]> asedeno.scripts.mit.edu Git - linux.git/blob - fs/xfs/libxfs/xfs_attr_leaf.c
Linux 5.6-rc7
[linux.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_remote.h"
23 #include "xfs_attr.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
31
32 /*
33  * xfs_attr_leaf.c
34  *
35  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
36  */
37
38 /*========================================================================
39  * Function prototypes for the kernel.
40  *========================================================================*/
41
42 /*
43  * Routines used for growing the Btree.
44  */
45 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
46                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
47 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
48                                    struct xfs_attr3_icleaf_hdr *ichdr,
49                                    struct xfs_da_args *args, int freemap_index);
50 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
51                                    struct xfs_attr3_icleaf_hdr *ichdr,
52                                    struct xfs_buf *leaf_buffer);
53 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
54                                                    xfs_da_state_blk_t *blk1,
55                                                    xfs_da_state_blk_t *blk2);
56 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
57                         xfs_da_state_blk_t *leaf_blk_1,
58                         struct xfs_attr3_icleaf_hdr *ichdr1,
59                         xfs_da_state_blk_t *leaf_blk_2,
60                         struct xfs_attr3_icleaf_hdr *ichdr2,
61                         int *number_entries_in_blk1,
62                         int *number_usedbytes_in_blk1);
63
64 /*
65  * Utility routines.
66  */
67 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
68                         struct xfs_attr_leafblock *src_leaf,
69                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
70                         struct xfs_attr_leafblock *dst_leaf,
71                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
72                         int move_count);
73 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
74
75 /*
76  * attr3 block 'firstused' conversion helpers.
77  *
78  * firstused refers to the offset of the first used byte of the nameval region
79  * of an attr leaf block. The region starts at the tail of the block and expands
80  * backwards towards the middle. As such, firstused is initialized to the block
81  * size for an empty leaf block and is reduced from there.
82  *
83  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
84  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
85  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
86  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
87  * the attr block size. The following helpers manage the conversion between the
88  * in-core and on-disk formats.
89  */
90
91 static void
92 xfs_attr3_leaf_firstused_from_disk(
93         struct xfs_da_geometry          *geo,
94         struct xfs_attr3_icleaf_hdr     *to,
95         struct xfs_attr_leafblock       *from)
96 {
97         struct xfs_attr3_leaf_hdr       *hdr3;
98
99         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
100                 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
101                 to->firstused = be16_to_cpu(hdr3->firstused);
102         } else {
103                 to->firstused = be16_to_cpu(from->hdr.firstused);
104         }
105
106         /*
107          * Convert from the magic fsb size value to actual blocksize. This
108          * should only occur for empty blocks when the block size overflows
109          * 16-bits.
110          */
111         if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
112                 ASSERT(!to->count && !to->usedbytes);
113                 ASSERT(geo->blksize > USHRT_MAX);
114                 to->firstused = geo->blksize;
115         }
116 }
117
118 static void
119 xfs_attr3_leaf_firstused_to_disk(
120         struct xfs_da_geometry          *geo,
121         struct xfs_attr_leafblock       *to,
122         struct xfs_attr3_icleaf_hdr     *from)
123 {
124         struct xfs_attr3_leaf_hdr       *hdr3;
125         uint32_t                        firstused;
126
127         /* magic value should only be seen on disk */
128         ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
129
130         /*
131          * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
132          * value. This only overflows at the max supported value of 64k. Use the
133          * magic on-disk value to represent block size in this case.
134          */
135         firstused = from->firstused;
136         if (firstused > USHRT_MAX) {
137                 ASSERT(from->firstused == geo->blksize);
138                 firstused = XFS_ATTR3_LEAF_NULLOFF;
139         }
140
141         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
142                 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
143                 hdr3->firstused = cpu_to_be16(firstused);
144         } else {
145                 to->hdr.firstused = cpu_to_be16(firstused);
146         }
147 }
148
149 void
150 xfs_attr3_leaf_hdr_from_disk(
151         struct xfs_da_geometry          *geo,
152         struct xfs_attr3_icleaf_hdr     *to,
153         struct xfs_attr_leafblock       *from)
154 {
155         int     i;
156
157         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
158                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
159
160         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
161                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
162
163                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
164                 to->back = be32_to_cpu(hdr3->info.hdr.back);
165                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
166                 to->count = be16_to_cpu(hdr3->count);
167                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
168                 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
169                 to->holes = hdr3->holes;
170
171                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
172                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
173                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
174                 }
175                 return;
176         }
177         to->forw = be32_to_cpu(from->hdr.info.forw);
178         to->back = be32_to_cpu(from->hdr.info.back);
179         to->magic = be16_to_cpu(from->hdr.info.magic);
180         to->count = be16_to_cpu(from->hdr.count);
181         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
182         xfs_attr3_leaf_firstused_from_disk(geo, to, from);
183         to->holes = from->hdr.holes;
184
185         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
186                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
187                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
188         }
189 }
190
191 void
192 xfs_attr3_leaf_hdr_to_disk(
193         struct xfs_da_geometry          *geo,
194         struct xfs_attr_leafblock       *to,
195         struct xfs_attr3_icleaf_hdr     *from)
196 {
197         int                             i;
198
199         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
200                from->magic == XFS_ATTR3_LEAF_MAGIC);
201
202         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
203                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
204
205                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
206                 hdr3->info.hdr.back = cpu_to_be32(from->back);
207                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
208                 hdr3->count = cpu_to_be16(from->count);
209                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
210                 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
211                 hdr3->holes = from->holes;
212                 hdr3->pad1 = 0;
213
214                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
215                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
216                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
217                 }
218                 return;
219         }
220         to->hdr.info.forw = cpu_to_be32(from->forw);
221         to->hdr.info.back = cpu_to_be32(from->back);
222         to->hdr.info.magic = cpu_to_be16(from->magic);
223         to->hdr.count = cpu_to_be16(from->count);
224         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
225         xfs_attr3_leaf_firstused_to_disk(geo, to, from);
226         to->hdr.holes = from->holes;
227         to->hdr.pad1 = 0;
228
229         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
230                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
231                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
232         }
233 }
234
235 static xfs_failaddr_t
236 xfs_attr3_leaf_verify_entry(
237         struct xfs_mount                        *mp,
238         char                                    *buf_end,
239         struct xfs_attr_leafblock               *leaf,
240         struct xfs_attr3_icleaf_hdr             *leafhdr,
241         struct xfs_attr_leaf_entry              *ent,
242         int                                     idx,
243         __u32                                   *last_hashval)
244 {
245         struct xfs_attr_leaf_name_local         *lentry;
246         struct xfs_attr_leaf_name_remote        *rentry;
247         char                                    *name_end;
248         unsigned int                            nameidx;
249         unsigned int                            namesize;
250         __u32                                   hashval;
251
252         /* hash order check */
253         hashval = be32_to_cpu(ent->hashval);
254         if (hashval < *last_hashval)
255                 return __this_address;
256         *last_hashval = hashval;
257
258         nameidx = be16_to_cpu(ent->nameidx);
259         if (nameidx < leafhdr->firstused || nameidx >= mp->m_attr_geo->blksize)
260                 return __this_address;
261
262         /*
263          * Check the name information.  The namelen fields are u8 so we can't
264          * possibly exceed the maximum name length of 255 bytes.
