continue;
723 if (args->flags & ATTR_KERNOVAL) {
724 args->valuelen = sfe->valuelen;
725 return -EEXIST;
726 }
727 if (args->valuelen < sfe->valuelen) {
728 args->valuelen = sfe->valuelen;
729 return -ERANGE;
730 }
731 args->valuelen = sfe->valuelen;
732 memcpy(args->value, &sfe->nameval[args->namelen],
733 args->valuelen);
734 return -EEXIST;
735 }
736 return -ENOATTR;
737 }
738
739 /*
740 * Convert from using the shortform to the leaf. On success, return the
741 * buffer so that we can keep it locked until we're totally done with it.
742 */
743 int
744 xfs_attr_shortform_to_leaf(
745 struct xfs_da_args *args,
746 struct xfs_buf **leaf_bp)
747 {
748 xfs_inode_t *dp;
749 xfs_attr_shortform_t *sf;
750 xfs_attr_sf_entry_t *sfe;
751 xfs_da_args_t nargs;
752 char *tmpbuffer;
753 int error, i, size;
754 xfs_dablk_t blkno;
755 struct xfs_buf *bp;
756 xfs_ifork_t *ifp;
757
758 trace_xfs_attr_sf_to_leaf(args);
759
760 dp = args->dp;
761 ifp = dp->i_afp;
762 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
763 size = be16_to_cpu(sf->hdr.totsize);
764 tmpbuffer = kmem_alloc(size, KM_SLEEP);
765 ASSERT(tmpbuffer != NULL);
766 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
767 sf = (xfs_attr_shortform_t *)tmpbuffer;
768
769 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
770 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
771
772 bp = NULL;
773 error = xfs_da_grow_inode(args, &blkno);
774 if (error) {
775 /*
776 * If we hit an IO error middle of the transaction inside
777 * grow_inode(), we may have inconsistent data. Bail out.
778 */
779 if (error == -EIO)
780 goto out;
781 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
782 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
783 goto out;
784 }
785
786 ASSERT(blkno == 0);
787 error = xfs_attr3_leaf_create(args, blkno, &bp);
788 if (error) {
789 /* xfs_attr3_leaf_create may not have instantiated a block */
790 if (bp && (xfs_da_shrink_inode(args, 0, bp) != 0))
791 goto out;
792 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
793 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
794 goto out;
795 }
796
797 memset((char *)&nargs, 0, sizeof(nargs));
798 nargs.dp = dp;
799 nargs.geo = args->geo;
800 nargs.firstblock = args->firstblock;
801 nargs.dfops = args->dfops;
802 nargs.total = args->total;
803 nargs.whichfork = XFS_ATTR_FORK;
804 nargs.trans = args->trans;
805 nargs.op_flags = XFS_DA_OP_OKNOENT;
806
807 sfe = &sf->list[0];
808 for (i = 0; i < sf->hdr.count; i++) {
809 nargs.name = sfe->nameval;
810 nargs.namelen = sfe->namelen;
811 nargs.value = &sfe->nameval[nargs.namelen];
812 nargs.valuelen = sfe->valuelen;
813 nargs.hashval = xfs_da_hashname(sfe->nameval,
814 sfe->namelen);
815 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
816 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
817 ASSERT(error == -ENOATTR);
818 error = xfs_attr3_leaf_add(bp, &nargs);
819 ASSERT(error != -ENOSPC);
820 if (error)
821 goto out;
822 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
823 }
824 error = 0;
825 *leaf_bp = bp;
826 out:
827 kmem_free(tmpbuffer);
828 return error;
829 }
830
831 /*
832 * Check a leaf attribute block to see if all the entries would fit into
833 * a shortform attribute list.
834 */
835 int
836 xfs_attr_shortform_allfit(
837 struct xfs_buf *bp,
838 struct xfs_inode *dp)
839 {
840 struct xfs_attr_leafblock *leaf;
841 struct xfs_attr_leaf_entry *entry;
842 xfs_attr_leaf_name_local_t *name_loc;
843 struct xfs_attr3_icleaf_hdr leafhdr;
844 int bytes;
845 int i;
846 struct xfs_mount *mp = bp->b_target->bt_mount;
847
848 leaf = bp->b_addr;
849 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
850 entry = xfs_attr3_leaf_entryp(leaf);
851
852 bytes = sizeof(struct xfs_attr_sf_hdr);
853 for (i = 0; i < leafhdr.count; entry++, i++) {
854 if (entry->flags & XFS_ATTR_INCOMPLETE)
855 continue; /* don't copy partial entries */
856 if (!(entry->flags & XFS_ATTR_LOCAL))
857 return 0;
858 name_loc = xfs_attr3_leaf_name_local(leaf, i);
859 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
860 return 0;
861 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
862 return 0;
863 bytes += sizeof(struct xfs_attr_sf_entry) - 1
864 + name_loc->namelen
865 + be16_to_cpu(name_loc->valuelen);
866 }
867 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
868 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
869 (bytes == sizeof(struct xfs_attr_sf_hdr)))
870 return -1;
871 return xfs_attr_shortform_bytesfit(dp, bytes);
872 }
873
874 /* Verify the consistency of an inline attribute fork. */
875 xfs_failaddr_t
876 xfs_attr_shortform_verify(
877 struct xfs_inode *ip)
878 {
879 struct xfs_attr_shortform *sfp;
880 struct xfs_attr_sf_entry *sfep;
881 struct xfs_attr_sf_entry *next_sfep;
882 char *endp;
883 struct xfs_ifork *ifp;
884 int i;
885 int size;
886
887 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL);
888 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
889 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
890 size = ifp->if_bytes;
891
892 /*
893 * Give up if the attribute is way too short.
894 */
895 if (size < sizeof(struct xfs_attr_sf_hdr))
896 return __this_address;
897
898 endp = (char *)sfp + size;
899
900 /* Check all reported entries */
901 sfep = &sfp->list[0];
902 for (i = 0; i < sfp->hdr.count; i++) {
903 /*
904 * struct xfs_attr_sf_entry has a variable length.
905 * Check the fixed-offset parts of the structure are
906 * within the data buffer.
907 */
908 if (((char *)sfep + sizeof(*sfep)) >= endp)
909 return __this_address;
910
911 /* Don't allow names with known bad length. */
912 if (sfep->namelen == 0)
913 return __this_address;
914
915 /*
916 * Check that the variable-length part of the structure is
917 * within the data buffer. The next entry starts after the
918 * name component, so nextentry is an acceptable test.
919 */
920 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
921 if ((char *)next_sfep > endp)
922 return __this_address;
923
924 /*
925 * Check for unknown flags. Short form doesn't support
926 * the incomplete or local bits, so we can use the namespace
927 * mask here.
928 */
929 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
930 return __this_address;
931
932 /*
933 * Check for invalid namespace combinations. We only allow
934 * one namespace flag per xattr, so we can just count the
935 * bits (i.e. hweight) here.
936 */
937 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
938 return __this_address;
939
940 sfep = next_sfep;
941 }
942 if ((void *)sfep != (void *)endp)
943 return __this_address;
944
945 return NULL;
946 }
947
948 /*
949 * Convert a leaf attribute list to shortform attribute list
950 */
951 int
952 xfs_attr3_leaf_to_shortform(
953 struct xfs_buf *bp,
954 struct xfs_da_args *args,
955 int forkoff)
956 {
957 struct xfs_attr_leafblock *leaf;
958 struct xfs_attr3_icleaf_hdr ichdr;
959 struct xfs_attr_leaf_entry *entry;
960 struct xfs_attr_leaf_name_local *name_loc;
961 struct xfs_da_args nargs;
962 struct xfs_inode *dp = args->dp;
963 char *tmpbuffer;
964 int error;
965 int i;
966
967 trace_xfs_attr_leaf_to_sf(args);
968
969 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
970 if (!tmpbuffer)
971 return -ENOMEM;
972
973 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
974
975 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
976 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
977 entry = xfs_attr3_leaf_entryp(leaf);
978
979 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
980 memset(bp->b_addr, 0, args->geo->blksize);
981
982 /*
983 * Clean out the prior contents of the attribute list.
