1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
17 #include "xfs_inode.h"
18 #include "xfs_trans.h"
19 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap.h"
21 #include "xfs_attr_sf.h"
22 #include "xfs_attr.h"
23 #include "xfs_attr_remote.h"
24 #include "xfs_attr_leaf.h"
25 #include "xfs_error.h"
26 #include "xfs_trace.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_dir2.h"
29 #include "xfs_log.h"
30 #include "xfs_ag.h"
31 #include "xfs_errortag.h"
32
33
34 /*
35 * xfs_attr_leaf.c
36 *
37 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
38 */
39
40 /*========================================================================
41 * Function prototypes for the kernel.
42 *========================================================================*/
43
44 /*
45 * Routines used for growing the Btree.
46 */
47 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
48 xfs_dablk_t which_block, struct xfs_buf **bpp);
49 STATIC void xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
50 struct xfs_attr3_icleaf_hdr *ichdr,
51 struct xfs_da_args *args, int freemap_index);
52 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
53 struct xfs_attr3_icleaf_hdr *ichdr,
54 struct xfs_buf *leaf_buffer);
55 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
56 xfs_da_state_blk_t *blk1,
57 xfs_da_state_blk_t *blk2);
58 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
59 xfs_da_state_blk_t *leaf_blk_1,
60 struct xfs_attr3_icleaf_hdr *ichdr1,
61 xfs_da_state_blk_t *leaf_blk_2,
62 struct xfs_attr3_icleaf_hdr *ichdr2,
63 int *number_entries_in_blk1,
64 int *number_usedbytes_in_blk1);
65
66 /*
67 * Utility routines.
68 */
69 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
70 struct xfs_attr_leafblock *src_leaf,
71 struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
72 struct xfs_attr_leafblock *dst_leaf,
73 struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
74 int move_count);
75 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
76
77 /*
78 * attr3 block 'firstused' conversion helpers.
79 *
80 * firstused refers to the offset of the first used byte of the nameval region
81 * of an attr leaf block. The region starts at the tail of the block and expands
82 * backwards towards the middle. As such, firstused is initialized to the block
83 * size for an empty leaf block and is reduced from there.
84 *
85 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
86 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
87 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
88 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
89 * the attr block size. The following helpers manage the conversion between the
90 * in-core and on-disk formats.
91 */
92
93 static void
xfs_attr3_leaf_firstused_from_disk(struct xfs_da_geometry * geo,struct xfs_attr3_icleaf_hdr * to,struct xfs_attr_leafblock * from)94 xfs_attr3_leaf_firstused_from_disk(
95 struct xfs_da_geometry *geo,
96 struct xfs_attr3_icleaf_hdr *to,
97 struct xfs_attr_leafblock *from)
98 {
99 struct xfs_attr3_leaf_hdr *hdr3;
100
101 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
102 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
103 to->firstused = be16_to_cpu(hdr3->firstused);
104 } else {
105 to->firstused = be16_to_cpu(from->hdr.firstused);
106 }
107
108 /*
109 * Convert from the magic fsb size value to actual blocksize. This
110 * should only occur for empty blocks when the block size overflows
111 * 16-bits.
112 */
113 if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
114 ASSERT(!to->count && !to->usedbytes);
115 ASSERT(geo->blksize > USHRT_MAX);
116 to->firstused = geo->blksize;
117 }
118 }
119
120 static void
xfs_attr3_leaf_firstused_to_disk(struct xfs_da_geometry * geo,struct xfs_attr_leafblock * to,struct xfs_attr3_icleaf_hdr * from)121 xfs_attr3_leaf_firstused_to_disk(
122 struct xfs_da_geometry *geo,
123 struct xfs_attr_leafblock *to,
124 struct xfs_attr3_icleaf_hdr *from)
125 {
126 struct xfs_attr3_leaf_hdr *hdr3;
127 uint32_t firstused;
128
129 /* magic value should only be seen on disk */
130 ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
131
132 /*
133 * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
134 * value. This only overflows at the max supported value of 64k. Use the
135 * magic on-disk value to represent block size in this case.
136 */
137 firstused = from->firstused;
138 if (firstused > USHRT_MAX) {
139 ASSERT(from->firstused == geo->blksize);
140 firstused = XFS_ATTR3_LEAF_NULLOFF;
141 }
142
143 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
144 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
145 hdr3->firstused = cpu_to_be16(firstused);
146 } else {
147 to->hdr.firstused = cpu_to_be16(firstused);
148 }
149 }
150
151 void
xfs_attr3_leaf_hdr_from_disk(struct xfs_da_geometry * geo,struct xfs_attr3_icleaf_hdr * to,struct xfs_attr_leafblock * from)152 xfs_attr3_leaf_hdr_from_disk(
153 struct xfs_da_geometry *geo,
154 struct xfs_attr3_icleaf_hdr *to,
155 struct xfs_attr_leafblock *from)
156 {
157 int i;
158
159 ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
160 from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
161
162 if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
163 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
164
165 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
166 to->back = be32_to_cpu(hdr3->info.hdr.back);
167 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
168 to->count = be16_to_cpu(hdr3->count);
169 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
170 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
171 to->holes = hdr3->holes;
172
173 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
174 to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
175 to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
176 }
177 return;
178 }
179 to->forw = be32_to_cpu(from->hdr.info.forw);
180 to->back = be32_to_cpu(from->hdr.info.back);
181 to->magic = be16_to_cpu(from->hdr.info.magic);
182 to->count = be16_to_cpu(from->hdr.count);
183 to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
184 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
185 to->holes = from->hdr.holes;
186
187 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
188 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
189 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
190 }
191 }
192
193 void
xfs_attr3_leaf_hdr_to_disk(struct xfs_da_geometry * geo,struct xfs_attr_leafblock * to,struct xfs_attr3_icleaf_hdr * from)194 xfs_attr3_leaf_hdr_to_disk(
195 struct xfs_da_geometry *geo,
196 struct xfs_attr_leafblock *to,
197 struct xfs_attr3_icleaf_hdr *from)
198 {
199 int i;
200
201 ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
202 from->magic == XFS_ATTR3_LEAF_MAGIC);
203
204 if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
205 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
206
207 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
208 hdr3->info.hdr.back = cpu_to_be32(from->back);
209 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
210 hdr3->count = cpu_to_be16(from->count);
211 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
212 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
213 hdr3->holes = from->holes;
214 hdr3->pad1 = 0;
215
216 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
217 hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
218 hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
219 }
220 return;
221 }
222 to->hdr.info.forw = cpu_to_be32(from->forw);
223 to->hdr.info.back = cpu_to_be32(from->back);
224 to->hdr.info.magic = cpu_to_be16(from->magic);
225 to->hdr.count = cpu_to_be16(from->count);
226 to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
227 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
228 to->hdr.holes = from->holes;
229 to->hdr.pad1 = 0;
230
231 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
232 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
233 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
234 }
235 }
236
237 static xfs_failaddr_t
xfs_attr3_leaf_verify_entry(struct xfs_mount * mp,char * buf_end,struct xfs_attr_leafblock * leaf,struct xfs_attr3_icleaf_hdr * leafhdr,struct xfs_attr_leaf_entry * ent,int idx,__u32 * last_hashval)238 xfs_attr3_leaf_verify_entry(
239 struct xfs_mount *mp,
240 char *buf_end,
241 struct xfs_attr_leafblock *leaf,
242 struct xfs_attr3_icleaf_hdr *leafhdr,
243 struct xfs_attr_leaf_entry *ent,
244 int idx,
245 __u32 *last_hashval)
246 {
247 struct xfs_attr_leaf_name_local *lentry;
248 struct xfs_attr_leaf_name_remote *rentry;
249 char *name_end;
250 unsigned int nameidx;
251 unsigned int namesize;
252 __u32 hashval;
253
254 /* hash order check */
255 hashval = be32_to_cpu(ent->hashval);
256 if (hashval < *last_hashval)
257 return __this_address;
258 *last_hashval = hashval;
259
260 nameidx = be16_to_cpu(ent->nameidx);
261 if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
262 return __this_address;
263
264 /*
265 * Check the name information. The namelen fields are u8 so we can't
266 * possibly exceed the maximum name length of 255 bytes.
267 */
268 if (ent->flags & XFS_ATTR_LOCAL) {
269 lentry = xfs_attr3_leaf_name_local(leaf, idx);
270 namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
271 be16_to_cpu(lentry->valuelen));
272 name_end = (char *)lentry + namesize;
273 if (lentry->namelen == 0)
274 return __this_address;
275 } else {
276 rentry = xfs_attr3_leaf_name_remote(leaf, idx);
277 namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
278 name_end = (char *)rentry + namesize;
279 if (rentry->namelen == 0)
280 return __this_address;
281 if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
282 rentry->valueblk == 0)
283 return __this_address;
284 }
285
286 if (name_end > buf_end)
287 return __this_address;
288
289 return NULL;
290 }
291
292 /*
293 * Validate an attribute leaf block.
294 *
295 * Empty leaf blocks can occur under the following circumstances:
296 *
297 * 1. setxattr adds a new extended attribute to a file;
298 * 2. The file has zero existing attributes;
299 * 3. The attribute is too large to fit in the attribute fork;
300 * 4. The attribute is small enough to fit in a leaf block;
301 * 5. A log flush occurs after committing the transaction that creates
302 * the (empty) leaf block; and
303 * 6. The filesystem goes down after the log flush but before the new
304 * attribute can be committed to the leaf block.
305 *
306 * Hence we need to ensure that we don't fail the validation purely
307 * because the leaf is empty.
308 */
309 static xfs_failaddr_t
xfs_attr3_leaf_verify(struct xfs_buf * bp)310 xfs_attr3_leaf_verify(
311 struct xfs_buf *bp)
312 {
313 struct xfs_attr3_icleaf_hdr ichdr;
314 struct xfs_mount *mp = bp->b_mount;
315 struct xfs_attr_leafblock *leaf = bp->b_addr;
316 struct xfs_attr_leaf_entry *entries;
317 struct xfs_attr_leaf_entry *ent;
318 char *buf_end;
319 uint32_t end; /* must be 32bit - see below */
320 __u32 last_hashval = 0;
321 int i;
322 xfs_failaddr_t fa;
323
324 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
325
326 fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
327 if (fa)
328 return fa;
329
330 /*
331 * firstused is the block offset of the first name info structure.
332 * Make sure it doesn't go off the block or crash into the header.
333 */
334 if (ichdr.firstused > mp->m_attr_geo->blksize)
335 return __this_address;
336 if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
337 return __this_address;
338
339 /* Make sure the entries array doesn't crash into the name info. */
340 entries = xfs_attr3_leaf_entryp(bp->b_addr);
341 if ((char *)&entries[ichdr.count] >
342 (char *)bp->b_addr + ichdr.firstused)
343 return __this_address;
344
345 /*
346 * NOTE: This verifier historically failed empty leaf buffers because
347 * we expect the fork to be in another format. Empty attr fork format
348 * conversions are possible during xattr set, however, and format
349 * conversion is not atomic with the xattr set that triggers it. We
350 * cannot assume leaf blocks are non-empty until that is addressed.
351 */
352 buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
353 for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
354 fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
355 ent, i, &last_hashval);
356 if (fa)
357 return fa;
358 }
359
360 /*
361 * Quickly check the freemap information. Attribute data has to be
362 * aligned to 4-byte boundaries, and likewise for the free space.
363 *
364 * Note that for 64k block size filesystems, the freemap entries cannot
365 * overflow as they are only be16 fields. However, when checking end
366 * pointer of the freemap, we have to be careful to detect overflows and
367 * so use uint32_t for those checks.
