]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/xfs/libxfs/xfs_attr_leaf.c
Merge branches 'for-4.4/upstream-fixes', 'for-4.5/async-suspend', 'for-4.5/container...
[mirror_ubuntu-artful-kernel.git] / fs / xfs / libxfs / xfs_attr_leaf.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_bmap.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_attr_remote.h"
37 #include "xfs_attr.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_error.h"
40 #include "xfs_trace.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_cksum.h"
43 #include "xfs_dir2.h"
44 #include "xfs_log.h"
45
46
47 /*
48 * xfs_attr_leaf.c
49 *
50 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
51 */
52
53 /*========================================================================
54 * Function prototypes for the kernel.
55 *========================================================================*/
56
57 /*
58 * Routines used for growing the Btree.
59 */
60 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
61 xfs_dablk_t which_block, struct xfs_buf **bpp);
62 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
63 struct xfs_attr3_icleaf_hdr *ichdr,
64 struct xfs_da_args *args, int freemap_index);
65 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
66 struct xfs_attr3_icleaf_hdr *ichdr,
67 struct xfs_buf *leaf_buffer);
68 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
69 xfs_da_state_blk_t *blk1,
70 xfs_da_state_blk_t *blk2);
71 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
72 xfs_da_state_blk_t *leaf_blk_1,
73 struct xfs_attr3_icleaf_hdr *ichdr1,
74 xfs_da_state_blk_t *leaf_blk_2,
75 struct xfs_attr3_icleaf_hdr *ichdr2,
76 int *number_entries_in_blk1,
77 int *number_usedbytes_in_blk1);
78
79 /*
80 * Utility routines.
81 */
82 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
83 struct xfs_attr_leafblock *src_leaf,
84 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
85 struct xfs_attr_leafblock *dst_leaf,
86 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87 int move_count);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
89
90 /*
91 * attr3 block 'firstused' conversion helpers.
92 *
93 * firstused refers to the offset of the first used byte of the nameval region
94 * of an attr leaf block. The region starts at the tail of the block and expands
95 * backwards towards the middle. As such, firstused is initialized to the block
96 * size for an empty leaf block and is reduced from there.
97 *
98 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
99 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
100 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
101 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
102 * the attr block size. The following helpers manage the conversion between the
103 * in-core and on-disk formats.
104 */
105
106 static void
107 xfs_attr3_leaf_firstused_from_disk(
108 struct xfs_da_geometry *geo,
109 struct xfs_attr3_icleaf_hdr *to,
110 struct xfs_attr_leafblock *from)
111 {
112 struct xfs_attr3_leaf_hdr *hdr3;
113
114 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
115 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
116 to->firstused = be16_to_cpu(hdr3->firstused);
117 } else {
118 to->firstused = be16_to_cpu(from->hdr.firstused);
119 }
120
121 /*
122 * Convert from the magic fsb size value to actual blocksize. This
123 * should only occur for empty blocks when the block size overflows
124 * 16-bits.
125 */
126 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
127 ASSERT(!to->count && !to->usedbytes);
128 ASSERT(geo->blksize > USHRT_MAX);
129 to->firstused = geo->blksize;
130 }
131 }
132
133 static void
134 xfs_attr3_leaf_firstused_to_disk(
135 struct xfs_da_geometry *geo,
136 struct xfs_attr_leafblock *to,
137 struct xfs_attr3_icleaf_hdr *from)
138 {
139 struct xfs_attr3_leaf_hdr *hdr3;
140 uint32_t firstused;
141
142 /* magic value should only be seen on disk */
143 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
144
145 /*
146 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
147 * value. This only overflows at the max supported value of 64k. Use the
148 * magic on-disk value to represent block size in this case.
149 */
150 firstused = from->firstused;
151 if (firstused > USHRT_MAX) {
152 ASSERT(from->firstused == geo->blksize);
153 firstused = XFS_ATTR3_LEAF_NULLOFF;
154 }
155
156 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
157 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
158 hdr3->firstused = cpu_to_be16(firstused);
159 } else {
160 to->hdr.firstused = cpu_to_be16(firstused);
161 }
162 }
163
164 void
165 xfs_attr3_leaf_hdr_from_disk(
166 struct xfs_da_geometry *geo,
167 struct xfs_attr3_icleaf_hdr *to,
168 struct xfs_attr_leafblock *from)
169 {
170 int i;
171
172 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
173 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
174
175 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
176 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
177
178 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
179 to->back = be32_to_cpu(hdr3->info.hdr.back);
180 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
181 to->count = be16_to_cpu(hdr3->count);
182 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
183 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
184 to->holes = hdr3->holes;
185
186 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
187 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
188 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
189 }
190 return;
191 }
192 to->forw = be32_to_cpu(from->hdr.info.forw);
193 to->back = be32_to_cpu(from->hdr.info.back);
194 to->magic = be16_to_cpu(from->hdr.info.magic);
195 to->count = be16_to_cpu(from->hdr.count);
196 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
197 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
198 to->holes = from->hdr.holes;
199
200 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
201 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
202 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
203 }
204 }
205
206 void
207 xfs_attr3_leaf_hdr_to_disk(
208 struct xfs_da_geometry *geo,
209 struct xfs_attr_leafblock *to,
210 struct xfs_attr3_icleaf_hdr *from)
211 {
212 int i;
213
214 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
215 from->magic == XFS_ATTR3_LEAF_MAGIC);
216
217 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
218 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
219
220 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
221 hdr3->info.hdr.back = cpu_to_be32(from->back);
222 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
223 hdr3->count = cpu_to_be16(from->count);
224 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
225 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
226 hdr3->holes = from->holes;
227 hdr3->pad1 = 0;
228
229 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
230 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
231 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
232 }
233 return;
234 }
235 to->hdr.info.forw = cpu_to_be32(from->forw);
236 to->hdr.info.back = cpu_to_be32(from->back);
237 to->hdr.info.magic = cpu_to_be16(from->magic);
238 to->hdr.count = cpu_to_be16(from->count);
239 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
240 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
241 to->hdr.holes = from->holes;
242 to->hdr.pad1 = 0;
243
244 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
245 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
246 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
247 }
248 }
249
250 static bool
251 xfs_attr3_leaf_verify(
252 struct xfs_buf *bp)
253 {
254 struct xfs_mount *mp = bp->b_target->bt_mount;
255 struct xfs_attr_leafblock *leaf = bp->b_addr;
256 struct xfs_attr3_icleaf_hdr ichdr;
257
258 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
259
260 if (xfs_sb_version_hascrc(&mp->m_sb)) {
261 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
262
263 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
264 return false;
265
266 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
267 return false;
268 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
269 return false;
270 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
271 return false;
272 } else {
273 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
274 return false;
275 }
276 if (ichdr.count == 0)
277 return false;
278
279 /* XXX: need to range check rest of attr header values */
280 /* XXX: hash order check? */
281
282 return true;
283 }
284
285 static void
286 xfs_attr3_leaf_write_verify(
287 struct xfs_buf *bp)
288 {
289 struct xfs_mount *mp = bp->b_target->bt_mount;
290 struct xfs_buf_log_item *bip = bp->b_fspriv;
291 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
292
293 if (!xfs_attr3_leaf_verify(bp)) {
294 xfs_buf_ioerror(bp, -EFSCORRUPTED);
295 xfs_verifier_error(bp);
296 return;
297 }
298
299 if (!xfs_sb_version_hascrc(&mp->m_sb))
300 return;
301
302 if (bip)
303 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
304
305 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
306 }
307
308 /*
309 * leaf/node format detection on trees is sketchy, so a node read can be done on
310 * leaf level blocks when detection identifies the tree as a node format tree
311 * incorrectly. In this case, we need to swap the verifier to match the correct
312 * format of the block being read.
313 */
314 static void
315 xfs_attr3_leaf_read_verify(
316 struct xfs_buf *bp)
317 {
318 struct xfs_mount *mp = bp->b_target->bt_mount;
319
320 if (xfs_sb_version_hascrc(&mp->m_sb) &&
321 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
322 xfs_buf_ioerror(bp, -EFSBADCRC);
323 else if (!xfs_attr3_leaf_verify(bp))
324 xfs_buf_ioerror(bp, -EFSCORRUPTED);
325
326 if (bp->b_error)
327 xfs_verifier_error(bp);
328 }
329
330 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
331 .verify_read = xfs_attr3_leaf_read_verify,
332 .verify_write = xfs_attr3_leaf_write_verify,
333 };
334
335 int
336 xfs_attr3_leaf_read(
337 struct xfs_trans *tp,
338 struct xfs_inode *dp,
339 xfs_dablk_t bno,
340 xfs_daddr_t mappedbno,
341 struct xfs_buf **bpp)
342 {
343 int err;
344
345 err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
346 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
347 if (!err && tp)
348 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
349 return err;
350 }
351
352 /*========================================================================
353 * Namespace helper routines
354 *========================================================================*/
355
356 /*
357 * If namespace bits don't match return 0.
358 * If all match then return 1.
359 */
360 STATIC int
361 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
362 {
363 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
364 }
365
366
367 /*========================================================================
368 * External routines when attribute fork size < XFS_LITINO(mp).
369 *========================================================================*/
370
371 /*
372 * Query whether the requested number of additional bytes of extended
373 * attribute space will be able to fit inline.
374 *
375 * Returns zero if not, else the di_forkoff fork offset to be used in the
376 * literal area for attribute data once the new bytes have been added.
377 *
378 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
379 * special case for dev/uuid inodes, they have fixed size data forks.
380 */
381 int
382 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
383 {
384 int offset;
385 int minforkoff; /* lower limit on valid forkoff locations */
386 int maxforkoff; /* upper limit on valid forkoff locations */
387 int dsize;
388 xfs_mount_t *mp = dp->i_mount;
389
390 /* rounded down */
391 offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
392
393 switch (dp->i_d.di_format) {
394 case XFS_DINODE_FMT_DEV:
395 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
396 return (offset >= minforkoff) ? minforkoff : 0;
397 case XFS_DINODE_FMT_UUID:
398 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
399 return (offset >= minforkoff) ? minforkoff : 0;
400 }
401
402 /*
403 * If the requested numbers of bytes is smaller or equal to the
404 * current attribute fork size we can always proceed.