984 */
985 error = xfs_da_shrink_inode(args, 0, bp);
986 if (error)
987 goto out;
988
989 if (forkoff == -1) {
990 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
991 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
992 xfs_attr_fork_remove(dp, args->trans);
993 goto out;
994 }
995
996 xfs_attr_shortform_create(args);
997
998 /*
999 * Copy the attributes
1000 */
1001 memset((char *)&nargs, 0, sizeof(nargs));
1002 nargs.geo = args->geo;
1003 nargs.dp = dp;
1004 nargs.firstblock = args->firstblock;
1005 nargs.dfops = args->dfops;
1006 nargs.total = args->total;
1007 nargs.whichfork = XFS_ATTR_FORK;
1008 nargs.trans = args->trans;
1009 nargs.op_flags = XFS_DA_OP_OKNOENT;
1010
1011 for (i = 0; i < ichdr.count; entry++, i++) {
1012 if (entry->flags & XFS_ATTR_INCOMPLETE)
1013 continue; /* don't copy partial entries */
1014 if (!entry->nameidx)
1015 continue;
1016 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1017 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1018 nargs.name = name_loc->nameval;
1019 nargs.namelen = name_loc->namelen;
1020 nargs.value = &name_loc->nameval[nargs.namelen];
1021 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1022 nargs.hashval = be32_to_cpu(entry->hashval);
1023 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1024 xfs_attr_shortform_add(&nargs, forkoff);
1025 }
1026 error = 0;
1027
1028 out:
1029 kmem_free(tmpbuffer);
1030 return error;
1031 }
1032
1033 /*
1034 * Convert from using a single leaf to a root node and a leaf.
1035 */
1036 int
1037 xfs_attr3_leaf_to_node(
1038 struct xfs_da_args *args)
1039 {
1040 struct xfs_attr_leafblock *leaf;
1041 struct xfs_attr3_icleaf_hdr icleafhdr;
1042 struct xfs_attr_leaf_entry *entries;
1043 struct xfs_da_node_entry *btree;
1044 struct xfs_da3_icnode_hdr icnodehdr;
1045 struct xfs_da_intnode *node;
1046 struct xfs_inode *dp = args->dp;
1047 struct xfs_mount *mp = dp->i_mount;
1048 struct xfs_buf *bp1 = NULL;
1049 struct xfs_buf *bp2 = NULL;
1050 xfs_dablk_t blkno;
1051 int error;
1052
1053 trace_xfs_attr_leaf_to_node(args);
1054
1055 error = xfs_da_grow_inode(args, &blkno);
1056 if (error)
1057 goto out;
1058 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1059 if (error)
1060 goto out;
1061
1062 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1063 if (error)
1064 goto out;
1065
1066 /* copy leaf to new buffer, update identifiers */
1067 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1068 bp2->b_ops = bp1->b_ops;
1069 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1070 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1071 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1072 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1073 }
1074 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1075
1076 /*
1077 * Set up the new root node.
1078 */
1079 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1080 if (error)
1081 goto out;
1082 node = bp1->b_addr;
1083 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1084 btree = dp->d_ops->node_tree_p(node);
1085
1086 leaf = bp2->b_addr;
1087 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1088 entries = xfs_attr3_leaf_entryp(leaf);
1089
1090 /* both on-disk, don't endian-flip twice */
1091 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1092 btree[0].before = cpu_to_be32(blkno);
1093 icnodehdr.count = 1;
1094 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1095 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1096 error = 0;
1097 out:
1098 return error;
1099 }
1100
1101 /*========================================================================
1102 * Routines used for growing the Btree.
1103 *========================================================================*/
1104
1105 /*
1106 * Create the initial contents of a leaf attribute list
1107 * or a leaf in a node attribute list.
1108 */
1109 STATIC int
1110 xfs_attr3_leaf_create(
1111 struct xfs_da_args *args,
1112 xfs_dablk_t blkno,
1113 struct xfs_buf **bpp)
1114 {
1115 struct xfs_attr_leafblock *leaf;
1116 struct xfs_attr3_icleaf_hdr ichdr;
1117 struct xfs_inode *dp = args->dp;
1118 struct xfs_mount *mp = dp->i_mount;
1119 struct xfs_buf *bp;
1120 int error;
1121
1122 trace_xfs_attr_leaf_create(args);
1123
1124 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1125 XFS_ATTR_FORK);
1126 if (error)
1127 return error;
1128 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1129 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1130 leaf = bp->b_addr;
1131 memset(leaf, 0, args->geo->blksize);
1132
1133 memset(&ichdr, 0, sizeof(ichdr));
1134 ichdr.firstused = args->geo->blksize;
1135
1136 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1137 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1138
1139 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1140
1141 hdr3->blkno = cpu_to_be64(bp->b_bn);
1142 hdr3->owner = cpu_to_be64(dp->i_ino);
1143 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1144
1145 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1146 } else {
1147 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1148 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1149 }
1150 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1151
1152 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1153 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1154
1155 *bpp = bp;
1156 return 0;
1157 }
1158
1159 /*
1160 * Split the leaf node, rebalance, then add the new entry.
1161 */
1162 int
1163 xfs_attr3_leaf_split(
1164 struct xfs_da_state *state,
1165 struct xfs_da_state_blk *oldblk,
1166 struct xfs_da_state_blk *newblk)
1167 {
1168 xfs_dablk_t blkno;
1169 int error;
1170
1171 trace_xfs_attr_leaf_split(state->args);
1172
1173 /*
1174 * Allocate space for a new leaf node.
1175 */
1176 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1177 error = xfs_da_grow_inode(state->args, &blkno);
1178 if (error)
1179 return error;
1180 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1181 if (error)
1182 return error;
1183 newblk->blkno = blkno;
1184 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1185
1186 /*
1187 * Rebalance the entries across the two leaves.
1188 * NOTE: rebalance() currently depends on the 2nd block being empty.
1189 */
1190 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1191 error = xfs_da3_blk_link(state, oldblk, newblk);
1192 if (error)
1193 return error;
1194
1195 /*
1196 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1197 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1198 * "new" attrs info. Will need the "old" info to remove it later.
1199 *
1200 * Insert the "new" entry in the correct block.
1201 */
1202 if (state->inleaf) {
1203 trace_xfs_attr_leaf_add_old(state->args);
1204 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1205 } else {
1206 trace_xfs_attr_leaf_add_new(state->args);
1207 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1208 }
1209
1210 /*
1211 * Update last hashval in each block since we added the name.
1212 */
1213 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1214 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1215 return error;
1216 }
1217
1218 /*
1219 * Add a name to the leaf attribute list structure.
1220 */
1221 int
1222 xfs_attr3_leaf_add(
1223 struct xfs_buf *bp,
1224 struct xfs_da_args *args)
1225 {
1226 struct xfs_attr_leafblock *leaf;
1227 struct xfs_attr3_icleaf_hdr ichdr;
1228 int tablesize;
1229 int entsize;
1230 int sum;
1231 int tmp;
1232 int i;
1233
1234 trace_xfs_attr_leaf_add(args);
1235
1236 leaf = bp->b_addr;
1237 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1238 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1239 entsize = xfs_attr_leaf_newentsize(args, NULL);
1240
1241 /*
1242 * Search through freemap for first-fit on new name length.
1243 * (may need to figure in size of entry struct too)
1244 */
1245 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1246 + xfs_attr3_leaf_hdr_size(leaf);
1247 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1248 if (tablesize > ichdr.firstused) {
1249 sum += ichdr.freemap[i].size;
1250 continue;
1251 }
1252 if (!ichdr.freemap[i].size)
1253 continue; /* no space in this map */
1254 tmp = entsize;
1255 if (ichdr.freemap[i].base < ichdr.firstused)
1256 tmp += sizeof(xfs_attr_leaf_entry_t);
1257 if (ichdr.freemap[i].size >= tmp) {
1258 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1259 goto out_log_hdr;
1260 }
1261 sum += ichdr.freemap[i].size;
1262 }
1263
1264 /*
1265 * If there are no holes in the address space of the block,
1266 * and we don't have enough freespace, then compaction will do us
1267 * no good and we should just give up.
1268 */
1269 if (!ichdr.holes && sum < entsize)
1270 return -ENOSPC;
1271
1272 /*
1273 * Compact the entries to coalesce free space.