368 */
369 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
370 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
371 return __this_address;
372 if (ichdr.freemap[i].base & 0x3)
373 return __this_address;
374 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
375 return __this_address;
376 if (ichdr.freemap[i].size & 0x3)
377 return __this_address;
378
379 /* be care of 16 bit overflows here */
380 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
381 if (end < ichdr.freemap[i].base)
382 return __this_address;
383 if (end > mp->m_attr_geo->blksize)
384 return __this_address;
385 }
386
387 return NULL;
388 }
389
390 static void
xfs_attr3_leaf_write_verify(struct xfs_buf * bp)391 xfs_attr3_leaf_write_verify(
392 struct xfs_buf *bp)
393 {
394 struct xfs_mount *mp = bp->b_mount;
395 struct xfs_buf_log_item *bip = bp->b_log_item;
396 struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
397 xfs_failaddr_t fa;
398
399 fa = xfs_attr3_leaf_verify(bp);
400 if (fa) {
401 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
402 return;
403 }
404
405 if (!xfs_has_crc(mp))
406 return;
407
408 if (bip)
409 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
410
411 xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
412 }
413
414 /*
415 * leaf/node format detection on trees is sketchy, so a node read can be done on
416 * leaf level blocks when detection identifies the tree as a node format tree
417 * incorrectly. In this case, we need to swap the verifier to match the correct
418 * format of the block being read.
419 */
420 static void
xfs_attr3_leaf_read_verify(struct xfs_buf * bp)421 xfs_attr3_leaf_read_verify(
422 struct xfs_buf *bp)
423 {
424 struct xfs_mount *mp = bp->b_mount;
425 xfs_failaddr_t fa;
426
427 if (xfs_has_crc(mp) &&
428 !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
429 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
430 else {
431 fa = xfs_attr3_leaf_verify(bp);
432 if (fa)
433 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
434 }
435 }
436
437 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
438 .name = "xfs_attr3_leaf",
439 .magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
440 cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
441 .verify_read = xfs_attr3_leaf_read_verify,
442 .verify_write = xfs_attr3_leaf_write_verify,
443 .verify_struct = xfs_attr3_leaf_verify,
444 };
445
446 int
xfs_attr3_leaf_read(struct xfs_trans * tp,struct xfs_inode * dp,xfs_dablk_t bno,struct xfs_buf ** bpp)447 xfs_attr3_leaf_read(
448 struct xfs_trans *tp,
449 struct xfs_inode *dp,
450 xfs_dablk_t bno,
451 struct xfs_buf **bpp)
452 {
453 int err;
454
455 err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
456 &xfs_attr3_leaf_buf_ops);
457 if (!err && tp && *bpp)
458 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
459 return err;
460 }
461
462 /*========================================================================
463 * Namespace helper routines
464 *========================================================================*/
465
466 /*
467 * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
468 * flag on disk - if there's an incomplete attr then recovery needs to tear it
469 * down. If there's no incomplete attr, then recovery needs to tear that attr
470 * down to replace it with the attr that has been logged. In this case, the
471 * INCOMPLETE flag will not be set in attr->attr_filter, but rather
472 * XFS_DA_OP_RECOVERY will be set in args->op_flags.
473 */
474 static bool
xfs_attr_match(struct xfs_da_args * args,uint8_t namelen,unsigned char * name,int flags)475 xfs_attr_match(
476 struct xfs_da_args *args,
477 uint8_t namelen,
478 unsigned char *name,
479 int flags)
480 {
481
482 if (args->namelen != namelen)
483 return false;
484 if (memcmp(args->name, name, namelen) != 0)
485 return false;
486
487 /* Recovery ignores the INCOMPLETE flag. */
488 if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
489 args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
490 return true;
491
492 /* All remaining matches need to be filtered by INCOMPLETE state. */
493 if (args->attr_filter !=
494 (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
495 return false;
496 return true;
497 }
498
499 static int
xfs_attr_copy_value(struct xfs_da_args * args,unsigned char * value,int valuelen)500 xfs_attr_copy_value(
501 struct xfs_da_args *args,
502 unsigned char *value,
503 int valuelen)
504 {
505 /*
506 * No copy if all we have to do is get the length
507 */
508 if (!args->valuelen) {
509 args->valuelen = valuelen;
510 return 0;
511 }
512
513 /*
514 * No copy if the length of the existing buffer is too small
515 */
516 if (args->valuelen < valuelen) {
517 args->valuelen = valuelen;
518 return -ERANGE;
519 }
520
521 if (!args->value) {
522 args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
523 if (!args->value)
524 return -ENOMEM;
525 }
526 args->valuelen = valuelen;
527
528 /* remote block xattr requires IO for copy-in */
529 if (args->rmtblkno)
530 return xfs_attr_rmtval_get(args);
531
532 /*
533 * This is to prevent a GCC warning because the remote xattr case
534 * doesn't have a value to pass in. In that case, we never reach here,
535 * but GCC can't work that out and so throws a "passing NULL to
536 * memcpy" warning.
537 */
538 if (!value)
539 return -EINVAL;
540 memcpy(args->value, value, valuelen);
541 return 0;
542 }
543
544 /*========================================================================
545 * External routines when attribute fork size < XFS_LITINO(mp).
546 *========================================================================*/
547
548 /*
549 * Query whether the total requested number of attr fork bytes of extended
550 * attribute space will be able to fit inline.
551 *
552 * Returns zero if not, else the i_forkoff fork offset to be used in the
553 * literal area for attribute data once the new bytes have been added.
554 *
555 * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
556 * special case for dev/uuid inodes, they have fixed size data forks.
557 */
558 int
xfs_attr_shortform_bytesfit(struct xfs_inode * dp,int bytes)559 xfs_attr_shortform_bytesfit(
560 struct xfs_inode *dp,
561 int bytes)
562 {
563 struct xfs_mount *mp = dp->i_mount;
564 int64_t dsize;
565 int minforkoff;
566 int maxforkoff;
567 int offset;
568
569 /*
570 * Check if the new size could fit at all first:
571 */
572 if (bytes > XFS_LITINO(mp))
573 return 0;
574
575 /* rounded down */
576 offset = (XFS_LITINO(mp) - bytes) >> 3;
577
578 if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
579 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
580 return (offset >= minforkoff) ? minforkoff : 0;
581 }
582
583 /*
584 * If the requested numbers of bytes is smaller or equal to the
585 * current attribute fork size we can always proceed.
586 *
587 * Note that if_bytes in the data fork might actually be larger than
588 * the current data fork size is due to delalloc extents. In that
589 * case either the extent count will go down when they are converted
590 * to real extents, or the delalloc conversion will take care of the
591 * literal area rebalancing.
592 */
593 if (bytes <= xfs_inode_attr_fork_size(dp))
594 return dp->i_forkoff;
595
596 /*
597 * For attr2 we can try to move the forkoff if there is space in the
598 * literal area, but for the old format we are done if there is no
599 * space in the fixed attribute fork.
600 */
601 if (!xfs_has_attr2(mp))
602 return 0;
603
604 dsize = dp->i_df.if_bytes;
605
606 switch (dp->i_df.if_format) {
607 case XFS_DINODE_FMT_EXTENTS:
608 /*
609 * If there is no attr fork and the data fork is extents,
610 * determine if creating the default attr fork will result
611 * in the extents form migrating to btree. If so, the
612 * minimum offset only needs to be the space required for
613 * the btree root.
614 */
615 if (!dp->i_forkoff && dp->i_df.if_bytes >
616 xfs_default_attroffset(dp))
617 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
618 break;
619 case XFS_DINODE_FMT_BTREE:
620 /*
621 * If we have a data btree then keep forkoff if we have one,
622 * otherwise we are adding a new attr, so then we set
623 * minforkoff to where the btree root can finish so we have
624 * plenty of room for attrs
625 */
626 if (dp->i_forkoff) {
627 if (offset < dp->i_forkoff)
628 return 0;
629 return dp->i_forkoff;
630 }
631 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
632 break;
633 }
634
635 /*
636 * A data fork btree root must have space for at least
637 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
638 */
639 minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
640 minforkoff = roundup(minforkoff, 8) >> 3;
641
642 /* attr fork btree root can have at least this many key/ptr pairs */
643 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
644 maxforkoff = maxforkoff >> 3; /* rounded down */
645
646 if (offset >= maxforkoff)
647 return maxforkoff;
648 if (offset >= minforkoff)
649 return offset;
650 return 0;
651 }
652
653 /*
654 * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
655 * - noattr2 mount option is set,
656 * - on-disk version bit says it is already set, or
657 * - the attr2 mount option is not set to enable automatic upgrade from attr1.
658 */
659 STATIC void
xfs_sbversion_add_attr2(struct xfs_mount * mp,struct xfs_trans * tp)660 xfs_sbversion_add_attr2(
661 struct xfs_mount *mp,
662 struct xfs_trans *tp)
663 {
664 if (xfs_has_noattr2(mp))
665 return;
666 if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
667 return;
668 if (!xfs_has_attr2(mp))
669 return;
670
671 spin_lock(&mp->m_sb_lock);
672 xfs_add_attr2(mp);
673 spin_unlock(&mp->m_sb_lock);
674 xfs_log_sb(tp);
675 }
676
677 /*
678 * Create the initial contents of a shortform attribute list.
679 */
680 void
xfs_attr_shortform_create(struct xfs_da_args * args)681 xfs_attr_shortform_create(
682 struct xfs_da_args *args)
683 {
684 struct xfs_inode *dp = args->dp;
685 struct xfs_ifork *ifp = &dp->i_af;
686 struct xfs_attr_sf_hdr *hdr;
687
688 trace_xfs_attr_sf_create(args);
689
690 ASSERT(ifp->if_bytes == 0);
691 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
692 ifp->if_format = XFS_DINODE_FMT_LOCAL;
693 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
694 hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
695 memset(hdr, 0, sizeof(*hdr));
696 hdr->totsize = cpu_to_be16(sizeof(*hdr));
697 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
698 }
699
700 /*
701 * Return -EEXIST if attr is found, or -ENOATTR if not
702 * args: args containing attribute name and namelen
703 * sfep: If not null, pointer will be set to the last attr entry found on
704 -EEXIST. On -ENOATTR pointer is left at the last entry in the list
705 * basep: If not null, pointer is set to the byte offset of the entry in the
706 * list on -EEXIST. On -ENOATTR, pointer is left at the byte offset of
707 * the last entry in the list
708 */
709 int
xfs_attr_sf_findname(struct xfs_da_args * args,struct xfs_attr_sf_entry ** sfep,unsigned int * basep)710 xfs_attr_sf_findname(
711 struct xfs_da_args *args,
712 struct xfs_attr_sf_entry **sfep,
713 unsigned int *basep)
714 {
715 struct xfs_attr_shortform *sf;
716 struct xfs_attr_sf_entry *sfe;
717 unsigned int base = sizeof(struct xfs_attr_sf_hdr);
718 int size = 0;
719 int end;
720 int i;
721
722 sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
723 sfe = &sf->list[0];
724 end = sf->hdr.count;
725 for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
726 base += size, i++) {
727 size = xfs_attr_sf_entsize(sfe);
728 if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
729 sfe->flags))
730 continue;
731 break;
732 }
733
734 if (sfep != NULL)
735 *sfep = sfe;
736
737 if (basep != NULL)
738 *basep = base;
739
740 if (i == end)
741 return -ENOATTR;
742 return -EEXIST;
743 }
744
745 /*
746 * Add a name/value pair to the shortform attribute list.
747 * Overflow from the inode has already been checked for.
748 */
749 void
xfs_attr_shortform_add(struct xfs_da_args * args,int forkoff)750 xfs_attr_shortform_add(
751 struct xfs_da_args *args,
752 int forkoff)
753 {
754 struct xfs_attr_shortform *sf;
755 struct xfs_attr_sf_entry *sfe;
756 int offset, size;
757 struct xfs_mount *mp;
758 struct xfs_inode *dp;
759 struct xfs_ifork *ifp;
760
761 trace_xfs_attr_sf_add(args);
762
763 dp = args->dp;
764 mp = dp->i_mount;
765 dp->i_forkoff = forkoff;
766
767 ifp = &dp->i_af;
768 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
769 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
770 if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
771 ASSERT(0);
772
773 offset = (char *)sfe - (char *)sf;
774 size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
775 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
776 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
777 sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
778
779 sfe->namelen = args->namelen;
780 sfe->valuelen = args->valuelen;
781 sfe->flags = args->attr_filter;
782 memcpy(sfe->nameval, args->name, args->namelen);
783 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
784 sf->hdr.count++;
785 be16_add_cpu(&sf->hdr.totsize, size);
786 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
787
788 xfs_sbversion_add_attr2(mp, args->trans);
789 }
790
791 /*
792 * After the last attribute is removed revert to original inode format,
793 * making all literal area available to the data fork once more.