265          */
266         if (ent->flags & XFS_ATTR_LOCAL) {
267                 lentry = xfs_attr3_leaf_name_local(leaf, idx);
268                 namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
269                                 be16_to_cpu(lentry->valuelen));
270                 name_end = (char *)lentry + namesize;
271                 if (lentry->namelen == 0)
272                         return __this_address;
273         } else {
274                 rentry = xfs_attr3_leaf_name_remote(leaf, idx);
275                 namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
276                 name_end = (char *)rentry + namesize;
277                 if (rentry->namelen == 0)
278                         return __this_address;
279                 if (!(ent->flags & XFS_ATTR_INCOMPLETE) &&
280                     rentry->valueblk == 0)
281                         return __this_address;
282         }
283
284         if (name_end > buf_end)
285                 return __this_address;
286
287         return NULL;
288 }
289
290 static xfs_failaddr_t
291 xfs_attr3_leaf_verify(
292         struct xfs_buf                  *bp)
293 {
294         struct xfs_attr3_icleaf_hdr     ichdr;
295         struct xfs_mount                *mp = bp->b_mount;
296         struct xfs_attr_leafblock       *leaf = bp->b_addr;
297         struct xfs_attr_leaf_entry      *entries;
298         struct xfs_attr_leaf_entry      *ent;
299         char                            *buf_end;
300         uint32_t                        end;    /* must be 32bit - see below */
301         __u32                           last_hashval = 0;
302         int                             i;
303         xfs_failaddr_t                  fa;
304
305         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
306
307         fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
308         if (fa)
309                 return fa;
310
311         /*
312          * In recovery there is a transient state where count == 0 is valid
313          * because we may have transitioned an empty shortform attr to a leaf
314          * if the attr didn't fit in shortform.
315          */
316         if (!xfs_log_in_recovery(mp) && ichdr.count == 0)
317                 return __this_address;
318
319         /*
320          * firstused is the block offset of the first name info structure.
321          * Make sure it doesn't go off the block or crash into the header.
322          */
323         if (ichdr.firstused > mp->m_attr_geo->blksize)
324                 return __this_address;
325         if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
326                 return __this_address;
327
328         /* Make sure the entries array doesn't crash into the name info. */
329         entries = xfs_attr3_leaf_entryp(bp->b_addr);
330         if ((char *)&entries[ichdr.count] >
331             (char *)bp->b_addr + ichdr.firstused)
332                 return __this_address;
333
334         buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
335         for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
336                 fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
337                                 ent, i, &last_hashval);
338                 if (fa)
339                         return fa;
340         }
341
342         /*
343          * Quickly check the freemap information.  Attribute data has to be
344          * aligned to 4-byte boundaries, and likewise for the free space.
345          *
346          * Note that for 64k block size filesystems, the freemap entries cannot
347          * overflow as they are only be16 fields. However, when checking end
348          * pointer of the freemap, we have to be careful to detect overflows and
349          * so use uint32_t for those checks.
350          */
351         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
352                 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
353                         return __this_address;
354                 if (ichdr.freemap[i].base & 0x3)
355                         return __this_address;
356                 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
357                         return __this_address;
358                 if (ichdr.freemap[i].size & 0x3)
359                         return __this_address;
360
361                 /* be care of 16 bit overflows here */
362                 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
363                 if (end < ichdr.freemap[i].base)
364                         return __this_address;
365                 if (end > mp->m_attr_geo->blksize)
366                         return __this_address;
367         }
368
369         return NULL;
370 }
371
372 static void
373 xfs_attr3_leaf_write_verify(
374         struct xfs_buf  *bp)
375 {
376         struct xfs_mount        *mp = bp->b_mount;
377         struct xfs_buf_log_item *bip = bp->b_log_item;
378         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
379         xfs_failaddr_t          fa;
380
381         fa = xfs_attr3_leaf_verify(bp);
382         if (fa) {
383                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
384                 return;
385         }
386
387         if (!xfs_sb_version_hascrc(&mp->m_sb))
388                 return;
389
390         if (bip)
391                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
392
393         xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
394 }
395
396 /*
397  * leaf/node format detection on trees is sketchy, so a node read can be done on
398  * leaf level blocks when detection identifies the tree as a node format tree
399  * incorrectly. In this case, we need to swap the verifier to match the correct
400  * format of the block being read.
401  */
402 static void
403 xfs_attr3_leaf_read_verify(
404         struct xfs_buf          *bp)
405 {
406         struct xfs_mount        *mp = bp->b_mount;
407         xfs_failaddr_t          fa;
408
409         if (xfs_sb_version_hascrc(&mp->m_sb) &&
410              !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
411                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
412         else {
413                 fa = xfs_attr3_leaf_verify(bp);
414                 if (fa)
415                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
416         }
417 }
418
419 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
420         .name = "xfs_attr3_leaf",
421         .magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
422                      cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
423         .verify_read = xfs_attr3_leaf_read_verify,
424         .verify_write = xfs_attr3_leaf_write_verify,
425         .verify_struct = xfs_attr3_leaf_verify,
426 };
427
428 int
429 xfs_attr3_leaf_read(
430         struct xfs_trans        *tp,
431         struct xfs_inode        *dp,
432         xfs_dablk_t             bno,
433         struct xfs_buf          **bpp)
434 {
435         int                     err;
436
437         err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
438                         &xfs_attr3_leaf_buf_ops);
439         if (!err && tp && *bpp)
440                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
441         return err;
442 }
443
444 /*========================================================================
445  * Namespace helper routines
446  *========================================================================*/
447
448 /*
449  * If namespace bits don't match return 0.
450  * If all match then return 1.