405 *
406 * Note that if_bytes in the data fork might actually be larger than
407 * the current data fork size is due to delalloc extents. In that
408 * case either the extent count will go down when they are converted
409 * to real extents, or the delalloc conversion will take care of the
410 * literal area rebalancing.
411 */
412 if (bytes <= XFS_IFORK_ASIZE(dp))
413 return dp->i_d.di_forkoff;
414
415 /*
416 * For attr2 we can try to move the forkoff if there is space in the
417 * literal area, but for the old format we are done if there is no
418 * space in the fixed attribute fork.
419 */
420 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
421 return 0;
422
423 dsize = dp->i_df.if_bytes;
424
425 switch (dp->i_d.di_format) {
426 case XFS_DINODE_FMT_EXTENTS:
427 /*
428 * If there is no attr fork and the data fork is extents,
429 * determine if creating the default attr fork will result
430 * in the extents form migrating to btree. If so, the
431 * minimum offset only needs to be the space required for
432 * the btree root.
433 */
434 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
435 xfs_default_attroffset(dp))
436 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
437 break;
438 case XFS_DINODE_FMT_BTREE:
439 /*
440 * If we have a data btree then keep forkoff if we have one,
441 * otherwise we are adding a new attr, so then we set
442 * minforkoff to where the btree root can finish so we have
443 * plenty of room for attrs
444 */
445 if (dp->i_d.di_forkoff) {
446 if (offset < dp->i_d.di_forkoff)
447 return 0;
448 return dp->i_d.di_forkoff;
449 }
450 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
451 break;
452 }
453
454 /*
455 * A data fork btree root must have space for at least
456 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
457 */
458 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
459 minforkoff = roundup(minforkoff, 8) >> 3;
460
461 /* attr fork btree root can have at least this many key/ptr pairs */
462 maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
463 XFS_BMDR_SPACE_CALC(MINABTPTRS);
464 maxforkoff = maxforkoff >> 3; /* rounded down */
465
466 if (offset >= maxforkoff)
467 return maxforkoff;
468 if (offset >= minforkoff)
469 return offset;
470 return 0;
471 }
472
473 /*
474 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
475 */
476 STATIC void
477 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
478 {
479 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
480 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
481 spin_lock(&mp->m_sb_lock);
482 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
483 xfs_sb_version_addattr2(&mp->m_sb);
484 spin_unlock(&mp->m_sb_lock);
485 xfs_log_sb(tp);
486 } else
487 spin_unlock(&mp->m_sb_lock);
488 }
489 }
490
491 /*
492 * Create the initial contents of a shortform attribute list.
493 */
494 void
495 xfs_attr_shortform_create(xfs_da_args_t *args)
496 {
497 xfs_attr_sf_hdr_t *hdr;
498 xfs_inode_t *dp;
499 xfs_ifork_t *ifp;
500
501 trace_xfs_attr_sf_create(args);
502
503 dp = args->dp;
504 ASSERT(dp != NULL);
505 ifp = dp->i_afp;
506 ASSERT(ifp != NULL);
507 ASSERT(ifp->if_bytes == 0);
508 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
509 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
510 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
511 ifp->if_flags |= XFS_IFINLINE;
512 } else {
513 ASSERT(ifp->if_flags & XFS_IFINLINE);
514 }
515 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
516 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
517 hdr->count = 0;
518 hdr->totsize = cpu_to_be16(sizeof(*hdr));
519 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
520 }
521
522 /*
523 * Add a name/value pair to the shortform attribute list.
524 * Overflow from the inode has already been checked for.
525 */
526 void
527 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
528 {
529 xfs_attr_shortform_t *sf;
530 xfs_attr_sf_entry_t *sfe;
531 int i, offset, size;
532 xfs_mount_t *mp;
533 xfs_inode_t *dp;
534 xfs_ifork_t *ifp;
535
536 trace_xfs_attr_sf_add(args);
537
538 dp = args->dp;
539 mp = dp->i_mount;
540 dp->i_d.di_forkoff = forkoff;
541
542 ifp = dp->i_afp;
543 ASSERT(ifp->if_flags & XFS_IFINLINE);
544 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
545 sfe = &sf->list[0];
546 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
547 #ifdef DEBUG
548 if (sfe->namelen != args->namelen)
549 continue;
550 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
551 continue;
552 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
553 continue;
554 ASSERT(0);
555 #endif
556 }
557
558 offset = (char *)sfe - (char *)sf;
559 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
560 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
561 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
562 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
563
564 sfe->namelen = args->namelen;
565 sfe->valuelen = args->valuelen;
566 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
567 memcpy(sfe->nameval, args->name, args->namelen);
568 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
569 sf->hdr.count++;
570 be16_add_cpu(&sf->hdr.totsize, size);
571 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
572
573 xfs_sbversion_add_attr2(mp, args->trans);
574 }
575
576 /*
577 * After the last attribute is removed revert to original inode format,
578 * making all literal area available to the data fork once more.
579 */
580 void
581 xfs_attr_fork_remove(
582 struct xfs_inode *ip,
583 struct xfs_trans *tp)
584 {
585 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
586 ip->i_d.di_forkoff = 0;
587 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
588
589 ASSERT(ip->i_d.di_anextents == 0);
590 ASSERT(ip->i_afp == NULL);
591
592 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
593 }
594
595 /*
596 * Remove an attribute from the shortform attribute list structure.
597 */
598 int
599 xfs_attr_shortform_remove(xfs_da_args_t *args)
600 {
601 xfs_attr_shortform_t *sf;
602 xfs_attr_sf_entry_t *sfe;
603 int base, size=0, end, totsize, i;
604 xfs_mount_t *mp;
605 xfs_inode_t *dp;
606
607 trace_xfs_attr_sf_remove(args);
608
609 dp = args->dp;
610 mp = dp->i_mount;
611 base = sizeof(xfs_attr_sf_hdr_t);
612 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
613 sfe = &sf->list[0];
614 end = sf->hdr.count;
615 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
616 base += size, i++) {
617 size = XFS_ATTR_SF_ENTSIZE(sfe);
618 if (sfe->namelen != args->namelen)
619 continue;
620 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
621 continue;
622 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
623 continue;
624 break;
625 }
626 if (i == end)
627 return -ENOATTR;
628
629 /*
630 * Fix up the attribute fork data, covering the hole
631 */
632 end = base + size;
633 totsize = be16_to_cpu(sf->hdr.totsize);
634 if (end != totsize)
635 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
636 sf->hdr.count--;
637 be16_add_cpu(&sf->hdr.totsize, -size);
638
639 /*
640 * Fix up the start offset of the attribute fork
641 */
642 totsize -= size;
643 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
644 (mp->m_flags & XFS_MOUNT_ATTR2) &&
645 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
646 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
647 xfs_attr_fork_remove(dp, args->trans);
648 } else {
649 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
650 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
651 ASSERT(dp->i_d.di_forkoff);
652 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
653 (args->op_flags & XFS_DA_OP_ADDNAME) ||
654 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
655 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
656 xfs_trans_log_inode(args->trans, dp,
657 XFS_ILOG_CORE | XFS_ILOG_ADATA);
658 }
659
660 xfs_sbversion_add_attr2(mp, args->trans);
661
662 return 0;
663 }
664
665 /*
666 * Look up a name in a shortform attribute list structure.
667 */
668 /*ARGSUSED*/
669 int
670 xfs_attr_shortform_lookup(xfs_da_args_t *args)
671 {
672 xfs_attr_shortform_t *sf;
673 xfs_attr_sf_entry_t *sfe;
674 int i;
675 xfs_ifork_t *ifp;
676
677 trace_xfs_attr_sf_lookup(args);
678
679 ifp = args->dp->i_afp;
680 ASSERT(ifp->if_flags & XFS_IFINLINE);
681 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
682 sfe = &sf->list[0];
683 for (i = 0; i < sf->hdr.count;
684 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
685 if (sfe->namelen != args->namelen)
686 continue;
687 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
688 continue;
689 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
690 continue;
691 return -EEXIST;
692 }
693 return -ENOATTR;
694 }
695
696 /*
697 * Look up a name in a shortform attribute list structure.
698 */
699 /*ARGSUSED*/
700 int
701 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
702 {
703 xfs_attr_shortform_t *sf;
704 xfs_attr_sf_entry_t *sfe;
705 int i;
706
707 ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
708 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
709 sfe = &sf->list[0];
710 for (i = 0; i < sf->hdr.count;
711 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
712 if (sfe->namelen != args->namelen)
713 continue;
714 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
715 continue;
716 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
717 continue;
718 if (args->flags & ATTR_KERNOVAL) {
719 args->valuelen = sfe->valuelen;
720 return -EEXIST;
721 }
722 if (args->valuelen < sfe->valuelen) {
723 args->valuelen = sfe->valuelen;
724 return -ERANGE;
725 }
726 args->valuelen = sfe->valuelen;
727 memcpy(args->value, &sfe->nameval[args->namelen],
728 args->valuelen);
729 return -EEXIST;
730 }
731 return -ENOATTR;
732 }
733
734 /*
735 * Convert from using the shortform to the leaf.
736 */
737 int
738 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
739 {
740 xfs_inode_t *dp;
741 xfs_attr_shortform_t *sf;
742 xfs_attr_sf_entry_t *sfe;
743 xfs_da_args_t nargs;
744 char *tmpbuffer;
745 int error, i, size;
746 xfs_dablk_t blkno;
747 struct xfs_buf *bp;
748 xfs_ifork_t *ifp;
749
750 trace_xfs_attr_sf_to_leaf(args);
751
752 dp = args->dp;
753 ifp = dp->i_afp;
754 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
755 size = be16_to_cpu(sf->hdr.totsize);
756 tmpbuffer = kmem_alloc(size, KM_SLEEP);
757 ASSERT(tmpbuffer != NULL);
758 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
759 sf = (xfs_attr_shortform_t *)tmpbuffer;
760
761 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
762 xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
763
764 bp = NULL;
765 error = xfs_da_grow_inode(args, &blkno);
766 if (error) {
767 /*
768 * If we hit an IO error middle of the transaction inside
769 * grow_inode(), we may have inconsistent data. Bail out.
