fix short copy handling in copy_mc_pipe_to_iter()
[sfrench/cifs-2.6.git] / fs / xfs / libxfs / xfs_attr_leaf.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "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 int 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
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
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
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
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
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 static xfs_failaddr_t
293 xfs_attr3_leaf_verify(
294         struct xfs_buf                  *bp)
295 {
296         struct xfs_attr3_icleaf_hdr     ichdr;
297         struct xfs_mount                *mp = bp->b_mount;
298         struct xfs_attr_leafblock       *leaf = bp->b_addr;
299         struct xfs_attr_leaf_entry      *entries;
300         struct xfs_attr_leaf_entry      *ent;
301         char                            *buf_end;
302         uint32_t                        end;    /* must be 32bit - see below */
303         __u32                           last_hashval = 0;
304         int                             i;
305         xfs_failaddr_t                  fa;
306
307         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
308
309         fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
310         if (fa)
311                 return fa;
312
313         /*
314          * Empty leaf blocks should never occur;  they imply the existence of a
315          * software bug that needs fixing. xfs_repair also flags them as a
316          * corruption that needs fixing, so we should never let these go to
317          * disk.
318          */
319         if (ichdr.count == 0)
320                 return __this_address;
321
322         /*
323          * firstused is the block offset of the first name info structure.
324          * Make sure it doesn't go off the block or crash into the header.
325          */
326         if (ichdr.firstused > mp->m_attr_geo->blksize)
327                 return __this_address;
328         if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
329                 return __this_address;
330
331         /* Make sure the entries array doesn't crash into the name info. */
332         entries = xfs_attr3_leaf_entryp(bp->b_addr);
333         if ((char *)&entries[ichdr.count] >
334             (char *)bp->b_addr + ichdr.firstused)
335                 return __this_address;
336
337         /*
338          * NOTE: This verifier historically failed empty leaf buffers because
339          * we expect the fork to be in another format. Empty attr fork format
340          * conversions are possible during xattr set, however, and format
341          * conversion is not atomic with the xattr set that triggers it. We
342          * cannot assume leaf blocks are non-empty until that is addressed.
343         */
344         buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
345         for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
346                 fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
347                                 ent, i, &last_hashval);
348                 if (fa)
349                         return fa;
350         }
351
352         /*
353          * Quickly check the freemap information.  Attribute data has to be
354          * aligned to 4-byte boundaries, and likewise for the free space.
355          *
356          * Note that for 64k block size filesystems, the freemap entries cannot
357          * overflow as they are only be16 fields. However, when checking end
358          * pointer of the freemap, we have to be careful to detect overflows and
359          * so use uint32_t for those checks.
360          */
361         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
362                 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
363                         return __this_address;
364                 if (ichdr.freemap[i].base & 0x3)
365                         return __this_address;
366                 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
367                         return __this_address;
368                 if (ichdr.freemap[i].size & 0x3)
369                         return __this_address;
370
371                 /* be care of 16 bit overflows here */
372                 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
373                 if (end < ichdr.freemap[i].base)
374                         return __this_address;
375                 if (end > mp->m_attr_geo->blksize)
376                         return __this_address;
377         }
378
379         return NULL;
380 }
381
382 static void
383 xfs_attr3_leaf_write_verify(
384         struct xfs_buf  *bp)
385 {
386         struct xfs_mount        *mp = bp->b_mount;
387         struct xfs_buf_log_item *bip = bp->b_log_item;
388         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
389         xfs_failaddr_t          fa;
390
391         fa = xfs_attr3_leaf_verify(bp);
392         if (fa) {
393                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
394                 return;
395         }
396
397         if (!xfs_has_crc(mp))
398                 return;
399
400         if (bip)
401                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
402
403         xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
404 }
405
406 /*
407  * leaf/node format detection on trees is sketchy, so a node read can be done on
408  * leaf level blocks when detection identifies the tree as a node format tree
409  * incorrectly. In this case, we need to swap the verifier to match the correct
410  * format of the block being read.
411  */
412 static void
413 xfs_attr3_leaf_read_verify(
414         struct xfs_buf          *bp)
415 {
416         struct xfs_mount        *mp = bp->b_mount;
417         xfs_failaddr_t          fa;
418
419         if (xfs_has_crc(mp) &&
420              !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
421                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
422         else {
423                 fa = xfs_attr3_leaf_verify(bp);
424                 if (fa)
425                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
426         }
427 }
428
429 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
430         .name = "xfs_attr3_leaf",
431         .magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
432                      cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
433         .verify_read = xfs_attr3_leaf_read_verify,
434         .verify_write = xfs_attr3_leaf_write_verify,
435         .verify_struct = xfs_attr3_leaf_verify,
436 };
437
438 int
439 xfs_attr3_leaf_read(
440         struct xfs_trans        *tp,
441         struct xfs_inode        *dp,
442         xfs_dablk_t             bno,
443         struct xfs_buf          **bpp)
444 {
445         int                     err;
446
447         err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
448                         &xfs_attr3_leaf_buf_ops);
449         if (!err && tp && *bpp)
450                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
451         return err;
452 }
453
454 /*========================================================================
455  * Namespace helper routines
456  *========================================================================*/
457
458 /*
459  * If we are in log recovery, then we want the lookup to ignore the INCOMPLETE
460  * flag on disk - if there's an incomplete attr then recovery needs to tear it
461  * down. If there's no incomplete attr, then recovery needs to tear that attr
462  * down to replace it with the attr that has been logged. In this case, the
463  * INCOMPLETE flag will not be set in attr->attr_filter, but rather
464  * XFS_DA_OP_RECOVERY will be set in args->op_flags.
465  */
466 static bool
467 xfs_attr_match(
468         struct xfs_da_args      *args,
469         uint8_t                 namelen,
470         unsigned char           *name,
471         int                     flags)
472 {
473
474         if (args->namelen != namelen)
475                 return false;
476         if (memcmp(args->name, name, namelen) != 0)
477                 return false;
478
479         /* Recovery ignores the INCOMPLETE flag. */
480         if ((args->op_flags & XFS_DA_OP_RECOVERY) &&
481             args->attr_filter == (flags & XFS_ATTR_NSP_ONDISK_MASK))
482                 return true;
483
484         /* All remaining matches need to be filtered by INCOMPLETE state. */
485         if (args->attr_filter !=
486             (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
487                 return false;
488         return true;
489 }
490
491 static int
492 xfs_attr_copy_value(
493         struct xfs_da_args      *args,
494         unsigned char           *value,
495         int                     valuelen)
496 {
497         /*
498          * No copy if all we have to do is get the length
499          */
500         if (!args->valuelen) {
501                 args->valuelen = valuelen;
502                 return 0;
503         }
504
505         /*
506          * No copy if the length of the existing buffer is too small
507          */
508         if (args->valuelen < valuelen) {
509                 args->valuelen = valuelen;
510                 return -ERANGE;
511         }
512
513         if (!args->value) {
514                 args->value = kvmalloc(valuelen, GFP_KERNEL | __GFP_NOLOCKDEP);
515                 if (!args->value)
516                         return -ENOMEM;
517         }
518         args->valuelen = valuelen;
519
520         /* remote block xattr requires IO for copy-in */
521         if (args->rmtblkno)
522                 return xfs_attr_rmtval_get(args);
523
524         /*
525          * This is to prevent a GCC warning because the remote xattr case
526          * doesn't have a value to pass in. In that case, we never reach here,
527          * but GCC can't work that out and so throws a "passing NULL to
528          * memcpy" warning.
529          */
530         if (!value)
531                 return -EINVAL;
532         memcpy(args->value, value, valuelen);
533         return 0;
534 }
535
536 /*========================================================================
537  * External routines when attribute fork size < XFS_LITINO(mp).
538  *========================================================================*/
539
540 /*
541  * Query whether the total requested number of attr fork bytes of extended
542  * attribute space will be able to fit inline.
543  *
544  * Returns zero if not, else the i_forkoff fork offset to be used in the
545  * literal area for attribute data once the new bytes have been added.
546  *
547  * i_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
548  * special case for dev/uuid inodes, they have fixed size data forks.
549  */
550 int
551 xfs_attr_shortform_bytesfit(
552         struct xfs_inode        *dp,
553         int                     bytes)
554 {
555         struct xfs_mount        *mp = dp->i_mount;
556         int64_t                 dsize;
557         int                     minforkoff;
558         int                     maxforkoff;
559         int                     offset;
560
561         /*
562          * Check if the new size could fit at all first:
563          */
564         if (bytes > XFS_LITINO(mp))
565                 return 0;
566
567         /* rounded down */
568         offset = (XFS_LITINO(mp) - bytes) >> 3;
569
570         if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
571                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
572                 return (offset >= minforkoff) ? minforkoff : 0;
573         }
574
575         /*
576          * If the requested numbers of bytes is smaller or equal to the
577          * current attribute fork size we can always proceed.
578          *
579          * Note that if_bytes in the data fork might actually be larger than
580          * the current data fork size is due to delalloc extents. In that
581          * case either the extent count will go down when they are converted
582          * to real extents, or the delalloc conversion will take care of the
583          * literal area rebalancing.
584          */
585         if (bytes <= XFS_IFORK_ASIZE(dp))
586                 return dp->i_forkoff;
587
588         /*
589          * For attr2 we can try to move the forkoff if there is space in the
590          * literal area, but for the old format we are done if there is no
591          * space in the fixed attribute fork.
592          */
593         if (!xfs_has_attr2(mp))
594                 return 0;
595
596         dsize = dp->i_df.if_bytes;
597
598         switch (dp->i_df.if_format) {
599         case XFS_DINODE_FMT_EXTENTS:
600                 /*
601                  * If there is no attr fork and the data fork is extents,
602                  * determine if creating the default attr fork will result
603                  * in the extents form migrating to btree. If so, the
604                  * minimum offset only needs to be the space required for
605                  * the btree root.
606                  */
607                 if (!dp->i_forkoff && dp->i_df.if_bytes >
608                     xfs_default_attroffset(dp))
609                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
610                 break;
611         case XFS_DINODE_FMT_BTREE:
612                 /*
613                  * If we have a data btree then keep forkoff if we have one,
614                  * otherwise we are adding a new attr, so then we set
615                  * minforkoff to where the btree root can finish so we have
616                  * plenty of room for attrs
617                  */
618                 if (dp->i_forkoff) {
619                         if (offset < dp->i_forkoff)
620                                 return 0;
621                         return dp->i_forkoff;
622                 }
623                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
624                 break;
625         }
626
627         /*
628          * A data fork btree root must have space for at least
629          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
630          */
631         minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
632         minforkoff = roundup(minforkoff, 8) >> 3;
633
634         /* attr fork btree root can have at least this many key/ptr pairs */
635         maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
636         maxforkoff = maxforkoff >> 3;   /* rounded down */
637
638         if (offset >= maxforkoff)
639                 return maxforkoff;
640         if (offset >= minforkoff)
641                 return offset;
642         return 0;
643 }
644
645 /*
646  * Switch on the ATTR2 superblock bit (implies also FEATURES2) unless:
647  * - noattr2 mount option is set,
648  * - on-disk version bit says it is already set, or
649  * - the attr2 mount option is not set to enable automatic upgrade from attr1.
