]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blob - libxfs/xfs_attr_leaf.c
c200084976f88cd6aa10250d8e600778d41f725c
[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 error = xfs_da_shrink_inode(args, 0, bp);
790 bp = NULL;
791 if (error)
792 goto out;
793 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
794 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
795 goto out;
796 }
797
798 memset((char *)&nargs, 0, sizeof(nargs));
799 nargs.dp = dp;
800 nargs.geo = args->geo;
801 nargs.firstblock = args->firstblock;
802 nargs.dfops = args->dfops;
803 nargs.total = args->total;
804 nargs.whichfork = XFS_ATTR_FORK;
805 nargs.trans = args->trans;
806 nargs.op_flags = XFS_DA_OP_OKNOENT;
807
808 sfe = &sf->list[0];
809 for (i = 0; i < sf->hdr.count; i++) {
810 nargs.name = sfe->nameval;
811 nargs.namelen = sfe->namelen;
812 nargs.value = &sfe->nameval[nargs.namelen];
813 nargs.valuelen = sfe->valuelen;
814 nargs.hashval = xfs_da_hashname(sfe->nameval,
815 sfe->namelen);
816 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
817 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
818 ASSERT(error == -ENOATTR);
819 error = xfs_attr3_leaf_add(bp, &nargs);
820 ASSERT(error != -ENOSPC);
821 if (error)
822 goto out;
823 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
824 }
825 error = 0;
826 *leaf_bp = bp;
827 out:
828 kmem_free(tmpbuffer);
829 return error;
830 }
831
832 /*
833 * Check a leaf attribute block to see if all the entries would fit into
834 * a shortform attribute list.
835 */
836 int
837 xfs_attr_shortform_allfit(
838 struct xfs_buf *bp,
839 struct xfs_inode *dp)
840 {
841 struct xfs_attr_leafblock *leaf;
842 struct xfs_attr_leaf_entry *entry;
843 xfs_attr_leaf_name_local_t *name_loc;
844 struct xfs_attr3_icleaf_hdr leafhdr;
845 int bytes;
846 int i;
847 struct xfs_mount *mp = bp->b_target->bt_mount;
848
849 leaf = bp->b_addr;
850 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
851 entry = xfs_attr3_leaf_entryp(leaf);
852
853 bytes = sizeof(struct xfs_attr_sf_hdr);
854 for (i = 0; i < leafhdr.count; entry++, i++) {
855 if (entry->flags & XFS_ATTR_INCOMPLETE)
856 continue; /* don't copy partial entries */
857 if (!(entry->flags & XFS_ATTR_LOCAL))
858 return 0;
859 name_loc = xfs_attr3_leaf_name_local(leaf, i);
860 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
861 return 0;
862 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
863 return 0;
864 bytes += sizeof(struct xfs_attr_sf_entry) - 1
865 + name_loc->namelen
866 + be16_to_cpu(name_loc->valuelen);
867 }
868 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
869 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
870 (bytes == sizeof(struct xfs_attr_sf_hdr)))
871 return -1;
872 return xfs_attr_shortform_bytesfit(dp, bytes);
873 }
874
875 /* Verify the consistency of an inline attribute fork. */
876 xfs_failaddr_t
877 xfs_attr_shortform_verify(
878 struct xfs_inode *ip)
879 {
880 struct xfs_attr_shortform *sfp;
881 struct xfs_attr_sf_entry *sfep;
882 struct xfs_attr_sf_entry *next_sfep;
883 char *endp;
884 struct xfs_ifork *ifp;
885 int i;
886 int size;
887
888 ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL);
889 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
890 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
891 size = ifp->if_bytes;
892
893 /*
894 * Give up if the attribute is way too short.
895 */
896 if (size < sizeof(struct xfs_attr_sf_hdr))
897 return __this_address;
898
899 endp = (char *)sfp + size;
900
901 /* Check all reported entries */
902 sfep = &sfp->list[0];
903 for (i = 0; i < sfp->hdr.count; i++) {
904 /*
905 * struct xfs_attr_sf_entry has a variable length.
906 * Check the fixed-offset parts of the structure are
907 * within the data buffer.
908 */
909 if (((char *)sfep + sizeof(*sfep)) >= endp)
910 return __this_address;
911
912 /* Don't allow names with known bad length. */
913 if (sfep->namelen == 0)
914 return __this_address;
915
916 /*
917 * Check that the variable-length part of the structure is
918 * within the data buffer. The next entry starts after the
919 * name component, so nextentry is an acceptable test.
920 */
921 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
922 if ((char *)next_sfep > endp)
923 return __this_address;
924
925 /*
926 * Check for unknown flags. Short form doesn't support
927 * the incomplete or local bits, so we can use the namespace
928 * mask here.
929 */
930 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
931 return __this_address;
932
933 /*
934 * Check for invalid namespace combinations. We only allow
935 * one namespace flag per xattr, so we can just count the
936 * bits (i.e. hweight) here.
937 */
938 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
939 return __this_address;
940
941 sfep = next_sfep;
942 }
943 if ((void *)sfep != (void *)endp)
944 return __this_address;
945
946 return NULL;
947 }
948
949 /*
950 * Convert a leaf attribute list to shortform attribute list
951 */
952 int
953 xfs_attr3_leaf_to_shortform(
954 struct xfs_buf *bp,
955 struct xfs_da_args *args,
956 int forkoff)
957 {
958 struct xfs_attr_leafblock *leaf;
959 struct xfs_attr3_icleaf_hdr ichdr;
960 struct xfs_attr_leaf_entry *entry;
961 struct xfs_attr_leaf_name_local *name_loc;
962 struct xfs_da_args nargs;
963 struct xfs_inode *dp = args->dp;
964 char *tmpbuffer;
965 int error;
966 int i;
967
968 trace_xfs_attr_leaf_to_sf(args);
969
970 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
971 if (!tmpbuffer)
972 return -ENOMEM;
973
974 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
975
976 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
977 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
978 entry = xfs_attr3_leaf_entryp(leaf);
979
980 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
981 memset(bp->b_addr, 0, args->geo->blksize);
982
983 /*
984 * Clean out the prior contents of the attribute list.
985 */
986 error = xfs_da_shrink_inode(args, 0, bp);
987 if (error)
988 goto out;
989
990 if (forkoff == -1) {
991 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
992 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
993 xfs_attr_fork_remove(dp, args->trans);
994 goto out;
995 }
996
997 xfs_attr_shortform_create(args);
998
999 /*
1000 * Copy the attributes
1001 */
1002 memset((char *)&nargs, 0, sizeof(nargs));
1003 nargs.geo = args->geo;
1004 nargs.dp = dp;
1005 nargs.firstblock = args->firstblock;
1006 nargs.dfops = args->dfops;
1007 nargs.total = args->total;
1008 nargs.whichfork = XFS_ATTR_FORK;
1009 nargs.trans = args->trans;
1010 nargs.op_flags = XFS_DA_OP_OKNOENT;
1011
1012 for (i = 0; i < ichdr.count; entry++, i++) {
1013 if (entry->flags & XFS_ATTR_INCOMPLETE)
1014 continue; /* don't copy partial entries */
1015 if (!entry->nameidx)
1016 continue;
1017 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1018 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1019 nargs.name = name_loc->nameval;
1020 nargs.namelen = name_loc->namelen;
1021 nargs.value = &name_loc->nameval[nargs.namelen];
1022 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1023 nargs.hashval = be32_to_cpu(entry->hashval);
1024 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1025 xfs_attr_shortform_add(&nargs, forkoff);
1026 }
1027 error = 0;
1028
1029 out:
1030 kmem_free(tmpbuffer);
1031 return error;
1032 }
1033
1034 /*
1035 * Convert from using a single leaf to a root node and a leaf.
