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