650  */
651 STATIC void
652 xfs_sbversion_add_attr2(
653         struct xfs_mount        *mp,
654         struct xfs_trans        *tp)
655 {
656         if (xfs_has_noattr2(mp))
657                 return;
658         if (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)
659                 return;
660         if (!xfs_has_attr2(mp))
661                 return;
662
663         spin_lock(&mp->m_sb_lock);
664         xfs_add_attr2(mp);
665         spin_unlock(&mp->m_sb_lock);
666         xfs_log_sb(tp);
667 }
668
669 /*
670  * Create the initial contents of a shortform attribute list.
671  */
672 void
673 xfs_attr_shortform_create(
674         struct xfs_da_args      *args)
675 {
676         struct xfs_inode        *dp = args->dp;
677         struct xfs_ifork        *ifp = dp->i_afp;
678         struct xfs_attr_sf_hdr  *hdr;
679
680         trace_xfs_attr_sf_create(args);
681
682         ASSERT(ifp->if_bytes == 0);
683         if (ifp->if_format == XFS_DINODE_FMT_EXTENTS)
684                 ifp->if_format = XFS_DINODE_FMT_LOCAL;
685         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
686         hdr = (struct xfs_attr_sf_hdr *)ifp->if_u1.if_data;
687         memset(hdr, 0, sizeof(*hdr));
688         hdr->totsize = cpu_to_be16(sizeof(*hdr));
689         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
690 }
691
692 /*
693  * Return -EEXIST if attr is found, or -ENOATTR if not
694  * args:  args containing attribute name and namelen
695  * sfep:  If not null, pointer will be set to the last attr entry found on
696           -EEXIST.  On -ENOATTR pointer is left at the last entry in the list
697  * basep: If not null, pointer is set to the byte offset of the entry in the
698  *        list on -EEXIST.  On -ENOATTR, pointer is left at the byte offset of
699  *        the last entry in the list
700  */
701 int
702 xfs_attr_sf_findname(
703         struct xfs_da_args       *args,
704         struct xfs_attr_sf_entry **sfep,
705         unsigned int             *basep)
706 {
707         struct xfs_attr_shortform *sf;
708         struct xfs_attr_sf_entry *sfe;
709         unsigned int            base = sizeof(struct xfs_attr_sf_hdr);
710         int                     size = 0;
711         int                     end;
712         int                     i;
713
714         sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
715         sfe = &sf->list[0];
716         end = sf->hdr.count;
717         for (i = 0; i < end; sfe = xfs_attr_sf_nextentry(sfe),
718                              base += size, i++) {
719                 size = xfs_attr_sf_entsize(sfe);
720                 if (!xfs_attr_match(args, sfe->namelen, sfe->nameval,
721                                     sfe->flags))
722                         continue;
723                 break;
724         }
725
726         if (sfep != NULL)
727                 *sfep = sfe;
728
729         if (basep != NULL)
730                 *basep = base;
731
732         if (i == end)
733                 return -ENOATTR;
734         return -EEXIST;
735 }
736
737 /*
738  * Add a name/value pair to the shortform attribute list.
739  * Overflow from the inode has already been checked for.
740  */
741 void
742 xfs_attr_shortform_add(
743         struct xfs_da_args              *args,
744         int                             forkoff)
745 {
746         struct xfs_attr_shortform       *sf;
747         struct xfs_attr_sf_entry        *sfe;
748         int                             offset, size;
749         struct xfs_mount                *mp;
750         struct xfs_inode                *dp;
751         struct xfs_ifork                *ifp;
752
753         trace_xfs_attr_sf_add(args);
754
755         dp = args->dp;
756         mp = dp->i_mount;
757         dp->i_forkoff = forkoff;
758
759         ifp = dp->i_afp;
760         ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
761         sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
762         if (xfs_attr_sf_findname(args, &sfe, NULL) == -EEXIST)
763                 ASSERT(0);
764
765         offset = (char *)sfe - (char *)sf;
766         size = xfs_attr_sf_entsize_byname(args->namelen, args->valuelen);
767         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
768         sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
769         sfe = (struct xfs_attr_sf_entry *)((char *)sf + offset);
770
771         sfe->namelen = args->namelen;
772         sfe->valuelen = args->valuelen;
773         sfe->flags = args->attr_filter;
774         memcpy(sfe->nameval, args->name, args->namelen);
775         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
776         sf->hdr.count++;
777         be16_add_cpu(&sf->hdr.totsize, size);
778         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
779
780         xfs_sbversion_add_attr2(mp, args->trans);
781 }
782
783 /*
784  * After the last attribute is removed revert to original inode format,
785  * making all literal area available to the data fork once more.
786  */
787 void
788 xfs_attr_fork_remove(
789         struct xfs_inode        *ip,
790         struct xfs_trans        *tp)
791 {
792         ASSERT(ip->i_afp->if_nextents == 0);
793
794         xfs_idestroy_fork(ip->i_afp);
795         kmem_cache_free(xfs_ifork_cache, ip->i_afp);
796         ip->i_afp = NULL;
797         ip->i_forkoff = 0;
798         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
799 }
800
801 /*
802  * Remove an attribute from the shortform attribute list structure.
803  */
804 int
805 xfs_attr_sf_removename(
806         struct xfs_da_args              *args)
807 {
808         struct xfs_attr_shortform       *sf;
809         struct xfs_attr_sf_entry        *sfe;
810         int                             size = 0, end, totsize;
811         unsigned int                    base;
812         struct xfs_mount                *mp;
813         struct xfs_inode                *dp;
814         int                             error;
815
816         trace_xfs_attr_sf_remove(args);
817
818         dp = args->dp;
819         mp = dp->i_mount;
820         sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data;
821
822         error = xfs_attr_sf_findname(args, &sfe, &base);
823
824         /*
825          * If we are recovering an operation, finding nothing to
826          * remove is not an error - it just means there was nothing
827          * to clean up.
828          */
829         if (error == -ENOATTR && (args->op_flags & XFS_DA_OP_RECOVERY))
830                 return 0;
831         if (error != -EEXIST)
832                 return error;
833         size = xfs_attr_sf_entsize(sfe);
834
835         /*
836          * Fix up the attribute fork data, covering the hole
837          */
838         end = base + size;
839         totsize = be16_to_cpu(sf->hdr.totsize);
840         if (end != totsize)
841                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
842         sf->hdr.count--;
843         be16_add_cpu(&sf->hdr.totsize, -size);
844
845         /*
846          * Fix up the start offset of the attribute fork
847          */
848         totsize -= size;
849         if (totsize == sizeof(xfs_attr_sf_hdr_t) && xfs_has_attr2(mp) &&
850             (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
851             !(args->op_flags & (XFS_DA_OP_ADDNAME | XFS_DA_OP_REPLACE))) {
852                 xfs_attr_fork_remove(dp, args->trans);
853         } else {
854                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
855                 dp->i_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
856                 ASSERT(dp->i_forkoff);
857                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
858                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
859                                 !xfs_has_attr2(mp) ||
860                                 dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
861                 xfs_trans_log_inode(args->trans, dp,
862                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
863         }
864
865         xfs_sbversion_add_attr2(mp, args->trans);
866
867         return 0;
868 }
869
870 /*
871  * Look up a name in a shortform attribute list structure.
872  */
873 /*ARGSUSED*/
874 int
875 xfs_attr_shortform_lookup(xfs_da_args_t *args)
876 {
877         struct xfs_attr_shortform *sf;
878         struct xfs_attr_sf_entry *sfe;
879         int i;
880         struct xfs_ifork *ifp;
881
882         trace_xfs_attr_sf_lookup(args);
883
884         ifp = args->dp->i_afp;
885         ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
886         sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
887         sfe = &sf->list[0];
888         for (i = 0; i < sf->hdr.count;
889                                 sfe = xfs_attr_sf_nextentry(sfe), i++) {
890                 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
891                                 sfe->flags))
892                         return -EEXIST;
893         }
894         return -ENOATTR;
895 }
896
897 /*
898  * Retrieve the attribute value and length.
899  *
900  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
901  * lookup, we only return an error if the attribute does not exist or we can't
902  * retrieve the value.
903  */
904 int
905 xfs_attr_shortform_getvalue(
906         struct xfs_da_args      *args)
907 {
908         struct xfs_attr_shortform *sf;
909         struct xfs_attr_sf_entry *sfe;
910         int                     i;
911
912         ASSERT(args->dp->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
913         sf = (struct xfs_attr_shortform *)args->dp->i_afp->if_u1.if_data;
914         sfe = &sf->list[0];
915         for (i = 0; i < sf->hdr.count;
916                                 sfe = xfs_attr_sf_nextentry(sfe), i++) {
917                 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
918                                 sfe->flags))
919                         return xfs_attr_copy_value(args,
920                                 &sfe->nameval[args->namelen], sfe->valuelen);
921         }
922         return -ENOATTR;
923 }
924
925 /*
926  * Convert from using the shortform to the leaf.  On success, return the
927  * buffer so that we can keep it locked until we're totally done with it.
