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