1274 * This may change the hdr->count via dropping INCOMPLETE entries.
1275 */
1276 xfs_attr3_leaf_compact(args, &ichdr, bp);
1277
1278 /*
1279 * After compaction, the block is guaranteed to have only one
1280 * free region, in freemap[0]. If it is not big enough, give up.
1281 */
1282 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1283 tmp = -ENOSPC;
1284 goto out_log_hdr;
1285 }
1286
1287 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1288
1289 out_log_hdr:
1290 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1291 xfs_trans_log_buf(args->trans, bp,
1292 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1293 xfs_attr3_leaf_hdr_size(leaf)));
1294 return tmp;
1295 }
1296
1297 /*
1298 * Add a name to a leaf attribute list structure.
1299 */
1300 STATIC int
1301 xfs_attr3_leaf_add_work(
1302 struct xfs_buf *bp,
1303 struct xfs_attr3_icleaf_hdr *ichdr,
1304 struct xfs_da_args *args,
1305 int mapindex)
1306 {
1307 struct xfs_attr_leafblock *leaf;
1308 struct xfs_attr_leaf_entry *entry;
1309 struct xfs_attr_leaf_name_local *name_loc;
1310 struct xfs_attr_leaf_name_remote *name_rmt;
1311 struct xfs_mount *mp;
1312 int tmp;
1313 int i;
1314
1315 trace_xfs_attr_leaf_add_work(args);
1316
1317 leaf = bp->b_addr;
1318 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1319 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1320
1321 /*
1322 * Force open some space in the entry array and fill it in.
1323 */
1324 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1325 if (args->index < ichdr->count) {
1326 tmp = ichdr->count - args->index;
1327 tmp *= sizeof(xfs_attr_leaf_entry_t);
1328 memmove(entry + 1, entry, tmp);
1329 xfs_trans_log_buf(args->trans, bp,
1330 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1331 }
1332 ichdr->count++;
1333
1334 /*
1335 * Allocate space for the new string (at the end of the run).
1336 */
1337 mp = args->trans->t_mountp;
1338 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1339 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1340 ASSERT(ichdr->freemap[mapindex].size >=
1341 xfs_attr_leaf_newentsize(args, NULL));
1342 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1343 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1344
1345 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1346
1347 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1348 ichdr->freemap[mapindex].size);
1349 entry->hashval = cpu_to_be32(args->hashval);
1350 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1351 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1352 if (args->op_flags & XFS_DA_OP_RENAME) {
1353 entry->flags |= XFS_ATTR_INCOMPLETE;
1354 if ((args->blkno2 == args->blkno) &&
1355 (args->index2 <= args->index)) {
1356 args->index2++;
1357 }
1358 }
1359 xfs_trans_log_buf(args->trans, bp,
1360 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1361 ASSERT((args->index == 0) ||
1362 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1363 ASSERT((args->index == ichdr->count - 1) ||
1364 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1365
1366 /*
1367 * For "remote" attribute values, simply note that we need to
1368 * allocate space for the "remote" value. We can't actually
1369 * allocate the extents in this transaction, and we can't decide
1370 * which blocks they should be as we might allocate more blocks
1371 * as part of this transaction (a split operation for example).
1372 */
1373 if (entry->flags & XFS_ATTR_LOCAL) {
1374 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1375 name_loc->namelen = args->namelen;
1376 name_loc->valuelen = cpu_to_be16(args->valuelen);
1377 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1378 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1379 be16_to_cpu(name_loc->valuelen));
1380 } else {
1381 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1382 name_rmt->namelen = args->namelen;
1383 memcpy((char *)name_rmt->name, args->name, args->namelen);
1384 entry->flags |= XFS_ATTR_INCOMPLETE;
1385 /* just in case */
1386 name_rmt->valuelen = 0;
1387 name_rmt->valueblk = 0;
1388 args->rmtblkno = 1;
1389 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1390 args->rmtvaluelen = args->valuelen;
1391 }
1392 xfs_trans_log_buf(args->trans, bp,
1393 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1394 xfs_attr_leaf_entsize(leaf, args->index)));
1395
1396 /*
1397 * Update the control info for this leaf node
1398 */
1399 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1400 ichdr->firstused = be16_to_cpu(entry->nameidx);
1401
1402 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1403 + xfs_attr3_leaf_hdr_size(leaf));
1404 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1405 + xfs_attr3_leaf_hdr_size(leaf);
1406
1407 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1408 if (ichdr->freemap[i].base == tmp) {
1409 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1410 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1411 }
1412 }
1413 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1414 return 0;
1415 }
1416
1417 /*
1418 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1419 */
1420 STATIC void
1421 xfs_attr3_leaf_compact(
1422 struct xfs_da_args *args,
1423 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1424 struct xfs_buf *bp)
1425 {
1426 struct xfs_attr_leafblock *leaf_src;
1427 struct xfs_attr_leafblock *leaf_dst;
1428 struct xfs_attr3_icleaf_hdr ichdr_src;
1429 struct xfs_trans *trans = args->trans;
1430 char *tmpbuffer;
1431
1432 trace_xfs_attr_leaf_compact(args);
1433
1434 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1435 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1436 memset(bp->b_addr, 0, args->geo->blksize);
1437 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1438 leaf_dst = bp->b_addr;
1439
1440 /*
1441 * Copy the on-disk header back into the destination buffer to ensure
1442 * all the information in the header that is not part of the incore
1443 * header structure is preserved.
1444 */
1445 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1446
1447 /* Initialise the incore headers */
1448 ichdr_src = *ichdr_dst; /* struct copy */
1449 ichdr_dst->firstused = args->geo->blksize;
1450 ichdr_dst->usedbytes = 0;
1451 ichdr_dst->count = 0;
1452 ichdr_dst->holes = 0;
1453 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1454 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1455 ichdr_dst->freemap[0].base;
1456
1457 /* write the header back to initialise the underlying buffer */
1458 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1459
1460 /*
1461 * Copy all entry's in the same (sorted) order,
1462 * but allocate name/value pairs packed and in sequence.
1463 */
1464 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1465 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1466 /*
1467 * this logs the entire buffer, but the caller must write the header
1468 * back to the buffer when it is finished modifying it.
1469 */
1470 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1471
1472 kmem_free(tmpbuffer);
1473 }
1474
1475 /*
1476 * Compare two leaf blocks "order".
1477 * Return 0 unless leaf2 should go before leaf1.
1478 */
1479 static int
1480 xfs_attr3_leaf_order(
1481 struct xfs_buf *leaf1_bp,
1482 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1483 struct xfs_buf *leaf2_bp,
1484 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1485 {
1486 struct xfs_attr_leaf_entry *entries1;
1487 struct xfs_attr_leaf_entry *entries2;
1488
1489 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1490 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1491 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1492 ((be32_to_cpu(entries2[0].hashval) <
1493 be32_to_cpu(entries1[0].hashval)) ||
1494 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1495 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1496 return 1;
1497 }
1498 return 0;
1499 }
1500
1501 int
1502 xfs_attr_leaf_order(
1503 struct xfs_buf *leaf1_bp,
1504 struct xfs_buf *leaf2_bp)
1505 {
1506 struct xfs_attr3_icleaf_hdr ichdr1;
1507 struct xfs_attr3_icleaf_hdr ichdr2;
1508 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1509
1510 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1511 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1512 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1513 }
1514
1515 /*
1516 * Redistribute the attribute list entries between two leaf nodes,
1517 * taking into account the size of the new entry.
1518 *
1519 * NOTE: if new block is empty, then it will get the upper half of the
1520 * old block. At present, all (one) callers pass in an empty second block.
1521 *
1522 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1523 * to match what it is doing in splitting the attribute leaf block. Those
1524 * values are used in "atomic rename" operations on attributes. Note that
1525 * the "new" and "old" values can end up in different blocks.
1526 */
1527 STATIC void
1528 xfs_attr3_leaf_rebalance(
1529 struct xfs_da_state *state,
1530 struct xfs_da_state_blk *blk1,
1531 struct xfs_da_state_blk *blk2)
1532 {
1533 struct xfs_da_args *args;
1534 struct xfs_attr_leafblock *leaf1;
1535 struct xfs_attr_leafblock *leaf2;
1536 struct xfs_attr3_icleaf_hdr ichdr1;
1537 struct xfs_attr3_icleaf_hdr ichdr2;
1538 struct xfs_attr_leaf_entry *entries1;
1539 struct xfs_attr_leaf_entry *entries2;
1540 int count;
1541 int totallen;
1542 int max;
1543 int space;
1544 int swap;
1545
1546 /*
1547 * Set up environment.