928  */
929 int
930 xfs_attr_shortform_to_leaf(
931         struct xfs_da_args              *args,
932         struct xfs_buf                  **leaf_bp)
933 {
934         struct xfs_inode                *dp;
935         struct xfs_attr_shortform       *sf;
936         struct xfs_attr_sf_entry        *sfe;
937         struct xfs_da_args              nargs;
938         char                            *tmpbuffer;
939         int                             error, i, size;
940         xfs_dablk_t                     blkno;
941         struct xfs_buf                  *bp;
942         struct xfs_ifork                *ifp;
943
944         trace_xfs_attr_sf_to_leaf(args);
945
946         dp = args->dp;
947         ifp = dp->i_afp;
948         sf = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
949         size = be16_to_cpu(sf->hdr.totsize);
950         tmpbuffer = kmem_alloc(size, 0);
951         ASSERT(tmpbuffer != NULL);
952         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
953         sf = (struct xfs_attr_shortform *)tmpbuffer;
954
955         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
956         xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
957
958         bp = NULL;
959         error = xfs_da_grow_inode(args, &blkno);
960         if (error)
961                 goto out;
962
963         ASSERT(blkno == 0);
964         error = xfs_attr3_leaf_create(args, blkno, &bp);
965         if (error)
966                 goto out;
967
968         memset((char *)&nargs, 0, sizeof(nargs));
969         nargs.dp = dp;
970         nargs.geo = args->geo;
971         nargs.total = args->total;
972         nargs.whichfork = XFS_ATTR_FORK;
973         nargs.trans = args->trans;
974         nargs.op_flags = XFS_DA_OP_OKNOENT;
975
976         sfe = &sf->list[0];
977         for (i = 0; i < sf->hdr.count; i++) {
978                 nargs.name = sfe->nameval;
979                 nargs.namelen = sfe->namelen;
980                 nargs.value = &sfe->nameval[nargs.namelen];
981                 nargs.valuelen = sfe->valuelen;
982                 nargs.hashval = xfs_da_hashname(sfe->nameval,
983                                                 sfe->namelen);
984                 nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
985                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
986                 ASSERT(error == -ENOATTR);
987                 error = xfs_attr3_leaf_add(bp, &nargs);
988                 ASSERT(error != -ENOSPC);
989                 if (error)
990                         goto out;
991                 sfe = xfs_attr_sf_nextentry(sfe);
992         }
993         error = 0;
994         *leaf_bp = bp;
995 out:
996         kmem_free(tmpbuffer);
997         return error;
998 }
999
1000 /*
1001  * Check a leaf attribute block to see if all the entries would fit into
1002  * a shortform attribute list.
1003  */
1004 int
1005 xfs_attr_shortform_allfit(
1006         struct xfs_buf          *bp,
1007         struct xfs_inode        *dp)
1008 {
1009         struct xfs_attr_leafblock *leaf;
1010         struct xfs_attr_leaf_entry *entry;
1011         xfs_attr_leaf_name_local_t *name_loc;
1012         struct xfs_attr3_icleaf_hdr leafhdr;
1013         int                     bytes;
1014         int                     i;
1015         struct xfs_mount        *mp = bp->b_mount;
1016
1017         leaf = bp->b_addr;
1018         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
1019         entry = xfs_attr3_leaf_entryp(leaf);
1020
1021         bytes = sizeof(struct xfs_attr_sf_hdr);
1022         for (i = 0; i < leafhdr.count; entry++, i++) {
1023                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1024                         continue;               /* don't copy partial entries */
1025                 if (!(entry->flags & XFS_ATTR_LOCAL))
1026                         return 0;
1027                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1028                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
1029                         return 0;
1030                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
1031                         return 0;
1032                 bytes += xfs_attr_sf_entsize_byname(name_loc->namelen,
1033                                         be16_to_cpu(name_loc->valuelen));
1034         }
1035         if (xfs_has_attr2(dp->i_mount) &&
1036             (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
1037             (bytes == sizeof(struct xfs_attr_sf_hdr)))
1038                 return -1;
1039         return xfs_attr_shortform_bytesfit(dp, bytes);
1040 }
1041
1042 /* Verify the consistency of an inline attribute fork. */
1043 xfs_failaddr_t
1044 xfs_attr_shortform_verify(
1045         struct xfs_inode                *ip)
1046 {
1047         struct xfs_attr_shortform       *sfp;
1048         struct xfs_attr_sf_entry        *sfep;
1049         struct xfs_attr_sf_entry        *next_sfep;
1050         char                            *endp;
1051         struct xfs_ifork                *ifp;
1052         int                             i;
1053         int64_t                         size;
1054
1055         ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
1056         ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
1057         sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
1058         size = ifp->if_bytes;
1059
1060         /*
1061          * Give up if the attribute is way too short.
1062          */
1063         if (size < sizeof(struct xfs_attr_sf_hdr))
1064                 return __this_address;
1065
1066         endp = (char *)sfp + size;
1067
1068         /* Check all reported entries */
1069         sfep = &sfp->list[0];
1070         for (i = 0; i < sfp->hdr.count; i++) {
1071                 /*
1072                  * struct xfs_attr_sf_entry has a variable length.
1073                  * Check the fixed-offset parts of the structure are
1074                  * within the data buffer.
1075                  * xfs_attr_sf_entry is defined with a 1-byte variable
1076                  * array at the end, so we must subtract that off.
1077                  */
1078                 if (((char *)sfep + sizeof(*sfep)) >= endp)
1079                         return __this_address;
1080
1081                 /* Don't allow names with known bad length. */
1082                 if (sfep->namelen == 0)
1083                         return __this_address;
1084
1085                 /*
1086                  * Check that the variable-length part of the structure is
1087                  * within the data buffer.  The next entry starts after the
1088                  * name component, so nextentry is an acceptable test.
1089                  */
1090                 next_sfep = xfs_attr_sf_nextentry(sfep);
1091                 if ((char *)next_sfep > endp)
1092                         return __this_address;
1093
1094                 /*
1095                  * Check for unknown flags.  Short form doesn't support
1096                  * the incomplete or local bits, so we can use the namespace
1097                  * mask here.
1098                  */
1099                 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1100                         return __this_address;
1101
1102                 /*
1103                  * Check for invalid namespace combinations.  We only allow
1104                  * one namespace flag per xattr, so we can just count the
1105                  * bits (i.e. hweight) here.
1106                  */
1107                 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1108                         return __this_address;
1109
1110                 sfep = next_sfep;
1111         }
1112         if ((void *)sfep != (void *)endp)
1113                 return __this_address;
1114
1115         return NULL;
1116 }
1117
1118 /*
1119  * Convert a leaf attribute list to shortform attribute list
1120  */
1121 int
1122 xfs_attr3_leaf_to_shortform(
1123         struct xfs_buf          *bp,
1124         struct xfs_da_args      *args,
1125         int                     forkoff)
1126 {
1127         struct xfs_attr_leafblock *leaf;
1128         struct xfs_attr3_icleaf_hdr ichdr;
1129         struct xfs_attr_leaf_entry *entry;
1130         struct xfs_attr_leaf_name_local *name_loc;
1131         struct xfs_da_args      nargs;
1132         struct xfs_inode        *dp = args->dp;
1133         char                    *tmpbuffer;
1134         int                     error;
1135         int                     i;
1136
1137         trace_xfs_attr_leaf_to_sf(args);
1138
1139         tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1140         if (!tmpbuffer)
1141                 return -ENOMEM;
1142
1143         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1144
1145         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1146         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1147         entry = xfs_attr3_leaf_entryp(leaf);
1148
1149         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1150         memset(bp->b_addr, 0, args->geo->blksize);
1151
1152         /*
1153          * Clean out the prior contents of the attribute list.
1154          */
1155         error = xfs_da_shrink_inode(args, 0, bp);
1156         if (error)
1157                 goto out;
1158
1159         if (forkoff == -1) {
1160                 /*
1161                  * Don't remove the attr fork if this operation is the first
1162                  * part of a attr replace operations. We're going to add a new
1163                  * attr immediately, so we need to keep the attr fork around in
1164                  * this case.
1165                  */
1166                 if (!(args->op_flags & XFS_DA_OP_REPLACE)) {
1167                         ASSERT(xfs_has_attr2(dp->i_mount));
1168                         ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1169                         xfs_attr_fork_remove(dp, args->trans);
1170                 }
1171                 goto out;
1172         }
1173
1174         xfs_attr_shortform_create(args);
1175
1176         /*
1177          * Copy the attributes
1178          */
1179         memset((char *)&nargs, 0, sizeof(nargs));
1180         nargs.geo = args->geo;
1181         nargs.dp = dp;
1182         nargs.total = args->total;
1183         nargs.whichfork = XFS_ATTR_FORK;
1184         nargs.trans = args->trans;
1185         nargs.op_flags = XFS_DA_OP_OKNOENT;
1186
1187         for (i = 0; i < ichdr.count; entry++, i++) {
1188                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1189                         continue;       /* don't copy partial entries */
1190                 if (!entry->nameidx)
1191                         continue;
1192                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1193                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1194                 nargs.name = name_loc->nameval;
1195                 nargs.namelen = name_loc->namelen;
1196                 nargs.value = &name_loc->nameval[nargs.namelen];
1197                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1198                 nargs.hashval = be32_to_cpu(entry->hashval);
1199                 nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1200                 xfs_attr_shortform_add(&nargs, forkoff);
1201         }
1202         error = 0;
1203
1204 out:
1205         kmem_free(tmpbuffer);
1206         return error;
1207 }
1208
1209 /*
1210  * Convert from using a single leaf to a root node and a leaf.
1211  */
1212 int
1213 xfs_attr3_leaf_to_node(
1214         struct xfs_da_args      *args)
1215 {
1216         struct xfs_attr_leafblock *leaf;
1217         struct xfs_attr3_icleaf_hdr icleafhdr;
1218         struct xfs_attr_leaf_entry *entries;
1219         struct xfs_da3_icnode_hdr icnodehdr;
1220         struct xfs_da_intnode   *node;
1221         struct xfs_inode        *dp = args->dp;
1222         struct xfs_mount        *mp = dp->i_mount;
1223         struct xfs_buf          *bp1 = NULL;
1224         struct xfs_buf          *bp2 = NULL;
1225         xfs_dablk_t             blkno;
1226         int                     error;
1227
1228         trace_xfs_attr_leaf_to_node(args);
1229
1230         if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_ATTR_LEAF_TO_NODE)) {
1231                 error = -EIO;
1232                 goto out;
1233         }
1234
1235         error = xfs_da_grow_inode(args, &blkno);
1236         if (error)
1237                 goto out;
1238         error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1239         if (error)
1240                 goto out;
1241
1242         error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1243         if (error)
1244                 goto out;
1245
1246         /* copy leaf to new buffer, update identifiers */
1247         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1248         bp2->b_ops = bp1->b_ops;
1249         memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1250         if (xfs_has_crc(mp)) {
1251                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1252                 hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp2));
1253         }
1254         xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1255
1256         /*
1257          * Set up the new root node.
1258          */
1259         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1260         if (error)
1261                 goto out;
1262         node = bp1->b_addr;
1263         xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1264
1265         leaf = bp2->b_addr;
1266         xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1267         entries = xfs_attr3_leaf_entryp(leaf);
1268
1269         /* both on-disk, don't endian-flip twice */
1270         icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1271         icnodehdr.btree[0].before = cpu_to_be32(blkno);
1272         icnodehdr.count = 1;
1273         xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1274         xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1275         error = 0;
1276 out:
1277         return error;
1278 }
1279
1280 /*========================================================================
1281  * Routines used for growing the Btree.
1282  *========================================================================*/
1283
1284 /*
1285  * Create the initial contents of a leaf attribute list
1286  * or a leaf in a node attribute list.