794 */
795 void
xfs_attr_fork_remove(struct xfs_inode * ip,struct xfs_trans * tp)796 xfs_attr_fork_remove(
797 struct xfs_inode *ip,
798 struct xfs_trans *tp)
799 {
800 ASSERT(ip->i_af.if_nextents == 0);
801
802 xfs_ifork_zap_attr(ip);
803 ip->i_forkoff = 0;
804 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
805 }
806
807 /*
808 * Remove an attribute from the shortform attribute list structure.
809 */
810 int
xfs_attr_sf_removename(struct xfs_da_args * args)811 xfs_attr_sf_removename(
812 struct xfs_da_args *args)
813 {
814 struct xfs_attr_shortform *sf;
815 struct xfs_attr_sf_entry *sfe;
816 int size = 0, end, totsize;
817 unsigned int base;
818 struct xfs_mount *mp;
819 struct xfs_inode *dp;
820 int error;
821
822 trace_xfs_attr_sf_remove(args);
823
824 dp = args->dp;
825 mp = dp->i_mount;
826 sf = (struct xfs_attr_shortform *)dp->i_af.if_u1.if_data;
827
828 error = xfs_attr_sf_findname(args, &sfe, &base);
829
830 /*
831 * If we are recovering an operation, finding nothing to
832 * remove is not an error - it just means there was nothing
833 * to clean up.
834 */
835 if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
836 return 0;
837 if (error != -EEXIST)
838 return error;
839 size = xfs_attr_sf_entsize(sfe);
840
841 /*
842 * Fix up the attribute fork data, covering the hole
843 */
844 end = base + size;
845 totsize = be16_to_cpu(sf->hdr.totsize);
846 if (end != totsize)
847 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
848 sf->hdr.count--;
849 be16_add_cpu(&sf->hdr.totsize, -size);
850
851 /*
852 * Fix up the start offset of the attribute fork
853 */
854 totsize -= size;
855 if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
856 (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
857 !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
858 xfs_attr_fork_remove(dp, args->trans);
859 } else {
860 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
861 dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
862 ASSERT(dp->i_forkoff);
863 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
864 (args->op_flags & XFS_DA_OP_ADDNAME) ||
865 !xfs_has_attr2(mp) ||
866 dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
867 xfs_trans_log_inode(args->trans, dp,
868 XFS_ILOG_CORE | XFS_ILOG_ADATA);
869 }
870
871 xfs_sbversion_add_attr2(mp, args->trans);
872
873 return 0;
874 }
875
876 /*
877 * Look up a name in a shortform attribute list structure.
878 */
879 /*ARGSUSED*/
880 int
xfs_attr_shortform_lookup(xfs_da_args_t * args)881 xfs_attr_shortform_lookup(xfs_da_args_t *args)
882 {
883 struct xfs_attr_shortform *sf;
884 struct xfs_attr_sf_entry *sfe;
885 int i;
886 struct xfs_ifork *ifp;
887
888 trace_xfs_attr_sf_lookup(args);
889
890 ifp = &args->dp->i_af;
891 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
892 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
893 sfe = &sf->list[0];
894 for (i = 0; i < sf->hdr.count;
895 sfe = xfs_attr_sf_nextentry(sfe), i++) {
896 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
897 sfe->flags))
898 return -EEXIST;
899 }
900 return -ENOATTR;
901 }
902
903 /*
904 * Retrieve the attribute value and length.
905 *
906 * If args->valuelen is zero, only the length needs to be returned. Unlike a
907 * lookup, we only return an error if the attribute does not exist or we can't
908 * retrieve the value.
909 */
910 int
xfs_attr_shortform_getvalue(struct xfs_da_args * args)911 xfs_attr_shortform_getvalue(
912 struct xfs_da_args *args)
913 {
914 struct xfs_attr_shortform *sf;
915 struct xfs_attr_sf_entry *sfe;
916 int i;
917
918 ASSERT(args->dp->i_af.if_format == XFS_DINODE_FMT_LOCAL);
919 sf = (struct xfs_attr_shortform *)args->dp->i_af.if_u1.if_data;
920 sfe = &sf->list[0];
921 for (i = 0; i < sf->hdr.count;
922 sfe = xfs_attr_sf_nextentry(sfe), i++) {
923 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
924 sfe->flags))
925 return xfs_attr_copy_value(args,
926 &sfe->nameval[args->namelen], sfe->valuelen);
927 }
928 return -ENOATTR;
929 }
930
931 /* Convert from using the shortform to the leaf format. */
932 int
xfs_attr_shortform_to_leaf(struct xfs_da_args * args)933 xfs_attr_shortform_to_leaf(
934 struct xfs_da_args *args)
935 {
936 struct xfs_inode *dp;
937 struct xfs_attr_shortform *sf;
938 struct xfs_attr_sf_entry *sfe;
939 struct xfs_da_args nargs;
940 char *tmpbuffer;
941 int error, i, size;
942 xfs_dablk_t blkno;
943 struct xfs_buf *bp;
944 struct xfs_ifork *ifp;
945
946 trace_xfs_attr_sf_to_leaf(args);
947
948 dp = args->dp;
949 ifp = &dp->i_af;
950 sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
951 size = be16_to_cpu(sf->hdr.totsize);
952 tmpbuffer = kmem_alloc(size, 0);
953 ASSERT(tmpbuffer != NULL);
954 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
955 sf = (struct xfs_attr_shortform *)tmpbuffer;
956
957 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
958 xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
959
960 bp = NULL;
961 error = xfs_da_grow_inode(args, &blkno);
962 if (error)
963 goto out;
964
965 ASSERT(blkno == 0);
966 error = xfs_attr3_leaf_create(args, blkno, &bp);
967 if (error)
968 goto out;
969
970 memset((char *)&nargs, 0, sizeof(nargs));
971 nargs.dp = dp;
972 nargs.geo = args->geo;
973 nargs.total = args->total;
974 nargs.whichfork = XFS_ATTR_FORK;
975 nargs.trans = args->trans;
976 nargs.op_flags = XFS_DA_OP_OKNOENT;
977
978 sfe = &sf->list[0];
979 for (i = 0; i < sf->hdr.count; i++) {
980 nargs.name = sfe->nameval;
981 nargs.namelen = sfe->namelen;
982 nargs.value = &sfe->nameval[nargs.namelen];
983 nargs.valuelen = sfe->valuelen;
984 nargs.hashval = xfs_da_hashname(sfe->nameval,
985 sfe->namelen);
986 nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
987 if (!xfs_attr_check_namespace(sfe->flags)) {
988 error = -EFSCORRUPTED;
989 goto out;
990 }
991 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
992 ASSERT(error == -ENOATTR);
993 if (!xfs_attr3_leaf_add(bp, &nargs))
994 ASSERT(0);
995 sfe = xfs_attr_sf_nextentry(sfe);
996 }
997 error = 0;
998 out:
999 kmem_free(tmpbuffer);
1000 return error;
1001 }
1002
1003 /*
1004 * Check a leaf attribute block to see if all the entries would fit into
1005 * a shortform attribute list.
1006 */
1007 int
xfs_attr_shortform_allfit(struct xfs_buf * bp,struct xfs_inode * dp)1008 xfs_attr_shortform_allfit(
1009 struct xfs_buf *bp,
1010 struct xfs_inode *dp)
1011 {
1012 struct xfs_attr_leafblock *leaf;
1013 struct xfs_attr_leaf_entry *entry;
1014 xfs_attr_leaf_name_local_t *name_loc;
1015 struct xfs_attr3_icleaf_hdr leafhdr;
1016 int bytes;
1017 int i;
1018 struct xfs_mount *mp = bp->b_mount;
1019
1020 leaf = bp->b_addr;
1021 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1022 entry = xfs_attr3_leaf_entryp(leaf);
1023
1024 bytes = sizeof(struct xfs_attr_sf_hdr);
1025 for (i = 0; i < leafhdr.count; entry++, i++) {
1026 if (entry->flags & XFS_ATTR_INCOMPLETE)
1027 continue; /* don't copy partial entries */
1028 if (!(entry->flags & XFS_ATTR_LOCAL))
1029 return 0;
1030 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1031 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1032 return 0;
1033 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1034 return 0;
1035 bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1036 be16_to_cpu(name_loc->valuelen));
1037 }
1038 if (xfs_has_attr2(dp->i_mount) &&
1039 (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1040 (bytes == sizeof(struct xfs_attr_sf_hdr)))
1041 return -1;
1042 return xfs_attr_shortform_bytesfit(dp, bytes);
1043 }
1044
1045 /* Verify the consistency of an inline attribute fork. */
1046 xfs_failaddr_t
xfs_attr_shortform_verify(struct xfs_inode * ip)1047 xfs_attr_shortform_verify(
1048 struct xfs_inode *ip)
1049 {
1050 struct xfs_attr_shortform *sfp;
1051 struct xfs_attr_sf_entry *sfep;
1052 struct xfs_attr_sf_entry *next_sfep;
1053 char *endp;
1054 struct xfs_ifork *ifp;
1055 int i;
1056 int64_t size;
1057
1058 ASSERT(ip->i_af.if_format == XFS_DINODE_FMT_LOCAL);
1059 ifp = xfs_ifork_ptr(ip, XFS_ATTR_FORK);
1060 sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1061 size = ifp->if_bytes;
1062
1063 /*
1064 * Give up if the attribute is way too short.
1065 */
1066 if (size < sizeof(struct xfs_attr_sf_hdr))
1067 return __this_address;
1068
1069 endp = (char *)sfp + size;
1070
1071 /* Check all reported entries */
1072 sfep = &sfp->list[0];
1073 for (i = 0; i < sfp->hdr.count; i++) {
1074 /*
1075 * struct xfs_attr_sf_entry has a variable length.
1076 * Check the fixed-offset parts of the structure are
1077 * within the data buffer.
1078 * xfs_attr_sf_entry is defined with a 1-byte variable
1079 * array at the end, so we must subtract that off.
1080 */
1081 if (((char *)sfep + sizeof(*sfep)) >= endp)
1082 return __this_address;
1083
1084 /* Don't allow names with known bad length. */
1085 if (sfep->namelen == 0)
1086 return __this_address;
1087
1088 /*
1089 * Check that the variable-length part of the structure is
1090 * within the data buffer. The next entry starts after the
1091 * name component, so nextentry is an acceptable test.
1092 */
1093 next_sfep = xfs_attr_sf_nextentry(sfep);
1094 if ((char *)next_sfep > endp)
1095 return __this_address;
1096
1097 /*
1098 * Check for unknown flags. Short form doesn't support
1099 * the incomplete or local bits, so we can use the namespace
1100 * mask here.
1101 */
1102 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1103 return __this_address;
1104
1105 /*
1106 * Check for invalid namespace combinations. We only allow
1107 * one namespace flag per xattr, so we can just count the
1108 * bits (i.e. hweight) here.
1109 */
1110 if (!xfs_attr_check_namespace(sfep->flags))
1111 return __this_address;
1112
1113 sfep = next_sfep;
1114 }
1115 if ((void *)sfep != (void *)endp)
1116 return __this_address;
1117
1118 return NULL;
1119 }
1120
1121 /*
1122 * Convert a leaf attribute list to shortform attribute list
1123 */
1124 int
xfs_attr3_leaf_to_shortform(struct xfs_buf * bp,struct xfs_da_args * args,int forkoff)1125 xfs_attr3_leaf_to_shortform(
1126 struct xfs_buf *bp,
1127 struct xfs_da_args *args,
1128 int forkoff)
1129 {
1130 struct xfs_attr_leafblock *leaf;
1131 struct xfs_attr3_icleaf_hdr ichdr;
1132 struct xfs_attr_leaf_entry *entry;
1133 struct xfs_attr_leaf_name_local *name_loc;
1134 struct xfs_da_args nargs;
1135 struct xfs_inode *dp = args->dp;
1136 char *tmpbuffer;
1137 int error;
1138 int i;
1139
1140 trace_xfs_attr_leaf_to_sf(args);
1141
1142 tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1143 if (!tmpbuffer)
1144 return -ENOMEM;
1145
1146 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1147
1148 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1149 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1150 entry = xfs_attr3_leaf_entryp(leaf);
1151
1152 /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1153 memset(bp->b_addr, 0, args->geo->blksize);
1154
1155 /*
1156 * Clean out the prior contents of the attribute list.