1548 */
1549 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1550 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1551 leaf1 = blk1->bp->b_addr;
1552 leaf2 = blk2->bp->b_addr;
1553 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1554 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1555 ASSERT(ichdr2.count == 0);
1556 args = state->args;
1557
1558 trace_xfs_attr_leaf_rebalance(args);
1559
1560 /*
1561 * Check ordering of blocks, reverse if it makes things simpler.
1562 *
1563 * NOTE: Given that all (current) callers pass in an empty
1564 * second block, this code should never set "swap".
1565 */
1566 swap = 0;
1567 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1568 struct xfs_da_state_blk *tmp_blk;
1569 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1570
1571 tmp_blk = blk1;
1572 blk1 = blk2;
1573 blk2 = tmp_blk;
1574
1575 /* struct copies to swap them rather than reconverting */
1576 tmp_ichdr = ichdr1;
1577 ichdr1 = ichdr2;
1578 ichdr2 = tmp_ichdr;
1579
1580 leaf1 = blk1->bp->b_addr;
1581 leaf2 = blk2->bp->b_addr;
1582 swap = 1;
1583 }
1584
1585 /*
1586 * Examine entries until we reduce the absolute difference in
1587 * byte usage between the two blocks to a minimum. Then get
1588 * the direction to copy and the number of elements to move.
1589 *
1590 * "inleaf" is true if the new entry should be inserted into blk1.
1591 * If "swap" is also true, then reverse the sense of "inleaf".
1592 */
1593 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1594 blk2, &ichdr2,
1595 &count, &totallen);
1596 if (swap)
1597 state->inleaf = !state->inleaf;
1598
1599 /*
1600 * Move any entries required from leaf to leaf:
1601 */
1602 if (count < ichdr1.count) {
1603 /*
1604 * Figure the total bytes to be added to the destination leaf.
1605 */
1606 /* number entries being moved */
1607 count = ichdr1.count - count;
1608 space = ichdr1.usedbytes - totallen;
1609 space += count * sizeof(xfs_attr_leaf_entry_t);
1610
1611 /*
1612 * leaf2 is the destination, compact it if it looks tight.
1613 */
1614 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1615 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1616 if (space > max)
1617 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1618
1619 /*
1620 * Move high entries from leaf1 to low end of leaf2.
1621 */
1622 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1623 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1624
1625 } else if (count > ichdr1.count) {
1626 /*
1627 * I assert that since all callers pass in an empty
1628 * second buffer, this code should never execute.
1629 */
1630 ASSERT(0);
1631
1632 /*
1633 * Figure the total bytes to be added to the destination leaf.
1634 */
1635 /* number entries being moved */
1636 count -= ichdr1.count;
1637 space = totallen - ichdr1.usedbytes;
1638 space += count * sizeof(xfs_attr_leaf_entry_t);
1639
1640 /*
1641 * leaf1 is the destination, compact it if it looks tight.
1642 */
1643 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1644 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1645 if (space > max)
1646 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1647
1648 /*
1649 * Move low entries from leaf2 to high end of leaf1.
1650 */
1651 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1652 ichdr1.count, count);
1653 }
1654
1655 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1656 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1657 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1658 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1659
1660 /*
1661 * Copy out last hashval in each block for B-tree code.
1662 */
1663 entries1 = xfs_attr3_leaf_entryp(leaf1);
1664 entries2 = xfs_attr3_leaf_entryp(leaf2);
1665 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1666 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1667
1668 /*
1669 * Adjust the expected index for insertion.
1670 * NOTE: this code depends on the (current) situation that the
1671 * second block was originally empty.
1672 *
1673 * If the insertion point moved to the 2nd block, we must adjust
1674 * the index. We must also track the entry just following the
1675 * new entry for use in an "atomic rename" operation, that entry
1676 * is always the "old" entry and the "new" entry is what we are
1677 * inserting. The index/blkno fields refer to the "old" entry,
1678 * while the index2/blkno2 fields refer to the "new" entry.
1679 */
1680 if (blk1->index > ichdr1.count) {
1681 ASSERT(state->inleaf == 0);
1682 blk2->index = blk1->index - ichdr1.count;
1683 args->index = args->index2 = blk2->index;
1684 args->blkno = args->blkno2 = blk2->blkno;
1685 } else if (blk1->index == ichdr1.count) {
1686 if (state->inleaf) {
1687 args->index = blk1->index;
1688 args->blkno = blk1->blkno;
1689 args->index2 = 0;
1690 args->blkno2 = blk2->blkno;
1691 } else {
1692 /*
1693 * On a double leaf split, the original attr location
1694 * is already stored in blkno2/index2, so don't
1695 * overwrite it overwise we corrupt the tree.
1696 */
1697 blk2->index = blk1->index - ichdr1.count;
1698 args->index = blk2->index;
1699 args->blkno = blk2->blkno;
1700 if (!state->extravalid) {
1701 /*
1702 * set the new attr location to match the old
1703 * one and let the higher level split code
1704 * decide where in the leaf to place it.
1705 */
1706 args->index2 = blk2->index;
1707 args->blkno2 = blk2->blkno;
1708 }
1709 }
1710 } else {
1711 ASSERT(state->inleaf == 1);
1712 args->index = args->index2 = blk1->index;
1713 args->blkno = args->blkno2 = blk1->blkno;
1714 }
1715 }
1716
1717 /*
1718 * Examine entries until we reduce the absolute difference in
1719 * byte usage between the two blocks to a minimum.
1720 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1721 * GROT: there will always be enough room in either block for a new entry.
1722 * GROT: Do a double-split for this case?
1723 */
1724 STATIC int
1725 xfs_attr3_leaf_figure_balance(
1726 struct xfs_da_state *state,
1727 struct xfs_da_state_blk *blk1,
1728 struct xfs_attr3_icleaf_hdr *ichdr1,
1729 struct xfs_da_state_blk *blk2,
1730 struct xfs_attr3_icleaf_hdr *ichdr2,
1731 int *countarg,
1732 int *usedbytesarg)
1733 {
1734 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1735 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1736 struct xfs_attr_leaf_entry *entry;
1737 int count;
1738 int max;
1739 int index;
1740 int totallen = 0;
1741 int half;
1742 int lastdelta;
1743 int foundit = 0;
1744 int tmp;
1745
1746 /*
1747 * Examine entries until we reduce the absolute difference in
1748 * byte usage between the two blocks to a minimum.
1749 */
1750 max = ichdr1->count + ichdr2->count;
1751 half = (max + 1) * sizeof(*entry);
1752 half += ichdr1->usedbytes + ichdr2->usedbytes +
1753 xfs_attr_leaf_newentsize(state->args, NULL);
1754 half /= 2;
1755 lastdelta = state->args->geo->blksize;
1756 entry = xfs_attr3_leaf_entryp(leaf1);
1757 for (count = index = 0; count < max; entry++, index++, count++) {
1758
1759 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1760 /*
1761 * The new entry is in the first block, account for it.
1762 */
1763 if (count == blk1->index) {
1764 tmp = totallen + sizeof(*entry) +
1765 xfs_attr_leaf_newentsize(state->args, NULL);
1766 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1767 break;
1768 lastdelta = XFS_ATTR_ABS(half - tmp);
1769 totallen = tmp;
1770 foundit = 1;
1771 }
1772
1773 /*
1774 * Wrap around into the second block if necessary.
1775 */
1776 if (count == ichdr1->count) {
1777 leaf1 = leaf2;
1778 entry = xfs_attr3_leaf_entryp(leaf1);
1779 index = 0;
1780 }
1781
1782 /*
1783 * Figure out if next leaf entry would be too much.
1784 */
1785 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1786 index);
1787 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1788 break;
1789 lastdelta = XFS_ATTR_ABS(half - tmp);
1790 totallen = tmp;
1791 #undef XFS_ATTR_ABS
1792 }
1793
1794 /*
1795 * Calculate the number of usedbytes that will end up in lower block.