770 */
771 if (error == -EIO)
772 goto out;
773 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
774 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
775 goto out;
776 }
777
778 ASSERT(blkno == 0);
779 error = xfs_attr3_leaf_create(args, blkno, &bp);
780 if (error) {
781 error = xfs_da_shrink_inode(args, 0, bp);
782 bp = NULL;
783 if (error)
784 goto out;
785 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
786 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
787 goto out;
788 }
789
790 memset((char *)&nargs, 0, sizeof(nargs));
791 nargs.dp = dp;
792 nargs.geo = args->geo;
793 nargs.firstblock = args->firstblock;
794 nargs.flist = args->flist;
795 nargs.total = args->total;
796 nargs.whichfork = XFS_ATTR_FORK;
797 nargs.trans = args->trans;
798 nargs.op_flags = XFS_DA_OP_OKNOENT;
799
800 sfe = &sf->list[0];
801 for (i = 0; i < sf->hdr.count; i++) {
802 nargs.name = sfe->nameval;
803 nargs.namelen = sfe->namelen;
804 nargs.value = &sfe->nameval[nargs.namelen];
805 nargs.valuelen = sfe->valuelen;
806 nargs.hashval = xfs_da_hashname(sfe->nameval,
807 sfe->namelen);
808 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
809 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
810 ASSERT(error == -ENOATTR);
811 error = xfs_attr3_leaf_add(bp, &nargs);
812 ASSERT(error != -ENOSPC);
813 if (error)
814 goto out;
815 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
816 }
817 error = 0;
818
819 out:
820 kmem_free(tmpbuffer);
821 return error;
822 }
823
824 /*
825 * Check a leaf attribute block to see if all the entries would fit into
826 * a shortform attribute list.
827 */
828 int
829 xfs_attr_shortform_allfit(
830 struct xfs_buf *bp,
831 struct xfs_inode *dp)
832 {
833 struct xfs_attr_leafblock *leaf;
834 struct xfs_attr_leaf_entry *entry;
835 xfs_attr_leaf_name_local_t *name_loc;
836 struct xfs_attr3_icleaf_hdr leafhdr;
837 int bytes;
838 int i;
839 struct xfs_mount *mp = bp->b_target->bt_mount;
840
841 leaf = bp->b_addr;
842 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
843 entry = xfs_attr3_leaf_entryp(leaf);
844
845 bytes = sizeof(struct xfs_attr_sf_hdr);
846 for (i = 0; i < leafhdr.count; entry++, i++) {
847 if (entry->flags & XFS_ATTR_INCOMPLETE)
848 continue; /* don't copy partial entries */
849 if (!(entry->flags & XFS_ATTR_LOCAL))
850 return 0;
851 name_loc = xfs_attr3_leaf_name_local(leaf, i);
852 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
853 return 0;
854 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
855 return 0;
856 bytes += sizeof(struct xfs_attr_sf_entry) - 1
857 + name_loc->namelen
858 + be16_to_cpu(name_loc->valuelen);
859 }
860 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
861 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
862 (bytes == sizeof(struct xfs_attr_sf_hdr)))
863 return -1;
864 return xfs_attr_shortform_bytesfit(dp, bytes);
865 }
866
867 /*
868 * Convert a leaf attribute list to shortform attribute list
869 */
870 int
871 xfs_attr3_leaf_to_shortform(
872 struct xfs_buf *bp,
873 struct xfs_da_args *args,
874 int forkoff)
875 {
876 struct xfs_attr_leafblock *leaf;
877 struct xfs_attr3_icleaf_hdr ichdr;
878 struct xfs_attr_leaf_entry *entry;
879 struct xfs_attr_leaf_name_local *name_loc;
880 struct xfs_da_args nargs;
881 struct xfs_inode *dp = args->dp;
882 char *tmpbuffer;
883 int error;
884 int i;
885
886 trace_xfs_attr_leaf_to_sf(args);
887
888 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
889 if (!tmpbuffer)
890 return -ENOMEM;
891
892 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
893
894 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
895 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
896 entry = xfs_attr3_leaf_entryp(leaf);
897
898 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
899 memset(bp->b_addr, 0, args->geo->blksize);
900
901 /*
902 * Clean out the prior contents of the attribute list.
903 */
904 error = xfs_da_shrink_inode(args, 0, bp);
905 if (error)
906 goto out;
907
908 if (forkoff == -1) {
909 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
910 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
911 xfs_attr_fork_remove(dp, args->trans);
912 goto out;
913 }
914
915 xfs_attr_shortform_create(args);
916
917 /*
918 * Copy the attributes
919 */
920 memset((char *)&nargs, 0, sizeof(nargs));
921 nargs.geo = args->geo;
922 nargs.dp = dp;
923 nargs.firstblock = args->firstblock;
924 nargs.flist = args->flist;
925 nargs.total = args->total;
926 nargs.whichfork = XFS_ATTR_FORK;
927 nargs.trans = args->trans;
928 nargs.op_flags = XFS_DA_OP_OKNOENT;
929
930 for (i = 0; i < ichdr.count; entry++, i++) {
931 if (entry->flags & XFS_ATTR_INCOMPLETE)
932 continue; /* don't copy partial entries */
933 if (!entry->nameidx)
934 continue;
935 ASSERT(entry->flags & XFS_ATTR_LOCAL);
936 name_loc = xfs_attr3_leaf_name_local(leaf, i);
937 nargs.name = name_loc->nameval;
938 nargs.namelen = name_loc->namelen;
939 nargs.value = &name_loc->nameval[nargs.namelen];
940 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
941 nargs.hashval = be32_to_cpu(entry->hashval);
942 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
943 xfs_attr_shortform_add(&nargs, forkoff);
944 }
945 error = 0;
946
947 out:
948 kmem_free(tmpbuffer);
949 return error;
950 }
951
952 /*
953 * Convert from using a single leaf to a root node and a leaf.
954 */
955 int
956 xfs_attr3_leaf_to_node(
957 struct xfs_da_args *args)
958 {
959 struct xfs_attr_leafblock *leaf;
960 struct xfs_attr3_icleaf_hdr icleafhdr;
961 struct xfs_attr_leaf_entry *entries;
962 struct xfs_da_node_entry *btree;
963 struct xfs_da3_icnode_hdr icnodehdr;
964 struct xfs_da_intnode *node;
965 struct xfs_inode *dp = args->dp;
966 struct xfs_mount *mp = dp->i_mount;
967 struct xfs_buf *bp1 = NULL;
968 struct xfs_buf *bp2 = NULL;
969 xfs_dablk_t blkno;
970 int error;
971
972 trace_xfs_attr_leaf_to_node(args);
973
974 error = xfs_da_grow_inode(args, &blkno);
975 if (error)
976 goto out;
977 error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
978 if (error)
979 goto out;
980
981 error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
982 if (error)
983 goto out;
984
985 /* copy leaf to new buffer, update identifiers */
986 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
987 bp2->b_ops = bp1->b_ops;
988 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
989 if (xfs_sb_version_hascrc(&mp->m_sb)) {
990 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
991 hdr3->blkno = cpu_to_be64(bp2->b_bn);
992 }
993 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
994
995 /*
996 * Set up the new root node.
997 */
998 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
999 if (error)
1000 goto out;
1001 node = bp1->b_addr;
1002 dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1003 btree = dp->d_ops->node_tree_p(node);
1004
1005 leaf = bp2->b_addr;
1006 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1007 entries = xfs_attr3_leaf_entryp(leaf);
1008
1009 /* both on-disk, don't endian-flip twice */
1010 btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1011 btree[0].before = cpu_to_be32(blkno);
1012 icnodehdr.count = 1;
1013 dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1014 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1015 error = 0;
1016 out:
1017 return error;
1018 }
1019
1020 /*========================================================================
1021 * Routines used for growing the Btree.
1022 *========================================================================*/
1023
1024 /*
1025 * Create the initial contents of a leaf attribute list
1026 * or a leaf in a node attribute list.
1027 */
1028 STATIC int
1029 xfs_attr3_leaf_create(
1030 struct xfs_da_args *args,
1031 xfs_dablk_t blkno,
1032 struct xfs_buf **bpp)
1033 {
1034 struct xfs_attr_leafblock *leaf;
1035 struct xfs_attr3_icleaf_hdr ichdr;
1036 struct xfs_inode *dp = args->dp;
1037 struct xfs_mount *mp = dp->i_mount;
1038 struct xfs_buf *bp;
1039 int error;
1040
1041 trace_xfs_attr_leaf_create(args);
1042
1043 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1044 XFS_ATTR_FORK);
1045 if (error)
1046 return error;
1047 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1048 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1049 leaf = bp->b_addr;
1050 memset(leaf, 0, args->geo->blksize);
1051
1052 memset(&ichdr, 0, sizeof(ichdr));
1053 ichdr.firstused = args->geo->blksize;
1054
1055 if (xfs_sb_version_hascrc(&mp->m_sb)) {
1056 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1057
1058 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1059
1060 hdr3->blkno = cpu_to_be64(bp->b_bn);
1061 hdr3->owner = cpu_to_be64(dp->i_ino);
1062 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1063
1064 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1065 } else {
1066 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1067 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1068 }
1069 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1070
1071 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1072 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1073
1074 *bpp = bp;
1075 return 0;
1076 }
1077
1078 /*
1079 * Split the leaf node, rebalance, then add the new entry.
1080 */
1081 int
1082 xfs_attr3_leaf_split(
1083 struct xfs_da_state *state,
1084 struct xfs_da_state_blk *oldblk,
1085 struct xfs_da_state_blk *newblk)
1086 {
1087 xfs_dablk_t blkno;
1088 int error;
1089
1090 trace_xfs_attr_leaf_split(state->args);
1091
1092 /*
1093 * Allocate space for a new leaf node.
1094 */
1095 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1096 error = xfs_da_grow_inode(state->args, &blkno);
1097 if (error)
1098 return error;
1099 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1100 if (error)
1101 return error;
1102 newblk->blkno = blkno;
1103 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1104
1105 /*
1106 * Rebalance the entries across the two leaves.
1107 * NOTE: rebalance() currently depends on the 2nd block being empty.
1108 */
1109 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1110 error = xfs_da3_blk_link(state, oldblk, newblk);
1111 if (error)
1112 return error;
1113
1114 /*
1115 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1116 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1117 * "new" attrs info. Will need the "old" info to remove it later.