1036 */
1037 int
1038 xfs_attr3_leaf_to_node(
1039 struct xfs_da_args *args)
1040 {
1041 struct xfs_attr_leafblock *leaf;
1042 struct xfs_attr3_icleaf_hdr icleafhdr;
1043 struct xfs_attr_leaf_entry *entries;
1044 struct xfs_da_node_entry *btree;
1045 struct xfs_da3_icnode_hdr icnodehdr;
1046 struct xfs_da_intnode *node;
1047 struct xfs_inode *dp = args->dp;
1048 struct xfs_mount *mp = dp->i_mount;
1049 struct xfs_buf *bp1 = NULL;
1050 struct xfs_buf *bp2 = NULL;
1051 xfs_dablk_t blkno;
1052 int error;
1053
1054 trace_xfs_attr_leaf_to_node(args);
1055
1056 error = xfs_da_grow_inode(args, &blkno);
1057 if (error)
1058 goto out;
1059 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1060 if (error)
1061 goto out;
1062
1063 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1064 if (error)
1065 goto out;
1066
1067 /* copy leaf to new buffer, update identifiers */
1068 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1069 bp2->b_ops = bp1->b_ops;
1070 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1071 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1072 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1073 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1074 }
1075 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1076
1077 /*
1078 * Set up the new root node.
1079 */
1080 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1081 if (error)
1082 goto out;
1083 node = bp1->b_addr;
1084 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1085 btree = dp->d_ops->node_tree_p(node);
1086
1087 leaf = bp2->b_addr;
1088 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1089 entries = xfs_attr3_leaf_entryp(leaf);
1090
1091 /* both on-disk, don't endian-flip twice */
1092 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1093 btree[0].before = cpu_to_be32(blkno);
1094 icnodehdr.count = 1;
1095 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1096 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1097 error = 0;
1098 out:
1099 return error;
1100 }
1101
1102 /*========================================================================
1103 * Routines used for growing the Btree.
1104 *========================================================================*/
1105
1106 /*
1107 * Create the initial contents of a leaf attribute list
1108 * or a leaf in a node attribute list.
1109 */
1110 STATIC int
1111 xfs_attr3_leaf_create(
1112 struct xfs_da_args *args,
1113 xfs_dablk_t blkno,
1114 struct xfs_buf **bpp)
1115 {
1116 struct xfs_attr_leafblock *leaf;
1117 struct xfs_attr3_icleaf_hdr ichdr;
1118 struct xfs_inode *dp = args->dp;
1119 struct xfs_mount *mp = dp->i_mount;
1120 struct xfs_buf *bp;
1121 int error;
1122
1123 trace_xfs_attr_leaf_create(args);
1124
1125 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1126 XFS_ATTR_FORK);
1127 if (error)
1128 return error;
1129 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1130 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1131 leaf = bp->b_addr;
1132 memset(leaf, 0, args->geo->blksize);
1133
1134 memset(&ichdr, 0, sizeof(ichdr));
1135 ichdr.firstused = args->geo->blksize;
1136
1137 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1138 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1139
1140 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1141
1142 hdr3->blkno = cpu_to_be64(bp->b_bn);
1143 hdr3->owner = cpu_to_be64(dp->i_ino);
1144 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1145
1146 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1147 } else {
1148 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1149 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1150 }
1151 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1152
1153 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1154 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1155
1156 *bpp = bp;
1157 return 0;
1158 }
1159
1160 /*
1161 * Split the leaf node, rebalance, then add the new entry.
1162 */
1163 int
1164 xfs_attr3_leaf_split(
1165 struct xfs_da_state *state,
1166 struct xfs_da_state_blk *oldblk,
1167 struct xfs_da_state_blk *newblk)
1168 {
1169 xfs_dablk_t blkno;
1170 int error;
1171
1172 trace_xfs_attr_leaf_split(state->args);
1173
1174 /*
1175 * Allocate space for a new leaf node.
1176 */
1177 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1178 error = xfs_da_grow_inode(state->args, &blkno);
1179 if (error)
1180 return error;
1181 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1182 if (error)
1183 return error;
1184 newblk->blkno = blkno;
1185 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1186
1187 /*
1188 * Rebalance the entries across the two leaves.
1189 * NOTE: rebalance() currently depends on the 2nd block being empty.
1190 */
1191 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1192 error = xfs_da3_blk_link(state, oldblk, newblk);
1193 if (error)
1194 return error;
1195
1196 /*
1197 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1198 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1199 * "new" attrs info. Will need the "old" info to remove it later.
1200 *
1201 * Insert the "new" entry in the correct block.
1202 */
1203 if (state->inleaf) {
1204 trace_xfs_attr_leaf_add_old(state->args);
1205 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1206 } else {
1207 trace_xfs_attr_leaf_add_new(state->args);
1208 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1209 }
1210
1211 /*
1212 * Update last hashval in each block since we added the name.
1213 */
1214 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1215 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1216 return error;
1217 }
1218
1219 /*
1220 * Add a name to the leaf attribute list structure.
1221 */
1222 int
1223 xfs_attr3_leaf_add(
1224 struct xfs_buf *bp,
1225 struct xfs_da_args *args)
1226 {
1227 struct xfs_attr_leafblock *leaf;
1228 struct xfs_attr3_icleaf_hdr ichdr;
1229 int tablesize;
1230 int entsize;
1231 int sum;
1232 int tmp;
1233 int i;
1234
1235 trace_xfs_attr_leaf_add(args);
1236
1237 leaf = bp->b_addr;
1238 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1239 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1240 entsize = xfs_attr_leaf_newentsize(args, NULL);
1241
1242 /*
1243 * Search through freemap for first-fit on new name length.
1244 * (may need to figure in size of entry struct too)
1245 */
1246 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1247 + xfs_attr3_leaf_hdr_size(leaf);
1248 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1249 if (tablesize > ichdr.firstused) {
1250 sum += ichdr.freemap[i].size;
1251 continue;
1252 }
1253 if (!ichdr.freemap[i].size)
1254 continue; /* no space in this map */
1255 tmp = entsize;
1256 if (ichdr.freemap[i].base < ichdr.firstused)
1257 tmp += sizeof(xfs_attr_leaf_entry_t);
1258 if (ichdr.freemap[i].size >= tmp) {
1259 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1260 goto out_log_hdr;
1261 }
1262 sum += ichdr.freemap[i].size;
1263 }
1264
1265 /*
1266 * If there are no holes in the address space of the block,
1267 * and we don't have enough freespace, then compaction will do us
1268 * no good and we should just give up.
1269 */
1270 if (!ichdr.holes && sum < entsize)
1271 return -ENOSPC;
1272
1273 /*
1274 * Compact the entries to coalesce free space.
1275 * This may change the hdr->count via dropping INCOMPLETE entries.
1276 */
1277 xfs_attr3_leaf_compact(args, &ichdr, bp);
1278
1279 /*
1280 * After compaction, the block is guaranteed to have only one
1281 * free region, in freemap[0]. If it is not big enough, give up.
1282 */
1283 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1284 tmp = -ENOSPC;
1285 goto out_log_hdr;
1286 }
1287
1288 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1289
1290 out_log_hdr:
1291 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1292 xfs_trans_log_buf(args->trans, bp,
1293 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1294 xfs_attr3_leaf_hdr_size(leaf)));
1295 return tmp;
1296 }
1297
1298 /*
1299 * Add a name to a leaf attribute list structure.
1300 */
1301 STATIC int
1302 xfs_attr3_leaf_add_work(
1303 struct xfs_buf *bp,
1304 struct xfs_attr3_icleaf_hdr *ichdr,
1305 struct xfs_da_args *args,
1306 int mapindex)
1307 {
1308 struct xfs_attr_leafblock *leaf;
1309 struct xfs_attr_leaf_entry *entry;
1310 struct xfs_attr_leaf_name_local *name_loc;
1311 struct xfs_attr_leaf_name_remote *name_rmt;
1312 struct xfs_mount *mp;
1313 int tmp;
1314 int i;
1315
1316 trace_xfs_attr_leaf_add_work(args);
1317
1318 leaf = bp->b_addr;
1319 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1320 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1321
1322 /*
1323 * Force open some space in the entry array and fill it in.