1796 * If new entry not in lower block, fix up the count.
1797 */
1798 totallen -= count * sizeof(*entry);
1799 if (foundit) {
1800 totallen -= sizeof(*entry) +
1801 xfs_attr_leaf_newentsize(state->args, NULL);
1802 }
1803
1804 *countarg = count;
1805 *usedbytesarg = totallen;
1806 return foundit;
1807 }
1808
1809 /*========================================================================
1810 * Routines used for shrinking the Btree.
1811 *========================================================================*/
1812
1813 /*
1814 * Check a leaf block and its neighbors to see if the block should be
1815 * collapsed into one or the other neighbor. Always keep the block
1816 * with the smaller block number.
1817 * If the current block is over 50% full, don't try to join it, return 0.
1818 * If the block is empty, fill in the state structure and return 2.
1819 * If it can be collapsed, fill in the state structure and return 1.
1820 * If nothing can be done, return 0.
1821 *
1822 * GROT: allow for INCOMPLETE entries in calculation.
1823 */
1824 int
1825 xfs_attr3_leaf_toosmall(
1826 struct xfs_da_state *state,
1827 int *action)
1828 {
1829 struct xfs_attr_leafblock *leaf;
1830 struct xfs_da_state_blk *blk;
1831 struct xfs_attr3_icleaf_hdr ichdr;
1832 struct xfs_buf *bp;
1833 xfs_dablk_t blkno;
1834 int bytes;
1835 int forward;
1836 int error;
1837 int retval;
1838 int i;
1839
1840 trace_xfs_attr_leaf_toosmall(state->args);
1841
1842 /*
1843 * Check for the degenerate case of the block being over 50% full.
1844 * If so, it's not worth even looking to see if we might be able
1845 * to coalesce with a sibling.
1846 */
1847 blk = &state->path.blk[ state->path.active-1 ];
1848 leaf = blk->bp->b_addr;
1849 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1850 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1851 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1852 ichdr.usedbytes;
1853 if (bytes > (state->args->geo->blksize >> 1)) {
1854 *action = 0; /* blk over 50%, don't try to join */
1855 return 0;
1856 }
1857
1858 /*
1859 * Check for the degenerate case of the block being empty.
1860 * If the block is empty, we'll simply delete it, no need to
1861 * coalesce it with a sibling block. We choose (arbitrarily)
1862 * to merge with the forward block unless it is NULL.
1863 */
1864 if (ichdr.count == 0) {
1865 /*
1866 * Make altpath point to the block we want to keep and
1867 * path point to the block we want to drop (this one).
1868 */
1869 forward = (ichdr.forw != 0);
1870 memcpy(&state->altpath, &state->path, sizeof(state->path));
1871 error = xfs_da3_path_shift(state, &state->altpath, forward,
1872 0, &retval);
1873 if (error)
1874 return error;
1875 if (retval) {
1876 *action = 0;
1877 } else {
1878 *action = 2;
1879 }
1880 return 0;
1881 }
1882
1883 /*
1884 * Examine each sibling block to see if we can coalesce with
1885 * at least 25% free space to spare. We need to figure out
1886 * whether to merge with the forward or the backward block.
1887 * We prefer coalescing with the lower numbered sibling so as
1888 * to shrink an attribute list over time.
1889 */
1890 /* start with smaller blk num */
1891 forward = ichdr.forw < ichdr.back;
1892 for (i = 0; i < 2; forward = !forward, i++) {
1893 struct xfs_attr3_icleaf_hdr ichdr2;
1894 if (forward)
1895 blkno = ichdr.forw;
1896 else
1897 blkno = ichdr.back;
1898 if (blkno == 0)
1899 continue;
1900 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1901 blkno, -1, &bp);
1902 if (error)
1903 return error;
1904
1905 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1906
1907 bytes = state->args->geo->blksize -
1908 (state->args->geo->blksize >> 2) -
1909 ichdr.usedbytes - ichdr2.usedbytes -
1910 ((ichdr.count + ichdr2.count) *
1911 sizeof(xfs_attr_leaf_entry_t)) -
1912 xfs_attr3_leaf_hdr_size(leaf);
1913
1914 xfs_trans_brelse(state->args->trans, bp);
1915 if (bytes >= 0)
1916 break; /* fits with at least 25% to spare */
1917 }
1918 if (i >= 2) {
1919 *action = 0;
1920 return 0;
1921 }
1922
1923 /*
1924 * Make altpath point to the block we want to keep (the lower
1925 * numbered block) and path point to the block we want to drop.
1926 */
1927 memcpy(&state->altpath, &state->path, sizeof(state->path));
1928 if (blkno < blk->blkno) {
1929 error = xfs_da3_path_shift(state, &state->altpath, forward,
1930 0, &retval);
1931 } else {
1932 error = xfs_da3_path_shift(state, &state->path, forward,
1933 0, &retval);
1934 }
1935 if (error)
1936 return error;
1937 if (retval) {
1938 *action = 0;
1939 } else {
1940 *action = 1;
1941 }
1942 return 0;
1943 }
1944
1945 /*
1946 * Remove a name from the leaf attribute list structure.
1947 *
1948 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1949 * If two leaves are 37% full, when combined they will leave 25% free.
1950 */
1951 int
1952 xfs_attr3_leaf_remove(
1953 struct xfs_buf *bp,
1954 struct xfs_da_args *args)
1955 {
1956 struct xfs_attr_leafblock *leaf;
1957 struct xfs_attr3_icleaf_hdr ichdr;
1958 struct xfs_attr_leaf_entry *entry;
1959 int before;
1960 int after;
1961 int smallest;
1962 int entsize;
1963 int tablesize;
1964 int tmp;
1965 int i;
1966
1967 trace_xfs_attr_leaf_remove(args);
1968
1969 leaf = bp->b_addr;
1970 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1971
1972 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1973 ASSERT(args->index >= 0 && args->index < ichdr.count);
1974 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1975 xfs_attr3_leaf_hdr_size(leaf));
1976
1977 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1978
1979 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1980 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1981
1982 /*
1983 * Scan through free region table:
1984 * check for adjacency of free'd entry with an existing one,
1985 * find smallest free region in case we need to replace it,
1986 * adjust any map that borders the entry table,
1987 */
1988 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1989 + xfs_attr3_leaf_hdr_size(leaf);
1990 tmp = ichdr.freemap[0].size;
1991 before = after = -1;
1992 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1993 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1994 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1995 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1996 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1997 if (ichdr.freemap[i].base == tablesize) {
1998 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1999 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2000 }
2001
2002 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2003 be16_to_cpu(entry->nameidx)) {
2004 before = i;
2005 } else if (ichdr.freemap[i].base ==
2006 (be16_to_cpu(entry->nameidx) + entsize)) {
2007 after = i;
2008 } else if (ichdr.freemap[i].size < tmp) {
2009 tmp = ichdr.freemap[i].size;
2010 smallest = i;
2011 }
2012 }
2013
2014 /*
2015 * Coalesce adjacent freemap regions,
2016 * or replace the smallest region.
2017 */
2018 if ((before >= 0) || (after >= 0)) {
2019 if ((before >= 0) && (after >= 0)) {
2020 ichdr.freemap[before].size += entsize;
2021 ichdr.freemap[before].size += ichdr.freemap[after].size;
2022 ichdr.freemap[after].base = 0;
2023 ichdr.freemap[after].size = 0;
2024 } else if (before >= 0) {
2025 ichdr.freemap[before].size += entsize;
2026 } else {
2027 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2028 ichdr.freemap[after].size += entsize;
2029 }
2030 } else {
2031 /*
2032 * Replace smallest region (if it is smaller than free'd entry)
2033 */
2034 if (ichdr.freemap[smallest].size < entsize) {
2035 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2036 ichdr.freemap[smallest].size = entsize;
2037 }
2038 }
2039
2040 /*
2041 * Did we remove the first entry?
2042 */
2043 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2044 smallest = 1;
2045 else
2046 smallest = 0;
2047
2048 /*
2049 * Compress the remaining entries and zero out the removed stuff.