1118 *
1119 * Insert the "new" entry in the correct block.
1120 */
1121 if (state->inleaf) {
1122 trace_xfs_attr_leaf_add_old(state->args);
1123 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1124 } else {
1125 trace_xfs_attr_leaf_add_new(state->args);
1126 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1127 }
1128
1129 /*
1130 * Update last hashval in each block since we added the name.
1131 */
1132 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1133 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1134 return error;
1135 }
1136
1137 /*
1138 * Add a name to the leaf attribute list structure.
1139 */
1140 int
1141 xfs_attr3_leaf_add(
1142 struct xfs_buf *bp,
1143 struct xfs_da_args *args)
1144 {
1145 struct xfs_attr_leafblock *leaf;
1146 struct xfs_attr3_icleaf_hdr ichdr;
1147 int tablesize;
1148 int entsize;
1149 int sum;
1150 int tmp;
1151 int i;
1152
1153 trace_xfs_attr_leaf_add(args);
1154
1155 leaf = bp->b_addr;
1156 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1157 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1158 entsize = xfs_attr_leaf_newentsize(args, NULL);
1159
1160 /*
1161 * Search through freemap for first-fit on new name length.
1162 * (may need to figure in size of entry struct too)
1163 */
1164 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1165 + xfs_attr3_leaf_hdr_size(leaf);
1166 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1167 if (tablesize > ichdr.firstused) {
1168 sum += ichdr.freemap[i].size;
1169 continue;
1170 }
1171 if (!ichdr.freemap[i].size)
1172 continue; /* no space in this map */
1173 tmp = entsize;
1174 if (ichdr.freemap[i].base < ichdr.firstused)
1175 tmp += sizeof(xfs_attr_leaf_entry_t);
1176 if (ichdr.freemap[i].size >= tmp) {
1177 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1178 goto out_log_hdr;
1179 }
1180 sum += ichdr.freemap[i].size;
1181 }
1182
1183 /*
1184 * If there are no holes in the address space of the block,
1185 * and we don't have enough freespace, then compaction will do us
1186 * no good and we should just give up.
1187 */
1188 if (!ichdr.holes && sum < entsize)
1189 return -ENOSPC;
1190
1191 /*
1192 * Compact the entries to coalesce free space.
1193 * This may change the hdr->count via dropping INCOMPLETE entries.
1194 */
1195 xfs_attr3_leaf_compact(args, &ichdr, bp);
1196
1197 /*
1198 * After compaction, the block is guaranteed to have only one
1199 * free region, in freemap[0]. If it is not big enough, give up.
1200 */
1201 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1202 tmp = -ENOSPC;
1203 goto out_log_hdr;
1204 }
1205
1206 tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1207
1208 out_log_hdr:
1209 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1210 xfs_trans_log_buf(args->trans, bp,
1211 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1212 xfs_attr3_leaf_hdr_size(leaf)));
1213 return tmp;
1214 }
1215
1216 /*
1217 * Add a name to a leaf attribute list structure.
1218 */
1219 STATIC int
1220 xfs_attr3_leaf_add_work(
1221 struct xfs_buf *bp,
1222 struct xfs_attr3_icleaf_hdr *ichdr,
1223 struct xfs_da_args *args,
1224 int mapindex)
1225 {
1226 struct xfs_attr_leafblock *leaf;
1227 struct xfs_attr_leaf_entry *entry;
1228 struct xfs_attr_leaf_name_local *name_loc;
1229 struct xfs_attr_leaf_name_remote *name_rmt;
1230 struct xfs_mount *mp;
1231 int tmp;
1232 int i;
1233
1234 trace_xfs_attr_leaf_add_work(args);
1235
1236 leaf = bp->b_addr;
1237 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1238 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1239
1240 /*
1241 * Force open some space in the entry array and fill it in.
1242 */
1243 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1244 if (args->index < ichdr->count) {
1245 tmp = ichdr->count - args->index;
1246 tmp *= sizeof(xfs_attr_leaf_entry_t);
1247 memmove(entry + 1, entry, tmp);
1248 xfs_trans_log_buf(args->trans, bp,
1249 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1250 }
1251 ichdr->count++;
1252
1253 /*
1254 * Allocate space for the new string (at the end of the run).
1255 */
1256 mp = args->trans->t_mountp;
1257 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1258 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1259 ASSERT(ichdr->freemap[mapindex].size >=
1260 xfs_attr_leaf_newentsize(args, NULL));
1261 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1262 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1263
1264 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1265
1266 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1267 ichdr->freemap[mapindex].size);
1268 entry->hashval = cpu_to_be32(args->hashval);
1269 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1270 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1271 if (args->op_flags & XFS_DA_OP_RENAME) {
1272 entry->flags |= XFS_ATTR_INCOMPLETE;
1273 if ((args->blkno2 == args->blkno) &&
1274 (args->index2 <= args->index)) {
1275 args->index2++;
1276 }
1277 }
1278 xfs_trans_log_buf(args->trans, bp,
1279 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1280 ASSERT((args->index == 0) ||
1281 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1282 ASSERT((args->index == ichdr->count - 1) ||
1283 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1284
1285 /*
1286 * For "remote" attribute values, simply note that we need to
1287 * allocate space for the "remote" value. We can't actually
1288 * allocate the extents in this transaction, and we can't decide
1289 * which blocks they should be as we might allocate more blocks
1290 * as part of this transaction (a split operation for example).
1291 */
1292 if (entry->flags & XFS_ATTR_LOCAL) {
1293 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1294 name_loc->namelen = args->namelen;
1295 name_loc->valuelen = cpu_to_be16(args->valuelen);
1296 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1297 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1298 be16_to_cpu(name_loc->valuelen));
1299 } else {
1300 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1301 name_rmt->namelen = args->namelen;
1302 memcpy((char *)name_rmt->name, args->name, args->namelen);
1303 entry->flags |= XFS_ATTR_INCOMPLETE;
1304 /* just in case */
1305 name_rmt->valuelen = 0;
1306 name_rmt->valueblk = 0;
1307 args->rmtblkno = 1;
1308 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1309 args->rmtvaluelen = args->valuelen;
1310 }
1311 xfs_trans_log_buf(args->trans, bp,
1312 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1313 xfs_attr_leaf_entsize(leaf, args->index)));
1314
1315 /*
1316 * Update the control info for this leaf node
1317 */
1318 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1319 ichdr->firstused = be16_to_cpu(entry->nameidx);
1320
1321 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1322 + xfs_attr3_leaf_hdr_size(leaf));
1323 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1324 + xfs_attr3_leaf_hdr_size(leaf);
1325
1326 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1327 if (ichdr->freemap[i].base == tmp) {
1328 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1329 ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1330 }
1331 }
1332 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1333 return 0;
1334 }
1335
1336 /*
1337 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1338 */
1339 STATIC void
1340 xfs_attr3_leaf_compact(
1341 struct xfs_da_args *args,
1342 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1343 struct xfs_buf *bp)
1344 {
1345 struct xfs_attr_leafblock *leaf_src;
1346 struct xfs_attr_leafblock *leaf_dst;
1347 struct xfs_attr3_icleaf_hdr ichdr_src;
1348 struct xfs_trans *trans = args->trans;
1349 char *tmpbuffer;
1350
1351 trace_xfs_attr_leaf_compact(args);
1352
1353 tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1354 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1355 memset(bp->b_addr, 0, args->geo->blksize);
1356 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1357 leaf_dst = bp->b_addr;
1358
1359 /*
1360 * Copy the on-disk header back into the destination buffer to ensure
1361 * all the information in the header that is not part of the incore
1362 * header structure is preserved.
1363 */
1364 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1365
1366 /* Initialise the incore headers */
1367 ichdr_src = *ichdr_dst; /* struct copy */
1368 ichdr_dst->firstused = args->geo->blksize;
1369 ichdr_dst->usedbytes = 0;
1370 ichdr_dst->count = 0;
1371 ichdr_dst->holes = 0;
1372 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1373 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1374 ichdr_dst->freemap[0].base;
1375
1376 /* write the header back to initialise the underlying buffer */
1377 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1378
1379 /*
1380 * Copy all entry's in the same (sorted) order,
1381 * but allocate name/value pairs packed and in sequence.
1382 */
1383 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1384 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1385 /*
1386 * this logs the entire buffer, but the caller must write the header
1387 * back to the buffer when it is finished modifying it.
1388 */
1389 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1390
1391 kmem_free(tmpbuffer);
1392 }
1393
1394 /*
1395 * Compare two leaf blocks "order".
1396 * Return 0 unless leaf2 should go before leaf1.
1397 */
1398 static int
1399 xfs_attr3_leaf_order(
1400 struct xfs_buf *leaf1_bp,
1401 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1402 struct xfs_buf *leaf2_bp,
1403 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1404 {
1405 struct xfs_attr_leaf_entry *entries1;
1406 struct xfs_attr_leaf_entry *entries2;
1407
1408 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1409 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1410 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1411 ((be32_to_cpu(entries2[0].hashval) <
1412 be32_to_cpu(entries1[0].hashval)) ||
1413 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1414 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1415 return 1;
1416 }
1417 return 0;
1418 }
1419
1420 int
1421 xfs_attr_leaf_order(
1422 struct xfs_buf *leaf1_bp,
1423 struct xfs_buf *leaf2_bp)
1424 {
1425 struct xfs_attr3_icleaf_hdr ichdr1;
1426 struct xfs_attr3_icleaf_hdr ichdr2;
1427 struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1428
1429 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1430 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1431 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1432 }
1433
1434 /*
1435 * Redistribute the attribute list entries between two leaf nodes,
1436 * taking into account the size of the new entry.
1437 *
1438 * NOTE: if new block is empty, then it will get the upper half of the
1439 * old block. At present, all (one) callers pass in an empty second block.
1440 *
1441 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1442 * to match what it is doing in splitting the attribute leaf block. Those
1443 * values are used in "atomic rename" operations on attributes. Note that
1444 * the "new" and "old" values can end up in different blocks.