1287  */
1288 STATIC int
1289 xfs_attr3_leaf_create(
1290         struct xfs_da_args      *args,
1291         xfs_dablk_t             blkno,
1292         struct xfs_buf          **bpp)
1293 {
1294         struct xfs_attr_leafblock *leaf;
1295         struct xfs_attr3_icleaf_hdr ichdr;
1296         struct xfs_inode        *dp = args->dp;
1297         struct xfs_mount        *mp = dp->i_mount;
1298         struct xfs_buf          *bp;
1299         int                     error;
1300
1301         trace_xfs_attr_leaf_create(args);
1302
1303         error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1304                                             XFS_ATTR_FORK);
1305         if (error)
1306                 return error;
1307         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1308         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1309         leaf = bp->b_addr;
1310         memset(leaf, 0, args->geo->blksize);
1311
1312         memset(&ichdr, 0, sizeof(ichdr));
1313         ichdr.firstused = args->geo->blksize;
1314
1315         if (xfs_has_crc(mp)) {
1316                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1317
1318                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1319
1320                 hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
1321                 hdr3->owner = cpu_to_be64(dp->i_ino);
1322                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1323
1324                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1325         } else {
1326                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1327                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1328         }
1329         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1330
1331         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1332         xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1333
1334         *bpp = bp;
1335         return 0;
1336 }
1337
1338 /*
1339  * Split the leaf node, rebalance, then add the new entry.
1340  */
1341 int
1342 xfs_attr3_leaf_split(
1343         struct xfs_da_state     *state,
1344         struct xfs_da_state_blk *oldblk,
1345         struct xfs_da_state_blk *newblk)
1346 {
1347         xfs_dablk_t blkno;
1348         int error;
1349
1350         trace_xfs_attr_leaf_split(state->args);
1351
1352         /*
1353          * Allocate space for a new leaf node.
1354          */
1355         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1356         error = xfs_da_grow_inode(state->args, &blkno);
1357         if (error)
1358                 return error;
1359         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1360         if (error)
1361                 return error;
1362         newblk->blkno = blkno;
1363         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1364
1365         /*
1366          * Rebalance the entries across the two leaves.
1367          * NOTE: rebalance() currently depends on the 2nd block being empty.
1368          */
1369         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1370         error = xfs_da3_blk_link(state, oldblk, newblk);
1371         if (error)
1372                 return error;
1373
1374         /*
1375          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1376          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1377          * "new" attrs info.  Will need the "old" info to remove it later.
1378          *
1379          * Insert the "new" entry in the correct block.
1380          */
1381         if (state->inleaf) {
1382                 trace_xfs_attr_leaf_add_old(state->args);
1383                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1384         } else {
1385                 trace_xfs_attr_leaf_add_new(state->args);
1386                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1387         }
1388
1389         /*
1390          * Update last hashval in each block since we added the name.
1391          */
1392         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1393         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1394         return error;
1395 }
1396
1397 /*
1398  * Add a name to the leaf attribute list structure.
1399  */
1400 int
1401 xfs_attr3_leaf_add(
1402         struct xfs_buf          *bp,
1403         struct xfs_da_args      *args)
1404 {
1405         struct xfs_attr_leafblock *leaf;
1406         struct xfs_attr3_icleaf_hdr ichdr;
1407         int                     tablesize;
1408         int                     entsize;
1409         int                     sum;
1410         int                     tmp;
1411         int                     i;
1412
1413         trace_xfs_attr_leaf_add(args);
1414
1415         leaf = bp->b_addr;
1416         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1417         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1418         entsize = xfs_attr_leaf_newentsize(args, NULL);
1419
1420         /*
1421          * Search through freemap for first-fit on new name length.
1422          * (may need to figure in size of entry struct too)
1423          */
1424         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1425                                         + xfs_attr3_leaf_hdr_size(leaf);
1426         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1427                 if (tablesize > ichdr.firstused) {
1428                         sum += ichdr.freemap[i].size;
1429                         continue;
1430                 }
1431                 if (!ichdr.freemap[i].size)
1432                         continue;       /* no space in this map */
1433                 tmp = entsize;
1434                 if (ichdr.freemap[i].base < ichdr.firstused)
1435                         tmp += sizeof(xfs_attr_leaf_entry_t);
1436                 if (ichdr.freemap[i].size >= tmp) {
1437                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1438                         goto out_log_hdr;
1439                 }
1440                 sum += ichdr.freemap[i].size;
1441         }
1442
1443         /*
1444          * If there are no holes in the address space of the block,
1445          * and we don't have enough freespace, then compaction will do us
1446          * no good and we should just give up.
1447          */
1448         if (!ichdr.holes && sum < entsize)
1449                 return -ENOSPC;
1450
1451         /*
1452          * Compact the entries to coalesce free space.
1453          * This may change the hdr->count via dropping INCOMPLETE entries.
1454          */
1455         xfs_attr3_leaf_compact(args, &ichdr, bp);
1456
1457         /*
1458          * After compaction, the block is guaranteed to have only one
1459          * free region, in freemap[0].  If it is not big enough, give up.
1460          */
1461         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1462                 tmp = -ENOSPC;
1463                 goto out_log_hdr;
1464         }
1465
1466         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1467
1468 out_log_hdr:
1469         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1470         xfs_trans_log_buf(args->trans, bp,
1471                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1472                                 xfs_attr3_leaf_hdr_size(leaf)));
1473         return tmp;
1474 }
1475
1476 /*
1477  * Add a name to a leaf attribute list structure.
1478  */
1479 STATIC int
1480 xfs_attr3_leaf_add_work(
1481         struct xfs_buf          *bp,
1482         struct xfs_attr3_icleaf_hdr *ichdr,
1483         struct xfs_da_args      *args,
1484         int                     mapindex)
1485 {
1486         struct xfs_attr_leafblock *leaf;
1487         struct xfs_attr_leaf_entry *entry;
1488         struct xfs_attr_leaf_name_local *name_loc;
1489         struct xfs_attr_leaf_name_remote *name_rmt;
1490         struct xfs_mount        *mp;
1491         int                     tmp;
1492         int                     i;
1493
1494         trace_xfs_attr_leaf_add_work(args);
1495
1496         leaf = bp->b_addr;
1497         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1498         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1499
1500         /*
1501          * Force open some space in the entry array and fill it in.
1502          */
1503         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1504         if (args->index < ichdr->count) {
1505                 tmp  = ichdr->count - args->index;
1506                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1507                 memmove(entry + 1, entry, tmp);
1508                 xfs_trans_log_buf(args->trans, bp,
1509                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1510         }
1511         ichdr->count++;
1512
1513         /*
1514          * Allocate space for the new string (at the end of the run).
1515          */
1516         mp = args->trans->t_mountp;
1517         ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1518         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1519         ASSERT(ichdr->freemap[mapindex].size >=
1520                 xfs_attr_leaf_newentsize(args, NULL));
1521         ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1522         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1523
1524         ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1525
1526         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1527                                      ichdr->freemap[mapindex].size);
1528         entry->hashval = cpu_to_be32(args->hashval);
1529         entry->flags = args->attr_filter;
1530         if (tmp)
1531                 entry->flags |= XFS_ATTR_LOCAL;
1532         if (args->op_flags & XFS_DA_OP_REPLACE) {
1533                 if (!xfs_has_larp(mp))
1534                         entry->flags |= XFS_ATTR_INCOMPLETE;
1535                 if ((args->blkno2 == args->blkno) &&
1536                     (args->index2 <= args->index)) {
1537                         args->index2++;
1538                 }
1539         }
1540         xfs_trans_log_buf(args->trans, bp,
1541                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1542         ASSERT((args->index == 0) ||
1543                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1544         ASSERT((args->index == ichdr->count - 1) ||
1545                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1546
1547         /*
1548          * For "remote" attribute values, simply note that we need to
1549          * allocate space for the "remote" value.  We can't actually
1550          * allocate the extents in this transaction, and we can't decide
1551          * which blocks they should be as we might allocate more blocks
1552          * as part of this transaction (a split operation for example).
1553          */
1554         if (entry->flags & XFS_ATTR_LOCAL) {
1555                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1556                 name_loc->namelen = args->namelen;
1557                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1558                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1559                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1560                                    be16_to_cpu(name_loc->valuelen));
1561         } else {
1562                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1563                 name_rmt->namelen = args->namelen;
1564                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1565                 entry->flags |= XFS_ATTR_INCOMPLETE;
1566                 /* just in case */
1567                 name_rmt->valuelen = 0;
1568                 name_rmt->valueblk = 0;
1569                 args->rmtblkno = 1;
1570                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1571                 args->rmtvaluelen = args->valuelen;
1572         }
1573         xfs_trans_log_buf(args->trans, bp,
1574              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1575                                    xfs_attr_leaf_entsize(leaf, args->index)));
1576
1577         /*
1578          * Update the control info for this leaf node
1579          */
1580         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1581                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1582
1583         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1584                                         + xfs_attr3_leaf_hdr_size(leaf));
1585         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1586                                         + xfs_attr3_leaf_hdr_size(leaf);
1587
1588         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1589                 if (ichdr->freemap[i].base == tmp) {
1590                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1591                         ichdr->freemap[i].size -=
1592                                 min_t(uint16_t, ichdr->freemap[i].size,
1593                                                 sizeof(xfs_attr_leaf_entry_t));
1594                 }
1595         }
1596         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1597         return 0;
1598 }
1599
1600 /*
1601  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1602  */
1603 STATIC void
1604 xfs_attr3_leaf_compact(
1605         struct xfs_da_args      *args,
1606         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1607         struct xfs_buf          *bp)
1608 {
1609         struct xfs_attr_leafblock *leaf_src;
1610         struct xfs_attr_leafblock *leaf_dst;
1611         struct xfs_attr3_icleaf_hdr ichdr_src;
1612         struct xfs_trans        *trans = args->trans;
1613         char                    *tmpbuffer;
1614
1615         trace_xfs_attr_leaf_compact(args);
1616
1617         tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1618         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1619         memset(bp->b_addr, 0, args->geo->blksize);
1620         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1621         leaf_dst = bp->b_addr;
1622
1623         /*
1624          * Copy the on-disk header back into the destination buffer to ensure
1625          * all the information in the header that is not part of the incore
1626          * header structure is preserved.
1627          */
1628         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1629
1630         /* Initialise the incore headers */
1631         ichdr_src = *ichdr_dst; /* struct copy */
1632         ichdr_dst->firstused = args->geo->blksize;
1633         ichdr_dst->usedbytes = 0;
1634         ichdr_dst->count = 0;
1635         ichdr_dst->holes = 0;
1636         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1637         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1638                                                 ichdr_dst->freemap[0].base;
1639
1640         /* write the header back to initialise the underlying buffer */
1641         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1642
1643         /*
1644          * Copy all entry's in the same (sorted) order,
1645          * but allocate name/value pairs packed and in sequence.
