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