451  */
452 STATIC int
453 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
454 {
455         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
456 }
457
458 static int
459 xfs_attr_copy_value(
460         struct xfs_da_args      *args,
461         unsigned char           *value,
462         int                     valuelen)
463 {
464         /*
465          * No copy if all we have to do is get the length
466          */
467         if (args->flags & ATTR_KERNOVAL) {
468                 args->valuelen = valuelen;
469                 return 0;
470         }
471
472         /*
473          * No copy if the length of the existing buffer is too small
474          */
475         if (args->valuelen < valuelen) {
476                 args->valuelen = valuelen;
477                 return -ERANGE;
478         }
479
480         if (args->op_flags & XFS_DA_OP_ALLOCVAL) {
481                 args->value = kmem_alloc_large(valuelen, 0);
482                 if (!args->value)
483                         return -ENOMEM;
484         }
485         args->valuelen = valuelen;
486
487         /* remote block xattr requires IO for copy-in */
488         if (args->rmtblkno)
489                 return xfs_attr_rmtval_get(args);
490
491         /*
492          * This is to prevent a GCC warning because the remote xattr case
493          * doesn't have a value to pass in. In that case, we never reach here,
494          * but GCC can't work that out and so throws a "passing NULL to
495          * memcpy" warning.
496          */
497         if (!value)
498                 return -EINVAL;
499         memcpy(args->value, value, valuelen);
500         return 0;
501 }
502
503 /*========================================================================
504  * External routines when attribute fork size < XFS_LITINO(mp).
505  *========================================================================*/
506
507 /*
508  * Query whether the requested number of additional bytes of extended
509  * attribute space will be able to fit inline.
510  *
511  * Returns zero if not, else the di_forkoff fork offset to be used in the
512  * literal area for attribute data once the new bytes have been added.
513  *
514  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
515  * special case for dev/uuid inodes, they have fixed size data forks.
516  */
517 int
518 xfs_attr_shortform_bytesfit(
519         struct xfs_inode        *dp,
520         int                     bytes)
521 {
522         struct xfs_mount        *mp = dp->i_mount;
523         int64_t                 dsize;
524         int                     minforkoff;
525         int                     maxforkoff;
526         int                     offset;
527
528         /* rounded down */
529         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
530
531         if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
532                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
533                 return (offset >= minforkoff) ? minforkoff : 0;
534         }
535
536         /*
537          * If the requested numbers of bytes is smaller or equal to the
538          * current attribute fork size we can always proceed.
539          *
540          * Note that if_bytes in the data fork might actually be larger than
541          * the current data fork size is due to delalloc extents. In that
542          * case either the extent count will go down when they are converted
543          * to real extents, or the delalloc conversion will take care of the
544          * literal area rebalancing.
545          */
546         if (bytes <= XFS_IFORK_ASIZE(dp))
547                 return dp->i_d.di_forkoff;
548
549         /*
550          * For attr2 we can try to move the forkoff if there is space in the
551          * literal area, but for the old format we are done if there is no
552          * space in the fixed attribute fork.
553          */
554         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
555                 return 0;
556
557         dsize = dp->i_df.if_bytes;
558
559         switch (dp->i_d.di_format) {
560         case XFS_DINODE_FMT_EXTENTS:
561                 /*
562                  * If there is no attr fork and the data fork is extents, 
563                  * determine if creating the default attr fork will result
564                  * in the extents form migrating to btree. If so, the
565                  * minimum offset only needs to be the space required for
566                  * the btree root.
567                  */
568                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
569                     xfs_default_attroffset(dp))
570                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
571                 break;
572         case XFS_DINODE_FMT_BTREE:
573                 /*
574                  * If we have a data btree then keep forkoff if we have one,
575                  * otherwise we are adding a new attr, so then we set
576                  * minforkoff to where the btree root can finish so we have
577                  * plenty of room for attrs
578                  */
579                 if (dp->i_d.di_forkoff) {
580                         if (offset < dp->i_d.di_forkoff)
581                                 return 0;
582                         return dp->i_d.di_forkoff;
583                 }
584                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
585                 break;
586         }
587
588         /*
589          * A data fork btree root must have space for at least
590          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
591          */
592         minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
593         minforkoff = roundup(minforkoff, 8) >> 3;
594
595         /* attr fork btree root can have at least this many key/ptr pairs */
596         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
597                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
598         maxforkoff = maxforkoff >> 3;   /* rounded down */
599
600         if (offset >= maxforkoff)
601                 return maxforkoff;
602         if (offset >= minforkoff)
603                 return offset;
604         return 0;
605 }
606
607 /*
608  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
609  */
610 STATIC void
611 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
612 {
613         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
614             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
615                 spin_lock(&mp->m_sb_lock);
616                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
617                         xfs_sb_version_addattr2(&mp->m_sb);
618                         spin_unlock(&mp->m_sb_lock);
619                         xfs_log_sb(tp);
620                 } else
621                         spin_unlock(&mp->m_sb_lock);
622         }
623 }
624
625 /*
626  * Create the initial contents of a shortform attribute list.
627  */
628 void
629 xfs_attr_shortform_create(xfs_da_args_t *args)
630 {
631         xfs_attr_sf_hdr_t *hdr;
632         xfs_inode_t *dp;
633         struct xfs_ifork *ifp;
634
635         trace_xfs_attr_sf_create(args);
636
637         dp = args->dp;
638         ASSERT(dp != NULL);
639         ifp = dp->i_afp;
640         ASSERT(ifp != NULL);
641         ASSERT(ifp->if_bytes == 0);
642         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
643                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
644                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
645                 ifp->if_flags |= XFS_IFINLINE;
646         } else {
647                 ASSERT(ifp->if_flags & XFS_IFINLINE);
648         }
649         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
650         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
651         hdr->count = 0;
652         hdr->totsize = cpu_to_be16(sizeof(*hdr));
653         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
654 }
655
656 /*
657  * Add a name/value pair to the shortform attribute list.
658  * Overflow from the inode has already been checked for.
659  */
660 void
661 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
662 {
663         xfs_attr_shortform_t *sf;
664         xfs_attr_sf_entry_t *sfe;
665         int i, offset, size;
666         xfs_mount_t *mp;
667         xfs_inode_t *dp;
668         struct xfs_ifork *ifp;
669
670         trace_xfs_attr_sf_add(args);
671
672         dp = args->dp;
673         mp = dp->i_mount;
674         dp->i_d.di_forkoff = forkoff;
675
676         ifp = dp->i_afp;
677         ASSERT(ifp->if_flags & XFS_IFINLINE);
678         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
679         sfe = &sf->list[0];
680         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
681 #ifdef DEBUG
682                 if (sfe->namelen != args->namelen)
683                         continue;
684                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
685                         continue;
686                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
687                         continue;
688                 ASSERT(0);
689 #endif
690         }
691
692         offset = (char *)sfe - (char *)sf;
693         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
694         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
695         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
696         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
697
698         sfe->namelen = args->namelen;
699         sfe->valuelen = args->valuelen;
700         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
701         memcpy(sfe->nameval, args->name, args->namelen);
702         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
703         sf->hdr.count++;
704         be16_add_cpu(&sf->hdr.totsize, size);
705         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
706
707         xfs_sbversion_add_attr2(mp, args->trans);
708 }
709
710 /*
711  * After the last attribute is removed revert to original inode format,
712  * making all literal area available to the data fork once more.