1324 */
1325 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1326 if (args->index < ichdr->count) {
1327 tmp = ichdr->count - args->index;
1328 tmp *= sizeof(xfs_attr_leaf_entry_t);
1329 memmove(entry + 1, entry, tmp);
1330 xfs_trans_log_buf(args->trans, bp,
1331 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1332 }
1333 ichdr->count++;
1334
1335 /*
1336 * Allocate space for the new string (at the end of the run).
1337 */
1338 mp = args->trans->t_mountp;
1339 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1340 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1341 ASSERT(ichdr->freemap[mapindex].size >=
1342 xfs_attr_leaf_newentsize(args, NULL));
1343 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1344 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1345
1346 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1347
1348 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1349 ichdr->freemap[mapindex].size);
1350 entry->hashval = cpu_to_be32(args->hashval);
1351 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1352 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1353 if (args->op_flags & XFS_DA_OP_RENAME) {
1354 entry->flags |= XFS_ATTR_INCOMPLETE;
1355 if ((args->blkno2 == args->blkno) &&
1356 (args->index2 <= args->index)) {
1357 args->index2++;
1358 }
1359 }
1360 xfs_trans_log_buf(args->trans, bp,
1361 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1362 ASSERT((args->index == 0) ||
1363 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1364 ASSERT((args->index == ichdr->count - 1) ||
1365 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1366
1367 /*
1368 * For "remote" attribute values, simply note that we need to
1369 * allocate space for the "remote" value. We can't actually
1370 * allocate the extents in this transaction, and we can't decide
1371 * which blocks they should be as we might allocate more blocks
1372 * as part of this transaction (a split operation for example).
1373 */
1374 if (entry->flags & XFS_ATTR_LOCAL) {
1375 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1376 name_loc->namelen = args->namelen;
1377 name_loc->valuelen = cpu_to_be16(args->valuelen);
1378 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1379 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1380 be16_to_cpu(name_loc->valuelen));
1381 } else {
1382 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1383 name_rmt->namelen = args->namelen;
1384 memcpy((char *)name_rmt->name, args->name, args->namelen);
1385 entry->flags |= XFS_ATTR_INCOMPLETE;
1386 /* just in case */
1387 name_rmt->valuelen = 0;
1388 name_rmt->valueblk = 0;
1389 args->rmtblkno = 1;
1390 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1391 args->rmtvaluelen = args->valuelen;
1392 }
1393 xfs_trans_log_buf(args->trans, bp,
1394 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1395 xfs_attr_leaf_entsize(leaf, args->index)));
1396
1397 /*
1398 * Update the control info for this leaf node
1399 */
1400 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1401 ichdr->firstused = be16_to_cpu(entry->nameidx);
1402
1403 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1404 + xfs_attr3_leaf_hdr_size(leaf));
1405 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1406 + xfs_attr3_leaf_hdr_size(leaf);
1407
1408 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1409 if (ichdr->freemap[i].base == tmp) {
1410 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1411 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1412 }
1413 }
1414 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1415 return 0;
1416 }
1417
1418 /*
1419 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1420 */
1421 STATIC void
1422 xfs_attr3_leaf_compact(
1423 struct xfs_da_args *args,
1424 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1425 struct xfs_buf *bp)
1426 {
1427 struct xfs_attr_leafblock *leaf_src;
1428 struct xfs_attr_leafblock *leaf_dst;
1429 struct xfs_attr3_icleaf_hdr ichdr_src;
1430 struct xfs_trans *trans = args->trans;
1431 char *tmpbuffer;
1432
1433 trace_xfs_attr_leaf_compact(args);
1434
1435 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1436 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1437 memset(bp->b_addr, 0, args->geo->blksize);
1438 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1439 leaf_dst = bp->b_addr;
1440
1441 /*
1442 * Copy the on-disk header back into the destination buffer to ensure
1443 * all the information in the header that is not part of the incore
1444 * header structure is preserved.
1445 */
1446 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1447
1448 /* Initialise the incore headers */
1449 ichdr_src = *ichdr_dst; /* struct copy */
1450 ichdr_dst->firstused = args->geo->blksize;
1451 ichdr_dst->usedbytes = 0;
1452 ichdr_dst->count = 0;
1453 ichdr_dst->holes = 0;
1454 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1455 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1456 ichdr_dst->freemap[0].base;
1457
1458 /* write the header back to initialise the underlying buffer */
1459 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1460
1461 /*
1462 * Copy all entry's in the same (sorted) order,
1463 * but allocate name/value pairs packed and in sequence.
1464 */
1465 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1466 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1467 /*
1468 * this logs the entire buffer, but the caller must write the header
1469 * back to the buffer when it is finished modifying it.
1470 */
1471 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1472
1473 kmem_free(tmpbuffer);
1474 }
1475
1476 /*
1477 * Compare two leaf blocks "order".
1478 * Return 0 unless leaf2 should go before leaf1.
1479 */
1480 static int
1481 xfs_attr3_leaf_order(
1482 struct xfs_buf *leaf1_bp,
1483 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1484 struct xfs_buf *leaf2_bp,
1485 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1486 {
1487 struct xfs_attr_leaf_entry *entries1;
1488 struct xfs_attr_leaf_entry *entries2;
1489
1490 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1491 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1492 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1493 ((be32_to_cpu(entries2[0].hashval) <
1494 be32_to_cpu(entries1[0].hashval)) ||
1495 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1496 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1497 return 1;
1498 }
1499 return 0;
1500 }
1501
1502 int
1503 xfs_attr_leaf_order(
1504 struct xfs_buf *leaf1_bp,
1505 struct xfs_buf *leaf2_bp)
1506 {
1507 struct xfs_attr3_icleaf_hdr ichdr1;
1508 struct xfs_attr3_icleaf_hdr ichdr2;
1509 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1510
1511 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1512 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1513 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1514 }
1515
1516 /*
1517 * Redistribute the attribute list entries between two leaf nodes,
1518 * taking into account the size of the new entry.
1519 *
1520 * NOTE: if new block is empty, then it will get the upper half of the
1521 * old block. At present, all (one) callers pass in an empty second block.
1522 *
1523 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1524 * to match what it is doing in splitting the attribute leaf block. Those
1525 * values are used in "atomic rename" operations on attributes. Note that
1526 * the "new" and "old" values can end up in different blocks.
1527 */
1528 STATIC void
1529 xfs_attr3_leaf_rebalance(
1530 struct xfs_da_state *state,
1531 struct xfs_da_state_blk *blk1,
1532 struct xfs_da_state_blk *blk2)
1533 {
1534 struct xfs_da_args *args;
1535 struct xfs_attr_leafblock *leaf1;
1536 struct xfs_attr_leafblock *leaf2;
1537 struct xfs_attr3_icleaf_hdr ichdr1;
1538 struct xfs_attr3_icleaf_hdr ichdr2;
1539 struct xfs_attr_leaf_entry *entries1;
1540 struct xfs_attr_leaf_entry *entries2;
1541 int count;
1542 int totallen;
1543 int max;
1544 int space;
1545 int swap;
1546
1547 /*
1548 * Set up environment.
1549 */
1550 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1551 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1552 leaf1 = blk1->bp->b_addr;
1553 leaf2 = blk2->bp->b_addr;
1554 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1555 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1556 ASSERT(ichdr2.count == 0);
1557 args = state->args;
1558
1559 trace_xfs_attr_leaf_rebalance(args);
1560
1561 /*
1562 * Check ordering of blocks, reverse if it makes things simpler.
1563 *
1564 * NOTE: Given that all (current) callers pass in an empty
1565 * second block, this code should never set "swap".