1646          */
1647         xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1648                                 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1649         /*
1650          * this logs the entire buffer, but the caller must write the header
1651          * back to the buffer when it is finished modifying it.
1652          */
1653         xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1654
1655         kmem_free(tmpbuffer);
1656 }
1657
1658 /*
1659  * Compare two leaf blocks "order".
1660  * Return 0 unless leaf2 should go before leaf1.
1661  */
1662 static int
1663 xfs_attr3_leaf_order(
1664         struct xfs_buf  *leaf1_bp,
1665         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1666         struct xfs_buf  *leaf2_bp,
1667         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1668 {
1669         struct xfs_attr_leaf_entry *entries1;
1670         struct xfs_attr_leaf_entry *entries2;
1671
1672         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1673         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1674         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1675             ((be32_to_cpu(entries2[0].hashval) <
1676               be32_to_cpu(entries1[0].hashval)) ||
1677              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1678               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1679                 return 1;
1680         }
1681         return 0;
1682 }
1683
1684 int
1685 xfs_attr_leaf_order(
1686         struct xfs_buf  *leaf1_bp,
1687         struct xfs_buf  *leaf2_bp)
1688 {
1689         struct xfs_attr3_icleaf_hdr ichdr1;
1690         struct xfs_attr3_icleaf_hdr ichdr2;
1691         struct xfs_mount *mp = leaf1_bp->b_mount;
1692
1693         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1694         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1695         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1696 }
1697
1698 /*
1699  * Redistribute the attribute list entries between two leaf nodes,
1700  * taking into account the size of the new entry.
1701  *
1702  * NOTE: if new block is empty, then it will get the upper half of the
1703  * old block.  At present, all (one) callers pass in an empty second block.
1704  *
1705  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1706  * to match what it is doing in splitting the attribute leaf block.  Those
1707  * values are used in "atomic rename" operations on attributes.  Note that
1708  * the "new" and "old" values can end up in different blocks.
1709  */
1710 STATIC void
1711 xfs_attr3_leaf_rebalance(
1712         struct xfs_da_state     *state,
1713         struct xfs_da_state_blk *blk1,
1714         struct xfs_da_state_blk *blk2)
1715 {
1716         struct xfs_da_args      *args;
1717         struct xfs_attr_leafblock *leaf1;
1718         struct xfs_attr_leafblock *leaf2;
1719         struct xfs_attr3_icleaf_hdr ichdr1;
1720         struct xfs_attr3_icleaf_hdr ichdr2;
1721         struct xfs_attr_leaf_entry *entries1;
1722         struct xfs_attr_leaf_entry *entries2;
1723         int                     count;
1724         int                     totallen;
1725         int                     max;
1726         int                     space;
1727         int                     swap;
1728
1729         /*
1730          * Set up environment.
1731          */
1732         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1733         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1734         leaf1 = blk1->bp->b_addr;
1735         leaf2 = blk2->bp->b_addr;
1736         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1737         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1738         ASSERT(ichdr2.count == 0);
1739         args = state->args;
1740
1741         trace_xfs_attr_leaf_rebalance(args);
1742
1743         /*
1744          * Check ordering of blocks, reverse if it makes things simpler.
1745          *
1746          * NOTE: Given that all (current) callers pass in an empty
1747          * second block, this code should never set "swap".
1748          */
1749         swap = 0;
1750         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1751                 swap(blk1, blk2);
1752
1753                 /* swap structures rather than reconverting them */
1754                 swap(ichdr1, ichdr2);
1755
1756                 leaf1 = blk1->bp->b_addr;
1757                 leaf2 = blk2->bp->b_addr;
1758                 swap = 1;
1759         }
1760
1761         /*
1762          * Examine entries until we reduce the absolute difference in
1763          * byte usage between the two blocks to a minimum.  Then get
1764          * the direction to copy and the number of elements to move.
1765          *
1766          * "inleaf" is true if the new entry should be inserted into blk1.
1767          * If "swap" is also true, then reverse the sense of "inleaf".
1768          */
1769         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1770                                                       blk2, &ichdr2,
1771                                                       &count, &totallen);
1772         if (swap)
1773                 state->inleaf = !state->inleaf;
1774
1775         /*
1776          * Move any entries required from leaf to leaf:
1777          */
1778         if (count < ichdr1.count) {
1779                 /*
1780                  * Figure the total bytes to be added to the destination leaf.
1781                  */
1782                 /* number entries being moved */
1783                 count = ichdr1.count - count;
1784                 space  = ichdr1.usedbytes - totallen;
1785                 space += count * sizeof(xfs_attr_leaf_entry_t);
1786
1787                 /*
1788                  * leaf2 is the destination, compact it if it looks tight.
1789                  */
1790                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1791                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1792                 if (space > max)
1793                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1794
1795                 /*
1796                  * Move high entries from leaf1 to low end of leaf2.
1797                  */
1798                 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1799                                 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1800
1801         } else if (count > ichdr1.count) {
1802                 /*
1803                  * I assert that since all callers pass in an empty
1804                  * second buffer, this code should never execute.
1805                  */
1806                 ASSERT(0);
1807
1808                 /*
1809                  * Figure the total bytes to be added to the destination leaf.
1810                  */
1811                 /* number entries being moved */
1812                 count -= ichdr1.count;
1813                 space  = totallen - ichdr1.usedbytes;
1814                 space += count * sizeof(xfs_attr_leaf_entry_t);
1815
1816                 /*
1817                  * leaf1 is the destination, compact it if it looks tight.
1818                  */
1819                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1820                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1821                 if (space > max)
1822                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1823
1824                 /*
1825                  * Move low entries from leaf2 to high end of leaf1.
1826                  */
1827                 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1828                                         ichdr1.count, count);
1829         }
1830
1831         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1832         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1833         xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1834         xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1835
1836         /*
1837          * Copy out last hashval in each block for B-tree code.
1838          */
1839         entries1 = xfs_attr3_leaf_entryp(leaf1);
1840         entries2 = xfs_attr3_leaf_entryp(leaf2);
1841         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1842         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1843
1844         /*
1845          * Adjust the expected index for insertion.
1846          * NOTE: this code depends on the (current) situation that the
1847          * second block was originally empty.
1848          *
1849          * If the insertion point moved to the 2nd block, we must adjust
1850          * the index.  We must also track the entry just following the
1851          * new entry for use in an "atomic rename" operation, that entry
1852          * is always the "old" entry and the "new" entry is what we are
1853          * inserting.  The index/blkno fields refer to the "old" entry,
1854          * while the index2/blkno2 fields refer to the "new" entry.
1855          */
1856         if (blk1->index > ichdr1.count) {
1857                 ASSERT(state->inleaf == 0);
1858                 blk2->index = blk1->index - ichdr1.count;
1859                 args->index = args->index2 = blk2->index;
1860                 args->blkno = args->blkno2 = blk2->blkno;
1861         } else if (blk1->index == ichdr1.count) {
1862                 if (state->inleaf) {
1863                         args->index = blk1->index;
1864                         args->blkno = blk1->blkno;
1865                         args->index2 = 0;
1866                         args->blkno2 = blk2->blkno;
1867                 } else {
1868                         /*
1869                          * On a double leaf split, the original attr location
1870                          * is already stored in blkno2/index2, so don't
1871                          * overwrite it overwise we corrupt the tree.
1872                          */
1873                         blk2->index = blk1->index - ichdr1.count;
1874                         args->index = blk2->index;
1875                         args->blkno = blk2->blkno;
1876                         if (!state->extravalid) {
1877                                 /*
1878                                  * set the new attr location to match the old
1879                                  * one and let the higher level split code
1880                                  * decide where in the leaf to place it.
1881                                  */
1882                                 args->index2 = blk2->index;
1883                                 args->blkno2 = blk2->blkno;
1884                         }
1885                 }
1886         } else {
1887                 ASSERT(state->inleaf == 1);
1888                 args->index = args->index2 = blk1->index;
1889                 args->blkno = args->blkno2 = blk1->blkno;
1890         }
1891 }
1892
1893 /*
1894  * Examine entries until we reduce the absolute difference in
1895  * byte usage between the two blocks to a minimum.
1896  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1897  * GROT: there will always be enough room in either block for a new entry.
1898  * GROT: Do a double-split for this case?
1899  */
1900 STATIC int
1901 xfs_attr3_leaf_figure_balance(
1902         struct xfs_da_state             *state,
1903         struct xfs_da_state_blk         *blk1,
1904         struct xfs_attr3_icleaf_hdr     *ichdr1,
1905         struct xfs_da_state_blk         *blk2,
1906         struct xfs_attr3_icleaf_hdr     *ichdr2,
1907         int                             *countarg,
1908         int                             *usedbytesarg)
1909 {
1910         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1911         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1912         struct xfs_attr_leaf_entry      *entry;
1913         int                             count;
1914         int                             max;
1915         int                             index;
1916         int                             totallen = 0;
1917         int                             half;
1918         int                             lastdelta;
1919         int                             foundit = 0;
1920         int                             tmp;
1921
1922         /*
1923          * Examine entries until we reduce the absolute difference in
1924          * byte usage between the two blocks to a minimum.
1925          */
1926         max = ichdr1->count + ichdr2->count;
1927         half = (max + 1) * sizeof(*entry);
1928         half += ichdr1->usedbytes + ichdr2->usedbytes +
1929                         xfs_attr_leaf_newentsize(state->args, NULL);
1930         half /= 2;
1931         lastdelta = state->args->geo->blksize;
1932         entry = xfs_attr3_leaf_entryp(leaf1);
1933         for (count = index = 0; count < max; entry++, index++, count++) {
1934
1935 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1936                 /*
1937                  * The new entry is in the first block, account for it.
1938                  */
1939                 if (count == blk1->index) {
1940                         tmp = totallen + sizeof(*entry) +
1941                                 xfs_attr_leaf_newentsize(state->args, NULL);
1942                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1943                                 break;
1944                         lastdelta = XFS_ATTR_ABS(half - tmp);
1945                         totallen = tmp;
1946                         foundit = 1;
1947                 }
1948
1949                 /*
1950                  * Wrap around into the second block if necessary.
1951                  */
1952                 if (count == ichdr1->count) {
1953                         leaf1 = leaf2;
1954                         entry = xfs_attr3_leaf_entryp(leaf1);
1955                         index = 0;
1956                 }
1957
1958                 /*
1959                  * Figure out if next leaf entry would be too much.
1960                  */
1961                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1962                                                                         index);
1963                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1964                         break;
1965                 lastdelta = XFS_ATTR_ABS(half - tmp);
1966                 totallen = tmp;
1967 #undef XFS_ATTR_ABS
1968         }
1969
1970         /*
1971          * Calculate the number of usedbytes that will end up in lower block.