1445 */
1446 STATIC void
1447 xfs_attr3_leaf_rebalance(
1448 struct xfs_da_state *state,
1449 struct xfs_da_state_blk *blk1,
1450 struct xfs_da_state_blk *blk2)
1451 {
1452 struct xfs_da_args *args;
1453 struct xfs_attr_leafblock *leaf1;
1454 struct xfs_attr_leafblock *leaf2;
1455 struct xfs_attr3_icleaf_hdr ichdr1;
1456 struct xfs_attr3_icleaf_hdr ichdr2;
1457 struct xfs_attr_leaf_entry *entries1;
1458 struct xfs_attr_leaf_entry *entries2;
1459 int count;
1460 int totallen;
1461 int max;
1462 int space;
1463 int swap;
1464
1465 /*
1466 * Set up environment.
1467 */
1468 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1469 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1470 leaf1 = blk1->bp->b_addr;
1471 leaf2 = blk2->bp->b_addr;
1472 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1473 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1474 ASSERT(ichdr2.count == 0);
1475 args = state->args;
1476
1477 trace_xfs_attr_leaf_rebalance(args);
1478
1479 /*
1480 * Check ordering of blocks, reverse if it makes things simpler.
1481 *
1482 * NOTE: Given that all (current) callers pass in an empty
1483 * second block, this code should never set "swap".
1484 */
1485 swap = 0;
1486 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1487 struct xfs_da_state_blk *tmp_blk;
1488 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1489
1490 tmp_blk = blk1;
1491 blk1 = blk2;
1492 blk2 = tmp_blk;
1493
1494 /* struct copies to swap them rather than reconverting */
1495 tmp_ichdr = ichdr1;
1496 ichdr1 = ichdr2;
1497 ichdr2 = tmp_ichdr;
1498
1499 leaf1 = blk1->bp->b_addr;
1500 leaf2 = blk2->bp->b_addr;
1501 swap = 1;
1502 }
1503
1504 /*
1505 * Examine entries until we reduce the absolute difference in
1506 * byte usage between the two blocks to a minimum. Then get
1507 * the direction to copy and the number of elements to move.
1508 *
1509 * "inleaf" is true if the new entry should be inserted into blk1.
1510 * If "swap" is also true, then reverse the sense of "inleaf".
1511 */
1512 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1513 blk2, &ichdr2,
1514 &count, &totallen);
1515 if (swap)
1516 state->inleaf = !state->inleaf;
1517
1518 /*
1519 * Move any entries required from leaf to leaf:
1520 */
1521 if (count < ichdr1.count) {
1522 /*
1523 * Figure the total bytes to be added to the destination leaf.
1524 */
1525 /* number entries being moved */
1526 count = ichdr1.count - count;
1527 space = ichdr1.usedbytes - totallen;
1528 space += count * sizeof(xfs_attr_leaf_entry_t);
1529
1530 /*
1531 * leaf2 is the destination, compact it if it looks tight.
1532 */
1533 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1534 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1535 if (space > max)
1536 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1537
1538 /*
1539 * Move high entries from leaf1 to low end of leaf2.
1540 */
1541 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1542 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1543
1544 } else if (count > ichdr1.count) {
1545 /*
1546 * I assert that since all callers pass in an empty
1547 * second buffer, this code should never execute.
1548 */
1549 ASSERT(0);
1550
1551 /*
1552 * Figure the total bytes to be added to the destination leaf.
1553 */
1554 /* number entries being moved */
1555 count -= ichdr1.count;
1556 space = totallen - ichdr1.usedbytes;
1557 space += count * sizeof(xfs_attr_leaf_entry_t);
1558
1559 /*
1560 * leaf1 is the destination, compact it if it looks tight.
1561 */
1562 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1563 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1564 if (space > max)
1565 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1566
1567 /*
1568 * Move low entries from leaf2 to high end of leaf1.
1569 */
1570 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1571 ichdr1.count, count);
1572 }
1573
1574 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1575 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1576 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1577 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1578
1579 /*
1580 * Copy out last hashval in each block for B-tree code.
1581 */
1582 entries1 = xfs_attr3_leaf_entryp(leaf1);
1583 entries2 = xfs_attr3_leaf_entryp(leaf2);
1584 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1585 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1586
1587 /*
1588 * Adjust the expected index for insertion.
1589 * NOTE: this code depends on the (current) situation that the
1590 * second block was originally empty.
1591 *
1592 * If the insertion point moved to the 2nd block, we must adjust
1593 * the index. We must also track the entry just following the
1594 * new entry for use in an "atomic rename" operation, that entry
1595 * is always the "old" entry and the "new" entry is what we are
1596 * inserting. The index/blkno fields refer to the "old" entry,
1597 * while the index2/blkno2 fields refer to the "new" entry.
1598 */
1599 if (blk1->index > ichdr1.count) {
1600 ASSERT(state->inleaf == 0);
1601 blk2->index = blk1->index - ichdr1.count;
1602 args->index = args->index2 = blk2->index;
1603 args->blkno = args->blkno2 = blk2->blkno;
1604 } else if (blk1->index == ichdr1.count) {
1605 if (state->inleaf) {
1606 args->index = blk1->index;
1607 args->blkno = blk1->blkno;
1608 args->index2 = 0;
1609 args->blkno2 = blk2->blkno;
1610 } else {
1611 /*
1612 * On a double leaf split, the original attr location
1613 * is already stored in blkno2/index2, so don't
1614 * overwrite it overwise we corrupt the tree.
1615 */
1616 blk2->index = blk1->index - ichdr1.count;
1617 args->index = blk2->index;
1618 args->blkno = blk2->blkno;
1619 if (!state->extravalid) {
1620 /*
1621 * set the new attr location to match the old
1622 * one and let the higher level split code
1623 * decide where in the leaf to place it.
1624 */
1625 args->index2 = blk2->index;
1626 args->blkno2 = blk2->blkno;
1627 }
1628 }
1629 } else {
1630 ASSERT(state->inleaf == 1);
1631 args->index = args->index2 = blk1->index;
1632 args->blkno = args->blkno2 = blk1->blkno;
1633 }
1634 }
1635
1636 /*
1637 * Examine entries until we reduce the absolute difference in
1638 * byte usage between the two blocks to a minimum.
1639 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1640 * GROT: there will always be enough room in either block for a new entry.
1641 * GROT: Do a double-split for this case?
1642 */
1643 STATIC int
1644 xfs_attr3_leaf_figure_balance(
1645 struct xfs_da_state *state,
1646 struct xfs_da_state_blk *blk1,
1647 struct xfs_attr3_icleaf_hdr *ichdr1,
1648 struct xfs_da_state_blk *blk2,
1649 struct xfs_attr3_icleaf_hdr *ichdr2,
1650 int *countarg,
1651 int *usedbytesarg)
1652 {
1653 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1654 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1655 struct xfs_attr_leaf_entry *entry;
1656 int count;
1657 int max;
1658 int index;
1659 int totallen = 0;
1660 int half;
1661 int lastdelta;
1662 int foundit = 0;
1663 int tmp;
1664
1665 /*
1666 * Examine entries until we reduce the absolute difference in
1667 * byte usage between the two blocks to a minimum.
1668 */
1669 max = ichdr1->count + ichdr2->count;
1670 half = (max + 1) * sizeof(*entry);
1671 half += ichdr1->usedbytes + ichdr2->usedbytes +
1672 xfs_attr_leaf_newentsize(state->args, NULL);
1673 half /= 2;
1674 lastdelta = state->args->geo->blksize;
1675 entry = xfs_attr3_leaf_entryp(leaf1);
1676 for (count = index = 0; count < max; entry++, index++, count++) {
1677
1678 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1679 /*
1680 * The new entry is in the first block, account for it.
1681 */
1682 if (count == blk1->index) {
1683 tmp = totallen + sizeof(*entry) +
1684 xfs_attr_leaf_newentsize(state->args, NULL);
1685 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1686 break;
1687 lastdelta = XFS_ATTR_ABS(half - tmp);
1688 totallen = tmp;
1689 foundit = 1;
1690 }
1691
1692 /*
1693 * Wrap around into the second block if necessary.
1694 */
1695 if (count == ichdr1->count) {
1696 leaf1 = leaf2;
1697 entry = xfs_attr3_leaf_entryp(leaf1);
1698 index = 0;
1699 }
1700
1701 /*
1702 * Figure out if next leaf entry would be too much.
1703 */
1704 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1705 index);
1706 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1707 break;
1708 lastdelta = XFS_ATTR_ABS(half - tmp);
1709 totallen = tmp;
1710 #undef XFS_ATTR_ABS
1711 }
1712
1713 /*
1714 * Calculate the number of usedbytes that will end up in lower block.
1715 * If new entry not in lower block, fix up the count.
1716 */
1717 totallen -= count * sizeof(*entry);
1718 if (foundit) {
1719 totallen -= sizeof(*entry) +
1720 xfs_attr_leaf_newentsize(state->args, NULL);
1721 }
1722
1723 *countarg = count;
1724 *usedbytesarg = totallen;
1725 return foundit;
1726 }
1727
1728 /*========================================================================
1729 * Routines used for shrinking the Btree.
1730 *========================================================================*/
1731
1732 /*
1733 * Check a leaf block and its neighbors to see if the block should be
1734 * collapsed into one or the other neighbor. Always keep the block
1735 * with the smaller block number.
1736 * If the current block is over 50% full, don't try to join it, return 0.
1737 * If the block is empty, fill in the state structure and return 2.
1738 * If it can be collapsed, fill in the state structure and return 1.
1739 * If nothing can be done, return 0.
1740 *
1741 * GROT: allow for INCOMPLETE entries in calculation.
1742 */
1743 int
1744 xfs_attr3_leaf_toosmall(
1745 struct xfs_da_state *state,
1746 int *action)
1747 {
1748 struct xfs_attr_leafblock *leaf;
1749 struct xfs_da_state_blk *blk;
1750 struct xfs_attr3_icleaf_hdr ichdr;
1751 struct xfs_buf *bp;
1752 xfs_dablk_t blkno;
1753 int bytes;
1754 int forward;
1755 int error;
1756 int retval;
1757 int i;
1758
1759 trace_xfs_attr_leaf_toosmall(state->args);
1760
1761 /*
1762 * Check for the degenerate case of the block being over 50% full.
1763 * If so, it's not worth even looking to see if we might be able
1764 * to coalesce with a sibling.