1157 */
1158 error = xfs_da_shrink_inode(args, 0, bp);
1159 if (error)
1160 goto out;
1161
1162 if (forkoff == -1) {
1163 /*
1164 * Don't remove the attr fork if this operation is the first
1165 * part of a attr replace operations. We're going to add a new
1166 * attr immediately, so we need to keep the attr fork around in
1167 * this case.
1168 */
1169 if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1170 ASSERT(xfs_has_attr2(dp->i_mount));
1171 ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1172 xfs_attr_fork_remove(dp, args->trans);
1173 }
1174 goto out;
1175 }
1176
1177 xfs_attr_shortform_create(args);
1178
1179 /*
1180 * Copy the attributes
1181 */
1182 memset((char *)&nargs, 0, sizeof(nargs));
1183 nargs.geo = args->geo;
1184 nargs.dp = dp;
1185 nargs.total = args->total;
1186 nargs.whichfork = XFS_ATTR_FORK;
1187 nargs.trans = args->trans;
1188 nargs.op_flags = XFS_DA_OP_OKNOENT;
1189
1190 for (i = 0; i < ichdr.count; entry++, i++) {
1191 if (entry->flags & XFS_ATTR_INCOMPLETE)
1192 continue; /* don't copy partial entries */
1193 if (!entry->nameidx)
1194 continue;
1195 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1196 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1197 nargs.name = name_loc->nameval;
1198 nargs.namelen = name_loc->namelen;
1199 nargs.value = &name_loc->nameval[nargs.namelen];
1200 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1201 nargs.hashval = be32_to_cpu(entry->hashval);
1202 nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1203 xfs_attr_shortform_add(&nargs, forkoff);
1204 }
1205 error = 0;
1206
1207 out:
1208 kmem_free(tmpbuffer);
1209 return error;
1210 }
1211
1212 /*
1213 * Convert from using a single leaf to a root node and a leaf.
1214 */
1215 int
xfs_attr3_leaf_to_node(struct xfs_da_args * args)1216 xfs_attr3_leaf_to_node(
1217 struct xfs_da_args *args)
1218 {
1219 struct xfs_attr_leafblock *leaf;
1220 struct xfs_attr3_icleaf_hdr icleafhdr;
1221 struct xfs_attr_leaf_entry *entries;
1222 struct xfs_da3_icnode_hdr icnodehdr;
1223 struct xfs_da_intnode *node;
1224 struct xfs_inode *dp = args->dp;
1225 struct xfs_mount *mp = dp->i_mount;
1226 struct xfs_buf *bp1 = NULL;
1227 struct xfs_buf *bp2 = NULL;
1228 xfs_dablk_t blkno;
1229 int error;
1230
1231 trace_xfs_attr_leaf_to_node(args);
1232
1233 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1234 error = -EIO;
1235 goto out;
1236 }
1237
1238 error = xfs_da_grow_inode(args, &blkno);
1239 if (error)
1240 goto out;
1241 error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1242 if (error)
1243 goto out;
1244
1245 error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1246 if (error)
1247 goto out;
1248
1249 /* copy leaf to new buffer, update identifiers */
1250 xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1251 bp2->b_ops = bp1->b_ops;
1252 memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1253 if (xfs_has_crc(mp)) {
1254 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1255 hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1256 }
1257 xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1258
1259 /*
1260 * Set up the new root node.
1261 */
1262 error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1263 if (error)
1264 goto out;
1265 node = bp1->b_addr;
1266 xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1267
1268 leaf = bp2->b_addr;
1269 xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1270 entries = xfs_attr3_leaf_entryp(leaf);
1271
1272 /* both on-disk, don't endian-flip twice */
1273 icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1274 icnodehdr.btree[0].before = cpu_to_be32(blkno);
1275 icnodehdr.count = 1;
1276 xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1277 xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1278 error = 0;
1279 out:
1280 return error;
1281 }
1282
1283 /*========================================================================
1284 * Routines used for growing the Btree.
1285 *========================================================================*/
1286
1287 /*
1288 * Create the initial contents of a leaf attribute list
1289 * or a leaf in a node attribute list.
1290 */
1291 STATIC int
xfs_attr3_leaf_create(struct xfs_da_args * args,xfs_dablk_t blkno,struct xfs_buf ** bpp)1292 xfs_attr3_leaf_create(
1293 struct xfs_da_args *args,
1294 xfs_dablk_t blkno,
1295 struct xfs_buf **bpp)
1296 {
1297 struct xfs_attr_leafblock *leaf;
1298 struct xfs_attr3_icleaf_hdr ichdr;
1299 struct xfs_inode *dp = args->dp;
1300 struct xfs_mount *mp = dp->i_mount;
1301 struct xfs_buf *bp;
1302 int error;
1303
1304 trace_xfs_attr_leaf_create(args);
1305
1306 error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1307 XFS_ATTR_FORK);
1308 if (error)
1309 return error;
1310 bp->b_ops = &xfs_attr3_leaf_buf_ops;
1311 xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1312 leaf = bp->b_addr;
1313 memset(leaf, 0, args->geo->blksize);
1314
1315 memset(&ichdr, 0, sizeof(ichdr));
1316 ichdr.firstused = args->geo->blksize;
1317
1318 if (xfs_has_crc(mp)) {
1319 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1320
1321 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1322
1323 hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1324 hdr3->owner = cpu_to_be64(dp->i_ino);
1325 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1326
1327 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1328 } else {
1329 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1330 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1331 }
1332 ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1333
1334 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1335 xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1336
1337 *bpp = bp;
1338 return 0;
1339 }
1340
1341 /*
1342 * Split the leaf node, rebalance, then add the new entry.
1343 *
1344 * Returns 0 if the entry was added, 1 if a further split is needed or a
1345 * negative error number otherwise.
1346 */
1347 int
xfs_attr3_leaf_split(struct xfs_da_state * state,struct xfs_da_state_blk * oldblk,struct xfs_da_state_blk * newblk)1348 xfs_attr3_leaf_split(
1349 struct xfs_da_state *state,
1350 struct xfs_da_state_blk *oldblk,
1351 struct xfs_da_state_blk *newblk)
1352 {
1353 bool added;
1354 xfs_dablk_t blkno;
1355 int error;
1356
1357 trace_xfs_attr_leaf_split(state->args);
1358
1359 /*
1360 * Allocate space for a new leaf node.
1361 */
1362 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1363 error = xfs_da_grow_inode(state->args, &blkno);
1364 if (error)
1365 return error;
1366 error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1367 if (error)
1368 return error;
1369 newblk->blkno = blkno;
1370 newblk->magic = XFS_ATTR_LEAF_MAGIC;
1371
1372 /*
1373 * Rebalance the entries across the two leaves.
1374 * NOTE: rebalance() currently depends on the 2nd block being empty.
1375 */
1376 xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1377 error = xfs_da3_blk_link(state, oldblk, newblk);
1378 if (error)
1379 return error;
1380
1381 /*
1382 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1383 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1384 * "new" attrs info. Will need the "old" info to remove it later.
1385 *
1386 * Insert the "new" entry in the correct block.
1387 */
1388 if (state->inleaf) {
1389 trace_xfs_attr_leaf_add_old(state->args);
1390 added = xfs_attr3_leaf_add(oldblk->bp, state->args);
1391 } else {
1392 trace_xfs_attr_leaf_add_new(state->args);
1393 added = xfs_attr3_leaf_add(newblk->bp, state->args);
1394 }
1395
1396 /*
1397 * Update last hashval in each block since we added the name.
1398 */
1399 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1400 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1401 if (!added)
1402 return 1;
1403 return 0;
1404 }
1405
1406 /*
1407 * Add a name to the leaf attribute list structure.
1408 */
1409 bool
xfs_attr3_leaf_add(struct xfs_buf * bp,struct xfs_da_args * args)1410 xfs_attr3_leaf_add(
1411 struct xfs_buf *bp,
1412 struct xfs_da_args *args)
1413 {
1414 struct xfs_attr_leafblock *leaf;
1415 struct xfs_attr3_icleaf_hdr ichdr;
1416 int tablesize;
1417 int entsize;
1418 bool added = true;
1419 int sum;
1420 int tmp;
1421 int i;
1422
1423 trace_xfs_attr_leaf_add(args);
1424
1425 leaf = bp->b_addr;
1426 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1427 ASSERT(args->index >= 0 && args->index <= ichdr.count);
1428 entsize = xfs_attr_leaf_newentsize(args, NULL);
1429
1430 /*
1431 * Search through freemap for first-fit on new name length.
1432 * (may need to figure in size of entry struct too)
1433 */
1434 tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1435 + xfs_attr3_leaf_hdr_size(leaf);
1436 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1437 if (tablesize > ichdr.firstused) {
1438 sum += ichdr.freemap[i].size;
1439 continue;
1440 }
1441 if (!ichdr.freemap[i].size)
1442 continue; /* no space in this map */
1443 tmp = entsize;
1444 if (ichdr.freemap[i].base < ichdr.firstused)
1445 tmp += sizeof(xfs_attr_leaf_entry_t);
1446 if (ichdr.freemap[i].size >= tmp) {
1447 xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1448 goto out_log_hdr;
1449 }
1450 sum += ichdr.freemap[i].size;
1451 }
1452
1453 /*
1454 * If there are no holes in the address space of the block,
1455 * and we don't have enough freespace, then compaction will do us
1456 * no good and we should just give up.
1457 */
1458 if (!ichdr.holes && sum < entsize)
1459 return false;
1460
1461 /*
1462 * Compact the entries to coalesce free space.
1463 * This may change the hdr->count via dropping INCOMPLETE entries.
1464 */
1465 xfs_attr3_leaf_compact(args, &ichdr, bp);
1466
1467 /*
1468 * After compaction, the block is guaranteed to have only one
1469 * free region, in freemap[0]. If it is not big enough, give up.
1470 */
1471 if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1472 added = false;
1473 goto out_log_hdr;
1474 }
1475
1476 xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1477
1478 out_log_hdr:
1479 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1480 xfs_trans_log_buf(args->trans, bp,
1481 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1482 xfs_attr3_leaf_hdr_size(leaf)));
1483 return added;
1484 }
1485
1486 /*
1487 * Add a name to a leaf attribute list structure.
1488 */
1489 STATIC void
xfs_attr3_leaf_add_work(struct xfs_buf * bp,struct xfs_attr3_icleaf_hdr * ichdr,struct xfs_da_args * args,int mapindex)1490 xfs_attr3_leaf_add_work(
1491 struct xfs_buf *bp,
1492 struct xfs_attr3_icleaf_hdr *ichdr,
1493 struct xfs_da_args *args,
1494 int mapindex)
1495 {
1496 struct xfs_attr_leafblock *leaf;
1497 struct xfs_attr_leaf_entry *entry;
1498 struct xfs_attr_leaf_name_local *name_loc;
1499 struct xfs_attr_leaf_name_remote *name_rmt;
1500 struct xfs_mount *mp;
1501 int tmp;
1502 int i;
1503
1504 trace_xfs_attr_leaf_add_work(args);
1505
1506 leaf = bp->b_addr;
1507 ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1508 ASSERT(args->index >= 0 && args->index <= ichdr->count);
1509
1510 /*
1511 * Force open some space in the entry array and fill it in.
1512 */
1513 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1514 if (args->index < ichdr->count) {
1515 tmp = ichdr->count - args->index;
1516 tmp *= sizeof(xfs_attr_leaf_entry_t);
1517 memmove(entry + 1, entry, tmp);
1518 xfs_trans_log_buf(args->trans, bp,
1519 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1520 }
1521 ichdr->count++;
1522
1523 /*
1524 * Allocate space for the new string (at the end of the run).