2050 */
2051 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2052 ichdr.usedbytes -= entsize;
2053 xfs_trans_log_buf(args->trans, bp,
2054 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2055 entsize));
2056
2057 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2058 memmove(entry, entry + 1, tmp);
2059 ichdr.count--;
2060 xfs_trans_log_buf(args->trans, bp,
2061 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2062
2063 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2064 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2065
2066 /*
2067 * If we removed the first entry, re-find the first used byte
2068 * in the name area. Note that if the entry was the "firstused",
2069 * then we don't have a "hole" in our block resulting from
2070 * removing the name.
2071 */
2072 if (smallest) {
2073 tmp = args->geo->blksize;
2074 entry = xfs_attr3_leaf_entryp(leaf);
2075 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2076 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2077 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2078
2079 if (be16_to_cpu(entry->nameidx) < tmp)
2080 tmp = be16_to_cpu(entry->nameidx);
2081 }
2082 ichdr.firstused = tmp;
2083 ASSERT(ichdr.firstused != 0);
2084 } else {
2085 ichdr.holes = 1; /* mark as needing compaction */
2086 }
2087 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2088 xfs_trans_log_buf(args->trans, bp,
2089 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2090 xfs_attr3_leaf_hdr_size(leaf)));
2091
2092 /*
2093 * Check if leaf is less than 50% full, caller may want to
2094 * "join" the leaf with a sibling if so.
2095 */
2096 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2097 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2098
2099 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2100 }
2101
2102 /*
2103 * Move all the attribute list entries from drop_leaf into save_leaf.
2104 */
2105 void
2106 xfs_attr3_leaf_unbalance(
2107 struct xfs_da_state *state,
2108 struct xfs_da_state_blk *drop_blk,
2109 struct xfs_da_state_blk *save_blk)
2110 {
2111 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2112 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2113 struct xfs_attr3_icleaf_hdr drophdr;
2114 struct xfs_attr3_icleaf_hdr savehdr;
2115 struct xfs_attr_leaf_entry *entry;
2116
2117 trace_xfs_attr_leaf_unbalance(state->args);
2118
2119 drop_leaf = drop_blk->bp->b_addr;
2120 save_leaf = save_blk->bp->b_addr;
2121 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2122 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2123 entry = xfs_attr3_leaf_entryp(drop_leaf);
2124
2125 /*
2126 * Save last hashval from dying block for later Btree fixup.
2127 */
2128 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2129
2130 /*
2131 * Check if we need a temp buffer, or can we do it in place.
2132 * Note that we don't check "leaf" for holes because we will
2133 * always be dropping it, toosmall() decided that for us already.
2134 */
2135 if (savehdr.holes == 0) {
2136 /*
2137 * dest leaf has no holes, so we add there. May need
2138 * to make some room in the entry array.
2139 */
2140 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2141 drop_blk->bp, &drophdr)) {
2142 xfs_attr3_leaf_moveents(state->args,
2143 drop_leaf, &drophdr, 0,
2144 save_leaf, &savehdr, 0,
2145 drophdr.count);
2146 } else {
2147 xfs_attr3_leaf_moveents(state->args,
2148 drop_leaf, &drophdr, 0,
2149 save_leaf, &savehdr,
2150 savehdr.count, drophdr.count);
2151 }
2152 } else {
2153 /*
2154 * Destination has holes, so we make a temporary copy
2155 * of the leaf and add them both to that.
2156 */
2157 struct xfs_attr_leafblock *tmp_leaf;
2158 struct xfs_attr3_icleaf_hdr tmphdr;
2159
2160 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2161
2162 /*
2163 * Copy the header into the temp leaf so that all the stuff
2164 * not in the incore header is present and gets copied back in
2165 * once we've moved all the entries.
2166 */
2167 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2168
2169 memset(&tmphdr, 0, sizeof(tmphdr));
2170 tmphdr.magic = savehdr.magic;
2171 tmphdr.forw = savehdr.forw;
2172 tmphdr.back = savehdr.back;
2173 tmphdr.firstused = state->args->geo->blksize;
2174
2175 /* write the header to the temp buffer to initialise it */
2176 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2177
2178 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2179 drop_blk->bp, &drophdr)) {
2180 xfs_attr3_leaf_moveents(state->args,
2181 drop_leaf, &drophdr, 0,
2182 tmp_leaf, &tmphdr, 0,
2183 drophdr.count);
2184 xfs_attr3_leaf_moveents(state->args,
2185 save_leaf, &savehdr, 0,
2186 tmp_leaf, &tmphdr, tmphdr.count,
2187 savehdr.count);
2188 } else {
2189 xfs_attr3_leaf_moveents(state->args,
2190 save_leaf, &savehdr, 0,
2191 tmp_leaf, &tmphdr, 0,
2192 savehdr.count);
2193 xfs_attr3_leaf_moveents(state->args,
2194 drop_leaf, &drophdr, 0,
2195 tmp_leaf, &tmphdr, tmphdr.count,
2196 drophdr.count);
2197 }
2198 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2199 savehdr = tmphdr; /* struct copy */
2200 kmem_free(tmp_leaf);
2201 }
2202
2203 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2204 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2205 state->args->geo->blksize - 1);
2206
2207 /*
2208 * Copy out last hashval in each block for B-tree code.
2209 */
2210 entry = xfs_attr3_leaf_entryp(save_leaf);
2211 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2212 }
2213
2214 /*========================================================================
2215 * Routines used for finding things in the Btree.
2216 *========================================================================*/
2217
2218 /*
2219 * Look up a name in a leaf attribute list structure.
2220 * This is the internal routine, it uses the caller's buffer.
2221 *
2222 * Note that duplicate keys are allowed, but only check within the
2223 * current leaf node. The Btree code must check in adjacent leaf nodes.
2224 *
2225 * Return in args->index the index into the entry[] array of either
2226 * the found entry, or where the entry should have been (insert before
2227 * that entry).
2228 *
2229 * Don't change the args->value unless we find the attribute.
2230 */
2231 int
2232 xfs_attr3_leaf_lookup_int(
2233 struct xfs_buf *bp,
2234 struct xfs_da_args *args)
2235 {
2236 struct xfs_attr_leafblock *leaf;
2237 struct xfs_attr3_icleaf_hdr ichdr;
2238 struct xfs_attr_leaf_entry *entry;
2239 struct xfs_attr_leaf_entry *entries;
2240 struct xfs_attr_leaf_name_local *name_loc;
2241 struct xfs_attr_leaf_name_remote *name_rmt;
2242 xfs_dahash_t hashval;
2243 int probe;
2244 int span;
2245
2246 trace_xfs_attr_leaf_lookup(args);
2247
2248 leaf = bp->b_addr;
2249 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2250 entries = xfs_attr3_leaf_entryp(leaf);
2251 if (ichdr.count >= args->geo->blksize / 8)
2252 return -EFSCORRUPTED;
2253
2254 /*
2255 * Binary search. (note: small blocks will skip this loop)
2256 */
2257 hashval = args->hashval;
2258 probe = span = ichdr.count / 2;
2259 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2260 span /= 2;
2261 if (be32_to_cpu(entry->hashval) < hashval)
2262 probe += span;
2263 else if (be32_to_cpu(entry->hashval) > hashval)
2264 probe -= span;
2265 else
2266 break;
2267 }
2268 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count)))
2269 return -EFSCORRUPTED;
2270 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval))
2271 return -EFSCORRUPTED;
2272
2273 /*
2274 * Since we may have duplicate hashval's, find the first matching
2275 * hashval in the leaf.
2276 */
2277 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2278 entry--;
2279 probe--;
2280 }
2281 while (probe < ichdr.count &&
2282 be32_to_cpu(entry->hashval) < hashval) {
2283 entry++;
2284 probe++;
2285 }
2286 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2287 args->index = probe;
2288 return -ENOATTR;
2289 }
2290
2291 /*
2292 * Duplicate keys may be present, so search all of them for a match.
2293 */
2294 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2295 entry++, probe++) {
2296 /*
2297 * GROT: Add code to remove incomplete entries.
2298 */
2299 /*
2300 * If we are looking for INCOMPLETE entries, show only those.
2301 * If we are looking for complete entries, show only those.