713  */
714 void
715 xfs_attr_fork_remove(
716         struct xfs_inode        *ip,
717         struct xfs_trans        *tp)
718 {
719         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
720         ip->i_d.di_forkoff = 0;
721         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
722
723         ASSERT(ip->i_d.di_anextents == 0);
724         ASSERT(ip->i_afp == NULL);
725
726         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
727 }
728
729 /*
730  * Remove an attribute from the shortform attribute list structure.
731  */
732 int
733 xfs_attr_shortform_remove(xfs_da_args_t *args)
734 {
735         xfs_attr_shortform_t *sf;
736         xfs_attr_sf_entry_t *sfe;
737         int base, size=0, end, totsize, i;
738         xfs_mount_t *mp;
739         xfs_inode_t *dp;
740
741         trace_xfs_attr_sf_remove(args);
742
743         dp = args->dp;
744         mp = dp->i_mount;
745         base = sizeof(xfs_attr_sf_hdr_t);
746         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
747         sfe = &sf->list[0];
748         end = sf->hdr.count;
749         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
750                                         base += size, i++) {
751                 size = XFS_ATTR_SF_ENTSIZE(sfe);
752                 if (sfe->namelen != args->namelen)
753                         continue;
754                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
755                         continue;
756                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
757                         continue;
758                 break;
759         }
760         if (i == end)
761                 return -ENOATTR;
762
763         /*
764          * Fix up the attribute fork data, covering the hole
765          */
766         end = base + size;
767         totsize = be16_to_cpu(sf->hdr.totsize);
768         if (end != totsize)
769                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
770         sf->hdr.count--;
771         be16_add_cpu(&sf->hdr.totsize, -size);
772
773         /*
774          * Fix up the start offset of the attribute fork
775          */
776         totsize -= size;
777         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
778             (mp->m_flags & XFS_MOUNT_ATTR2) &&
779             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
780             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
781                 xfs_attr_fork_remove(dp, args->trans);
782         } else {
783                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
784                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
785                 ASSERT(dp->i_d.di_forkoff);
786                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
787                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
788                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
789                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
790                 xfs_trans_log_inode(args->trans, dp,
791                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
792         }
793
794         xfs_sbversion_add_attr2(mp, args->trans);
795
796         return 0;
797 }
798
799 /*
800  * Look up a name in a shortform attribute list structure.
801  */
802 /*ARGSUSED*/
803 int
804 xfs_attr_shortform_lookup(xfs_da_args_t *args)
805 {
806         xfs_attr_shortform_t *sf;
807         xfs_attr_sf_entry_t *sfe;
808         int i;
809         struct xfs_ifork *ifp;
810
811         trace_xfs_attr_sf_lookup(args);
812
813         ifp = args->dp->i_afp;
814         ASSERT(ifp->if_flags & XFS_IFINLINE);
815         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
816         sfe = &sf->list[0];
817         for (i = 0; i < sf->hdr.count;
818                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
819                 if (sfe->namelen != args->namelen)
820                         continue;
821                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
822                         continue;
823                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
824                         continue;
825                 return -EEXIST;
826         }
827         return -ENOATTR;
828 }
829
830 /*
831  * Retrieve the attribute value and length.
832  *
833  * If ATTR_KERNOVAL is specified, only the length needs to be returned.
834  * Unlike a lookup, we only return an error if the attribute does not
835  * exist or we can't retrieve the value.
836  */
837 int
838 xfs_attr_shortform_getvalue(
839         struct xfs_da_args      *args)
840 {
841         struct xfs_attr_shortform *sf;
842         struct xfs_attr_sf_entry *sfe;
843         int                     i;
844
845         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
846         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
847         sfe = &sf->list[0];
848         for (i = 0; i < sf->hdr.count;
849                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
850                 if (sfe->namelen != args->namelen)
851                         continue;
852                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
853                         continue;
854                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
855                         continue;
856                 return xfs_attr_copy_value(args, &sfe->nameval[args->namelen],
857                                                 sfe->valuelen);
858         }
859         return -ENOATTR;
860 }
861
862 /*
863  * Convert from using the shortform to the leaf.  On success, return the
864  * buffer so that we can keep it locked until we're totally done with it.
865  */
866 int
867 xfs_attr_shortform_to_leaf(
868         struct xfs_da_args              *args,
869         struct xfs_buf                  **leaf_bp)
870 {
871         struct xfs_inode                *dp;
872         struct xfs_attr_shortform       *sf;
873         struct xfs_attr_sf_entry        *sfe;
874         struct xfs_da_args              nargs;
875         char                            *tmpbuffer;
876         int                             error, i, size;
877         xfs_dablk_t                     blkno;
878         struct xfs_buf                  *bp;
879         struct xfs_ifork                *ifp;
880
881         trace_xfs_attr_sf_to_leaf(args);
882
883         dp = args->dp;
884         ifp = dp->i_afp;
885         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
886         size = be16_to_cpu(sf->hdr.totsize);
887         tmpbuffer = kmem_alloc(size, 0);
888         ASSERT(tmpbuffer != NULL);
889         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
890         sf = (xfs_attr_shortform_t *)tmpbuffer;
891
892         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
893         xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
894
895         bp = NULL;
896         error = xfs_da_grow_inode(args, &blkno);
897         if (error)
898                 goto out;
899
900         ASSERT(blkno == 0);
901         error = xfs_attr3_leaf_create(args, blkno, &bp);
902         if (error)
903                 goto out;
904
905         memset((char *)&nargs, 0, sizeof(nargs));
906         nargs.dp = dp;
907         nargs.geo = args->geo;
908         nargs.total = args->total;
909         nargs.whichfork = XFS_ATTR_FORK;
910         nargs.trans = args->trans;
911         nargs.op_flags = XFS_DA_OP_OKNOENT;
912
913         sfe = &sf->list[0];
914         for (i = 0; i < sf->hdr.count; i++) {
915                 nargs.name = sfe->nameval;
916                 nargs.namelen = sfe->namelen;
917                 nargs.value = &sfe->nameval[nargs.namelen];
918                 nargs.valuelen = sfe->valuelen;
919                 nargs.hashval = xfs_da_hashname(sfe->nameval,
920                                                 sfe->namelen);
921                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
922                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
923                 ASSERT(error == -ENOATTR);
924                 error = xfs_attr3_leaf_add(bp, &nargs);
925                 ASSERT(error != -ENOSPC);
926                 if (error)
927                         goto out;
928                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
929         }
930         error = 0;
931         *leaf_bp = bp;
932 out:
933         kmem_free(tmpbuffer);
934         return error;
935 }
936
937 /*
938  * Check a leaf attribute block to see if all the entries would fit into
939  * a shortform attribute list.