1525 */
1526 mp = args->trans->t_mountp;
1527 ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1528 ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1529 ASSERT(ichdr->freemap[mapindex].size >=
1530 xfs_attr_leaf_newentsize(args, NULL));
1531 ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1532 ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1533
1534 ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1535
1536 entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1537 ichdr->freemap[mapindex].size);
1538 entry->hashval = cpu_to_be32(args->hashval);
1539 entry->flags = args->attr_filter;
1540 if (tmp)
1541 entry->flags |= XFS_ATTR_LOCAL;
1542 if (args->op_flags & XFS_DA_OP_REPLACE) {
1543 if (!(args->op_flags & XFS_DA_OP_LOGGED))
1544 entry->flags |= XFS_ATTR_INCOMPLETE;
1545 if ((args->blkno2 == args->blkno) &&
1546 (args->index2 <= args->index)) {
1547 args->index2++;
1548 }
1549 }
1550 xfs_trans_log_buf(args->trans, bp,
1551 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1552 ASSERT((args->index == 0) ||
1553 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1554 ASSERT((args->index == ichdr->count - 1) ||
1555 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1556
1557 /*
1558 * For "remote" attribute values, simply note that we need to
1559 * allocate space for the "remote" value. We can't actually
1560 * allocate the extents in this transaction, and we can't decide
1561 * which blocks they should be as we might allocate more blocks
1562 * as part of this transaction (a split operation for example).
1563 */
1564 if (entry->flags & XFS_ATTR_LOCAL) {
1565 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1566 name_loc->namelen = args->namelen;
1567 name_loc->valuelen = cpu_to_be16(args->valuelen);
1568 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1569 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1570 be16_to_cpu(name_loc->valuelen));
1571 } else {
1572 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1573 name_rmt->namelen = args->namelen;
1574 memcpy((char *)name_rmt->name, args->name, args->namelen);
1575 entry->flags |= XFS_ATTR_INCOMPLETE;
1576 /* just in case */
1577 name_rmt->valuelen = 0;
1578 name_rmt->valueblk = 0;
1579 args->rmtblkno = 1;
1580 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1581 args->rmtvaluelen = args->valuelen;
1582 }
1583 xfs_trans_log_buf(args->trans, bp,
1584 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1585 xfs_attr_leaf_entsize(leaf, args->index)));
1586
1587 /*
1588 * Update the control info for this leaf node
1589 */
1590 if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1591 ichdr->firstused = be16_to_cpu(entry->nameidx);
1592
1593 ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1594 + xfs_attr3_leaf_hdr_size(leaf));
1595 tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1596 + xfs_attr3_leaf_hdr_size(leaf);
1597
1598 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1599 if (ichdr->freemap[i].base == tmp) {
1600 ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1601 ichdr->freemap[i].size -=
1602 min_t(uint16_t, ichdr->freemap[i].size,
1603 sizeof(xfs_attr_leaf_entry_t));
1604 }
1605 }
1606 ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1607 }
1608
1609 /*
1610 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1611 */
1612 STATIC void
xfs_attr3_leaf_compact(struct xfs_da_args * args,struct xfs_attr3_icleaf_hdr * ichdr_dst,struct xfs_buf * bp)1613 xfs_attr3_leaf_compact(
1614 struct xfs_da_args *args,
1615 struct xfs_attr3_icleaf_hdr *ichdr_dst,
1616 struct xfs_buf *bp)
1617 {
1618 struct xfs_attr_leafblock *leaf_src;
1619 struct xfs_attr_leafblock *leaf_dst;
1620 struct xfs_attr3_icleaf_hdr ichdr_src;
1621 struct xfs_trans *trans = args->trans;
1622 char *tmpbuffer;
1623
1624 trace_xfs_attr_leaf_compact(args);
1625
1626 tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1627 memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1628 memset(bp->b_addr, 0, args->geo->blksize);
1629 leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1630 leaf_dst = bp->b_addr;
1631
1632 /*
1633 * Copy the on-disk header back into the destination buffer to ensure
1634 * all the information in the header that is not part of the incore
1635 * header structure is preserved.
1636 */
1637 memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1638
1639 /* Initialise the incore headers */
1640 ichdr_src = *ichdr_dst; /* struct copy */
1641 ichdr_dst->firstused = args->geo->blksize;
1642 ichdr_dst->usedbytes = 0;
1643 ichdr_dst->count = 0;
1644 ichdr_dst->holes = 0;
1645 ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1646 ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1647 ichdr_dst->freemap[0].base;
1648
1649 /* write the header back to initialise the underlying buffer */
1650 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1651
1652 /*
1653 * Copy all entry's in the same (sorted) order,
1654 * but allocate name/value pairs packed and in sequence.
1655 */
1656 xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1657 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1658 /*
1659 * this logs the entire buffer, but the caller must write the header
1660 * back to the buffer when it is finished modifying it.
1661 */
1662 xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1663
1664 kmem_free(tmpbuffer);
1665 }
1666
1667 /*
1668 * Compare two leaf blocks "order".
1669 * Return 0 unless leaf2 should go before leaf1.
1670 */
1671 static int
xfs_attr3_leaf_order(struct xfs_buf * leaf1_bp,struct xfs_attr3_icleaf_hdr * leaf1hdr,struct xfs_buf * leaf2_bp,struct xfs_attr3_icleaf_hdr * leaf2hdr)1672 xfs_attr3_leaf_order(
1673 struct xfs_buf *leaf1_bp,
1674 struct xfs_attr3_icleaf_hdr *leaf1hdr,
1675 struct xfs_buf *leaf2_bp,
1676 struct xfs_attr3_icleaf_hdr *leaf2hdr)
1677 {
1678 struct xfs_attr_leaf_entry *entries1;
1679 struct xfs_attr_leaf_entry *entries2;
1680
1681 entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1682 entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1683 if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1684 ((be32_to_cpu(entries2[0].hashval) <
1685 be32_to_cpu(entries1[0].hashval)) ||
1686 (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1687 be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1688 return 1;
1689 }
1690 return 0;
1691 }
1692
1693 int
xfs_attr_leaf_order(struct xfs_buf * leaf1_bp,struct xfs_buf * leaf2_bp)1694 xfs_attr_leaf_order(
1695 struct xfs_buf *leaf1_bp,
1696 struct xfs_buf *leaf2_bp)
1697 {
1698 struct xfs_attr3_icleaf_hdr ichdr1;
1699 struct xfs_attr3_icleaf_hdr ichdr2;
1700 struct xfs_mount *mp = leaf1_bp->b_mount;
1701
1702 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1703 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1704 return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1705 }
1706
1707 /*
1708 * Redistribute the attribute list entries between two leaf nodes,
1709 * taking into account the size of the new entry.
1710 *
1711 * NOTE: if new block is empty, then it will get the upper half of the
1712 * old block. At present, all (one) callers pass in an empty second block.
1713 *
1714 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1715 * to match what it is doing in splitting the attribute leaf block. Those
1716 * values are used in "atomic rename" operations on attributes. Note that
1717 * the "new" and "old" values can end up in different blocks.
1718 */
1719 STATIC void
xfs_attr3_leaf_rebalance(struct xfs_da_state * state,struct xfs_da_state_blk * blk1,struct xfs_da_state_blk * blk2)1720 xfs_attr3_leaf_rebalance(
1721 struct xfs_da_state *state,
1722 struct xfs_da_state_blk *blk1,
1723 struct xfs_da_state_blk *blk2)
1724 {
1725 struct xfs_da_args *args;
1726 struct xfs_attr_leafblock *leaf1;
1727 struct xfs_attr_leafblock *leaf2;
1728 struct xfs_attr3_icleaf_hdr ichdr1;
1729 struct xfs_attr3_icleaf_hdr ichdr2;
1730 struct xfs_attr_leaf_entry *entries1;
1731 struct xfs_attr_leaf_entry *entries2;
1732 int count;
1733 int totallen;
1734 int max;
1735 int space;
1736 int swap;
1737
1738 /*
1739 * Set up environment.
1740 */
1741 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1742 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1743 leaf1 = blk1->bp->b_addr;
1744 leaf2 = blk2->bp->b_addr;
1745 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1746 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1747 ASSERT(ichdr2.count == 0);
1748 args = state->args;
1749
1750 trace_xfs_attr_leaf_rebalance(args);
1751
1752 /*
1753 * Check ordering of blocks, reverse if it makes things simpler.
1754 *
1755 * NOTE: Given that all (current) callers pass in an empty
1756 * second block, this code should never set "swap".
1757 */
1758 swap = 0;
1759 if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1760 swap(blk1, blk2);
1761
1762 /* swap structures rather than reconverting them */
1763 swap(ichdr1, ichdr2);
1764
1765 leaf1 = blk1->bp->b_addr;
1766 leaf2 = blk2->bp->b_addr;
1767 swap = 1;
1768 }
1769
1770 /*
1771 * Examine entries until we reduce the absolute difference in
1772 * byte usage between the two blocks to a minimum. Then get
1773 * the direction to copy and the number of elements to move.
1774 *
1775 * "inleaf" is true if the new entry should be inserted into blk1.
1776 * If "swap" is also true, then reverse the sense of "inleaf".
1777 */
1778 state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1779 blk2, &ichdr2,
1780 &count, &totallen);
1781 if (swap)
1782 state->inleaf = !state->inleaf;
1783
1784 /*
1785 * Move any entries required from leaf to leaf:
1786 */
1787 if (count < ichdr1.count) {
1788 /*
1789 * Figure the total bytes to be added to the destination leaf.
1790 */
1791 /* number entries being moved */
1792 count = ichdr1.count - count;
1793 space = ichdr1.usedbytes - totallen;
1794 space += count * sizeof(xfs_attr_leaf_entry_t);
1795
1796 /*
1797 * leaf2 is the destination, compact it if it looks tight.
1798 */
1799 max = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1800 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1801 if (space > max)
1802 xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1803
1804 /*
1805 * Move high entries from leaf1 to low end of leaf2.
1806 */
1807 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1808 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1809
1810 } else if (count > ichdr1.count) {
1811 /*
1812 * I assert that since all callers pass in an empty
1813 * second buffer, this code should never execute.
1814 */
1815 ASSERT(0);
1816
1817 /*
1818 * Figure the total bytes to be added to the destination leaf.
1819 */
1820 /* number entries being moved */
1821 count -= ichdr1.count;
1822 space = totallen - ichdr1.usedbytes;
1823 space += count * sizeof(xfs_attr_leaf_entry_t);
1824
1825 /*
1826 * leaf1 is the destination, compact it if it looks tight.
1827 */
1828 max = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1829 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1830 if (space > max)
1831 xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1832
1833 /*
1834 * Move low entries from leaf2 to high end of leaf1.
1835 */
1836 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1837 ichdr1.count, count);
1838 }
1839
1840 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1841 xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1842 xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1843 xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1844
1845 /*
1846 * Copy out last hashval in each block for B-tree code.
1847 */
1848 entries1 = xfs_attr3_leaf_entryp(leaf1);
1849 entries2 = xfs_attr3_leaf_entryp(leaf2);
1850 blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1851 blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1852
1853 /*
1854 * Adjust the expected index for insertion.
1855 * NOTE: this code depends on the (current) situation that the
1856 * second block was originally empty.
1857 *
1858 * If the insertion point moved to the 2nd block, we must adjust
1859 * the index. We must also track the entry just following the
1860 * new entry for use in an "atomic rename" operation, that entry
1861 * is always the "old" entry and the "new" entry is what we are
1862 * inserting. The index/blkno fields refer to the "old" entry,
1863 * while the index2/blkno2 fields refer to the "new" entry.
1864 */
1865 if (blk1->index > ichdr1.count) {
1866 ASSERT(state->inleaf == 0);
1867 blk2->index = blk1->index - ichdr1.count;
1868 args->index = args->index2 = blk2->index;
1869 args->blkno = args->blkno2 = blk2->blkno;
1870 } else if (blk1->index == ichdr1.count) {
1871 if (state->inleaf) {
1872 args->index = blk1->index;
1873 args->blkno = blk1->blkno;
1874 args->index2 = 0;
1875 args->blkno2 = blk2->blkno;
1876 } else {
1877 /*
1878 * On a double leaf split, the original attr location
1879 * is already stored in blkno2/index2, so don't
1880 * overwrite it overwise we corrupt the tree.
1881 */
1882 blk2->index = blk1->index - ichdr1.count;
1883 args->index = blk2->index;
1884 args->blkno = blk2->blkno;
1885 if (!state->extravalid) {
1886 /*
1887 * set the new attr location to match the old
1888 * one and let the higher level split code
1889 * decide where in the leaf to place it.