2302 */
2303 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2304 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2305 continue;
2306 }
2307 if (entry->flags & XFS_ATTR_LOCAL) {
2308 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2309 if (name_loc->namelen != args->namelen)
2310 continue;
2311 if (memcmp(args->name, name_loc->nameval,
2312 args->namelen) != 0)
2313 continue;
2314 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2315 continue;
2316 args->index = probe;
2317 return -EEXIST;
2318 } else {
2319 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2320 if (name_rmt->namelen != args->namelen)
2321 continue;
2322 if (memcmp(args->name, name_rmt->name,
2323 args->namelen) != 0)
2324 continue;
2325 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2326 continue;
2327 args->index = probe;
2328 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2329 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2330 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2331 args->dp->i_mount,
2332 args->rmtvaluelen);
2333 return -EEXIST;
2334 }
2335 }
2336 args->index = probe;
2337 return -ENOATTR;
2338 }
2339
2340 /*
2341 * Get the value associated with an attribute name from a leaf attribute
2342 * list structure.
2343 */
2344 int
2345 xfs_attr3_leaf_getvalue(
2346 struct xfs_buf *bp,
2347 struct xfs_da_args *args)
2348 {
2349 struct xfs_attr_leafblock *leaf;
2350 struct xfs_attr3_icleaf_hdr ichdr;
2351 struct xfs_attr_leaf_entry *entry;
2352 struct xfs_attr_leaf_name_local *name_loc;
2353 struct xfs_attr_leaf_name_remote *name_rmt;
2354 int valuelen;
2355
2356 leaf = bp->b_addr;
2357 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2358 ASSERT(ichdr.count < args->geo->blksize / 8);
2359 ASSERT(args->index < ichdr.count);
2360
2361 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2362 if (entry->flags & XFS_ATTR_LOCAL) {
2363 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2364 ASSERT(name_loc->namelen == args->namelen);
2365 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2366 valuelen = be16_to_cpu(name_loc->valuelen);
2367 if (args->flags & ATTR_KERNOVAL) {
2368 args->valuelen = valuelen;
2369 return 0;
2370 }
2371 if (args->valuelen < valuelen) {
2372 args->valuelen = valuelen;
2373 return -ERANGE;
2374 }
2375 args->valuelen = valuelen;
2376 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2377 } else {
2378 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2379 ASSERT(name_rmt->namelen == args->namelen);
2380 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2381 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2382 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2383 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2384 args->rmtvaluelen);
2385 if (args->flags & ATTR_KERNOVAL) {
2386 args->valuelen = args->rmtvaluelen;
2387 return 0;
2388 }
2389 if (args->valuelen < args->rmtvaluelen) {
2390 args->valuelen = args->rmtvaluelen;
2391 return -ERANGE;
2392 }
2393 args->valuelen = args->rmtvaluelen;
2394 }
2395 return 0;
2396 }
2397
2398 /*========================================================================
2399 * Utility routines.
2400 *========================================================================*/
2401
2402 /*
2403 * Move the indicated entries from one leaf to another.
2404 * NOTE: this routine modifies both source and destination leaves.
2405 */
2406 /*ARGSUSED*/
2407 STATIC void
2408 xfs_attr3_leaf_moveents(
2409 struct xfs_da_args *args,
2410 struct xfs_attr_leafblock *leaf_s,
2411 struct xfs_attr3_icleaf_hdr *ichdr_s,
2412 int start_s,
2413 struct xfs_attr_leafblock *leaf_d,
2414 struct xfs_attr3_icleaf_hdr *ichdr_d,
2415 int start_d,
2416 int count)
2417 {
2418 struct xfs_attr_leaf_entry *entry_s;
2419 struct xfs_attr_leaf_entry *entry_d;
2420 int desti;
2421 int tmp;
2422 int i;
2423
2424 /*
2425 * Check for nothing to do.
2426 */
2427 if (count == 0)
2428 return;
2429
2430 /*
2431 * Set up environment.
2432 */
2433 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2434 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2435 ASSERT(ichdr_s->magic == ichdr_d->magic);
2436 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2437 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2438 + xfs_attr3_leaf_hdr_size(leaf_s));
2439 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2440 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2441 + xfs_attr3_leaf_hdr_size(leaf_d));
2442
2443 ASSERT(start_s < ichdr_s->count);
2444 ASSERT(start_d <= ichdr_d->count);
2445 ASSERT(count <= ichdr_s->count);
2446
2447
2448 /*
2449 * Move the entries in the destination leaf up to make a hole?
2450 */
2451 if (start_d < ichdr_d->count) {
2452 tmp = ichdr_d->count - start_d;
2453 tmp *= sizeof(xfs_attr_leaf_entry_t);
2454 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2455 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2456 memmove(entry_d, entry_s, tmp);
2457 }
2458
2459 /*
2460 * Copy all entry's in the same (sorted) order,
2461 * but allocate attribute info packed and in sequence.
2462 */
2463 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2464 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2465 desti = start_d;
2466 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2467 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2468 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2469 #ifdef GROT
2470 /*
2471 * Code to drop INCOMPLETE entries. Difficult to use as we
2472 * may also need to change the insertion index. Code turned
2473 * off for 6.2, should be revisited later.
2474 */
2475 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2476 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2477 ichdr_s->usedbytes -= tmp;
2478 ichdr_s->count -= 1;
2479 entry_d--; /* to compensate for ++ in loop hdr */
2480 desti--;
2481 if ((start_s + i) < offset)
2482 result++; /* insertion index adjustment */
2483 } else {
2484 #endif /* GROT */
2485 ichdr_d->firstused -= tmp;
2486 /* both on-disk, don't endian flip twice */
2487 entry_d->hashval = entry_s->hashval;
2488 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2489 entry_d->flags = entry_s->flags;
2490 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2491 <= args->geo->blksize);
2492 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2493 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2494 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2495 <= args->geo->blksize);
2496 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2497 ichdr_s->usedbytes -= tmp;
2498 ichdr_d->usedbytes += tmp;
2499 ichdr_s->count -= 1;
2500 ichdr_d->count += 1;
2501 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2502 + xfs_attr3_leaf_hdr_size(leaf_d);
2503 ASSERT(ichdr_d->firstused >= tmp);
2504 #ifdef GROT
2505 }
2506 #endif /* GROT */
2507 }
2508
2509 /*
2510 * Zero out the entries we just copied.
2511 */
2512 if (start_s == ichdr_s->count) {
2513 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2514 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2515 ASSERT(((char *)entry_s + tmp) <=
2516 ((char *)leaf_s + args->geo->blksize));
2517 memset(entry_s, 0, tmp);
2518 } else {
2519 /*
2520 * Move the remaining entries down to fill the hole,
2521 * then zero the entries at the top.
2522 */
2523 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2524 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2525 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2526 memmove(entry_d, entry_s, tmp);
2527
2528 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2529 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2530 ASSERT(((char *)entry_s + tmp) <=
2531 ((char *)leaf_s + args->geo->blksize));
2532 memset(entry_s, 0, tmp);
2533 }
2534
2535 /*
2536 * Fill in the freemap information
2537 */
2538 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2539 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2540 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2541 ichdr_d->freemap[1].base = 0;
2542 ichdr_d->freemap[2].base = 0;
2543 ichdr_d->freemap[1].size = 0;
2544 ichdr_d->freemap[2].size = 0;
2545 ichdr_s->holes = 1; /* leaf may not be compact */
2546 }
2547
2548 /*
2549 * Pick up the last hashvalue from a leaf block.
2550 */
2551 xfs_dahash_t
2552 xfs_attr_leaf_lasthash(
2553 struct xfs_buf *bp,
2554 int *count)
2555 {
2556 struct xfs_attr3_icleaf_hdr ichdr;
2557 struct xfs_attr_leaf_entry *entries;
2558 struct xfs_mount *mp = bp->b_target->bt_mount;
2559
2560 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2561 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2562 if (count)
2563 *count = ichdr.count;
2564 if (!ichdr.count)
2565 return 0;
2566 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2567 }
2568
2569 /*
2570 * Calculate the number of bytes used to store the indicated attribute
2571 * (whether local or remote only calculate bytes in this block).