1972          * If new entry not in lower block, fix up the count.
1973          */
1974         totallen -= count * sizeof(*entry);
1975         if (foundit) {
1976                 totallen -= sizeof(*entry) +
1977                                 xfs_attr_leaf_newentsize(state->args, NULL);
1978         }
1979
1980         *countarg = count;
1981         *usedbytesarg = totallen;
1982         return foundit;
1983 }
1984
1985 /*========================================================================
1986  * Routines used for shrinking the Btree.
1987  *========================================================================*/
1988
1989 /*
1990  * Check a leaf block and its neighbors to see if the block should be
1991  * collapsed into one or the other neighbor.  Always keep the block
1992  * with the smaller block number.
1993  * If the current block is over 50% full, don't try to join it, return 0.
1994  * If the block is empty, fill in the state structure and return 2.
1995  * If it can be collapsed, fill in the state structure and return 1.
1996  * If nothing can be done, return 0.
1997  *
1998  * GROT: allow for INCOMPLETE entries in calculation.
1999  */
2000 int
2001 xfs_attr3_leaf_toosmall(
2002         struct xfs_da_state     *state,
2003         int                     *action)
2004 {
2005         struct xfs_attr_leafblock *leaf;
2006         struct xfs_da_state_blk *blk;
2007         struct xfs_attr3_icleaf_hdr ichdr;
2008         struct xfs_buf          *bp;
2009         xfs_dablk_t             blkno;
2010         int                     bytes;
2011         int                     forward;
2012         int                     error;
2013         int                     retval;
2014         int                     i;
2015
2016         trace_xfs_attr_leaf_toosmall(state->args);
2017
2018         /*
2019          * Check for the degenerate case of the block being over 50% full.
2020          * If so, it's not worth even looking to see if we might be able
2021          * to coalesce with a sibling.
2022          */
2023         blk = &state->path.blk[ state->path.active-1 ];
2024         leaf = blk->bp->b_addr;
2025         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
2026         bytes = xfs_attr3_leaf_hdr_size(leaf) +
2027                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
2028                 ichdr.usedbytes;
2029         if (bytes > (state->args->geo->blksize >> 1)) {
2030                 *action = 0;    /* blk over 50%, don't try to join */
2031                 return 0;
2032         }
2033
2034         /*
2035          * Check for the degenerate case of the block being empty.
2036          * If the block is empty, we'll simply delete it, no need to
2037          * coalesce it with a sibling block.  We choose (arbitrarily)
2038          * to merge with the forward block unless it is NULL.
2039          */
2040         if (ichdr.count == 0) {
2041                 /*
2042                  * Make altpath point to the block we want to keep and
2043                  * path point to the block we want to drop (this one).
2044                  */
2045                 forward = (ichdr.forw != 0);
2046                 memcpy(&state->altpath, &state->path, sizeof(state->path));
2047                 error = xfs_da3_path_shift(state, &state->altpath, forward,
2048                                                  0, &retval);
2049                 if (error)
2050                         return error;
2051                 if (retval) {
2052                         *action = 0;
2053                 } else {
2054                         *action = 2;
2055                 }
2056                 return 0;
2057         }
2058
2059         /*
2060          * Examine each sibling block to see if we can coalesce with
2061          * at least 25% free space to spare.  We need to figure out
2062          * whether to merge with the forward or the backward block.
2063          * We prefer coalescing with the lower numbered sibling so as
2064          * to shrink an attribute list over time.
2065          */
2066         /* start with smaller blk num */
2067         forward = ichdr.forw < ichdr.back;
2068         for (i = 0; i < 2; forward = !forward, i++) {
2069                 struct xfs_attr3_icleaf_hdr ichdr2;
2070                 if (forward)
2071                         blkno = ichdr.forw;
2072                 else
2073                         blkno = ichdr.back;
2074                 if (blkno == 0)
2075                         continue;
2076                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
2077                                         blkno, &bp);
2078                 if (error)
2079                         return error;
2080
2081                 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2082
2083                 bytes = state->args->geo->blksize -
2084                         (state->args->geo->blksize >> 2) -
2085                         ichdr.usedbytes - ichdr2.usedbytes -
2086                         ((ichdr.count + ichdr2.count) *
2087                                         sizeof(xfs_attr_leaf_entry_t)) -
2088                         xfs_attr3_leaf_hdr_size(leaf);
2089
2090                 xfs_trans_brelse(state->args->trans, bp);
2091                 if (bytes >= 0)
2092                         break;  /* fits with at least 25% to spare */
2093         }
2094         if (i >= 2) {
2095                 *action = 0;
2096                 return 0;
2097         }
2098
2099         /*
2100          * Make altpath point to the block we want to keep (the lower
2101          * numbered block) and path point to the block we want to drop.
2102          */
2103         memcpy(&state->altpath, &state->path, sizeof(state->path));
2104         if (blkno < blk->blkno) {
2105                 error = xfs_da3_path_shift(state, &state->altpath, forward,
2106                                                  0, &retval);
2107         } else {
2108                 error = xfs_da3_path_shift(state, &state->path, forward,
2109                                                  0, &retval);
2110         }
2111         if (error)
2112                 return error;
2113         if (retval) {
2114                 *action = 0;
2115         } else {
2116                 *action = 1;
2117         }
2118         return 0;
2119 }
2120
2121 /*
2122  * Remove a name from the leaf attribute list structure.
2123  *
2124  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2125  * If two leaves are 37% full, when combined they will leave 25% free.
2126  */
2127 int
2128 xfs_attr3_leaf_remove(
2129         struct xfs_buf          *bp,
2130         struct xfs_da_args      *args)
2131 {
2132         struct xfs_attr_leafblock *leaf;
2133         struct xfs_attr3_icleaf_hdr ichdr;
2134         struct xfs_attr_leaf_entry *entry;
2135         int                     before;
2136         int                     after;
2137         int                     smallest;
2138         int                     entsize;
2139         int                     tablesize;
2140         int                     tmp;
2141         int                     i;
2142
2143         trace_xfs_attr_leaf_remove(args);
2144
2145         leaf = bp->b_addr;
2146         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2147
2148         ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2149         ASSERT(args->index >= 0 && args->index < ichdr.count);
2150         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2151                                         xfs_attr3_leaf_hdr_size(leaf));
2152
2153         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2154
2155         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2156         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2157
2158         /*
2159          * Scan through free region table:
2160          *    check for adjacency of free'd entry with an existing one,
2161          *    find smallest free region in case we need to replace it,
2162          *    adjust any map that borders the entry table,
2163          */
2164         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2165                                         + xfs_attr3_leaf_hdr_size(leaf);
2166         tmp = ichdr.freemap[0].size;
2167         before = after = -1;
2168         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2169         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2170         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2171                 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2172                 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2173                 if (ichdr.freemap[i].base == tablesize) {
2174                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2175                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2176                 }
2177
2178                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2179                                 be16_to_cpu(entry->nameidx)) {
2180                         before = i;
2181                 } else if (ichdr.freemap[i].base ==
2182                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2183                         after = i;
2184                 } else if (ichdr.freemap[i].size < tmp) {
2185                         tmp = ichdr.freemap[i].size;
2186                         smallest = i;
2187                 }
2188         }
2189
2190         /*
2191          * Coalesce adjacent freemap regions,
2192          * or replace the smallest region.
2193          */
2194         if ((before >= 0) || (after >= 0)) {
2195                 if ((before >= 0) && (after >= 0)) {
2196                         ichdr.freemap[before].size += entsize;
2197                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2198                         ichdr.freemap[after].base = 0;
2199                         ichdr.freemap[after].size = 0;
2200                 } else if (before >= 0) {
2201                         ichdr.freemap[before].size += entsize;
2202                 } else {
2203                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2204                         ichdr.freemap[after].size += entsize;
2205                 }
2206         } else {
2207                 /*
2208                  * Replace smallest region (if it is smaller than free'd entry)
2209                  */
2210                 if (ichdr.freemap[smallest].size < entsize) {
2211                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2212                         ichdr.freemap[smallest].size = entsize;
2213                 }
2214         }
2215
2216         /*
2217          * Did we remove the first entry?
2218          */
2219         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2220                 smallest = 1;
2221         else
2222                 smallest = 0;
2223
2224         /*
2225          * Compress the remaining entries and zero out the removed stuff.
2226          */
2227         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2228         ichdr.usedbytes -= entsize;
2229         xfs_trans_log_buf(args->trans, bp,
2230              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2231                                    entsize));
2232
2233         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2234         memmove(entry, entry + 1, tmp);
2235         ichdr.count--;
2236         xfs_trans_log_buf(args->trans, bp,
2237             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2238
2239         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2240         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2241
2242         /*
2243          * If we removed the first entry, re-find the first used byte
2244          * in the name area.  Note that if the entry was the "firstused",
2245          * then we don't have a "hole" in our block resulting from
2246          * removing the name.
2247          */
2248         if (smallest) {
2249                 tmp = args->geo->blksize;
2250                 entry = xfs_attr3_leaf_entryp(leaf);
2251                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2252                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2253                         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2254
2255                         if (be16_to_cpu(entry->nameidx) < tmp)
2256                                 tmp = be16_to_cpu(entry->nameidx);
2257                 }
2258                 ichdr.firstused = tmp;
2259                 ASSERT(ichdr.firstused != 0);
2260         } else {
2261                 ichdr.holes = 1;        /* mark as needing compaction */
2262         }
2263         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2264         xfs_trans_log_buf(args->trans, bp,
2265                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2266                                           xfs_attr3_leaf_hdr_size(leaf)));
2267
2268         /*
2269          * Check if leaf is less than 50% full, caller may want to
2270          * "join" the leaf with a sibling if so.
2271          */
2272         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2273               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2274
2275         return tmp < args->geo->magicpct; /* leaf is < 37% full */
2276 }
2277
2278 /*
2279  * Move all the attribute list entries from drop_leaf into save_leaf.
2280  */
2281 void
2282 xfs_attr3_leaf_unbalance(
2283         struct xfs_da_state     *state,
2284         struct xfs_da_state_blk *drop_blk,
2285         struct xfs_da_state_blk *save_blk)
2286 {
2287         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2288         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2289         struct xfs_attr3_icleaf_hdr drophdr;
2290         struct xfs_attr3_icleaf_hdr savehdr;
2291         struct xfs_attr_leaf_entry *entry;
2292
2293         trace_xfs_attr_leaf_unbalance(state->args);
2294
2295         drop_leaf = drop_blk->bp->b_addr;
2296         save_leaf = save_blk->bp->b_addr;
2297         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2298         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2299         entry = xfs_attr3_leaf_entryp(drop_leaf);
2300
2301         /*
2302          * Save last hashval from dying block for later Btree fixup.