1890 */
1891 args->index2 = blk2->index;
1892 args->blkno2 = blk2->blkno;
1893 }
1894 }
1895 } else {
1896 ASSERT(state->inleaf == 1);
1897 args->index = args->index2 = blk1->index;
1898 args->blkno = args->blkno2 = blk1->blkno;
1899 }
1900 }
1901
1902 /*
1903 * Examine entries until we reduce the absolute difference in
1904 * byte usage between the two blocks to a minimum.
1905 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1906 * GROT: there will always be enough room in either block for a new entry.
1907 * GROT: Do a double-split for this case?
1908 */
1909 STATIC int
xfs_attr3_leaf_figure_balance(struct xfs_da_state * state,struct xfs_da_state_blk * blk1,struct xfs_attr3_icleaf_hdr * ichdr1,struct xfs_da_state_blk * blk2,struct xfs_attr3_icleaf_hdr * ichdr2,int * countarg,int * usedbytesarg)1910 xfs_attr3_leaf_figure_balance(
1911 struct xfs_da_state *state,
1912 struct xfs_da_state_blk *blk1,
1913 struct xfs_attr3_icleaf_hdr *ichdr1,
1914 struct xfs_da_state_blk *blk2,
1915 struct xfs_attr3_icleaf_hdr *ichdr2,
1916 int *countarg,
1917 int *usedbytesarg)
1918 {
1919 struct xfs_attr_leafblock *leaf1 = blk1->bp->b_addr;
1920 struct xfs_attr_leafblock *leaf2 = blk2->bp->b_addr;
1921 struct xfs_attr_leaf_entry *entry;
1922 int count;
1923 int max;
1924 int index;
1925 int totallen = 0;
1926 int half;
1927 int lastdelta;
1928 int foundit = 0;
1929 int tmp;
1930
1931 /*
1932 * Examine entries until we reduce the absolute difference in
1933 * byte usage between the two blocks to a minimum.
1934 */
1935 max = ichdr1->count + ichdr2->count;
1936 half = (max + 1) * sizeof(*entry);
1937 half += ichdr1->usedbytes + ichdr2->usedbytes +
1938 xfs_attr_leaf_newentsize(state->args, NULL);
1939 half /= 2;
1940 lastdelta = state->args->geo->blksize;
1941 entry = xfs_attr3_leaf_entryp(leaf1);
1942 for (count = index = 0; count < max; entry++, index++, count++) {
1943
1944 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1945 /*
1946 * The new entry is in the first block, account for it.
1947 */
1948 if (count == blk1->index) {
1949 tmp = totallen + sizeof(*entry) +
1950 xfs_attr_leaf_newentsize(state->args, NULL);
1951 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1952 break;
1953 lastdelta = XFS_ATTR_ABS(half - tmp);
1954 totallen = tmp;
1955 foundit = 1;
1956 }
1957
1958 /*
1959 * Wrap around into the second block if necessary.
1960 */
1961 if (count == ichdr1->count) {
1962 leaf1 = leaf2;
1963 entry = xfs_attr3_leaf_entryp(leaf1);
1964 index = 0;
1965 }
1966
1967 /*
1968 * Figure out if next leaf entry would be too much.
1969 */
1970 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1971 index);
1972 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1973 break;
1974 lastdelta = XFS_ATTR_ABS(half - tmp);
1975 totallen = tmp;
1976 #undef XFS_ATTR_ABS
1977 }
1978
1979 /*
1980 * Calculate the number of usedbytes that will end up in lower block.
1981 * If new entry not in lower block, fix up the count.
1982 */
1983 totallen -= count * sizeof(*entry);
1984 if (foundit) {
1985 totallen -= sizeof(*entry) +
1986 xfs_attr_leaf_newentsize(state->args, NULL);
1987 }
1988
1989 *countarg = count;
1990 *usedbytesarg = totallen;
1991 return foundit;
1992 }
1993
1994 /*========================================================================
1995 * Routines used for shrinking the Btree.
1996 *========================================================================*/
1997
1998 /*
1999 * Check a leaf block and its neighbors to see if the block should be
2000 * collapsed into one or the other neighbor. Always keep the block
2001 * with the smaller block number.
2002 * If the current block is over 50% full, don't try to join it, return 0.
2003 * If the block is empty, fill in the state structure and return 2.
2004 * If it can be collapsed, fill in the state structure and return 1.
2005 * If nothing can be done, return 0.
2006 *
2007 * GROT: allow for INCOMPLETE entries in calculation.
2008 */
2009 int
xfs_attr3_leaf_toosmall(struct xfs_da_state * state,int * action)2010 xfs_attr3_leaf_toosmall(
2011 struct xfs_da_state *state,
2012 int *action)
2013 {
2014 struct xfs_attr_leafblock *leaf;
2015 struct xfs_da_state_blk *blk;
2016 struct xfs_attr3_icleaf_hdr ichdr;
2017 struct xfs_buf *bp;
2018 xfs_dablk_t blkno;
2019 int bytes;
2020 int forward;
2021 int error;
2022 int retval;
2023 int i;
2024
2025 trace_xfs_attr_leaf_toosmall(state->args);
2026
2027 /*
2028 * Check for the degenerate case of the block being over 50% full.
2029 * If so, it's not worth even looking to see if we might be able
2030 * to coalesce with a sibling.
2031 */
2032 blk = &state->path.blk[ state->path.active-1 ];
2033 leaf = blk->bp->b_addr;
2034 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2035 bytes = xfs_attr3_leaf_hdr_size(leaf) +
2036 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2037 ichdr.usedbytes;
2038 if (bytes > (state->args->geo->blksize >> 1)) {
2039 *action = 0; /* blk over 50%, don't try to join */
2040 return 0;
2041 }
2042
2043 /*
2044 * Check for the degenerate case of the block being empty.
2045 * If the block is empty, we'll simply delete it, no need to
2046 * coalesce it with a sibling block. We choose (arbitrarily)
2047 * to merge with the forward block unless it is NULL.
2048 */
2049 if (ichdr.count == 0) {
2050 /*
2051 * Make altpath point to the block we want to keep and
2052 * path point to the block we want to drop (this one).
2053 */
2054 forward = (ichdr.forw != 0);
2055 memcpy(&state->altpath, &state->path, sizeof(state->path));
2056 error = xfs_da3_path_shift(state, &state->altpath, forward,
2057 0, &retval);
2058 if (error)
2059 return error;
2060 if (retval) {
2061 *action = 0;
2062 } else {
2063 *action = 2;
2064 }
2065 return 0;
2066 }
2067
2068 /*
2069 * Examine each sibling block to see if we can coalesce with
2070 * at least 25% free space to spare. We need to figure out
2071 * whether to merge with the forward or the backward block.
2072 * We prefer coalescing with the lower numbered sibling so as
2073 * to shrink an attribute list over time.
2074 */
2075 /* start with smaller blk num */
2076 forward = ichdr.forw < ichdr.back;
2077 for (i = 0; i < 2; forward = !forward, i++) {
2078 struct xfs_attr3_icleaf_hdr ichdr2;
2079 if (forward)
2080 blkno = ichdr.forw;
2081 else
2082 blkno = ichdr.back;
2083 if (blkno == 0)
2084 continue;
2085 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2086 blkno, &bp);
2087 if (error)
2088 return error;
2089
2090 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2091
2092 bytes = state->args->geo->blksize -
2093 (state->args->geo->blksize >> 2) -
2094 ichdr.usedbytes - ichdr2.usedbytes -
2095 ((ichdr.count + ichdr2.count) *
2096 sizeof(xfs_attr_leaf_entry_t)) -
2097 xfs_attr3_leaf_hdr_size(leaf);
2098
2099 xfs_trans_brelse(state->args->trans, bp);
2100 if (bytes >= 0)
2101 break; /* fits with at least 25% to spare */
2102 }
2103 if (i >= 2) {
2104 *action = 0;
2105 return 0;
2106 }
2107
2108 /*
2109 * Make altpath point to the block we want to keep (the lower
2110 * numbered block) and path point to the block we want to drop.
2111 */
2112 memcpy(&state->altpath, &state->path, sizeof(state->path));
2113 if (blkno < blk->blkno) {
2114 error = xfs_da3_path_shift(state, &state->altpath, forward,
2115 0, &retval);
2116 } else {
2117 error = xfs_da3_path_shift(state, &state->path, forward,
2118 0, &retval);
2119 }
2120 if (error)
2121 return error;
2122 if (retval) {
2123 *action = 0;
2124 } else {
2125 *action = 1;
2126 }
2127 return 0;
2128 }
2129
2130 /*
2131 * Remove a name from the leaf attribute list structure.
2132 *
2133 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2134 * If two leaves are 37% full, when combined they will leave 25% free.
2135 */
2136 int
xfs_attr3_leaf_remove(struct xfs_buf * bp,struct xfs_da_args * args)2137 xfs_attr3_leaf_remove(
2138 struct xfs_buf *bp,
2139 struct xfs_da_args *args)
2140 {
2141 struct xfs_attr_leafblock *leaf;
2142 struct xfs_attr3_icleaf_hdr ichdr;
2143 struct xfs_attr_leaf_entry *entry;
2144 int before;
2145 int after;
2146 int smallest;
2147 int entsize;
2148 int tablesize;
2149 int tmp;
2150 int i;
2151
2152 trace_xfs_attr_leaf_remove(args);
2153
2154 leaf = bp->b_addr;
2155 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2156
2157 ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2158 ASSERT(args->index >= 0 && args->index < ichdr.count);
2159 ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2160 xfs_attr3_leaf_hdr_size(leaf));
2161
2162 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2163
2164 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2165 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2166
2167 /*
2168 * Scan through free region table:
2169 * check for adjacency of free'd entry with an existing one,
2170 * find smallest free region in case we need to replace it,
2171 * adjust any map that borders the entry table,
2172 */
2173 tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2174 + xfs_attr3_leaf_hdr_size(leaf);
2175 tmp = ichdr.freemap[0].size;
2176 before = after = -1;
2177 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2178 entsize = xfs_attr_leaf_entsize(leaf, args->index);
2179 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2180 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2181 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2182 if (ichdr.freemap[i].base == tablesize) {
2183 ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2184 ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2185 }
2186
2187 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2188 be16_to_cpu(entry->nameidx)) {
2189 before = i;
2190 } else if (ichdr.freemap[i].base ==
2191 (be16_to_cpu(entry->nameidx) + entsize)) {
2192 after = i;
2193 } else if (ichdr.freemap[i].size < tmp) {
2194 tmp = ichdr.freemap[i].size;
2195 smallest = i;
2196 }
2197 }
2198
2199 /*
2200 * Coalesce adjacent freemap regions,
2201 * or replace the smallest region.
2202 */
2203 if ((before >= 0) || (after >= 0)) {
2204 if ((before >= 0) && (after >= 0)) {
2205 ichdr.freemap[before].size += entsize;
2206 ichdr.freemap[before].size += ichdr.freemap[after].size;
2207 ichdr.freemap[after].base = 0;
2208 ichdr.freemap[after].size = 0;
2209 } else if (before >= 0) {
2210 ichdr.freemap[before].size += entsize;
2211 } else {
2212 ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2213 ichdr.freemap[after].size += entsize;
2214 }
2215 } else {
2216 /*
2217 * Replace smallest region (if it is smaller than free'd entry)
2218 */
2219 if (ichdr.freemap[smallest].size < entsize) {
2220 ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2221 ichdr.freemap[smallest].size = entsize;
2222 }
2223 }
2224
2225 /*
2226 * Did we remove the first entry?
2227 */
2228 if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2229 smallest = 1;
2230 else
2231 smallest = 0;
2232
2233 /*
2234 * Compress the remaining entries and zero out the removed stuff.
2235 */
2236 memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2237 ichdr.usedbytes -= entsize;
2238 xfs_trans_log_buf(args->trans, bp,
2239 XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2240 entsize));
2241
2242 tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2243 memmove(entry, entry + 1, tmp);
2244 ichdr.count--;
2245 xfs_trans_log_buf(args->trans, bp,
2246 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2247
2248 entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2249 memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2250
2251 /*
2252 * If we removed the first entry, re-find the first used byte
2253 * in the name area. Note that if the entry was the "firstused",
2254 * then we don't have a "hole" in our block resulting from
2255 * removing the name.