1566 */
1567 swap = 0;
1568 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1569 struct xfs_da_state_blk *tmp_blk;
1570 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1571
1572 tmp_blk = blk1;
1573 blk1 = blk2;
1574 blk2 = tmp_blk;
1575
1576 /* struct copies to swap them rather than reconverting */
1577 tmp_ichdr = ichdr1;
1578 ichdr1 = ichdr2;
1579 ichdr2 = tmp_ichdr;
1580
1581 leaf1 = blk1->bp->b_addr;
1582 leaf2 = blk2->bp->b_addr;
1583 swap = 1;
1584 }
1585
1586 /*
1587 * Examine entries until we reduce the absolute difference in
1588 * byte usage between the two blocks to a minimum. Then get
1589 * the direction to copy and the number of elements to move.
1590 *
1591 * "inleaf" is true if the new entry should be inserted into blk1.
1592 * If "swap" is also true, then reverse the sense of "inleaf".
1593 */
1594 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1595 blk2, &ichdr2,
1596 &count, &totallen);
1597 if (swap)
1598 state->inleaf = !state->inleaf;
1599
1600 /*
1601 * Move any entries required from leaf to leaf:
1602 */
1603 if (count < ichdr1.count) {
1604 /*
1605 * Figure the total bytes to be added to the destination leaf.
1606 */
1607 /* number entries being moved */
1608 count = ichdr1.count - count;
1609 space = ichdr1.usedbytes - totallen;
1610 space += count * sizeof(xfs_attr_leaf_entry_t);
1611
1612 /*
1613 * leaf2 is the destination, compact it if it looks tight.
1614 */
1615 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1616 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1617 if (space > max)
1618 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1619
1620 /*
1621 * Move high entries from leaf1 to low end of leaf2.
1622 */
1623 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1624 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1625
1626 } else if (count > ichdr1.count) {
1627 /*
1628 * I assert that since all callers pass in an empty
1629 * second buffer, this code should never execute.
1630 */
1631 ASSERT(0);
1632
1633 /*
1634 * Figure the total bytes to be added to the destination leaf.
1635 */
1636 /* number entries being moved */
1637 count -= ichdr1.count;
1638 space = totallen - ichdr1.usedbytes;
1639 space += count * sizeof(xfs_attr_leaf_entry_t);
1640
1641 /*
1642 * leaf1 is the destination, compact it if it looks tight.
1643 */
1644 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1645 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1646 if (space > max)
1647 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1648
1649 /*
1650 * Move low entries from leaf2 to high end of leaf1.
1651 */
1652 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1653 ichdr1.count, count);
1654 }
1655
1656 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1657 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1658 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1659 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1660
1661 /*
1662 * Copy out last hashval in each block for B-tree code.
1663 */
1664 entries1 = xfs_attr3_leaf_entryp(leaf1);
1665 entries2 = xfs_attr3_leaf_entryp(leaf2);
1666 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1667 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1668
1669 /*
1670 * Adjust the expected index for insertion.
1671 * NOTE: this code depends on the (current) situation that the
1672 * second block was originally empty.
1673 *
1674 * If the insertion point moved to the 2nd block, we must adjust
1675 * the index. We must also track the entry just following the
1676 * new entry for use in an "atomic rename" operation, that entry
1677 * is always the "old" entry and the "new" entry is what we are
1678 * inserting. The index/blkno fields refer to the "old" entry,
1679 * while the index2/blkno2 fields refer to the "new" entry.
1680 */
1681 if (blk1->index > ichdr1.count) {
1682 ASSERT(state->inleaf == 0);
1683 blk2->index = blk1->index - ichdr1.count;
1684 args->index = args->index2 = blk2->index;
1685 args->blkno = args->blkno2 = blk2->blkno;
1686 } else if (blk1->index == ichdr1.count) {
1687 if (state->inleaf) {
1688 args->index = blk1->index;
1689 args->blkno = blk1->blkno;
1690 args->index2 = 0;
1691 args->blkno2 = blk2->blkno;
1692 } else {
1693 /*
1694 * On a double leaf split, the original attr location
1695 * is already stored in blkno2/index2, so don't
1696 * overwrite it overwise we corrupt the tree.
1697 */
1698 blk2->index = blk1->index - ichdr1.count;
1699 args->index = blk2->index;
1700 args->blkno = blk2->blkno;
1701 if (!state->extravalid) {
1702 /*
1703 * set the new attr location to match the old
1704 * one and let the higher level split code
1705 * decide where in the leaf to place it.
1706 */
1707 args->index2 = blk2->index;
1708 args->blkno2 = blk2->blkno;
1709 }
1710 }
1711 } else {
1712 ASSERT(state->inleaf == 1);
1713 args->index = args->index2 = blk1->index;
1714 args->blkno = args->blkno2 = blk1->blkno;
1715 }
1716 }
1717
1718 /*
1719 * Examine entries until we reduce the absolute difference in
1720 * byte usage between the two blocks to a minimum.
1721 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1722 * GROT: there will always be enough room in either block for a new entry.
1723 * GROT: Do a double-split for this case?
1724 */
1725 STATIC int
1726 xfs_attr3_leaf_figure_balance(
1727 struct xfs_da_state *state,
1728 struct xfs_da_state_blk *blk1,
1729 struct xfs_attr3_icleaf_hdr *ichdr1,
1730 struct xfs_da_state_blk *blk2,
1731 struct xfs_attr3_icleaf_hdr *ichdr2,
1732 int *countarg,
1733 int *usedbytesarg)
1734 {
1735 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1736 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1737 struct xfs_attr_leaf_entry *entry;
1738 int count;
1739 int max;
1740 int index;
1741 int totallen = 0;
1742 int half;
1743 int lastdelta;
1744 int foundit = 0;
1745 int tmp;
1746
1747 /*
1748 * Examine entries until we reduce the absolute difference in
1749 * byte usage between the two blocks to a minimum.
1750 */
1751 max = ichdr1->count + ichdr2->count;
1752 half = (max + 1) * sizeof(*entry);
1753 half += ichdr1->usedbytes + ichdr2->usedbytes +
1754 xfs_attr_leaf_newentsize(state->args, NULL);
1755 half /= 2;
1756 lastdelta = state->args->geo->blksize;
1757 entry = xfs_attr3_leaf_entryp(leaf1);
1758 for (count = index = 0; count < max; entry++, index++, count++) {
1759
1760 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1761 /*
1762 * The new entry is in the first block, account for it.
1763 */
1764 if (count == blk1->index) {
1765 tmp = totallen + sizeof(*entry) +
1766 xfs_attr_leaf_newentsize(state->args, NULL);
1767 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1768 break;
1769 lastdelta = XFS_ATTR_ABS(half - tmp);
1770 totallen = tmp;
1771 foundit = 1;
1772 }
1773
1774 /*
1775 * Wrap around into the second block if necessary.
1776 */
1777 if (count == ichdr1->count) {
1778 leaf1 = leaf2;
1779 entry = xfs_attr3_leaf_entryp(leaf1);
1780 index = 0;
1781 }
1782
1783 /*
1784 * Figure out if next leaf entry would be too much.
1785 */
1786 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1787 index);
1788 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1789 break;
1790 lastdelta = XFS_ATTR_ABS(half - tmp);
1791 totallen = tmp;
1792 #undef XFS_ATTR_ABS
1793 }
1794
1795 /*
1796 * Calculate the number of usedbytes that will end up in lower block.
1797 * If new entry not in lower block, fix up the count.
1798 */
1799 totallen -= count * sizeof(*entry);
1800 if (foundit) {
1801 totallen -= sizeof(*entry) +
1802 xfs_attr_leaf_newentsize(state->args, NULL);
1803 }
1804
1805 *countarg = count;
1806 *usedbytesarg = totallen;
1807 return foundit;
1808 }
1809
1810 /*========================================================================
1811 * Routines used for shrinking the Btree.
1812 *========================================================================*/
1813
1814 /*
1815 * Check a leaf block and its neighbors to see if the block should be
1816 * collapsed into one or the other neighbor. Always keep the block
1817 * with the smaller block number.
1818 * If the current block is over 50% full, don't try to join it, return 0.
1819 * If the block is empty, fill in the state structure and return 2.
1820 * If it can be collapsed, fill in the state structure and return 1.
1821 * If nothing can be done, return 0.