1765 */
1766 blk = &state->path.blk[ state->path.active-1 ];
1767 leaf = blk->bp->b_addr;
1768 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1769 bytes = xfs_attr3_leaf_hdr_size(leaf) +
1770 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1771 ichdr.usedbytes;
1772 if (bytes > (state->args->geo->blksize >> 1)) {
1773 *action = 0; /* blk over 50%, don't try to join */
1774 return 0;
1775 }
1776
1777 /*
1778 * Check for the degenerate case of the block being empty.
1779 * If the block is empty, we'll simply delete it, no need to
1780 * coalesce it with a sibling block. We choose (arbitrarily)
1781 * to merge with the forward block unless it is NULL.
1782 */
1783 if (ichdr.count == 0) {
1784 /*
1785 * Make altpath point to the block we want to keep and
1786 * path point to the block we want to drop (this one).
1787 */
1788 forward = (ichdr.forw != 0);
1789 memcpy(&state->altpath, &state->path, sizeof(state->path));
1790 error = xfs_da3_path_shift(state, &state->altpath, forward,
1791 0, &retval);
1792 if (error)
1793 return error;
1794 if (retval) {
1795 *action = 0;
1796 } else {
1797 *action = 2;
1798 }
1799 return 0;
1800 }
1801
1802 /*
1803 * Examine each sibling block to see if we can coalesce with
1804 * at least 25% free space to spare. We need to figure out
1805 * whether to merge with the forward or the backward block.
1806 * We prefer coalescing with the lower numbered sibling so as
1807 * to shrink an attribute list over time.
1808 */
1809 /* start with smaller blk num */
1810 forward = ichdr.forw < ichdr.back;
1811 for (i = 0; i < 2; forward = !forward, i++) {
1812 struct xfs_attr3_icleaf_hdr ichdr2;
1813 if (forward)
1814 blkno = ichdr.forw;
1815 else
1816 blkno = ichdr.back;
1817 if (blkno == 0)
1818 continue;
1819 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1820 blkno, -1, &bp);
1821 if (error)
1822 return error;
1823
1824 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1825
1826 bytes = state->args->geo->blksize -
1827 (state->args->geo->blksize >> 2) -
1828 ichdr.usedbytes - ichdr2.usedbytes -
1829 ((ichdr.count + ichdr2.count) *
1830 sizeof(xfs_attr_leaf_entry_t)) -
1831 xfs_attr3_leaf_hdr_size(leaf);
1832
1833 xfs_trans_brelse(state->args->trans, bp);
1834 if (bytes >= 0)
1835 break; /* fits with at least 25% to spare */
1836 }
1837 if (i >= 2) {
1838 *action = 0;
1839 return 0;
1840 }
1841
1842 /*
1843 * Make altpath point to the block we want to keep (the lower
1844 * numbered block) and path point to the block we want to drop.
1845 */
1846 memcpy(&state->altpath, &state->path, sizeof(state->path));
1847 if (blkno < blk->blkno) {
1848 error = xfs_da3_path_shift(state, &state->altpath, forward,
1849 0, &retval);
1850 } else {
1851 error = xfs_da3_path_shift(state, &state->path, forward,
1852 0, &retval);
1853 }
1854 if (error)
1855 return error;
1856 if (retval) {
1857 *action = 0;
1858 } else {
1859 *action = 1;
1860 }
1861 return 0;
1862 }
1863
1864 /*
1865 * Remove a name from the leaf attribute list structure.
1866 *
1867 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1868 * If two leaves are 37% full, when combined they will leave 25% free.
1869 */
1870 int
1871 xfs_attr3_leaf_remove(
1872 struct xfs_buf *bp,
1873 struct xfs_da_args *args)
1874 {
1875 struct xfs_attr_leafblock *leaf;
1876 struct xfs_attr3_icleaf_hdr ichdr;
1877 struct xfs_attr_leaf_entry *entry;
1878 int before;
1879 int after;
1880 int smallest;
1881 int entsize;
1882 int tablesize;
1883 int tmp;
1884 int i;
1885
1886 trace_xfs_attr_leaf_remove(args);
1887
1888 leaf = bp->b_addr;
1889 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1890
1891 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1892 ASSERT(args->index >= 0 && args->index < ichdr.count);
1893 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1894 xfs_attr3_leaf_hdr_size(leaf));
1895
1896 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1897
1898 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1899 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1900
1901 /*
1902 * Scan through free region table:
1903 * check for adjacency of free'd entry with an existing one,
1904 * find smallest free region in case we need to replace it,
1905 * adjust any map that borders the entry table,
1906 */
1907 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1908 + xfs_attr3_leaf_hdr_size(leaf);
1909 tmp = ichdr.freemap[0].size;
1910 before = after = -1;
1911 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1912 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1913 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1914 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1915 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1916 if (ichdr.freemap[i].base == tablesize) {
1917 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1918 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1919 }
1920
1921 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1922 be16_to_cpu(entry->nameidx)) {
1923 before = i;
1924 } else if (ichdr.freemap[i].base ==
1925 (be16_to_cpu(entry->nameidx) + entsize)) {
1926 after = i;
1927 } else if (ichdr.freemap[i].size < tmp) {
1928 tmp = ichdr.freemap[i].size;
1929 smallest = i;
1930 }
1931 }
1932
1933 /*
1934 * Coalesce adjacent freemap regions,
1935 * or replace the smallest region.
1936 */
1937 if ((before >= 0) || (after >= 0)) {
1938 if ((before >= 0) && (after >= 0)) {
1939 ichdr.freemap[before].size += entsize;
1940 ichdr.freemap[before].size += ichdr.freemap[after].size;
1941 ichdr.freemap[after].base = 0;
1942 ichdr.freemap[after].size = 0;
1943 } else if (before >= 0) {
1944 ichdr.freemap[before].size += entsize;
1945 } else {
1946 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1947 ichdr.freemap[after].size += entsize;
1948 }
1949 } else {
1950 /*
1951 * Replace smallest region (if it is smaller than free'd entry)
1952 */
1953 if (ichdr.freemap[smallest].size < entsize) {
1954 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1955 ichdr.freemap[smallest].size = entsize;
1956 }
1957 }
1958
1959 /*
1960 * Did we remove the first entry?
1961 */
1962 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1963 smallest = 1;
1964 else
1965 smallest = 0;
1966
1967 /*
1968 * Compress the remaining entries and zero out the removed stuff.
1969 */
1970 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1971 ichdr.usedbytes -= entsize;
1972 xfs_trans_log_buf(args->trans, bp,
1973 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1974 entsize));
1975
1976 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1977 memmove(entry, entry + 1, tmp);
1978 ichdr.count--;
1979 xfs_trans_log_buf(args->trans, bp,
1980 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1981
1982 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1983 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1984
1985 /*
1986 * If we removed the first entry, re-find the first used byte
1987 * in the name area. Note that if the entry was the "firstused",
1988 * then we don't have a "hole" in our block resulting from
1989 * removing the name.
1990 */
1991 if (smallest) {
1992 tmp = args->geo->blksize;
1993 entry = xfs_attr3_leaf_entryp(leaf);
1994 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1995 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1996 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1997
1998 if (be16_to_cpu(entry->nameidx) < tmp)
1999 tmp = be16_to_cpu(entry->nameidx);
2000 }
2001 ichdr.firstused = tmp;
2002 ASSERT(ichdr.firstused != 0);
2003 } else {
2004 ichdr.holes = 1; /* mark as needing compaction */
2005 }
2006 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2007 xfs_trans_log_buf(args->trans, bp,
2008 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2009 xfs_attr3_leaf_hdr_size(leaf)));
2010
2011 /*
2012 * Check if leaf is less than 50% full, caller may want to
2013 * "join" the leaf with a sibling if so.
2014 */
2015 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2016 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2017
2018 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2019 }
2020
2021 /*
2022 * Move all the attribute list entries from drop_leaf into save_leaf.
2023 */
2024 void
2025 xfs_attr3_leaf_unbalance(
2026 struct xfs_da_state *state,
2027 struct xfs_da_state_blk *drop_blk,
2028 struct xfs_da_state_blk *save_blk)
2029 {
2030 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2031 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2032 struct xfs_attr3_icleaf_hdr drophdr;
2033 struct xfs_attr3_icleaf_hdr savehdr;
2034 struct xfs_attr_leaf_entry *entry;
2035
2036 trace_xfs_attr_leaf_unbalance(state->args);
2037
2038 drop_leaf = drop_blk->bp->b_addr;
2039 save_leaf = save_blk->bp->b_addr;
2040 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2041 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2042 entry = xfs_attr3_leaf_entryp(drop_leaf);
2043
2044 /*
2045 * Save last hashval from dying block for later Btree fixup.
2046 */
2047 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2048
2049 /*
2050 * Check if we need a temp buffer, or can we do it in place.
2051 * Note that we don't check "leaf" for holes because we will
2052 * always be dropping it, toosmall() decided that for us already.
2053 */
2054 if (savehdr.holes == 0) {
2055 /*
2056 * dest leaf has no holes, so we add there. May need
2057 * to make some room in the entry array.
2058 */
2059 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2060 drop_blk->bp, &drophdr)) {
2061 xfs_attr3_leaf_moveents(state->args,
2062 drop_leaf, &drophdr, 0,
2063 save_leaf, &savehdr, 0,
2064 drophdr.count);
2065 } else {
2066 xfs_attr3_leaf_moveents(state->args,
2067 drop_leaf, &drophdr, 0,
2068 save_leaf, &savehdr,
2069 savehdr.count, drophdr.count);
2070 }
2071 } else {
2072 /*
2073 * Destination has holes, so we make a temporary copy
2074 * of the leaf and add them both to that.
2075 */
2076 struct xfs_attr_leafblock *tmp_leaf;
2077 struct xfs_attr3_icleaf_hdr tmphdr;
2078
2079 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2080
2081 /*
2082 * Copy the header into the temp leaf so that all the stuff
2083 * not in the incore header is present and gets copied back in
2084 * once we've moved all the entries.