2256 */
2257 if (smallest) {
2258 tmp = args->geo->blksize;
2259 entry = xfs_attr3_leaf_entryp(leaf);
2260 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2261 ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2262 ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2263
2264 if (be16_to_cpu(entry->nameidx) < tmp)
2265 tmp = be16_to_cpu(entry->nameidx);
2266 }
2267 ichdr.firstused = tmp;
2268 ASSERT(ichdr.firstused != 0);
2269 } else {
2270 ichdr.holes = 1; /* mark as needing compaction */
2271 }
2272 xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2273 xfs_trans_log_buf(args->trans, bp,
2274 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2275 xfs_attr3_leaf_hdr_size(leaf)));
2276
2277 /*
2278 * Check if leaf is less than 50% full, caller may want to
2279 * "join" the leaf with a sibling if so.
2280 */
2281 tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2282 ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2283
2284 return tmp < args->geo->magicpct; /* leaf is < 37% full */
2285 }
2286
2287 /*
2288 * Move all the attribute list entries from drop_leaf into save_leaf.
2289 */
2290 void
xfs_attr3_leaf_unbalance(struct xfs_da_state * state,struct xfs_da_state_blk * drop_blk,struct xfs_da_state_blk * save_blk)2291 xfs_attr3_leaf_unbalance(
2292 struct xfs_da_state *state,
2293 struct xfs_da_state_blk *drop_blk,
2294 struct xfs_da_state_blk *save_blk)
2295 {
2296 struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2297 struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2298 struct xfs_attr3_icleaf_hdr drophdr;
2299 struct xfs_attr3_icleaf_hdr savehdr;
2300 struct xfs_attr_leaf_entry *entry;
2301
2302 trace_xfs_attr_leaf_unbalance(state->args);
2303
2304 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2305 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2306 entry = xfs_attr3_leaf_entryp(drop_leaf);
2307
2308 /*
2309 * Save last hashval from dying block for later Btree fixup.
2310 */
2311 drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2312
2313 /*
2314 * Check if we need a temp buffer, or can we do it in place.
2315 * Note that we don't check "leaf" for holes because we will
2316 * always be dropping it, toosmall() decided that for us already.
2317 */
2318 if (savehdr.holes == 0) {
2319 /*
2320 * dest leaf has no holes, so we add there. May need
2321 * to make some room in the entry array.
2322 */
2323 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2324 drop_blk->bp, &drophdr)) {
2325 xfs_attr3_leaf_moveents(state->args,
2326 drop_leaf, &drophdr, 0,
2327 save_leaf, &savehdr, 0,
2328 drophdr.count);
2329 } else {
2330 xfs_attr3_leaf_moveents(state->args,
2331 drop_leaf, &drophdr, 0,
2332 save_leaf, &savehdr,
2333 savehdr.count, drophdr.count);
2334 }
2335 } else {
2336 /*
2337 * Destination has holes, so we make a temporary copy
2338 * of the leaf and add them both to that.
2339 */
2340 struct xfs_attr_leafblock *tmp_leaf;
2341 struct xfs_attr3_icleaf_hdr tmphdr;
2342
2343 tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2344
2345 /*
2346 * Copy the header into the temp leaf so that all the stuff
2347 * not in the incore header is present and gets copied back in
2348 * once we've moved all the entries.
2349 */
2350 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2351
2352 memset(&tmphdr, 0, sizeof(tmphdr));
2353 tmphdr.magic = savehdr.magic;
2354 tmphdr.forw = savehdr.forw;
2355 tmphdr.back = savehdr.back;
2356 tmphdr.firstused = state->args->geo->blksize;
2357
2358 /* write the header to the temp buffer to initialise it */
2359 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2360
2361 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2362 drop_blk->bp, &drophdr)) {
2363 xfs_attr3_leaf_moveents(state->args,
2364 drop_leaf, &drophdr, 0,
2365 tmp_leaf, &tmphdr, 0,
2366 drophdr.count);
2367 xfs_attr3_leaf_moveents(state->args,
2368 save_leaf, &savehdr, 0,
2369 tmp_leaf, &tmphdr, tmphdr.count,
2370 savehdr.count);
2371 } else {
2372 xfs_attr3_leaf_moveents(state->args,
2373 save_leaf, &savehdr, 0,
2374 tmp_leaf, &tmphdr, 0,
2375 savehdr.count);
2376 xfs_attr3_leaf_moveents(state->args,
2377 drop_leaf, &drophdr, 0,
2378 tmp_leaf, &tmphdr, tmphdr.count,
2379 drophdr.count);
2380 }
2381 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2382 savehdr = tmphdr; /* struct copy */
2383 kmem_free(tmp_leaf);
2384 }
2385
2386 xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2387 xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2388 state->args->geo->blksize - 1);
2389
2390 /*
2391 * Copy out last hashval in each block for B-tree code.
2392 */
2393 entry = xfs_attr3_leaf_entryp(save_leaf);
2394 save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2395 }
2396
2397 /*========================================================================
2398 * Routines used for finding things in the Btree.
2399 *========================================================================*/
2400
2401 /*
2402 * Look up a name in a leaf attribute list structure.
2403 * This is the internal routine, it uses the caller's buffer.
2404 *
2405 * Note that duplicate keys are allowed, but only check within the
2406 * current leaf node. The Btree code must check in adjacent leaf nodes.
2407 *
2408 * Return in args->index the index into the entry[] array of either
2409 * the found entry, or where the entry should have been (insert before
2410 * that entry).
2411 *
2412 * Don't change the args->value unless we find the attribute.
2413 */
2414 int
xfs_attr3_leaf_lookup_int(struct xfs_buf * bp,struct xfs_da_args * args)2415 xfs_attr3_leaf_lookup_int(
2416 struct xfs_buf *bp,
2417 struct xfs_da_args *args)
2418 {
2419 struct xfs_attr_leafblock *leaf;
2420 struct xfs_attr3_icleaf_hdr ichdr;
2421 struct xfs_attr_leaf_entry *entry;
2422 struct xfs_attr_leaf_entry *entries;
2423 struct xfs_attr_leaf_name_local *name_loc;
2424 struct xfs_attr_leaf_name_remote *name_rmt;
2425 xfs_dahash_t hashval;
2426 int probe;
2427 int span;
2428
2429 trace_xfs_attr_leaf_lookup(args);
2430
2431 leaf = bp->b_addr;
2432 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2433 entries = xfs_attr3_leaf_entryp(leaf);
2434 if (ichdr.count >= args->geo->blksize / 8) {
2435 xfs_buf_mark_corrupt(bp);
2436 return -EFSCORRUPTED;
2437 }
2438
2439 /*
2440 * Binary search. (note: small blocks will skip this loop)
2441 */
2442 hashval = args->hashval;
2443 probe = span = ichdr.count / 2;
2444 for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2445 span /= 2;
2446 if (be32_to_cpu(entry->hashval) < hashval)
2447 probe += span;
2448 else if (be32_to_cpu(entry->hashval) > hashval)
2449 probe -= span;
2450 else
2451 break;
2452 }
2453 if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2454 xfs_buf_mark_corrupt(bp);
2455 return -EFSCORRUPTED;
2456 }
2457 if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2458 xfs_buf_mark_corrupt(bp);
2459 return -EFSCORRUPTED;
2460 }
2461
2462 /*
2463 * Since we may have duplicate hashval's, find the first matching
2464 * hashval in the leaf.
2465 */
2466 while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2467 entry--;
2468 probe--;
2469 }
2470 while (probe < ichdr.count &&
2471 be32_to_cpu(entry->hashval) < hashval) {
2472 entry++;
2473 probe++;
2474 }
2475 if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2476 args->index = probe;
2477 return -ENOATTR;
2478 }
2479
2480 /*
2481 * Duplicate keys may be present, so search all of them for a match.
2482 */
2483 for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2484 entry++, probe++) {
2485 /*
2486 * GROT: Add code to remove incomplete entries.
2487 */
2488 if (entry->flags & XFS_ATTR_LOCAL) {
2489 name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2490 if (!xfs_attr_match(args, name_loc->namelen,
2491 name_loc->nameval, entry->flags))
2492 continue;
2493 args->index = probe;
2494 return -EEXIST;
2495 } else {
2496 name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2497 if (!xfs_attr_match(args, name_rmt->namelen,
2498 name_rmt->name, entry->flags))
2499 continue;
2500 args->index = probe;
2501 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2502 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2503 args->rmtblkcnt = xfs_attr3_rmt_blocks(
2504 args->dp->i_mount,
2505 args->rmtvaluelen);
2506 return -EEXIST;
2507 }
2508 }
2509 args->index = probe;
2510 return -ENOATTR;
2511 }
2512
2513 /*
2514 * Get the value associated with an attribute name from a leaf attribute
2515 * list structure.
2516 *
2517 * If args->valuelen is zero, only the length needs to be returned. Unlike a
2518 * lookup, we only return an error if the attribute does not exist or we can't
2519 * retrieve the value.
2520 */
2521 int
xfs_attr3_leaf_getvalue(struct xfs_buf * bp,struct xfs_da_args * args)2522 xfs_attr3_leaf_getvalue(
2523 struct xfs_buf *bp,
2524 struct xfs_da_args *args)
2525 {
2526 struct xfs_attr_leafblock *leaf;
2527 struct xfs_attr3_icleaf_hdr ichdr;
2528 struct xfs_attr_leaf_entry *entry;
2529 struct xfs_attr_leaf_name_local *name_loc;
2530 struct xfs_attr_leaf_name_remote *name_rmt;
2531
2532 leaf = bp->b_addr;
2533 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2534 ASSERT(ichdr.count < args->geo->blksize / 8);
2535 ASSERT(args->index < ichdr.count);
2536
2537 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2538 if (entry->flags & XFS_ATTR_LOCAL) {
2539 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2540 ASSERT(name_loc->namelen == args->namelen);
2541 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2542 return xfs_attr_copy_value(args,
2543 &name_loc->nameval[args->namelen],
2544 be16_to_cpu(name_loc->valuelen));
2545 }
2546
2547 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2548 ASSERT(name_rmt->namelen == args->namelen);
2549 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2550 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2551 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2552 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2553 args->rmtvaluelen);
2554 return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2555 }
2556
2557 /*========================================================================
2558 * Utility routines.
2559 *========================================================================*/
2560
2561 /*
2562 * Move the indicated entries from one leaf to another.
2563 * NOTE: this routine modifies both source and destination leaves.
2564 */
2565 /*ARGSUSED*/
2566 STATIC void
xfs_attr3_leaf_moveents(struct xfs_da_args * args,struct xfs_attr_leafblock * leaf_s,struct xfs_attr3_icleaf_hdr * ichdr_s,int start_s,struct xfs_attr_leafblock * leaf_d,struct xfs_attr3_icleaf_hdr * ichdr_d,int start_d,int count)2567 xfs_attr3_leaf_moveents(
2568 struct xfs_da_args *args,
2569 struct xfs_attr_leafblock *leaf_s,
2570 struct xfs_attr3_icleaf_hdr *ichdr_s,
2571 int start_s,
2572 struct xfs_attr_leafblock *leaf_d,
2573 struct xfs_attr3_icleaf_hdr *ichdr_d,
2574 int start_d,
2575 int count)
2576 {
2577 struct xfs_attr_leaf_entry *entry_s;
2578 struct xfs_attr_leaf_entry *entry_d;
2579 int desti;
2580 int tmp;
2581 int i;
2582
2583 /*
2584 * Check for nothing to do.
2585 */
2586 if (count == 0)
2587 return;
2588
2589 /*
2590 * Set up environment.
2591 */
2592 ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2593 ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2594 ASSERT(ichdr_s->magic == ichdr_d->magic);
2595 ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2596 ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2597 + xfs_attr3_leaf_hdr_size(leaf_s));
2598 ASSERT(ichdr_d->count < args->geo->blksize / 8);
2599 ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2600 + xfs_attr3_leaf_hdr_size(leaf_d));
2601
2602 ASSERT(start_s < ichdr_s->count);
2603 ASSERT(start_d <= ichdr_d->count);
2604 ASSERT(count <= ichdr_s->count);
2605
2606
2607 /*
2608 * Move the entries in the destination leaf up to make a hole?