940  */
941 int
942 xfs_attr_shortform_allfit(
943         struct xfs_buf          *bp,
944         struct xfs_inode        *dp)
945 {
946         struct xfs_attr_leafblock *leaf;
947         struct xfs_attr_leaf_entry *entry;
948         xfs_attr_leaf_name_local_t *name_loc;
949         struct xfs_attr3_icleaf_hdr leafhdr;
950         int                     bytes;
951         int                     i;
952         struct xfs_mount        *mp = bp->b_mount;
953
954         leaf = bp->b_addr;
955         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
956         entry = xfs_attr3_leaf_entryp(leaf);
957
958         bytes = sizeof(struct xfs_attr_sf_hdr);
959         for (i = 0; i < leafhdr.count; entry++, i++) {
960                 if (entry->flags & XFS_ATTR_INCOMPLETE)
961                         continue;               /* don't copy partial entries */
962                 if (!(entry->flags & XFS_ATTR_LOCAL))
963                         return 0;
964                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
965                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
966                         return 0;
967                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
968                         return 0;
969                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
970                                 + name_loc->namelen
971                                 + be16_to_cpu(name_loc->valuelen);
972         }
973         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
974             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
975             (bytes == sizeof(struct xfs_attr_sf_hdr)))
976                 return -1;
977         return xfs_attr_shortform_bytesfit(dp, bytes);
978 }
979
980 /* Verify the consistency of an inline attribute fork. */
981 xfs_failaddr_t
982 xfs_attr_shortform_verify(
983         struct xfs_inode                *ip)
984 {
985         struct xfs_attr_shortform       *sfp;
986         struct xfs_attr_sf_entry        *sfep;
987         struct xfs_attr_sf_entry        *next_sfep;
988         char                            *endp;
989         struct xfs_ifork                *ifp;
990         int                             i;
991         int64_t                         size;
992
993         ASSERT(ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL);
994         ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
995         sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
996         size = ifp->if_bytes;
997
998         /*
999          * Give up if the attribute is way too short.
1000          */
1001         if (size < sizeof(struct xfs_attr_sf_hdr))
1002                 return __this_address;
1003
1004         endp = (char *)sfp + size;
1005
1006         /* Check all reported entries */
1007         sfep = &sfp->list[0];
1008         for (i = 0; i < sfp->hdr.count; i++) {
1009                 /*
1010                  * struct xfs_attr_sf_entry has a variable length.
1011                  * Check the fixed-offset parts of the structure are
1012                  * within the data buffer.
1013                  */
1014                 if (((char *)sfep + sizeof(*sfep)) >= endp)
1015                         return __this_address;
1016
1017                 /* Don't allow names with known bad length. */
1018                 if (sfep->namelen == 0)
1019                         return __this_address;
1020
1021                 /*
1022                  * Check that the variable-length part of the structure is
1023                  * within the data buffer.  The next entry starts after the
1024                  * name component, so nextentry is an acceptable test.
1025                  */
1026                 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
1027                 if ((char *)next_sfep > endp)
1028                         return __this_address;
1029
1030                 /*
1031                  * Check for unknown flags.  Short form doesn't support
1032                  * the incomplete or local bits, so we can use the namespace
1033                  * mask here.
1034                  */
1035                 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1036                         return __this_address;
1037
1038                 /*
1039                  * Check for invalid namespace combinations.  We only allow
1040                  * one namespace flag per xattr, so we can just count the
1041                  * bits (i.e. hweight) here.
1042                  */
1043                 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1044                         return __this_address;
1045
1046                 sfep = next_sfep;
1047         }
1048         if ((void *)sfep != (void *)endp)
1049                 return __this_address;
1050
1051         return NULL;
1052 }
1053
1054 /*
1055  * Convert a leaf attribute list to shortform attribute list
1056  */
1057 int
1058 xfs_attr3_leaf_to_shortform(
1059         struct xfs_buf          *bp,
1060         struct xfs_da_args      *args,
1061         int                     forkoff)
1062 {
1063         struct xfs_attr_leafblock *leaf;
1064         struct xfs_attr3_icleaf_hdr ichdr;
1065         struct xfs_attr_leaf_entry *entry;
1066         struct xfs_attr_leaf_name_local *name_loc;
1067         struct xfs_da_args      nargs;
1068         struct xfs_inode        *dp = args->dp;
1069         char                    *tmpbuffer;
1070         int                     error;
1071         int                     i;
1072
1073         trace_xfs_attr_leaf_to_sf(args);
1074
1075         tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1076         if (!tmpbuffer)
1077                 return -ENOMEM;
1078
1079         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1080
1081         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1082         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1083         entry = xfs_attr3_leaf_entryp(leaf);
1084
1085         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1086         memset(bp->b_addr, 0, args->geo->blksize);
1087
1088         /*
1089          * Clean out the prior contents of the attribute list.
1090          */
1091         error = xfs_da_shrink_inode(args, 0, bp);
1092         if (error)
1093                 goto out;
1094
1095         if (forkoff == -1) {
1096                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1097                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1098                 xfs_attr_fork_remove(dp, args->trans);
1099                 goto out;
1100         }
1101
1102         xfs_attr_shortform_create(args);
1103
1104         /*
1105          * Copy the attributes
1106          */
1107         memset((char *)&nargs, 0, sizeof(nargs));
1108         nargs.geo = args->geo;
1109         nargs.dp = dp;
1110         nargs.total = args->total;
1111         nargs.whichfork = XFS_ATTR_FORK;
1112         nargs.trans = args->trans;
1113         nargs.op_flags = XFS_DA_OP_OKNOENT;
1114
1115         for (i = 0; i < ichdr.count; entry++, i++) {
1116                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1117                         continue;       /* don't copy partial entries */
1118                 if (!entry->nameidx)
1119                         continue;
1120                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1121                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1122                 nargs.name = name_loc->nameval;
1123                 nargs.namelen = name_loc->namelen;
1124                 nargs.value = &name_loc->nameval[nargs.namelen];
1125                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1126                 nargs.hashval = be32_to_cpu(entry->hashval);
1127                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1128                 xfs_attr_shortform_add(&nargs, forkoff);
1129         }
1130         error = 0;
1131
1132 out:
1133         kmem_free(tmpbuffer);
1134         return error;
1135 }
1136
1137 /*
1138  * Convert from using a single leaf to a root node and a leaf.