2303          */
2304         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2305
2306         /*
2307          * Check if we need a temp buffer, or can we do it in place.
2308          * Note that we don't check "leaf" for holes because we will
2309          * always be dropping it, toosmall() decided that for us already.
2310          */
2311         if (savehdr.holes == 0) {
2312                 /*
2313                  * dest leaf has no holes, so we add there.  May need
2314                  * to make some room in the entry array.
2315                  */
2316                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2317                                          drop_blk->bp, &drophdr)) {
2318                         xfs_attr3_leaf_moveents(state->args,
2319                                                 drop_leaf, &drophdr, 0,
2320                                                 save_leaf, &savehdr, 0,
2321                                                 drophdr.count);
2322                 } else {
2323                         xfs_attr3_leaf_moveents(state->args,
2324                                                 drop_leaf, &drophdr, 0,
2325                                                 save_leaf, &savehdr,
2326                                                 savehdr.count, drophdr.count);
2327                 }
2328         } else {
2329                 /*
2330                  * Destination has holes, so we make a temporary copy
2331                  * of the leaf and add them both to that.
2332                  */
2333                 struct xfs_attr_leafblock *tmp_leaf;
2334                 struct xfs_attr3_icleaf_hdr tmphdr;
2335
2336                 tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2337
2338                 /*
2339                  * Copy the header into the temp leaf so that all the stuff
2340                  * not in the incore header is present and gets copied back in
2341                  * once we've moved all the entries.
2342                  */
2343                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2344
2345                 memset(&tmphdr, 0, sizeof(tmphdr));
2346                 tmphdr.magic = savehdr.magic;
2347                 tmphdr.forw = savehdr.forw;
2348                 tmphdr.back = savehdr.back;
2349                 tmphdr.firstused = state->args->geo->blksize;
2350
2351                 /* write the header to the temp buffer to initialise it */
2352                 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2353
2354                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2355                                          drop_blk->bp, &drophdr)) {
2356                         xfs_attr3_leaf_moveents(state->args,
2357                                                 drop_leaf, &drophdr, 0,
2358                                                 tmp_leaf, &tmphdr, 0,
2359                                                 drophdr.count);
2360                         xfs_attr3_leaf_moveents(state->args,
2361                                                 save_leaf, &savehdr, 0,
2362                                                 tmp_leaf, &tmphdr, tmphdr.count,
2363                                                 savehdr.count);
2364                 } else {
2365                         xfs_attr3_leaf_moveents(state->args,
2366                                                 save_leaf, &savehdr, 0,
2367                                                 tmp_leaf, &tmphdr, 0,
2368                                                 savehdr.count);
2369                         xfs_attr3_leaf_moveents(state->args,
2370                                                 drop_leaf, &drophdr, 0,
2371                                                 tmp_leaf, &tmphdr, tmphdr.count,
2372                                                 drophdr.count);
2373                 }
2374                 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2375                 savehdr = tmphdr; /* struct copy */
2376                 kmem_free(tmp_leaf);
2377         }
2378
2379         xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2380         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2381                                            state->args->geo->blksize - 1);
2382
2383         /*
2384          * Copy out last hashval in each block for B-tree code.
2385          */
2386         entry = xfs_attr3_leaf_entryp(save_leaf);
2387         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2388 }
2389
2390 /*========================================================================
2391  * Routines used for finding things in the Btree.
2392  *========================================================================*/
2393
2394 /*
2395  * Look up a name in a leaf attribute list structure.
2396  * This is the internal routine, it uses the caller's buffer.
2397  *
2398  * Note that duplicate keys are allowed, but only check within the
2399  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2400  *
2401  * Return in args->index the index into the entry[] array of either
2402  * the found entry, or where the entry should have been (insert before
2403  * that entry).
2404  *
2405  * Don't change the args->value unless we find the attribute.
2406  */
2407 int
2408 xfs_attr3_leaf_lookup_int(
2409         struct xfs_buf          *bp,
2410         struct xfs_da_args      *args)
2411 {
2412         struct xfs_attr_leafblock *leaf;
2413         struct xfs_attr3_icleaf_hdr ichdr;
2414         struct xfs_attr_leaf_entry *entry;
2415         struct xfs_attr_leaf_entry *entries;
2416         struct xfs_attr_leaf_name_local *name_loc;
2417         struct xfs_attr_leaf_name_remote *name_rmt;
2418         xfs_dahash_t            hashval;
2419         int                     probe;
2420         int                     span;
2421
2422         trace_xfs_attr_leaf_lookup(args);
2423
2424         leaf = bp->b_addr;
2425         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2426         entries = xfs_attr3_leaf_entryp(leaf);
2427         if (ichdr.count >= args->geo->blksize / 8) {
2428                 xfs_buf_mark_corrupt(bp);
2429                 return -EFSCORRUPTED;
2430         }
2431
2432         /*
2433          * Binary search.  (note: small blocks will skip this loop)
2434          */
2435         hashval = args->hashval;
2436         probe = span = ichdr.count / 2;
2437         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2438                 span /= 2;
2439                 if (be32_to_cpu(entry->hashval) < hashval)
2440                         probe += span;
2441                 else if (be32_to_cpu(entry->hashval) > hashval)
2442                         probe -= span;
2443                 else
2444                         break;
2445         }
2446         if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2447                 xfs_buf_mark_corrupt(bp);
2448                 return -EFSCORRUPTED;
2449         }
2450         if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2451                 xfs_buf_mark_corrupt(bp);
2452                 return -EFSCORRUPTED;
2453         }
2454
2455         /*
2456          * Since we may have duplicate hashval's, find the first matching
2457          * hashval in the leaf.
2458          */
2459         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2460                 entry--;
2461                 probe--;
2462         }
2463         while (probe < ichdr.count &&
2464                be32_to_cpu(entry->hashval) < hashval) {
2465                 entry++;
2466                 probe++;
2467         }
2468         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2469                 args->index = probe;
2470                 return -ENOATTR;
2471         }
2472
2473         /*
2474          * Duplicate keys may be present, so search all of them for a match.
2475          */
2476         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2477                         entry++, probe++) {
2478 /*
2479  * GROT: Add code to remove incomplete entries.
2480  */
2481                 if (entry->flags & XFS_ATTR_LOCAL) {
2482                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2483                         if (!xfs_attr_match(args, name_loc->namelen,
2484                                         name_loc->nameval, entry->flags))
2485                                 continue;
2486                         args->index = probe;
2487                         return -EEXIST;
2488                 } else {
2489                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2490                         if (!xfs_attr_match(args, name_rmt->namelen,
2491                                         name_rmt->name, entry->flags))
2492                                 continue;
2493                         args->index = probe;
2494                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2495                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2496                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2497                                                         args->dp->i_mount,
2498                                                         args->rmtvaluelen);
2499                         return -EEXIST;
2500                 }
2501         }
2502         args->index = probe;
2503         return -ENOATTR;
2504 }
2505
2506 /*
2507  * Get the value associated with an attribute name from a leaf attribute
2508  * list structure.
2509  *
2510  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2511  * lookup, we only return an error if the attribute does not exist or we can't
2512  * retrieve the value.
2513  */
2514 int
2515 xfs_attr3_leaf_getvalue(
2516         struct xfs_buf          *bp,
2517         struct xfs_da_args      *args)
2518 {
2519         struct xfs_attr_leafblock *leaf;
2520         struct xfs_attr3_icleaf_hdr ichdr;
2521         struct xfs_attr_leaf_entry *entry;
2522         struct xfs_attr_leaf_name_local *name_loc;
2523         struct xfs_attr_leaf_name_remote *name_rmt;
2524
2525         leaf = bp->b_addr;
2526         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2527         ASSERT(ichdr.count < args->geo->blksize / 8);
2528         ASSERT(args->index < ichdr.count);
2529
2530         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2531         if (entry->flags & XFS_ATTR_LOCAL) {
2532                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2533                 ASSERT(name_loc->namelen == args->namelen);
2534                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2535                 return xfs_attr_copy_value(args,
2536                                         &name_loc->nameval[args->namelen],
2537                                         be16_to_cpu(name_loc->valuelen));
2538         }
2539
2540         name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2541         ASSERT(name_rmt->namelen == args->namelen);
2542         ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2543         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2544         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2545         args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2546                                                args->rmtvaluelen);
2547         return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2548 }
2549
2550 /*========================================================================
2551  * Utility routines.
2552  *========================================================================*/
2553
2554 /*
2555  * Move the indicated entries from one leaf to another.
2556  * NOTE: this routine modifies both source and destination leaves.
2557  */
2558 /*ARGSUSED*/
2559 STATIC void
2560 xfs_attr3_leaf_moveents(
2561         struct xfs_da_args              *args,
2562         struct xfs_attr_leafblock       *leaf_s,
2563         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2564         int                             start_s,
2565         struct xfs_attr_leafblock       *leaf_d,
2566         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2567         int                             start_d,
2568         int                             count)
2569 {
2570         struct xfs_attr_leaf_entry      *entry_s;
2571         struct xfs_attr_leaf_entry      *entry_d;
2572         int                             desti;
2573         int                             tmp;
2574         int                             i;
2575
2576         /*
2577          * Check for nothing to do.
2578          */
2579         if (count == 0)
2580                 return;
2581
2582         /*
2583          * Set up environment.
2584          */
2585         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2586                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2587         ASSERT(ichdr_s->magic == ichdr_d->magic);
2588         ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2589         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2590                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2591         ASSERT(ichdr_d->count < args->geo->blksize / 8);
2592         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2593                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2594
2595         ASSERT(start_s < ichdr_s->count);
2596         ASSERT(start_d <= ichdr_d->count);
2597         ASSERT(count <= ichdr_s->count);
2598
2599
2600         /*
2601          * Move the entries in the destination leaf up to make a hole?
2602          */
2603         if (start_d < ichdr_d->count) {
2604                 tmp  = ichdr_d->count - start_d;
2605                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2606                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2607                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2608                 memmove(entry_d, entry_s, tmp);
2609         }
2610
2611         /*
2612          * Copy all entry's in the same (sorted) order,
2613          * but allocate attribute info packed and in sequence.
2614          */
2615         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2616         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2617         desti = start_d;
2618         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2619                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2620                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2621 #ifdef GROT
2622                 /*
2623                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2624                  * may also need to change the insertion index.  Code turned
2625                  * off for 6.2, should be revisited later.