2085 */
2086 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2087
2088 memset(&tmphdr, 0, sizeof(tmphdr));
2089 tmphdr.magic = savehdr.magic;
2090 tmphdr.forw = savehdr.forw;
2091 tmphdr.back = savehdr.back;
2092 tmphdr.firstused = state->args->geo->blksize;
2093
2094 /* write the header to the temp buffer to initialise it */
2095 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2096
2097 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2098 drop_blk->bp, &drophdr)) {
2099 xfs_attr3_leaf_moveents(state->args,
2100 drop_leaf, &drophdr, 0,
2101 tmp_leaf, &tmphdr, 0,
2102 drophdr.count);
2103 xfs_attr3_leaf_moveents(state->args,
2104 save_leaf, &savehdr, 0,
2105 tmp_leaf, &tmphdr, tmphdr.count,
2106 savehdr.count);
2107 } else {
2108 xfs_attr3_leaf_moveents(state->args,
2109 save_leaf, &savehdr, 0,
2110 tmp_leaf, &tmphdr, 0,
2111 savehdr.count);
2112 xfs_attr3_leaf_moveents(state->args,
2113 drop_leaf, &drophdr, 0,
2114 tmp_leaf, &tmphdr, tmphdr.count,
2115 drophdr.count);
2116 }
2117 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2118 savehdr = tmphdr; /* struct copy */
2119 kmem_free(tmp_leaf);
2120 }
2121
2122 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2123 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2124 state->args->geo->blksize - 1);
2125
2126 /*
2127 * Copy out last hashval in each block for B-tree code.
2128 */
2129 entry = xfs_attr3_leaf_entryp(save_leaf);
2130 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2131 }
2132
2133 /*========================================================================
2134 * Routines used for finding things in the Btree.
2135 *========================================================================*/
2136
2137 /*
2138 * Look up a name in a leaf attribute list structure.
2139 * This is the internal routine, it uses the caller's buffer.
2140 *
2141 * Note that duplicate keys are allowed, but only check within the
2142 * current leaf node. The Btree code must check in adjacent leaf nodes.
2143 *
2144 * Return in args->index the index into the entry[] array of either
2145 * the found entry, or where the entry should have been (insert before
2146 * that entry).
2147 *
2148 * Don't change the args->value unless we find the attribute.
2149 */
2150 int
2151 xfs_attr3_leaf_lookup_int(
2152 struct xfs_buf *bp,
2153 struct xfs_da_args *args)
2154 {
2155 struct xfs_attr_leafblock *leaf;
2156 struct xfs_attr3_icleaf_hdr ichdr;
2157 struct xfs_attr_leaf_entry *entry;
2158 struct xfs_attr_leaf_entry *entries;
2159 struct xfs_attr_leaf_name_local *name_loc;
2160 struct xfs_attr_leaf_name_remote *name_rmt;
2161 xfs_dahash_t hashval;
2162 int probe;
2163 int span;
2164
2165 trace_xfs_attr_leaf_lookup(args);
2166
2167 leaf = bp->b_addr;
2168 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2169 entries = xfs_attr3_leaf_entryp(leaf);
2170 ASSERT(ichdr.count < args->geo->blksize / 8);
2171
2172 /*
2173 * Binary search. (note: small blocks will skip this loop)
2174 */
2175 hashval = args->hashval;
2176 probe = span = ichdr.count / 2;
2177 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2178 span /= 2;
2179 if (be32_to_cpu(entry->hashval) < hashval)
2180 probe += span;
2181 else if (be32_to_cpu(entry->hashval) > hashval)
2182 probe -= span;
2183 else
2184 break;
2185 }
2186 ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2187 ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2188
2189 /*
2190 * Since we may have duplicate hashval's, find the first matching
2191 * hashval in the leaf.
2192 */
2193 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2194 entry--;
2195 probe--;
2196 }
2197 while (probe < ichdr.count &&
2198 be32_to_cpu(entry->hashval) < hashval) {
2199 entry++;
2200 probe++;
2201 }
2202 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2203 args->index = probe;
2204 return -ENOATTR;
2205 }
2206
2207 /*
2208 * Duplicate keys may be present, so search all of them for a match.
2209 */
2210 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2211 entry++, probe++) {
2212 /*
2213 * GROT: Add code to remove incomplete entries.
2214 */
2215 /*
2216 * If we are looking for INCOMPLETE entries, show only those.
2217 * If we are looking for complete entries, show only those.
2218 */
2219 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2220 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2221 continue;
2222 }
2223 if (entry->flags & XFS_ATTR_LOCAL) {
2224 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2225 if (name_loc->namelen != args->namelen)
2226 continue;
2227 if (memcmp(args->name, name_loc->nameval,
2228 args->namelen) != 0)
2229 continue;
2230 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2231 continue;
2232 args->index = probe;
2233 return -EEXIST;
2234 } else {
2235 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2236 if (name_rmt->namelen != args->namelen)
2237 continue;
2238 if (memcmp(args->name, name_rmt->name,
2239 args->namelen) != 0)
2240 continue;
2241 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2242 continue;
2243 args->index = probe;
2244 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2245 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2246 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2247 args->dp->i_mount,
2248 args->rmtvaluelen);
2249 return -EEXIST;
2250 }
2251 }
2252 args->index = probe;
2253 return -ENOATTR;
2254 }
2255
2256 /*
2257 * Get the value associated with an attribute name from a leaf attribute
2258 * list structure.
2259 */
2260 int
2261 xfs_attr3_leaf_getvalue(
2262 struct xfs_buf *bp,
2263 struct xfs_da_args *args)
2264 {
2265 struct xfs_attr_leafblock *leaf;
2266 struct xfs_attr3_icleaf_hdr ichdr;
2267 struct xfs_attr_leaf_entry *entry;
2268 struct xfs_attr_leaf_name_local *name_loc;
2269 struct xfs_attr_leaf_name_remote *name_rmt;
2270 int valuelen;
2271
2272 leaf = bp->b_addr;
2273 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2274 ASSERT(ichdr.count < args->geo->blksize / 8);
2275 ASSERT(args->index < ichdr.count);
2276
2277 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2278 if (entry->flags & XFS_ATTR_LOCAL) {
2279 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2280 ASSERT(name_loc->namelen == args->namelen);
2281 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2282 valuelen = be16_to_cpu(name_loc->valuelen);
2283 if (args->flags & ATTR_KERNOVAL) {
2284 args->valuelen = valuelen;
2285 return 0;
2286 }
2287 if (args->valuelen < valuelen) {
2288 args->valuelen = valuelen;
2289 return -ERANGE;
2290 }
2291 args->valuelen = valuelen;
2292 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2293 } else {
2294 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2295 ASSERT(name_rmt->namelen == args->namelen);
2296 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2297 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2298 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2299 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2300 args->rmtvaluelen);
2301 if (args->flags & ATTR_KERNOVAL) {
2302 args->valuelen = args->rmtvaluelen;
2303 return 0;
2304 }
2305 if (args->valuelen < args->rmtvaluelen) {
2306 args->valuelen = args->rmtvaluelen;
2307 return -ERANGE;
2308 }
2309 args->valuelen = args->rmtvaluelen;
2310 }
2311 return 0;
2312 }
2313
2314 /*========================================================================
2315 * Utility routines.
2316 *========================================================================*/
2317
2318 /*
2319 * Move the indicated entries from one leaf to another.
2320 * NOTE: this routine modifies both source and destination leaves.
2321 */
2322 /*ARGSUSED*/
2323 STATIC void
2324 xfs_attr3_leaf_moveents(
2325 struct xfs_da_args *args,
2326 struct xfs_attr_leafblock *leaf_s,
2327 struct xfs_attr3_icleaf_hdr *ichdr_s,
2328 int start_s,
2329 struct xfs_attr_leafblock *leaf_d,
2330 struct xfs_attr3_icleaf_hdr *ichdr_d,
2331 int start_d,
2332 int count)
2333 {
2334 struct xfs_attr_leaf_entry *entry_s;
2335 struct xfs_attr_leaf_entry *entry_d;
2336 int desti;
2337 int tmp;
2338 int i;
2339
2340 /*
2341 * Check for nothing to do.
2342 */
2343 if (count == 0)
2344 return;
2345
2346 /*
2347 * Set up environment.
2348 */
2349 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2350 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2351 ASSERT(ichdr_s->magic == ichdr_d->magic);
2352 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2353 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2354 + xfs_attr3_leaf_hdr_size(leaf_s));
2355 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2356 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2357 + xfs_attr3_leaf_hdr_size(leaf_d));
2358
2359 ASSERT(start_s < ichdr_s->count);
2360 ASSERT(start_d <= ichdr_d->count);
2361 ASSERT(count <= ichdr_s->count);
2362
2363
2364 /*
2365 * Move the entries in the destination leaf up to make a hole?
2366 */
2367 if (start_d < ichdr_d->count) {
2368 tmp = ichdr_d->count - start_d;
2369 tmp *= sizeof(xfs_attr_leaf_entry_t);
2370 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2371 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2372 memmove(entry_d, entry_s, tmp);
2373 }
2374
2375 /*
2376 * Copy all entry's in the same (sorted) order,
2377 * but allocate attribute info packed and in sequence.
2378 */
2379 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2380 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2381 desti = start_d;
2382 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2383 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2384 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2385 #ifdef GROT
2386 /*
2387 * Code to drop INCOMPLETE entries. Difficult to use as we
2388 * may also need to change the insertion index. Code turned
2389 * off for 6.2, should be revisited later.
2390 */
2391 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2392 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2393 ichdr_s->usedbytes -= tmp;
2394 ichdr_s->count -= 1;
2395 entry_d--; /* to compensate for ++ in loop hdr */
2396 desti--;
2397 if ((start_s + i) < offset)
2398 result++; /* insertion index adjustment */
2399 } else {
2400 #endif /* GROT */
2401 ichdr_d->firstused -= tmp;
2402 /* both on-disk, don't endian flip twice */
2403 entry_d->hashval = entry_s->hashval;
2404 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2405 entry_d->flags = entry_s->flags;
2406 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2407 <= args->geo->blksize);
2408 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2409 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2410 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2411 <= args->geo->blksize);
2412 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2413 ichdr_s->usedbytes -= tmp;
2414 ichdr_d->usedbytes += tmp;
2415 ichdr_s->count -= 1;
2416 ichdr_d->count += 1;
2417 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2418 + xfs_attr3_leaf_hdr_size(leaf_d);
2419 ASSERT(ichdr_d->firstused >= tmp);
2420 #ifdef GROT
2421 }
2422 #endif /* GROT */
2423 }
2424
2425 /*
2426 * Zero out the entries we just copied.