1139  */
1140 int
1141 xfs_attr3_leaf_to_node(
1142         struct xfs_da_args      *args)
1143 {
1144         struct xfs_attr_leafblock *leaf;
1145         struct xfs_attr3_icleaf_hdr icleafhdr;
1146         struct xfs_attr_leaf_entry *entries;
1147         struct xfs_da3_icnode_hdr icnodehdr;
1148         struct xfs_da_intnode   *node;
1149         struct xfs_inode        *dp = args->dp;
1150         struct xfs_mount        *mp = dp->i_mount;
1151         struct xfs_buf          *bp1 = NULL;
1152         struct xfs_buf          *bp2 = NULL;
1153         xfs_dablk_t             blkno;
1154         int                     error;
1155
1156         trace_xfs_attr_leaf_to_node(args);
1157
1158         error = xfs_da_grow_inode(args, &blkno);
1159         if (error)
1160                 goto out;
1161         error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1162         if (error)
1163                 goto out;
1164
1165         error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1166         if (error)
1167                 goto out;
1168
1169         /* copy leaf to new buffer, update identifiers */
1170         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1171         bp2->b_ops = bp1->b_ops;
1172         memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1173         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1174                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1175                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1176         }
1177         xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1178
1179         /*
1180          * Set up the new root node.
1181          */
1182         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1183         if (error)
1184                 goto out;
1185         node = bp1->b_addr;
1186         xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1187
1188         leaf = bp2->b_addr;
1189         xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1190         entries = xfs_attr3_leaf_entryp(leaf);
1191
1192         /* both on-disk, don't endian-flip twice */
1193         icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1194         icnodehdr.btree[0].before = cpu_to_be32(blkno);
1195         icnodehdr.count = 1;
1196         xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1197         xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1198         error = 0;
1199 out:
1200         return error;
1201 }
1202
1203 /*========================================================================
1204  * Routines used for growing the Btree.
1205  *========================================================================*/
1206
1207 /*
1208  * Create the initial contents of a leaf attribute list
1209  * or a leaf in a node attribute list.
1210  */
1211 STATIC int
1212 xfs_attr3_leaf_create(
1213         struct xfs_da_args      *args,
1214         xfs_dablk_t             blkno,
1215         struct xfs_buf          **bpp)
1216 {
1217         struct xfs_attr_leafblock *leaf;
1218         struct xfs_attr3_icleaf_hdr ichdr;
1219         struct xfs_inode        *dp = args->dp;
1220         struct xfs_mount        *mp = dp->i_mount;
1221         struct xfs_buf          *bp;
1222         int                     error;
1223
1224         trace_xfs_attr_leaf_create(args);
1225
1226         error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1227                                             XFS_ATTR_FORK);
1228         if (error)
1229                 return error;
1230         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1231         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1232         leaf = bp->b_addr;
1233         memset(leaf, 0, args->geo->blksize);
1234
1235         memset(&ichdr, 0, sizeof(ichdr));
1236         ichdr.firstused = args->geo->blksize;
1237
1238         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1239                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1240
1241                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1242
1243                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1244                 hdr3->owner = cpu_to_be64(dp->i_ino);
1245                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1246
1247                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1248         } else {
1249                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1250                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1251         }
1252         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1253
1254         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1255         xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1256
1257         *bpp = bp;
1258         return 0;
1259 }
1260
1261 /*
1262  * Split the leaf node, rebalance, then add the new entry.
1263  */
1264 int
1265 xfs_attr3_leaf_split(
1266         struct xfs_da_state     *state,
1267         struct xfs_da_state_blk *oldblk,
1268         struct xfs_da_state_blk *newblk)
1269 {
1270         xfs_dablk_t blkno;
1271         int error;
1272
1273         trace_xfs_attr_leaf_split(state->args);
1274
1275         /*
1276          * Allocate space for a new leaf node.
1277          */
1278         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1279         error = xfs_da_grow_inode(state->args, &blkno);
1280         if (error)
1281                 return error;
1282         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1283         if (error)
1284                 return error;
1285         newblk->blkno = blkno;
1286         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1287
1288         /*
1289          * Rebalance the entries across the two leaves.
1290          * NOTE: rebalance() currently depends on the 2nd block being empty.
1291          */
1292         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1293         error = xfs_da3_blk_link(state, oldblk, newblk);
1294         if (error)
1295                 return error;
1296
1297         /*
1298          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1299          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1300          * "new" attrs info.  Will need the "old" info to remove it later.
1301          *
1302          * Insert the "new" entry in the correct block.
1303          */
1304         if (state->inleaf) {
1305                 trace_xfs_attr_leaf_add_old(state->args);
1306                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1307         } else {
1308                 trace_xfs_attr_leaf_add_new(state->args);
1309                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1310         }
1311
1312         /*
1313          * Update last hashval in each block since we added the name.
1314          */
1315         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1316         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1317         return error;
1318 }
1319
1320 /*
1321  * Add a name to the leaf attribute list structure.
1322  */
1323 int
1324 xfs_attr3_leaf_add(
1325         struct xfs_buf          *bp,
1326         struct xfs_da_args      *args)
1327 {
1328         struct xfs_attr_leafblock *leaf;
1329         struct xfs_attr3_icleaf_hdr ichdr;
1330         int                     tablesize;
1331         int                     entsize;
1332         int                     sum;
1333         int                     tmp;
1334         int                     i;
1335
1336         trace_xfs_attr_leaf_add(args);
1337
1338         leaf = bp->b_addr;
1339         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1340         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1341         entsize = xfs_attr_leaf_newentsize(args, NULL);
1342
1343         /*
1344          * Search through freemap for first-fit on new name length.
1345          * (may need to figure in size of entry struct too)
1346          */
1347         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1348                                         + xfs_attr3_leaf_hdr_size(leaf);
1349         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1350                 if (tablesize > ichdr.firstused) {
1351                         sum += ichdr.freemap[i].size;
1352                         continue;
1353                 }
1354                 if (!ichdr.freemap[i].size)
1355                         continue;       /* no space in this map */
1356                 tmp = entsize;
1357                 if (ichdr.freemap[i].base < ichdr.firstused)
1358                         tmp += sizeof(xfs_attr_leaf_entry_t);
1359                 if (ichdr.freemap[i].size >= tmp) {
1360                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1361                         goto out_log_hdr;
1362                 }
1363                 sum += ichdr.freemap[i].size;
1364         }
1365
1366         /*
1367          * If there are no holes in the address space of the block,
1368          * and we don't have enough freespace, then compaction will do us
1369          * no good and we should just give up.