2626                  */
2627                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2628                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2629                         ichdr_s->usedbytes -= tmp;
2630                         ichdr_s->count -= 1;
2631                         entry_d--;      /* to compensate for ++ in loop hdr */
2632                         desti--;
2633                         if ((start_s + i) < offset)
2634                                 result++;       /* insertion index adjustment */
2635                 } else {
2636 #endif /* GROT */
2637                         ichdr_d->firstused -= tmp;
2638                         /* both on-disk, don't endian flip twice */
2639                         entry_d->hashval = entry_s->hashval;
2640                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2641                         entry_d->flags = entry_s->flags;
2642                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2643                                                         <= args->geo->blksize);
2644                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2645                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2646                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2647                                                         <= args->geo->blksize);
2648                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2649                         ichdr_s->usedbytes -= tmp;
2650                         ichdr_d->usedbytes += tmp;
2651                         ichdr_s->count -= 1;
2652                         ichdr_d->count += 1;
2653                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2654                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2655                         ASSERT(ichdr_d->firstused >= tmp);
2656 #ifdef GROT
2657                 }
2658 #endif /* GROT */
2659         }
2660
2661         /*
2662          * Zero out the entries we just copied.
2663          */
2664         if (start_s == ichdr_s->count) {
2665                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2666                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2667                 ASSERT(((char *)entry_s + tmp) <=
2668                        ((char *)leaf_s + args->geo->blksize));
2669                 memset(entry_s, 0, tmp);
2670         } else {
2671                 /*
2672                  * Move the remaining entries down to fill the hole,
2673                  * then zero the entries at the top.
2674                  */
2675                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2676                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2677                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2678                 memmove(entry_d, entry_s, tmp);
2679
2680                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2681                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2682                 ASSERT(((char *)entry_s + tmp) <=
2683                        ((char *)leaf_s + args->geo->blksize));
2684                 memset(entry_s, 0, tmp);
2685         }
2686
2687         /*
2688          * Fill in the freemap information
2689          */
2690         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2691         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2692         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2693         ichdr_d->freemap[1].base = 0;
2694         ichdr_d->freemap[2].base = 0;
2695         ichdr_d->freemap[1].size = 0;
2696         ichdr_d->freemap[2].size = 0;
2697         ichdr_s->holes = 1;     /* leaf may not be compact */
2698 }
2699
2700 /*
2701  * Pick up the last hashvalue from a leaf block.
2702  */
2703 xfs_dahash_t
2704 xfs_attr_leaf_lasthash(
2705         struct xfs_buf  *bp,
2706         int             *count)
2707 {
2708         struct xfs_attr3_icleaf_hdr ichdr;
2709         struct xfs_attr_leaf_entry *entries;
2710         struct xfs_mount *mp = bp->b_mount;
2711
2712         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2713         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2714         if (count)
2715                 *count = ichdr.count;
2716         if (!ichdr.count)
2717                 return 0;
2718         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2719 }
2720
2721 /*
2722  * Calculate the number of bytes used to store the indicated attribute
2723  * (whether local or remote only calculate bytes in this block).
2724  */
2725 STATIC int
2726 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2727 {
2728         struct xfs_attr_leaf_entry *entries;
2729         xfs_attr_leaf_name_local_t *name_loc;
2730         xfs_attr_leaf_name_remote_t *name_rmt;
2731         int size;
2732
2733         entries = xfs_attr3_leaf_entryp(leaf);
2734         if (entries[index].flags & XFS_ATTR_LOCAL) {
2735                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2736                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2737                                                    be16_to_cpu(name_loc->valuelen));
2738         } else {
2739                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2740                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2741         }
2742         return size;
2743 }
2744
2745 /*
2746  * Calculate the number of bytes that would be required to store the new
2747  * attribute (whether local or remote only calculate bytes in this block).
2748  * This routine decides as a side effect whether the attribute will be
2749  * a "local" or a "remote" attribute.
2750  */
2751 int
2752 xfs_attr_leaf_newentsize(
2753         struct xfs_da_args      *args,
2754         int                     *local)
2755 {
2756         int                     size;
2757
2758         size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2759         if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2760                 if (local)
2761                         *local = 1;
2762                 return size;
2763         }
2764         if (local)
2765                 *local = 0;
2766         return xfs_attr_leaf_entsize_remote(args->namelen);
2767 }
2768
2769
2770 /*========================================================================
2771  * Manage the INCOMPLETE flag in a leaf entry
2772  *========================================================================*/
2773
2774 /*
2775  * Clear the INCOMPLETE flag on an entry in a leaf block.
2776  */
2777 int
2778 xfs_attr3_leaf_clearflag(
2779         struct xfs_da_args      *args)
2780 {
2781         struct xfs_attr_leafblock *leaf;
2782         struct xfs_attr_leaf_entry *entry;
2783         struct xfs_attr_leaf_name_remote *name_rmt;
2784         struct xfs_buf          *bp;
2785         int                     error;
2786 #ifdef DEBUG
2787         struct xfs_attr3_icleaf_hdr ichdr;
2788         xfs_attr_leaf_name_local_t *name_loc;
2789         int namelen;
2790         char *name;
2791 #endif /* DEBUG */
2792
2793         trace_xfs_attr_leaf_clearflag(args);
2794         /*
2795          * Set up the operation.
2796          */
2797         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2798         if (error)
2799                 return error;
2800
2801         leaf = bp->b_addr;
2802         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2803         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2804
2805 #ifdef DEBUG
2806         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2807         ASSERT(args->index < ichdr.count);
2808         ASSERT(args->index >= 0);
2809
2810         if (entry->flags & XFS_ATTR_LOCAL) {
2811                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2812                 namelen = name_loc->namelen;
2813                 name = (char *)name_loc->nameval;
2814         } else {
2815                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2816                 namelen = name_rmt->namelen;
2817                 name = (char *)name_rmt->name;
2818         }
2819         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2820         ASSERT(namelen == args->namelen);
2821         ASSERT(memcmp(name, args->name, namelen) == 0);
2822 #endif /* DEBUG */
2823
2824         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2825         xfs_trans_log_buf(args->trans, bp,
2826                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2827
2828         if (args->rmtblkno) {
2829                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2830                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2831                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2832                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2833                 xfs_trans_log_buf(args->trans, bp,
2834                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2835         }
2836
2837         return 0;
2838 }
2839
2840 /*
2841  * Set the INCOMPLETE flag on an entry in a leaf block.
2842  */
2843 int
2844 xfs_attr3_leaf_setflag(
2845         struct xfs_da_args      *args)
2846 {
2847         struct xfs_attr_leafblock *leaf;
2848         struct xfs_attr_leaf_entry *entry;
2849         struct xfs_attr_leaf_name_remote *name_rmt;
2850         struct xfs_buf          *bp;
2851         int error;
2852 #ifdef DEBUG
2853         struct xfs_attr3_icleaf_hdr ichdr;
2854 #endif
2855
2856         trace_xfs_attr_leaf_setflag(args);
2857
2858         /*
2859          * Set up the operation.
2860          */
2861         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2862         if (error)
2863                 return error;
2864
2865         leaf = bp->b_addr;
2866 #ifdef DEBUG
2867         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2868         ASSERT(args->index < ichdr.count);
2869         ASSERT(args->index >= 0);
2870 #endif
2871         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2872
2873         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2874         entry->flags |= XFS_ATTR_INCOMPLETE;
2875         xfs_trans_log_buf(args->trans, bp,
2876                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2877         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2878                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2879                 name_rmt->valueblk = 0;
2880                 name_rmt->valuelen = 0;
2881                 xfs_trans_log_buf(args->trans, bp,
2882                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2883         }
2884
2885         return 0;
2886 }
2887
2888 /*
2889  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2890  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2891  * entry given by args->blkno2/index2.
2892  *
2893  * Note that they could be in different blocks, or in the same block.
2894  */
2895 int
2896 xfs_attr3_leaf_flipflags(
2897         struct xfs_da_args      *args)
2898 {
2899         struct xfs_attr_leafblock *leaf1;
2900         struct xfs_attr_leafblock *leaf2;
2901         struct xfs_attr_leaf_entry *entry1;
2902         struct xfs_attr_leaf_entry *entry2;
2903         struct xfs_attr_leaf_name_remote *name_rmt;
2904         struct xfs_buf          *bp1;
2905         struct xfs_buf          *bp2;
2906         int error;
2907 #ifdef DEBUG
2908         struct xfs_attr3_icleaf_hdr ichdr1;
2909         struct xfs_attr3_icleaf_hdr ichdr2;
2910         xfs_attr_leaf_name_local_t *name_loc;
2911         int namelen1, namelen2;
2912         char *name1, *name2;
2913 #endif /* DEBUG */
2914
2915         trace_xfs_attr_leaf_flipflags(args);
2916
2917         /*
2918          * Read the block containing the "old" attr
2919          */
2920         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2921         if (error)
2922                 return error;
2923
2924         /*
2925          * Read the block containing the "new" attr, if it is different
2926          */
2927         if (args->blkno2 != args->blkno) {
2928                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2929                                            &bp2);
2930                 if (error)
2931                         return error;
2932         } else {
2933                 bp2 = bp1;
2934         }
2935
2936         leaf1 = bp1->b_addr;
2937         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2938
2939         leaf2 = bp2->b_addr;
2940         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2941
2942 #ifdef DEBUG
2943         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2944         ASSERT(args->index < ichdr1.count);
2945         ASSERT(args->index >= 0);
2946
2947         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2948         ASSERT(args->index2 < ichdr2.count);
2949         ASSERT(args->index2 >= 0);
2950
2951         if (entry1->flags & XFS_ATTR_LOCAL) {
2952                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2953                 namelen1 = name_loc->namelen;
2954                 name1 = (char *)name_loc->nameval;
2955         } else {
2956                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2957                 namelen1 = name_rmt->namelen;
2958                 name1 = (char *)name_rmt->name;
2959         }
2960         if (entry2->flags & XFS_ATTR_LOCAL) {
2961                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2962                 namelen2 = name_loc->namelen;
2963                 name2 = (char *)name_loc->nameval;
2964         } else {
2965                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2966                 namelen2 = name_rmt->namelen;
2967                 name2 = (char *)name_rmt->name;
2968         }
2969         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2970         ASSERT(namelen1 == namelen2);
2971         ASSERT(memcmp(name1, name2, namelen1) == 0);
2972 #endif /* DEBUG */
2973
2974         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2975         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2976
2977         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2978         xfs_trans_log_buf(args->trans, bp1,
2979                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2980         if (args->rmtblkno) {
2981                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2982                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2983                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2984                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2985                 xfs_trans_log_buf(args->trans, bp1,
2986                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2987         }
2988
2989         entry2->flags |= XFS_ATTR_INCOMPLETE;
2990         xfs_trans_log_buf(args->trans, bp2,
2991                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2992         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2993                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2994                 name_rmt->valueblk = 0;
2995                 name_rmt->valuelen = 0;
2996                 xfs_trans_log_buf(args->trans, bp2,
2997                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2998         }
2999
3000         return 0;
3001 }