2427 */
2428 if (start_s == ichdr_s->count) {
2429 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2430 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2431 ASSERT(((char *)entry_s + tmp) <=
2432 ((char *)leaf_s + args->geo->blksize));
2433 memset(entry_s, 0, tmp);
2434 } else {
2435 /*
2436 * Move the remaining entries down to fill the hole,
2437 * then zero the entries at the top.
2438 */
2439 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2440 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2441 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2442 memmove(entry_d, entry_s, tmp);
2443
2444 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2445 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2446 ASSERT(((char *)entry_s + tmp) <=
2447 ((char *)leaf_s + args->geo->blksize));
2448 memset(entry_s, 0, tmp);
2449 }
2450
2451 /*
2452 * Fill in the freemap information
2453 */
2454 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2455 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2456 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2457 ichdr_d->freemap[1].base = 0;
2458 ichdr_d->freemap[2].base = 0;
2459 ichdr_d->freemap[1].size = 0;
2460 ichdr_d->freemap[2].size = 0;
2461 ichdr_s->holes = 1; /* leaf may not be compact */
2462 }
2463
2464 /*
2465 * Pick up the last hashvalue from a leaf block.
2466 */
2467 xfs_dahash_t
2468 xfs_attr_leaf_lasthash(
2469 struct xfs_buf *bp,
2470 int *count)
2471 {
2472 struct xfs_attr3_icleaf_hdr ichdr;
2473 struct xfs_attr_leaf_entry *entries;
2474 struct xfs_mount *mp = bp->b_target->bt_mount;
2475
2476 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2477 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2478 if (count)
2479 *count = ichdr.count;
2480 if (!ichdr.count)
2481 return 0;
2482 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2483 }
2484
2485 /*
2486 * Calculate the number of bytes used to store the indicated attribute
2487 * (whether local or remote only calculate bytes in this block).
2488 */
2489 STATIC int
2490 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2491 {
2492 struct xfs_attr_leaf_entry *entries;
2493 xfs_attr_leaf_name_local_t *name_loc;
2494 xfs_attr_leaf_name_remote_t *name_rmt;
2495 int size;
2496
2497 entries = xfs_attr3_leaf_entryp(leaf);
2498 if (entries[index].flags & XFS_ATTR_LOCAL) {
2499 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2500 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2501 be16_to_cpu(name_loc->valuelen));
2502 } else {
2503 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2504 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2505 }
2506 return size;
2507 }
2508
2509 /*
2510 * Calculate the number of bytes that would be required to store the new
2511 * attribute (whether local or remote only calculate bytes in this block).
2512 * This routine decides as a side effect whether the attribute will be
2513 * a "local" or a "remote" attribute.
2514 */
2515 int
2516 xfs_attr_leaf_newentsize(
2517 struct xfs_da_args *args,
2518 int *local)
2519 {
2520 int size;
2521
2522 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2523 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2524 if (local)
2525 *local = 1;
2526 return size;
2527 }
2528 if (local)
2529 *local = 0;
2530 return xfs_attr_leaf_entsize_remote(args->namelen);
2531 }
2532
2533
2534 /*========================================================================
2535 * Manage the INCOMPLETE flag in a leaf entry
2536 *========================================================================*/
2537
2538 /*
2539 * Clear the INCOMPLETE flag on an entry in a leaf block.
2540 */
2541 int
2542 xfs_attr3_leaf_clearflag(
2543 struct xfs_da_args *args)
2544 {
2545 struct xfs_attr_leafblock *leaf;
2546 struct xfs_attr_leaf_entry *entry;
2547 struct xfs_attr_leaf_name_remote *name_rmt;
2548 struct xfs_buf *bp;
2549 int error;
2550 #ifdef DEBUG
2551 struct xfs_attr3_icleaf_hdr ichdr;
2552 xfs_attr_leaf_name_local_t *name_loc;
2553 int namelen;
2554 char *name;
2555 #endif /* DEBUG */
2556
2557 trace_xfs_attr_leaf_clearflag(args);
2558 /*
2559 * Set up the operation.
2560 */
2561 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2562 if (error)
2563 return error;
2564
2565 leaf = bp->b_addr;
2566 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2567 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2568
2569 #ifdef DEBUG
2570 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2571 ASSERT(args->index < ichdr.count);
2572 ASSERT(args->index >= 0);
2573
2574 if (entry->flags & XFS_ATTR_LOCAL) {
2575 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2576 namelen = name_loc->namelen;
2577 name = (char *)name_loc->nameval;
2578 } else {
2579 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2580 namelen = name_rmt->namelen;
2581 name = (char *)name_rmt->name;
2582 }
2583 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2584 ASSERT(namelen == args->namelen);
2585 ASSERT(memcmp(name, args->name, namelen) == 0);
2586 #endif /* DEBUG */
2587
2588 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2589 xfs_trans_log_buf(args->trans, bp,
2590 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2591
2592 if (args->rmtblkno) {
2593 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2594 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2595 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2596 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2597 xfs_trans_log_buf(args->trans, bp,
2598 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2599 }
2600
2601 /*
2602 * Commit the flag value change and start the next trans in series.
2603 */
2604 return xfs_trans_roll(&args->trans, args->dp);
2605 }
2606
2607 /*
2608 * Set the INCOMPLETE flag on an entry in a leaf block.
2609 */
2610 int
2611 xfs_attr3_leaf_setflag(
2612 struct xfs_da_args *args)
2613 {
2614 struct xfs_attr_leafblock *leaf;
2615 struct xfs_attr_leaf_entry *entry;
2616 struct xfs_attr_leaf_name_remote *name_rmt;
2617 struct xfs_buf *bp;
2618 int error;
2619 #ifdef DEBUG
2620 struct xfs_attr3_icleaf_hdr ichdr;
2621 #endif
2622
2623 trace_xfs_attr_leaf_setflag(args);
2624
2625 /*
2626 * Set up the operation.
2627 */
2628 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2629 if (error)
2630 return error;
2631
2632 leaf = bp->b_addr;
2633 #ifdef DEBUG
2634 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2635 ASSERT(args->index < ichdr.count);
2636 ASSERT(args->index >= 0);
2637 #endif
2638 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2639
2640 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2641 entry->flags |= XFS_ATTR_INCOMPLETE;
2642 xfs_trans_log_buf(args->trans, bp,
2643 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2644 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2645 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2646 name_rmt->valueblk = 0;
2647 name_rmt->valuelen = 0;
2648 xfs_trans_log_buf(args->trans, bp,
2649 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2650 }
2651
2652 /*
2653 * Commit the flag value change and start the next trans in series.
2654 */
2655 return xfs_trans_roll(&args->trans, args->dp);
2656 }
2657
2658 /*
2659 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2660 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2661 * entry given by args->blkno2/index2.
2662 *
2663 * Note that they could be in different blocks, or in the same block.
2664 */
2665 int
2666 xfs_attr3_leaf_flipflags(
2667 struct xfs_da_args *args)
2668 {
2669 struct xfs_attr_leafblock *leaf1;
2670 struct xfs_attr_leafblock *leaf2;
2671 struct xfs_attr_leaf_entry *entry1;
2672 struct xfs_attr_leaf_entry *entry2;
2673 struct xfs_attr_leaf_name_remote *name_rmt;
2674 struct xfs_buf *bp1;
2675 struct xfs_buf *bp2;
2676 int error;
2677 #ifdef DEBUG
2678 struct xfs_attr3_icleaf_hdr ichdr1;
2679 struct xfs_attr3_icleaf_hdr ichdr2;
2680 xfs_attr_leaf_name_local_t *name_loc;
2681 int namelen1, namelen2;
2682 char *name1, *name2;
2683 #endif /* DEBUG */
2684
2685 trace_xfs_attr_leaf_flipflags(args);
2686
2687 /*
2688 * Read the block containing the "old" attr
2689 */
2690 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2691 if (error)
2692 return error;
2693
2694 /*
2695 * Read the block containing the "new" attr, if it is different
2696 */
2697 if (args->blkno2 != args->blkno) {
2698 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2699 -1, &bp2);
2700 if (error)
2701 return error;
2702 } else {
2703 bp2 = bp1;
2704 }
2705
2706 leaf1 = bp1->b_addr;
2707 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2708
2709 leaf2 = bp2->b_addr;
2710 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2711
2712 #ifdef DEBUG
2713 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2714 ASSERT(args->index < ichdr1.count);
2715 ASSERT(args->index >= 0);
2716
2717 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2718 ASSERT(args->index2 < ichdr2.count);
2719 ASSERT(args->index2 >= 0);
2720
2721 if (entry1->flags & XFS_ATTR_LOCAL) {
2722 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2723 namelen1 = name_loc->namelen;
2724 name1 = (char *)name_loc->nameval;
2725 } else {
2726 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2727 namelen1 = name_rmt->namelen;
2728 name1 = (char *)name_rmt->name;
2729 }
2730 if (entry2->flags & XFS_ATTR_LOCAL) {
2731 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2732 namelen2 = name_loc->namelen;
2733 name2 = (char *)name_loc->nameval;
2734 } else {
2735 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2736 namelen2 = name_rmt->namelen;
2737 name2 = (char *)name_rmt->name;
2738 }
2739 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2740 ASSERT(namelen1 == namelen2);
2741 ASSERT(memcmp(name1, name2, namelen1) == 0);
2742 #endif /* DEBUG */
2743
2744 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2745 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2746
2747 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2748 xfs_trans_log_buf(args->trans, bp1,
2749 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2750 if (args->rmtblkno) {
2751 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2752 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2753 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2754 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2755 xfs_trans_log_buf(args->trans, bp1,
2756 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2757 }
2758
2759 entry2->flags |= XFS_ATTR_INCOMPLETE;
2760 xfs_trans_log_buf(args->trans, bp2,
2761 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2762 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2763 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2764 name_rmt->valueblk = 0;
2765 name_rmt->valuelen = 0;
2766 xfs_trans_log_buf(args->trans, bp2,
2767 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2768 }
2769
2770 /*
2771 * Commit the flag value change and start the next trans in series.
2772 */
2773 error = xfs_trans_roll(&args->trans, args->dp);
2774
2775 return error;
2776 }