1370          */
1371         if (!ichdr.holes && sum < entsize)
1372                 return -ENOSPC;
1373
1374         /*
1375          * Compact the entries to coalesce free space.
1376          * This may change the hdr->count via dropping INCOMPLETE entries.
1377          */
1378         xfs_attr3_leaf_compact(args, &ichdr, bp);
1379
1380         /*
1381          * After compaction, the block is guaranteed to have only one
1382          * free region, in freemap[0].  If it is not big enough, give up.
1383          */
1384         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1385                 tmp = -ENOSPC;
1386                 goto out_log_hdr;
1387         }
1388
1389         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1390
1391 out_log_hdr:
1392         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1393         xfs_trans_log_buf(args->trans, bp,
1394                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1395                                 xfs_attr3_leaf_hdr_size(leaf)));
1396         return tmp;
1397 }
1398
1399 /*
1400  * Add a name to a leaf attribute list structure.
1401  */
1402 STATIC int
1403 xfs_attr3_leaf_add_work(
1404         struct xfs_buf          *bp,
1405         struct xfs_attr3_icleaf_hdr *ichdr,
1406         struct xfs_da_args      *args,
1407         int                     mapindex)
1408 {
1409         struct xfs_attr_leafblock *leaf;
1410         struct xfs_attr_leaf_entry *entry;
1411         struct xfs_attr_leaf_name_local *name_loc;
1412         struct xfs_attr_leaf_name_remote *name_rmt;
1413         struct xfs_mount        *mp;
1414         int                     tmp;
1415         int                     i;
1416
1417         trace_xfs_attr_leaf_add_work(args);
1418
1419         leaf = bp->b_addr;
1420         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1421         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1422
1423         /*
1424          * Force open some space in the entry array and fill it in.
1425          */
1426         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1427         if (args->index < ichdr->count) {
1428                 tmp  = ichdr->count - args->index;
1429                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1430                 memmove(entry + 1, entry, tmp);
1431                 xfs_trans_log_buf(args->trans, bp,
1432                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1433         }
1434         ichdr->count++;
1435
1436         /*
1437          * Allocate space for the new string (at the end of the run).
1438          */
1439         mp = args->trans->t_mountp;
1440         ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1441         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1442         ASSERT(ichdr->freemap[mapindex].size >=
1443                 xfs_attr_leaf_newentsize(args, NULL));
1444         ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1445         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1446
1447         ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1448
1449         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1450                                      ichdr->freemap[mapindex].size);
1451         entry->hashval = cpu_to_be32(args->hashval);
1452         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1453         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1454         if (args->op_flags & XFS_DA_OP_RENAME) {
1455                 entry->flags |= XFS_ATTR_INCOMPLETE;
1456                 if ((args->blkno2 == args->blkno) &&
1457                     (args->index2 <= args->index)) {
1458                         args->index2++;
1459                 }
1460         }
1461         xfs_trans_log_buf(args->trans, bp,
1462                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1463         ASSERT((args->index == 0) ||
1464                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1465         ASSERT((args->index == ichdr->count - 1) ||
1466                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1467
1468         /*
1469          * For "remote" attribute values, simply note that we need to
1470          * allocate space for the "remote" value.  We can't actually
1471          * allocate the extents in this transaction, and we can't decide
1472          * which blocks they should be as we might allocate more blocks
1473          * as part of this transaction (a split operation for example).
1474          */
1475         if (entry->flags & XFS_ATTR_LOCAL) {
1476                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1477                 name_loc->namelen = args->namelen;
1478                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1479                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1480                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1481                                    be16_to_cpu(name_loc->valuelen));
1482         } else {
1483                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1484                 name_rmt->namelen = args->namelen;
1485                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1486                 entry->flags |= XFS_ATTR_INCOMPLETE;
1487                 /* just in case */
1488                 name_rmt->valuelen = 0;
1489                 name_rmt->valueblk = 0;
1490                 args->rmtblkno = 1;
1491                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1492                 args->rmtvaluelen = args->valuelen;
1493         }
1494         xfs_trans_log_buf(args->trans, bp,
1495              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1496                                    xfs_attr_leaf_entsize(leaf, args->index)));
1497
1498         /*
1499          * Update the control info for this leaf node
1500          */
1501         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1502                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1503
1504         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1505                                         + xfs_attr3_leaf_hdr_size(leaf));
1506         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1507                                         + xfs_attr3_leaf_hdr_size(leaf);
1508
1509         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1510                 if (ichdr->freemap[i].base == tmp) {
1511                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1512                         ichdr->freemap[i].size -=
1513                                 min_t(uint16_t, ichdr->freemap[i].size,
1514                                                 sizeof(xfs_attr_leaf_entry_t));
1515                 }
1516         }
1517         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1518         return 0;
1519 }
1520
1521 /*
1522  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1523  */
1524 STATIC void
1525 xfs_attr3_leaf_compact(
1526         struct xfs_da_args      *args,
1527         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1528         struct xfs_buf          *bp)
1529 {
1530         struct xfs_attr_leafblock *leaf_src;
1531         struct xfs_attr_leafblock *leaf_dst;
1532         struct xfs_attr3_icleaf_hdr ichdr_src;
1533         struct xfs_trans        *trans = args->trans;
1534         char                    *tmpbuffer;
1535
1536         trace_xfs_attr_leaf_compact(args);
1537
1538         tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1539         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1540         memset(bp->b_addr, 0, args->geo->blksize);
1541         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1542         leaf_dst = bp->b_addr;
1543
1544         /*
1545          * Copy the on-disk header back into the destination buffer to ensure
1546          * all the information in the header that is not part of the incore
1547          * header structure is preserved.
1548          */
1549         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1550
1551         /* Initialise the incore headers */
1552         ichdr_src = *ichdr_dst; /* struct copy */
1553         ichdr_dst->firstused = args->geo->blksize;
1554         ichdr_dst->usedbytes = 0;
1555         ichdr_dst->count = 0;
1556         ichdr_dst->holes = 0;
1557         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1558         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1559                                                 ichdr_dst->freemap[0].base;
1560
1561         /* write the header back to initialise the underlying buffer */
1562         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1563
1564         /*
1565          * Copy all entry's in the same (sorted) order,
1566          * but allocate name/value pairs packed and in sequence.
1567          */
1568         xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1569                                 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1570         /*
1571          * this logs the entire buffer, but the caller must write the header
1572          * back to the buffer when it is finished modifying it.