1822 *
1823 * GROT: allow for INCOMPLETE entries in calculation.
1824 */
1825 int
1826 xfs_attr3_leaf_toosmall(
1827 struct xfs_da_state *state,
1828 int *action)
1829 {
1830 struct xfs_attr_leafblock *leaf;
1831 struct xfs_da_state_blk *blk;
1832 struct xfs_attr3_icleaf_hdr ichdr;
1833 struct xfs_buf *bp;
1834 xfs_dablk_t blkno;
1835 int bytes;
1836 int forward;
1837 int error;
1838 int retval;
1839 int i;
1840
1841 trace_xfs_attr_leaf_toosmall(state->args);
1842
1843 /*
1844 * Check for the degenerate case of the block being over 50% full.
1845 * If so, it's not worth even looking to see if we might be able
1846 * to coalesce with a sibling.
1847 */
1848 blk = &state->path.blk[ state->path.active-1 ];
1849 leaf = blk->bp->b_addr;
1850 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1851 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1852 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1853 ichdr.usedbytes;
1854 if (bytes > (state->args->geo->blksize >> 1)) {
1855 *action = 0; /* blk over 50%, don't try to join */
1856 return 0;
1857 }
1858
1859 /*
1860 * Check for the degenerate case of the block being empty.
1861 * If the block is empty, we'll simply delete it, no need to
1862 * coalesce it with a sibling block. We choose (arbitrarily)
1863 * to merge with the forward block unless it is NULL.
1864 */
1865 if (ichdr.count == 0) {
1866 /*
1867 * Make altpath point to the block we want to keep and
1868 * path point to the block we want to drop (this one).
1869 */
1870 forward = (ichdr.forw != 0);
1871 memcpy(&state->altpath, &state->path, sizeof(state->path));
1872 error = xfs_da3_path_shift(state, &state->altpath, forward,
1873 0, &retval);
1874 if (error)
1875 return error;
1876 if (retval) {
1877 *action = 0;
1878 } else {
1879 *action = 2;
1880 }
1881 return 0;
1882 }
1883
1884 /*
1885 * Examine each sibling block to see if we can coalesce with
1886 * at least 25% free space to spare. We need to figure out
1887 * whether to merge with the forward or the backward block.
1888 * We prefer coalescing with the lower numbered sibling so as
1889 * to shrink an attribute list over time.
1890 */
1891 /* start with smaller blk num */
1892 forward = ichdr.forw < ichdr.back;
1893 for (i = 0; i < 2; forward = !forward, i++) {
1894 struct xfs_attr3_icleaf_hdr ichdr2;
1895 if (forward)
1896 blkno = ichdr.forw;
1897 else
1898 blkno = ichdr.back;
1899 if (blkno == 0)
1900 continue;
1901 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1902 blkno, -1, &bp);
1903 if (error)
1904 return error;
1905
1906 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1907
1908 bytes = state->args->geo->blksize -
1909 (state->args->geo->blksize >> 2) -
1910 ichdr.usedbytes - ichdr2.usedbytes -
1911 ((ichdr.count + ichdr2.count) *
1912 sizeof(xfs_attr_leaf_entry_t)) -
1913 xfs_attr3_leaf_hdr_size(leaf);
1914
1915 xfs_trans_brelse(state->args->trans, bp);
1916 if (bytes >= 0)
1917 break; /* fits with at least 25% to spare */
1918 }
1919 if (i >= 2) {
1920 *action = 0;
1921 return 0;
1922 }
1923
1924 /*
1925 * Make altpath point to the block we want to keep (the lower
1926 * numbered block) and path point to the block we want to drop.
1927 */
1928 memcpy(&state->altpath, &state->path, sizeof(state->path));
1929 if (blkno < blk->blkno) {
1930 error = xfs_da3_path_shift(state, &state->altpath, forward,
1931 0, &retval);
1932 } else {
1933 error = xfs_da3_path_shift(state, &state->path, forward,
1934 0, &retval);
1935 }
1936 if (error)
1937 return error;
1938 if (retval) {
1939 *action = 0;
1940 } else {
1941 *action = 1;
1942 }
1943 return 0;
1944 }
1945
1946 /*
1947 * Remove a name from the leaf attribute list structure.
1948 *
1949 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1950 * If two leaves are 37% full, when combined they will leave 25% free.
1951 */
1952 int
1953 xfs_attr3_leaf_remove(
1954 struct xfs_buf *bp,
1955 struct xfs_da_args *args)
1956 {
1957 struct xfs_attr_leafblock *leaf;
1958 struct xfs_attr3_icleaf_hdr ichdr;
1959 struct xfs_attr_leaf_entry *entry;
1960 int before;
1961 int after;
1962 int smallest;
1963 int entsize;
1964 int tablesize;
1965 int tmp;
1966 int i;
1967
1968 trace_xfs_attr_leaf_remove(args);
1969
1970 leaf = bp->b_addr;
1971 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1972
1973 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1974 ASSERT(args->index >= 0 && args->index < ichdr.count);
1975 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1976 xfs_attr3_leaf_hdr_size(leaf));
1977
1978 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1979
1980 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1981 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1982
1983 /*
1984 * Scan through free region table:
1985 * check for adjacency of free'd entry with an existing one,
1986 * find smallest free region in case we need to replace it,
1987 * adjust any map that borders the entry table,
1988 */
1989 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1990 + xfs_attr3_leaf_hdr_size(leaf);
1991 tmp = ichdr.freemap[0].size;
1992 before = after = -1;
1993 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1994 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1995 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1996 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1997 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1998 if (ichdr.freemap[i].base == tablesize) {
1999 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2000 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2001 }
2002
2003 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2004 be16_to_cpu(entry->nameidx)) {
2005 before = i;
2006 } else if (ichdr.freemap[i].base ==
2007 (be16_to_cpu(entry->nameidx) + entsize)) {
2008 after = i;
2009 } else if (ichdr.freemap[i].size < tmp) {
2010 tmp = ichdr.freemap[i].size;
2011 smallest = i;
2012 }
2013 }
2014
2015 /*
2016 * Coalesce adjacent freemap regions,
2017 * or replace the smallest region.
2018 */
2019 if ((before >= 0) || (after >= 0)) {
2020 if ((before >= 0) && (after >= 0)) {
2021 ichdr.freemap[before].size += entsize;
2022 ichdr.freemap[before].size += ichdr.freemap[after].size;
2023 ichdr.freemap[after].base = 0;
2024 ichdr.freemap[after].size = 0;
2025 } else if (before >= 0) {
2026 ichdr.freemap[before].size += entsize;
2027 } else {
2028 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2029 ichdr.freemap[after].size += entsize;
2030 }
2031 } else {
2032 /*
2033 * Replace smallest region (if it is smaller than free'd entry)
2034 */
2035 if (ichdr.freemap[smallest].size < entsize) {
2036 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2037 ichdr.freemap[smallest].size = entsize;
2038 }
2039 }
2040
2041 /*
2042 * Did we remove the first entry?
2043 */
2044 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2045 smallest = 1;
2046 else
2047 smallest = 0;
2048
2049 /*
2050 * Compress the remaining entries and zero out the removed stuff.
2051 */
2052 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2053 ichdr.usedbytes -= entsize;
2054 xfs_trans_log_buf(args->trans, bp,
2055 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2056 entsize));
2057
2058 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2059 memmove(entry, entry + 1, tmp);
2060 ichdr.count--;
2061 xfs_trans_log_buf(args->trans, bp,
2062 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2063
2064 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2065 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2066
2067 /*
2068 * If we removed the first entry, re-find the first used byte
2069 * in the name area. Note that if the entry was the "firstused",
2070 * then we don't have a "hole" in our block resulting from
2071 * removing the name.