2609 */
2610 if (start_d < ichdr_d->count) {
2611 tmp = ichdr_d->count - start_d;
2612 tmp *= sizeof(xfs_attr_leaf_entry_t);
2613 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2614 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2615 memmove(entry_d, entry_s, tmp);
2616 }
2617
2618 /*
2619 * Copy all entry's in the same (sorted) order,
2620 * but allocate attribute info packed and in sequence.
2621 */
2622 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2623 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2624 desti = start_d;
2625 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2626 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2627 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2628 #ifdef GROT
2629 /*
2630 * Code to drop INCOMPLETE entries. Difficult to use as we
2631 * may also need to change the insertion index. Code turned
2632 * off for 6.2, should be revisited later.
2633 */
2634 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2635 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2636 ichdr_s->usedbytes -= tmp;
2637 ichdr_s->count -= 1;
2638 entry_d--; /* to compensate for ++ in loop hdr */
2639 desti--;
2640 if ((start_s + i) < offset)
2641 result++; /* insertion index adjustment */
2642 } else {
2643 #endif /* GROT */
2644 ichdr_d->firstused -= tmp;
2645 /* both on-disk, don't endian flip twice */
2646 entry_d->hashval = entry_s->hashval;
2647 entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2648 entry_d->flags = entry_s->flags;
2649 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2650 <= args->geo->blksize);
2651 memmove(xfs_attr3_leaf_name(leaf_d, desti),
2652 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2653 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2654 <= args->geo->blksize);
2655 memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2656 ichdr_s->usedbytes -= tmp;
2657 ichdr_d->usedbytes += tmp;
2658 ichdr_s->count -= 1;
2659 ichdr_d->count += 1;
2660 tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2661 + xfs_attr3_leaf_hdr_size(leaf_d);
2662 ASSERT(ichdr_d->firstused >= tmp);
2663 #ifdef GROT
2664 }
2665 #endif /* GROT */
2666 }
2667
2668 /*
2669 * Zero out the entries we just copied.
2670 */
2671 if (start_s == ichdr_s->count) {
2672 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2673 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2674 ASSERT(((char *)entry_s + tmp) <=
2675 ((char *)leaf_s + args->geo->blksize));
2676 memset(entry_s, 0, tmp);
2677 } else {
2678 /*
2679 * Move the remaining entries down to fill the hole,
2680 * then zero the entries at the top.
2681 */
2682 tmp = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2683 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2684 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2685 memmove(entry_d, entry_s, tmp);
2686
2687 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2688 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2689 ASSERT(((char *)entry_s + tmp) <=
2690 ((char *)leaf_s + args->geo->blksize));
2691 memset(entry_s, 0, tmp);
2692 }
2693
2694 /*
2695 * Fill in the freemap information
2696 */
2697 ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2698 ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2699 ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2700 ichdr_d->freemap[1].base = 0;
2701 ichdr_d->freemap[2].base = 0;
2702 ichdr_d->freemap[1].size = 0;
2703 ichdr_d->freemap[2].size = 0;
2704 ichdr_s->holes = 1; /* leaf may not be compact */
2705 }
2706
2707 /*
2708 * Pick up the last hashvalue from a leaf block.
2709 */
2710 xfs_dahash_t
xfs_attr_leaf_lasthash(struct xfs_buf * bp,int * count)2711 xfs_attr_leaf_lasthash(
2712 struct xfs_buf *bp,
2713 int *count)
2714 {
2715 struct xfs_attr3_icleaf_hdr ichdr;
2716 struct xfs_attr_leaf_entry *entries;
2717 struct xfs_mount *mp = bp->b_mount;
2718
2719 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2720 entries = xfs_attr3_leaf_entryp(bp->b_addr);
2721 if (count)
2722 *count = ichdr.count;
2723 if (!ichdr.count)
2724 return 0;
2725 return be32_to_cpu(entries[ichdr.count - 1].hashval);
2726 }
2727
2728 /*
2729 * Calculate the number of bytes used to store the indicated attribute
2730 * (whether local or remote only calculate bytes in this block).
2731 */
2732 STATIC int
xfs_attr_leaf_entsize(xfs_attr_leafblock_t * leaf,int index)2733 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2734 {
2735 struct xfs_attr_leaf_entry *entries;
2736 xfs_attr_leaf_name_local_t *name_loc;
2737 xfs_attr_leaf_name_remote_t *name_rmt;
2738 int size;
2739
2740 entries = xfs_attr3_leaf_entryp(leaf);
2741 if (entries[index].flags & XFS_ATTR_LOCAL) {
2742 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2743 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2744 be16_to_cpu(name_loc->valuelen));
2745 } else {
2746 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2747 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2748 }
2749 return size;
2750 }
2751
2752 /*
2753 * Calculate the number of bytes that would be required to store the new
2754 * attribute (whether local or remote only calculate bytes in this block).
2755 * This routine decides as a side effect whether the attribute will be
2756 * a "local" or a "remote" attribute.
2757 */
2758 int
xfs_attr_leaf_newentsize(struct xfs_da_args * args,int * local)2759 xfs_attr_leaf_newentsize(
2760 struct xfs_da_args *args,
2761 int *local)
2762 {
2763 int size;
2764
2765 size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2766 if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2767 if (local)
2768 *local = 1;
2769 return size;
2770 }
2771 if (local)
2772 *local = 0;
2773 return xfs_attr_leaf_entsize_remote(args->namelen);
2774 }
2775
2776
2777 /*========================================================================
2778 * Manage the INCOMPLETE flag in a leaf entry
2779 *========================================================================*/
2780
2781 /*
2782 * Clear the INCOMPLETE flag on an entry in a leaf block.
2783 */
2784 int
xfs_attr3_leaf_clearflag(struct xfs_da_args * args)2785 xfs_attr3_leaf_clearflag(
2786 struct xfs_da_args *args)
2787 {
2788 struct xfs_attr_leafblock *leaf;
2789 struct xfs_attr_leaf_entry *entry;
2790 struct xfs_attr_leaf_name_remote *name_rmt;
2791 struct xfs_buf *bp;
2792 int error;
2793 #ifdef DEBUG
2794 struct xfs_attr3_icleaf_hdr ichdr;
2795 xfs_attr_leaf_name_local_t *name_loc;
2796 int namelen;
2797 char *name;
2798 #endif /* DEBUG */
2799
2800 trace_xfs_attr_leaf_clearflag(args);
2801 /*
2802 * Set up the operation.
2803 */
2804 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2805 if (error)
2806 return error;
2807
2808 leaf = bp->b_addr;
2809 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2810 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2811
2812 #ifdef DEBUG
2813 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2814 ASSERT(args->index < ichdr.count);
2815 ASSERT(args->index >= 0);
2816
2817 if (entry->flags & XFS_ATTR_LOCAL) {
2818 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2819 namelen = name_loc->namelen;
2820 name = (char *)name_loc->nameval;
2821 } else {
2822 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2823 namelen = name_rmt->namelen;
2824 name = (char *)name_rmt->name;
2825 }
2826 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2827 ASSERT(namelen == args->namelen);
2828 ASSERT(memcmp(name, args->name, namelen) == 0);
2829 #endif /* DEBUG */
2830
2831 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2832 xfs_trans_log_buf(args->trans, bp,
2833 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2834
2835 if (args->rmtblkno) {
2836 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2837 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2838 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2839 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2840 xfs_trans_log_buf(args->trans, bp,
2841 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2842 }
2843
2844 return 0;
2845 }
2846
2847 /*
2848 * Set the INCOMPLETE flag on an entry in a leaf block.
2849 */
2850 int
xfs_attr3_leaf_setflag(struct xfs_da_args * args)2851 xfs_attr3_leaf_setflag(
2852 struct xfs_da_args *args)
2853 {
2854 struct xfs_attr_leafblock *leaf;
2855 struct xfs_attr_leaf_entry *entry;
2856 struct xfs_attr_leaf_name_remote *name_rmt;
2857 struct xfs_buf *bp;
2858 int error;
2859 #ifdef DEBUG
2860 struct xfs_attr3_icleaf_hdr ichdr;
2861 #endif
2862
2863 trace_xfs_attr_leaf_setflag(args);
2864
2865 /*
2866 * Set up the operation.
2867 */
2868 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2869 if (error)
2870 return error;
2871
2872 leaf = bp->b_addr;
2873 #ifdef DEBUG
2874 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2875 ASSERT(args->index < ichdr.count);
2876 ASSERT(args->index >= 0);
2877 #endif
2878 entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2879
2880 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2881 entry->flags |= XFS_ATTR_INCOMPLETE;
2882 xfs_trans_log_buf(args->trans, bp,
2883 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2884 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2885 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2886 name_rmt->valueblk = 0;
2887 name_rmt->valuelen = 0;
2888 xfs_trans_log_buf(args->trans, bp,
2889 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2890 }
2891
2892 return 0;
2893 }
2894
2895 /*
2896 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2897 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2898 * entry given by args->blkno2/index2.
2899 *
2900 * Note that they could be in different blocks, or in the same block.
2901 */
2902 int
xfs_attr3_leaf_flipflags(struct xfs_da_args * args)2903 xfs_attr3_leaf_flipflags(
2904 struct xfs_da_args *args)
2905 {
2906 struct xfs_attr_leafblock *leaf1;
2907 struct xfs_attr_leafblock *leaf2;
2908 struct xfs_attr_leaf_entry *entry1;
2909 struct xfs_attr_leaf_entry *entry2;
2910 struct xfs_attr_leaf_name_remote *name_rmt;
2911 struct xfs_buf *bp1;
2912 struct xfs_buf *bp2;
2913 int error;
2914 #ifdef DEBUG
2915 struct xfs_attr3_icleaf_hdr ichdr1;
2916 struct xfs_attr3_icleaf_hdr ichdr2;
2917 xfs_attr_leaf_name_local_t *name_loc;
2918 int namelen1, namelen2;
2919 char *name1, *name2;
2920 #endif /* DEBUG */
2921
2922 trace_xfs_attr_leaf_flipflags(args);
2923
2924 /*
2925 * Read the block containing the "old" attr
2926 */
2927 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2928 if (error)
2929 return error;
2930
2931 /*
2932 * Read the block containing the "new" attr, if it is different
2933 */
2934 if (args->blkno2 != args->blkno) {
2935 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2936 &bp2);
2937 if (error)
2938 return error;
2939 } else {
2940 bp2 = bp1;
2941 }
2942
2943 leaf1 = bp1->b_addr;
2944 entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2945
2946 leaf2 = bp2->b_addr;
2947 entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2948
2949 #ifdef DEBUG
2950 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2951 ASSERT(args->index < ichdr1.count);
2952 ASSERT(args->index >= 0);
2953
2954 xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2955 ASSERT(args->index2 < ichdr2.count);
2956 ASSERT(args->index2 >= 0);
2957
2958 if (entry1->flags & XFS_ATTR_LOCAL) {
2959 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2960 namelen1 = name_loc->namelen;
2961 name1 = (char *)name_loc->nameval;
2962 } else {
2963 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2964 namelen1 = name_rmt->namelen;
2965 name1 = (char *)name_rmt->name;
2966 }
2967 if (entry2->flags & XFS_ATTR_LOCAL) {
2968 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2969 namelen2 = name_loc->namelen;
2970 name2 = (char *)name_loc->nameval;
2971 } else {
2972 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2973 namelen2 = name_rmt->namelen;
2974 name2 = (char *)name_rmt->name;
2975 }
2976 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2977 ASSERT(namelen1 == namelen2);
2978 ASSERT(memcmp(name1, name2, namelen1) == 0);
2979 #endif /* DEBUG */
2980
2981 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2982 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2983
2984 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2985 xfs_trans_log_buf(args->trans, bp1,
2986 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2987 if (args->rmtblkno) {
2988 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2989 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2990 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2991 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2992 xfs_trans_log_buf(args->trans, bp1,
2993 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2994 }
2995
2996 entry2->flags |= XFS_ATTR_INCOMPLETE;
2997 xfs_trans_log_buf(args->trans, bp2,
2998 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2999 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
3000 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
3001 name_rmt->valueblk = 0;
3002 name_rmt->valuelen = 0;
3003 xfs_trans_log_buf(args->trans, bp2,
3004 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3005 }
3006
3007 return 0;
3008 }
3009