2572 */
2573 STATIC int
2574 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2575 {
2576 struct xfs_attr_leaf_entry *entries;
2577 xfs_attr_leaf_name_local_t *name_loc;
2578 xfs_attr_leaf_name_remote_t *name_rmt;
2579 int size;
2580
2581 entries = xfs_attr3_leaf_entryp(leaf);
2582 if (entries[index].flags & XFS_ATTR_LOCAL) {
2583 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2584 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2585 be16_to_cpu(name_loc->valuelen));
2586 } else {
2587 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2588 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2589 }
2590 return size;
2591 }
2592
2593 /*
2594 * Calculate the number of bytes that would be required to store the new
2595 * attribute (whether local or remote only calculate bytes in this block).
2596 * This routine decides as a side effect whether the attribute will be
2597 * a "local" or a "remote" attribute.
2598 */
2599 int
2600 xfs_attr_leaf_newentsize(
2601 struct xfs_da_args *args,
2602 int *local)
2603 {
2604 int size;
2605
2606 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2607 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2608 if (local)
2609 *local = 1;
2610 return size;
2611 }
2612 if (local)
2613 *local = 0;
2614 return xfs_attr_leaf_entsize_remote(args->namelen);
2615 }
2616
2617
2618 /*========================================================================
2619 * Manage the INCOMPLETE flag in a leaf entry
2620 *========================================================================*/
2621
2622 /*
2623 * Clear the INCOMPLETE flag on an entry in a leaf block.
2624 */
2625 int
2626 xfs_attr3_leaf_clearflag(
2627 struct xfs_da_args *args)
2628 {
2629 struct xfs_attr_leafblock *leaf;
2630 struct xfs_attr_leaf_entry *entry;
2631 struct xfs_attr_leaf_name_remote *name_rmt;
2632 struct xfs_buf *bp;
2633 int error;
2634 #ifdef DEBUG
2635 struct xfs_attr3_icleaf_hdr ichdr;
2636 xfs_attr_leaf_name_local_t *name_loc;
2637 int namelen;
2638 char *name;
2639 #endif /* DEBUG */
2640
2641 trace_xfs_attr_leaf_clearflag(args);
2642 /*
2643 * Set up the operation.
2644 */
2645 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2646 if (error)
2647 return error;
2648
2649 leaf = bp->b_addr;
2650 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2651 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2652
2653 #ifdef DEBUG
2654 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2655 ASSERT(args->index < ichdr.count);
2656 ASSERT(args->index >= 0);
2657
2658 if (entry->flags & XFS_ATTR_LOCAL) {
2659 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2660 namelen = name_loc->namelen;
2661 name = (char *)name_loc->nameval;
2662 } else {
2663 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2664 namelen = name_rmt->namelen;
2665 name = (char *)name_rmt->name;
2666 }
2667 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2668 ASSERT(namelen == args->namelen);
2669 ASSERT(memcmp(name, args->name, namelen) == 0);
2670 #endif /* DEBUG */
2671
2672 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2673 xfs_trans_log_buf(args->trans, bp,
2674 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2675
2676 if (args->rmtblkno) {
2677 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2678 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2679 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2680 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2681 xfs_trans_log_buf(args->trans, bp,
2682 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2683 }
2684
2685 /*
2686 * Commit the flag value change and start the next trans in series.
2687 */
2688 return xfs_trans_roll_inode(&args->trans, args->dp);
2689 }
2690
2691 /*
2692 * Set the INCOMPLETE flag on an entry in a leaf block.
2693 */
2694 int
2695 xfs_attr3_leaf_setflag(
2696 struct xfs_da_args *args)
2697 {
2698 struct xfs_attr_leafblock *leaf;
2699 struct xfs_attr_leaf_entry *entry;
2700 struct xfs_attr_leaf_name_remote *name_rmt;
2701 struct xfs_buf *bp;
2702 int error;
2703 #ifdef DEBUG
2704 struct xfs_attr3_icleaf_hdr ichdr;
2705 #endif
2706
2707 trace_xfs_attr_leaf_setflag(args);
2708
2709 /*
2710 * Set up the operation.
2711 */
2712 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2713 if (error)
2714 return error;
2715
2716 leaf = bp->b_addr;
2717 #ifdef DEBUG
2718 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2719 ASSERT(args->index < ichdr.count);
2720 ASSERT(args->index >= 0);
2721 #endif
2722 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2723
2724 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2725 entry->flags |= XFS_ATTR_INCOMPLETE;
2726 xfs_trans_log_buf(args->trans, bp,
2727 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2728 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2729 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2730 name_rmt->valueblk = 0;
2731 name_rmt->valuelen = 0;
2732 xfs_trans_log_buf(args->trans, bp,
2733 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2734 }
2735
2736 /*
2737 * Commit the flag value change and start the next trans in series.
2738 */
2739 return xfs_trans_roll_inode(&args->trans, args->dp);
2740 }
2741
2742 /*
2743 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2744 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2745 * entry given by args->blkno2/index2.
2746 *
2747 * Note that they could be in different blocks, or in the same block.
2748 */
2749 int
2750 xfs_attr3_leaf_flipflags(
2751 struct xfs_da_args *args)
2752 {
2753 struct xfs_attr_leafblock *leaf1;
2754 struct xfs_attr_leafblock *leaf2;
2755 struct xfs_attr_leaf_entry *entry1;
2756 struct xfs_attr_leaf_entry *entry2;
2757 struct xfs_attr_leaf_name_remote *name_rmt;
2758 struct xfs_buf *bp1;
2759 struct xfs_buf *bp2;
2760 int error;
2761 #ifdef DEBUG
2762 struct xfs_attr3_icleaf_hdr ichdr1;
2763 struct xfs_attr3_icleaf_hdr ichdr2;
2764 xfs_attr_leaf_name_local_t *name_loc;
2765 int namelen1, namelen2;
2766 char *name1, *name2;
2767 #endif /* DEBUG */
2768
2769 trace_xfs_attr_leaf_flipflags(args);
2770
2771 /*
2772 * Read the block containing the "old" attr
2773 */
2774 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2775 if (error)
2776 return error;
2777
2778 /*
2779 * Read the block containing the "new" attr, if it is different
2780 */
2781 if (args->blkno2 != args->blkno) {
2782 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2783 -1, &bp2);
2784 if (error)
2785 return error;
2786 } else {
2787 bp2 = bp1;
2788 }
2789
2790 leaf1 = bp1->b_addr;
2791 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2792
2793 leaf2 = bp2->b_addr;
2794 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2795
2796 #ifdef DEBUG
2797 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2798 ASSERT(args->index < ichdr1.count);
2799 ASSERT(args->index >= 0);
2800
2801 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2802 ASSERT(args->index2 < ichdr2.count);
2803 ASSERT(args->index2 >= 0);
2804
2805 if (entry1->flags & XFS_ATTR_LOCAL) {
2806 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2807 namelen1 = name_loc->namelen;
2808 name1 = (char *)name_loc->nameval;
2809 } else {
2810 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2811 namelen1 = name_rmt->namelen;
2812 name1 = (char *)name_rmt->name;
2813 }
2814 if (entry2->flags & XFS_ATTR_LOCAL) {
2815 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2816 namelen2 = name_loc->namelen;
2817 name2 = (char *)name_loc->nameval;
2818 } else {
2819 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2820 namelen2 = name_rmt->namelen;
2821 name2 = (char *)name_rmt->name;
2822 }
2823 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2824 ASSERT(namelen1 == namelen2);
2825 ASSERT(memcmp(name1, name2, namelen1) == 0);
2826 #endif /* DEBUG */
2827
2828 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2829 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2830
2831 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2832 xfs_trans_log_buf(args->trans, bp1,
2833 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2834 if (args->rmtblkno) {
2835 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2836 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2837 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2838 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2839 xfs_trans_log_buf(args->trans, bp1,
2840 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2841 }
2842
2843 entry2->flags |= XFS_ATTR_INCOMPLETE;
2844 xfs_trans_log_buf(args->trans, bp2,
2845 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2846 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2847 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2848 name_rmt->valueblk = 0;
2849 name_rmt->valuelen = 0;
2850 xfs_trans_log_buf(args->trans, bp2,
2851 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2852 }
2853
2854 /*
2855 * Commit the flag value change and start the next trans in series.
2856 */
2857 error = xfs_trans_roll_inode(&args->trans, args->dp);
2858
2859 return error;
2860 }