2072 */
2073 if (smallest) {
2074 tmp = args->geo->blksize;
2075 entry = xfs_attr3_leaf_entryp(leaf);
2076 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2077 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2078 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2079
2080 if (be16_to_cpu(entry->nameidx) < tmp)
2081 tmp = be16_to_cpu(entry->nameidx);
2082 }
2083 ichdr.firstused = tmp;
2084 ASSERT(ichdr.firstused != 0);
2085 } else {
2086 ichdr.holes = 1; /* mark as needing compaction */
2087 }
2088 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2089 xfs_trans_log_buf(args->trans, bp,
2090 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2091 xfs_attr3_leaf_hdr_size(leaf)));
2092
2093 /*
2094 * Check if leaf is less than 50% full, caller may want to
2095 * "join" the leaf with a sibling if so.
2096 */
2097 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2098 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2099
2100 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2101 }
2102
2103 /*
2104 * Move all the attribute list entries from drop_leaf into save_leaf.
2105 */
2106 void
2107 xfs_attr3_leaf_unbalance(
2108 struct xfs_da_state *state,
2109 struct xfs_da_state_blk *drop_blk,
2110 struct xfs_da_state_blk *save_blk)
2111 {
2112 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2113 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2114 struct xfs_attr3_icleaf_hdr drophdr;
2115 struct xfs_attr3_icleaf_hdr savehdr;
2116 struct xfs_attr_leaf_entry *entry;
2117
2118 trace_xfs_attr_leaf_unbalance(state->args);
2119
2120 drop_leaf = drop_blk->bp->b_addr;
2121 save_leaf = save_blk->bp->b_addr;
2122 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2123 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2124 entry = xfs_attr3_leaf_entryp(drop_leaf);
2125
2126 /*
2127 * Save last hashval from dying block for later Btree fixup.
2128 */
2129 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2130
2131 /*
2132 * Check if we need a temp buffer, or can we do it in place.
2133 * Note that we don't check "leaf" for holes because we will
2134 * always be dropping it, toosmall() decided that for us already.
2135 */
2136 if (savehdr.holes == 0) {
2137 /*
2138 * dest leaf has no holes, so we add there. May need
2139 * to make some room in the entry array.
2140 */
2141 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2142 drop_blk->bp, &drophdr)) {
2143 xfs_attr3_leaf_moveents(state->args,
2144 drop_leaf, &drophdr, 0,
2145 save_leaf, &savehdr, 0,
2146 drophdr.count);
2147 } else {
2148 xfs_attr3_leaf_moveents(state->args,
2149 drop_leaf, &drophdr, 0,
2150 save_leaf, &savehdr,
2151 savehdr.count, drophdr.count);
2152 }
2153 } else {
2154 /*
2155 * Destination has holes, so we make a temporary copy
2156 * of the leaf and add them both to that.
2157 */
2158 struct xfs_attr_leafblock *tmp_leaf;
2159 struct xfs_attr3_icleaf_hdr tmphdr;
2160
2161 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2162
2163 /*
2164 * Copy the header into the temp leaf so that all the stuff
2165 * not in the incore header is present and gets copied back in
2166 * once we've moved all the entries.
2167 */
2168 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2169
2170 memset(&tmphdr, 0, sizeof(tmphdr));
2171 tmphdr.magic = savehdr.magic;
2172 tmphdr.forw = savehdr.forw;
2173 tmphdr.back = savehdr.back;
2174 tmphdr.firstused = state->args->geo->blksize;
2175
2176 /* write the header to the temp buffer to initialise it */
2177 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2178
2179 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2180 drop_blk->bp, &drophdr)) {
2181 xfs_attr3_leaf_moveents(state->args,
2182 drop_leaf, &drophdr, 0,
2183 tmp_leaf, &tmphdr, 0,
2184 drophdr.count);
2185 xfs_attr3_leaf_moveents(state->args,
2186 save_leaf, &savehdr, 0,
2187 tmp_leaf, &tmphdr, tmphdr.count,
2188 savehdr.count);
2189 } else {
2190 xfs_attr3_leaf_moveents(state->args,
2191 save_leaf, &savehdr, 0,
2192 tmp_leaf, &tmphdr, 0,
2193 savehdr.count);
2194 xfs_attr3_leaf_moveents(state->args,
2195 drop_leaf, &drophdr, 0,
2196 tmp_leaf, &tmphdr, tmphdr.count,
2197 drophdr.count);
2198 }
2199 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2200 savehdr = tmphdr; /* struct copy */
2201 kmem_free(tmp_leaf);
2202 }
2203
2204 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2205 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2206 state->args->geo->blksize - 1);
2207
2208 /*
2209 * Copy out last hashval in each block for B-tree code.
2210 */
2211 entry = xfs_attr3_leaf_entryp(save_leaf);
2212 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2213 }
2214
2215 /*========================================================================
2216 * Routines used for finding things in the Btree.
2217 *========================================================================*/
2218
2219 /*
2220 * Look up a name in a leaf attribute list structure.
2221 * This is the internal routine, it uses the caller's buffer.
2222 *
2223 * Note that duplicate keys are allowed, but only check within the
2224 * current leaf node. The Btree code must check in adjacent leaf nodes.
2225 *
2226 * Return in args->index the index into the entry[] array of either
2227 * the found entry, or where the entry should have been (insert before
2228 * that entry).
2229 *
2230 * Don't change the args->value unless we find the attribute.
2231 */
2232 int
2233 xfs_attr3_leaf_lookup_int(
2234 struct xfs_buf *bp,
2235 struct xfs_da_args *args)
2236 {
2237 struct xfs_attr_leafblock *leaf;
2238 struct xfs_attr3_icleaf_hdr ichdr;
2239 struct xfs_attr_leaf_entry *entry;
2240 struct xfs_attr_leaf_entry *entries;
2241 struct xfs_attr_leaf_name_local *name_loc;
2242 struct xfs_attr_leaf_name_remote *name_rmt;
2243 xfs_dahash_t hashval;
2244 int probe;
2245 int span;
2246
2247 trace_xfs_attr_leaf_lookup(args);
2248
2249 leaf = bp->b_addr;
2250 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2251 entries = xfs_attr3_leaf_entryp(leaf);
2252 if (ichdr.count >= args->geo->blksize / 8)
2253 return -EFSCORRUPTED;
2254
2255 /*
2256 * Binary search. (note: small blocks will skip this loop)
2257 */
2258 hashval = args->hashval;
2259 probe = span = ichdr.count / 2;
2260 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2261 span /= 2;
2262 if (be32_to_cpu(entry->hashval) < hashval)
2263 probe += span;
2264 else if (be32_to_cpu(entry->hashval) > hashval)
2265 probe -= span;
2266 else
2267 break;
2268 }
2269 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count)))
2270 return -EFSCORRUPTED;
2271 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval))
2272 return -EFSCORRUPTED;
2273
2274 /*
2275 * Since we may have duplicate hashval's, find the first matching
2276 * hashval in the leaf.
2277 */
2278 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2279 entry--;
2280 probe--;
2281 }
2282 while (probe < ichdr.count &&
2283 be32_to_cpu(entry->hashval) < hashval) {
2284 entry++;
2285 probe++;
2286 }
2287 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2288 args->index = probe;
2289 return -ENOATTR;
2290 }
2291
2292 /*
2293 * Duplicate keys may be present, so search all of them for a match.
2294 */
2295 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2296 entry++, probe++) {
2297 /*
2298 * GROT: Add code to remove incomplete entries.
2299 */
2300 /*
2301 * If we are looking for INCOMPLETE entries, show only those.
2302 * If we are looking for complete entries, show only those.
2303 */
2304 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2305 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2306 continue;
2307 }
2308 if (entry->flags & XFS_ATTR_LOCAL) {
2309 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2310 if (name_loc->namelen != args->namelen)
2311 continue;
2312 if (memcmp(args->name, name_loc->nameval,
2313 args->namelen) != 0)
2314 continue;
2315 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2316 continue;
2317 args->index = probe;
2318 return -EEXIST;
2319 } else {
2320 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2321 if (name_rmt->namelen != args->namelen)
2322 continue;
2323 if (memcmp(args->name, name_rmt->name,
2324 args->namelen) != 0)
2325 continue;
2326 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2327 continue;
2328 args->index = probe;
2329 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2330 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2331 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2332 args->dp->i_mount,
2333 args->rmtvaluelen);
2334 return -EEXIST;
2335 }
2336 }
2337 args->index = probe;
2338 return -ENOATTR;
2339 }
2340
2341 /*
2342 * Get the value associated with an attribute name from a leaf attribute
2343 * list structure.
2344 */
2345 int
2346 xfs_attr3_leaf_getvalue(
2347 struct xfs_buf *bp,
2348 struct xfs_da_args *args)
2349 {
2350 struct xfs_attr_leafblock *leaf;
2351 struct xfs_attr3_icleaf_hdr ichdr;
2352 struct xfs_attr_leaf_entry *entry;
2353 struct xfs_attr_leaf_name_local *name_loc;
2354 struct xfs_attr_leaf_name_remote *name_rmt;
2355 int valuelen;
2356
2357 leaf = bp->b_addr;
2358 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2359 ASSERT(ichdr.count < args->geo->blksize / 8);
2360 ASSERT(args->index < ichdr.count);
2361
2362 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2363 if (entry->flags & XFS_ATTR_LOCAL) {
2364 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2365 ASSERT(name_loc->namelen == args->namelen);
2366 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2367 valuelen = be16_to_cpu(name_loc->valuelen);
2368 if (args->flags & ATTR_KERNOVAL) {
2369 args->valuelen = valuelen;
2370 return 0;
2371 }
2372 if (args->valuelen < valuelen) {
2373 args->valuelen = valuelen;
2374 return -ERANGE;
2375 }
2376 args->valuelen = valuelen;
2377 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2378 } else {
2379 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2380 ASSERT(name_rmt->namelen == args->namelen);
2381 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2382 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2383 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2384 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2385 args->rmtvaluelen);
2386 if (args->flags & ATTR_KERNOVAL) {
2387 args->valuelen = args->rmtvaluelen;
2388 return 0;
2389 }
2390 if (args->valuelen < args->rmtvaluelen) {
2391 args->valuelen = args->rmtvaluelen;
2392 return -ERANGE;
2393 }
2394 args->valuelen = args->rmtvaluelen;
2395 }
2396 return 0;
2397 }
2398
2399 /*========================================================================
2400 * Utility routines.
2401 *========================================================================*/
2402
2403 /*
2404 * Move the indicated entries from one leaf to another.
2405 * NOTE: this routine modifies both source and destination leaves.
2406 */
2407 /*ARGSUSED*/
2408 STATIC void
2409 xfs_attr3_leaf_moveents(
2410 struct xfs_da_args *args,
2411 struct xfs_attr_leafblock *leaf_s,
2412 struct xfs_attr3_icleaf_hdr *ichdr_s,
2413 int start_s,
2414 struct xfs_attr_leafblock *leaf_d,
2415 struct xfs_attr3_icleaf_hdr *ichdr_d,
2416 int start_d,
2417 int count)
2418 {
2419 struct xfs_attr_leaf_entry *entry_s;
2420 struct xfs_attr_leaf_entry *entry_d;
2421 int desti;
2422 int tmp;
2423 int i;
2424
2425 /*
2426 * Check for nothing to do.
2427 */
2428 if (count == 0)
2429 return;
2430
2431 /*
2432 * Set up environment.
2433 */
2434 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2435 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2436 ASSERT(ichdr_s->magic == ichdr_d->magic);
2437 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2438 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2439 + xfs_attr3_leaf_hdr_size(leaf_s));
2440 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2441 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2442 + xfs_attr3_leaf_hdr_size(leaf_d));
2443
2444 ASSERT(start_s < ichdr_s->count);
2445 ASSERT(start_d <= ichdr_d->count);
2446 ASSERT(count <= ichdr_s->count);
2447
2448
2449 /*
2450 * Move the entries in the destination leaf up to make a hole?
2451 */
2452 if (start_d < ichdr_d->count) {
2453 tmp = ichdr_d->count - start_d;
2454 tmp *= sizeof(xfs_attr_leaf_entry_t);
2455 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2456 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2457 memmove(entry_d, entry_s, tmp);
2458 }
2459
2460 /*
2461 * Copy all entry's in the same (sorted) order,
2462 * but allocate attribute info packed and in sequence.
2463 */
2464 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2465 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2466 desti = start_d;
2467 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2468 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2469 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2470 #ifdef GROT
2471 /*
2472 * Code to drop INCOMPLETE entries. Difficult to use as we
2473 * may also need to change the insertion index. Code turned
2474 * off for 6.2, should be revisited later.
2475 */
2476 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2477 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2478 ichdr_s->usedbytes -= tmp;
2479 ichdr_s->count -= 1;
2480 entry_d--; /* to compensate for ++ in loop hdr */
2481 desti--;
2482 if ((start_s + i) < offset)
2483 result++; /* insertion index adjustment */
2484 } else {
2485 #endif /* GROT */
2486 ichdr_d->firstused -= tmp;
2487 /* both on-disk, don't endian flip twice */
2488 entry_d->hashval = entry_s->hashval;
2489 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2490 entry_d->flags = entry_s->flags;
2491 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2492 <= args->geo->blksize);
2493 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2494 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2495 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2496 <= args->geo->blksize);
2497 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2498 ichdr_s->usedbytes -= tmp;
2499 ichdr_d->usedbytes += tmp;
2500 ichdr_s->count -= 1;
2501 ichdr_d->count += 1;
2502 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2503 + xfs_attr3_leaf_hdr_size(leaf_d);
2504 ASSERT(ichdr_d->firstused >= tmp);
2505 #ifdef GROT
2506 }
2507 #endif /* GROT */
2508 }
2509
2510 /*
2511 * Zero out the entries we just copied.
2512 */
2513 if (start_s == ichdr_s->count) {
2514 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2515 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2516 ASSERT(((char *)entry_s + tmp) <=
2517 ((char *)leaf_s + args->geo->blksize));
2518 memset(entry_s, 0, tmp);
2519 } else {
2520 /*
2521 * Move the remaining entries down to fill the hole,
2522 * then zero the entries at the top.
2523 */
2524 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2525 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2526 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2527 memmove(entry_d, entry_s, tmp);
2528
2529 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2530 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2531 ASSERT(((char *)entry_s + tmp) <=
2532 ((char *)leaf_s + args->geo->blksize));
2533 memset(entry_s, 0, tmp);
2534 }
2535
2536 /*
2537 * Fill in the freemap information
2538 */
2539 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2540 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2541 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2542 ichdr_d->freemap[1].base = 0;
2543 ichdr_d->freemap[2].base = 0;
2544 ichdr_d->freemap[1].size = 0;
2545 ichdr_d->freemap[2].size = 0;
2546 ichdr_s->holes = 1; /* leaf may not be compact */
2547 }
2548
2549 /*
2550 * Pick up the last hashvalue from a leaf block.
2551 */
2552 xfs_dahash_t
2553 xfs_attr_leaf_lasthash(
2554 struct xfs_buf *bp,
2555 int *count)
2556 {
2557 struct xfs_attr3_icleaf_hdr ichdr;
2558 struct xfs_attr_leaf_entry *entries;
2559 struct xfs_mount *mp = bp->b_target->bt_mount;
2560
2561 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2562 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2563 if (count)
2564 *count = ichdr.count;
2565 if (!ichdr.count)
2566 return 0;
2567 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2568 }
2569
2570 /*
2571 * Calculate the number of bytes used to store the indicated attribute
2572 * (whether local or remote only calculate bytes in this block).
2573 */
2574 STATIC int
2575 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2576 {
2577 struct xfs_attr_leaf_entry *entries;
2578 xfs_attr_leaf_name_local_t *name_loc;
2579 xfs_attr_leaf_name_remote_t *name_rmt;
2580 int size;
2581
2582 entries = xfs_attr3_leaf_entryp(leaf);
2583 if (entries[index].flags & XFS_ATTR_LOCAL) {
2584 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2585 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2586 be16_to_cpu(name_loc->valuelen));
2587 } else {
2588 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2589 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2590 }
2591 return size;
2592 }
2593
2594 /*
2595 * Calculate the number of bytes that would be required to store the new
2596 * attribute (whether local or remote only calculate bytes in this block).
2597 * This routine decides as a side effect whether the attribute will be
2598 * a "local" or a "remote" attribute.
2599 */
2600 int
2601 xfs_attr_leaf_newentsize(
2602 struct xfs_da_args *args,
2603 int *local)
2604 {
2605 int size;
2606
2607 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2608 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2609 if (local)
2610 *local = 1;
2611 return size;
2612 }
2613 if (local)
2614 *local = 0;
2615 return xfs_attr_leaf_entsize_remote(args->namelen);
2616 }
2617
2618
2619 /*========================================================================
2620 * Manage the INCOMPLETE flag in a leaf entry
2621 *========================================================================*/
2622
2623 /*
2624 * Clear the INCOMPLETE flag on an entry in a leaf block.
2625 */
2626 int
2627 xfs_attr3_leaf_clearflag(
2628 struct xfs_da_args *args)
2629 {
2630 struct xfs_attr_leafblock *leaf;
2631 struct xfs_attr_leaf_entry *entry;
2632 struct xfs_attr_leaf_name_remote *name_rmt;
2633 struct xfs_buf *bp;
2634 int error;
2635 #ifdef DEBUG
2636 struct xfs_attr3_icleaf_hdr ichdr;
2637 xfs_attr_leaf_name_local_t *name_loc;
2638 int namelen;
2639 char *name;
2640 #endif /* DEBUG */
2641
2642 trace_xfs_attr_leaf_clearflag(args);
2643 /*
2644 * Set up the operation.
2645 */
2646 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2647 if (error)
2648 return error;
2649
2650 leaf = bp->b_addr;
2651 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2652 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2653
2654 #ifdef DEBUG
2655 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2656 ASSERT(args->index < ichdr.count);
2657 ASSERT(args->index >= 0);
2658
2659 if (entry->flags & XFS_ATTR_LOCAL) {
2660 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2661 namelen = name_loc->namelen;
2662 name = (char *)name_loc->nameval;
2663 } else {
2664 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2665 namelen = name_rmt->namelen;
2666 name = (char *)name_rmt->name;
2667 }
2668 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2669 ASSERT(namelen == args->namelen);
2670 ASSERT(memcmp(name, args->name, namelen) == 0);
2671 #endif /* DEBUG */
2672
2673 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2674 xfs_trans_log_buf(args->trans, bp,
2675 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2676
2677 if (args->rmtblkno) {
2678 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2679 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2680 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2681 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2682 xfs_trans_log_buf(args->trans, bp,
2683 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2684 }
2685
2686 /*
2687 * Commit the flag value change and start the next trans in series.
2688 */
2689 return xfs_trans_roll_inode(&args->trans, args->dp);
2690 }
2691
2692 /*
2693 * Set the INCOMPLETE flag on an entry in a leaf block.
2694 */
2695 int
2696 xfs_attr3_leaf_setflag(
2697 struct xfs_da_args *args)
2698 {
2699 struct xfs_attr_leafblock *leaf;
2700 struct xfs_attr_leaf_entry *entry;
2701 struct xfs_attr_leaf_name_remote *name_rmt;
2702 struct xfs_buf *bp;
2703 int error;
2704 #ifdef DEBUG
2705 struct xfs_attr3_icleaf_hdr ichdr;
2706 #endif
2707
2708 trace_xfs_attr_leaf_setflag(args);
2709
2710 /*
2711 * Set up the operation.
2712 */
2713 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2714 if (error)
2715 return error;
2716
2717 leaf = bp->b_addr;
2718 #ifdef DEBUG
2719 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2720 ASSERT(args->index < ichdr.count);
2721 ASSERT(args->index >= 0);
2722 #endif
2723 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2724
2725 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2726 entry->flags |= XFS_ATTR_INCOMPLETE;
2727 xfs_trans_log_buf(args->trans, bp,
2728 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2729 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2730 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2731 name_rmt->valueblk = 0;
2732 name_rmt->valuelen = 0;
2733 xfs_trans_log_buf(args->trans, bp,
2734 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2735 }
2736
2737 /*
2738 * Commit the flag value change and start the next trans in series.
2739 */
2740 return xfs_trans_roll_inode(&args->trans, args->dp);
2741 }
2742
2743 /*
2744 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2745 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2746 * entry given by args->blkno2/index2.
2747 *
2748 * Note that they could be in different blocks, or in the same block.
2749 */
2750 int
2751 xfs_attr3_leaf_flipflags(
2752 struct xfs_da_args *args)
2753 {
2754 struct xfs_attr_leafblock *leaf1;
2755 struct xfs_attr_leafblock *leaf2;
2756 struct xfs_attr_leaf_entry *entry1;
2757 struct xfs_attr_leaf_entry *entry2;
2758 struct xfs_attr_leaf_name_remote *name_rmt;
2759 struct xfs_buf *bp1;
2760 struct xfs_buf *bp2;
2761 int error;
2762 #ifdef DEBUG
2763 struct xfs_attr3_icleaf_hdr ichdr1;
2764 struct xfs_attr3_icleaf_hdr ichdr2;
2765 xfs_attr_leaf_name_local_t *name_loc;
2766 int namelen1, namelen2;
2767 char *name1, *name2;
2768 #endif /* DEBUG */
2769
2770 trace_xfs_attr_leaf_flipflags(args);
2771
2772 /*
2773 * Read the block containing the "old" attr
2774 */
2775 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2776 if (error)
2777 return error;
2778
2779 /*
2780 * Read the block containing the "new" attr, if it is different
2781 */
2782 if (args->blkno2 != args->blkno) {
2783 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2784 -1, &bp2);
2785 if (error)
2786 return error;
2787 } else {
2788 bp2 = bp1;
2789 }
2790
2791 leaf1 = bp1->b_addr;
2792 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2793
2794 leaf2 = bp2->b_addr;
2795 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2796
2797 #ifdef DEBUG
2798 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2799 ASSERT(args->index < ichdr1.count);
2800 ASSERT(args->index >= 0);
2801
2802 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2803 ASSERT(args->index2 < ichdr2.count);
2804 ASSERT(args->index2 >= 0);
2805
2806 if (entry1->flags & XFS_ATTR_LOCAL) {
2807 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2808 namelen1 = name_loc->namelen;
2809 name1 = (char *)name_loc->nameval;
2810 } else {
2811 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2812 namelen1 = name_rmt->namelen;
2813 name1 = (char *)name_rmt->name;
2814 }
2815 if (entry2->flags & XFS_ATTR_LOCAL) {
2816 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2817 namelen2 = name_loc->namelen;
2818 name2 = (char *)name_loc->nameval;
2819 } else {
2820 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2821 namelen2 = name_rmt->namelen;
2822 name2 = (char *)name_rmt->name;
2823 }
2824 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2825 ASSERT(namelen1 == namelen2);
2826 ASSERT(memcmp(name1, name2, namelen1) == 0);
2827 #endif /* DEBUG */
2828
2829 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2830 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2831
2832 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2833 xfs_trans_log_buf(args->trans, bp1,
2834 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2835 if (args->rmtblkno) {
2836 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2837 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2838 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2839 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2840 xfs_trans_log_buf(args->trans, bp1,
2841 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2842 }
2843
2844 entry2->flags |= XFS_ATTR_INCOMPLETE;
2845 xfs_trans_log_buf(args->trans, bp2,
2846 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2847 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2848 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2849 name_rmt->valueblk = 0;
2850 name_rmt->valuelen = 0;
2851 xfs_trans_log_buf(args->trans, bp2,
2852 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2853 }
2854
2855 /*
2856 * Commit the flag value change and start the next trans in series.
2857 */
2858 error = xfs_trans_roll_inode(&args->trans, args->dp);
2859
2860 return error;
2861 }