1573          */
1574         xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1575
1576         kmem_free(tmpbuffer);
1577 }
1578
1579 /*
1580  * Compare two leaf blocks "order".
1581  * Return 0 unless leaf2 should go before leaf1.
1582  */
1583 static int
1584 xfs_attr3_leaf_order(
1585         struct xfs_buf  *leaf1_bp,
1586         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1587         struct xfs_buf  *leaf2_bp,
1588         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1589 {
1590         struct xfs_attr_leaf_entry *entries1;
1591         struct xfs_attr_leaf_entry *entries2;
1592
1593         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1594         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1595         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1596             ((be32_to_cpu(entries2[0].hashval) <
1597               be32_to_cpu(entries1[0].hashval)) ||
1598              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1599               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1600                 return 1;
1601         }
1602         return 0;
1603 }
1604
1605 int
1606 xfs_attr_leaf_order(
1607         struct xfs_buf  *leaf1_bp,
1608         struct xfs_buf  *leaf2_bp)
1609 {
1610         struct xfs_attr3_icleaf_hdr ichdr1;
1611         struct xfs_attr3_icleaf_hdr ichdr2;
1612         struct xfs_mount *mp = leaf1_bp->b_mount;
1613
1614         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1615         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1616         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1617 }
1618
1619 /*
1620  * Redistribute the attribute list entries between two leaf nodes,
1621  * taking into account the size of the new entry.
1622  *
1623  * NOTE: if new block is empty, then it will get the upper half of the
1624  * old block.  At present, all (one) callers pass in an empty second block.
1625  *
1626  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1627  * to match what it is doing in splitting the attribute leaf block.  Those
1628  * values are used in "atomic rename" operations on attributes.  Note that
1629  * the "new" and "old" values can end up in different blocks.
1630  */
1631 STATIC void
1632 xfs_attr3_leaf_rebalance(
1633         struct xfs_da_state     *state,
1634         struct xfs_da_state_blk *blk1,
1635         struct xfs_da_state_blk *blk2)
1636 {
1637         struct xfs_da_args      *args;
1638         struct xfs_attr_leafblock *leaf1;
1639         struct xfs_attr_leafblock *leaf2;
1640         struct xfs_attr3_icleaf_hdr ichdr1;
1641         struct xfs_attr3_icleaf_hdr ichdr2;
1642         struct xfs_attr_leaf_entry *entries1;
1643         struct xfs_attr_leaf_entry *entries2;
1644         int                     count;
1645         int                     totallen;
1646         int                     max;
1647         int                     space;
1648         int                     swap;
1649
1650         /*
1651          * Set up environment.
1652          */
1653         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1654         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1655         leaf1 = blk1->bp->b_addr;
1656         leaf2 = blk2->bp->b_addr;
1657         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1658         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1659         ASSERT(ichdr2.count == 0);
1660         args = state->args;
1661
1662         trace_xfs_attr_leaf_rebalance(args);
1663
1664         /*
1665          * Check ordering of blocks, reverse if it makes things simpler.
1666          *
1667          * NOTE: Given that all (current) callers pass in an empty
1668          * second block, this code should never set "swap".
1669          */
1670         swap = 0;
1671         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1672                 swap(blk1, blk2);
1673
1674                 /* swap structures rather than reconverting them */
1675                 swap(ichdr1, ichdr2);
1676
1677                 leaf1 = blk1->bp->b_addr;
1678                 leaf2 = blk2->bp->b_addr;
1679                 swap = 1;
1680         }
1681
1682         /*
1683          * Examine entries until we reduce the absolute difference in
1684          * byte usage between the two blocks to a minimum.  Then get
1685          * the direction to copy and the number of elements to move.
1686          *
1687          * "inleaf" is true if the new entry should be inserted into blk1.
1688          * If "swap" is also true, then reverse the sense of "inleaf".
1689          */
1690         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1691                                                       blk2, &ichdr2,
1692                                                       &count, &totallen);
1693         if (swap)
1694                 state->inleaf = !state->inleaf;
1695
1696         /*
1697          * Move any entries required from leaf to leaf:
1698          */
1699         if (count < ichdr1.count) {
1700                 /*
1701                  * Figure the total bytes to be added to the destination leaf.
1702                  */
1703                 /* number entries being moved */
1704                 count = ichdr1.count - count;
1705                 space  = ichdr1.usedbytes - totallen;
1706                 space += count * sizeof(xfs_attr_leaf_entry_t);
1707
1708                 /*
1709                  * leaf2 is the destination, compact it if it looks tight.
1710                  */
1711                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1712                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1713                 if (space > max)
1714                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1715
1716                 /*
1717                  * Move high entries from leaf1 to low end of leaf2.
1718                  */
1719                 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1720                                 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1721
1722         } else if (count > ichdr1.count) {
1723                 /*
1724                  * I assert that since all callers pass in an empty
1725                  * second buffer, this code should never execute.
1726                  */
1727                 ASSERT(0);
1728
1729                 /*
1730                  * Figure the total bytes to be added to the destination leaf.
1731                  */
1732                 /* number entries being moved */
1733                 count -= ichdr1.count;
1734                 space  = totallen - ichdr1.usedbytes;
1735                 space += count * sizeof(xfs_attr_leaf_entry_t);
1736
1737                 /*
1738                  * leaf1 is the destination, compact it if it looks tight.
1739                  */
1740                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1741                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1742                 if (space > max)
1743                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1744
1745                 /*
1746                  * Move low entries from leaf2 to high end of leaf1.
1747                  */
1748                 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1749                                         ichdr1.count, count);
1750         }
1751
1752         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1753         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1754         xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1755         xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1756
1757         /*
1758          * Copy out last hashval in each block for B-tree code.
1759          */
1760         entries1 = xfs_attr3_leaf_entryp(leaf1);
1761         entries2 = xfs_attr3_leaf_entryp(leaf2);
1762         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1763         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1764
1765         /*
1766          * Adjust the expected index for insertion.
1767          * NOTE: this code depends on the (current) situation that the
1768          * second block was originally empty.
1769          *
1770          * If the insertion point moved to the 2nd block, we must adjust
1771          * the index.  We must also track the entry just following the
1772          * new entry for use in an "atomic rename" operation, that entry
1773          * is always the "old" entry and the "new" entry is what we are
1774          * inserting.  The index/blkno fields refer to the "old" entry,
1775          * while the index2/blkno2 fields refer to the "new" entry.
1776          */
1777         if (blk1->index > ichdr1.count) {
1778                 ASSERT(state->inleaf == 0);
1779                 blk2->index = blk1->index - ichdr1.count;
1780                 args->index = args->index2 = blk2->index;
1781                 args->blkno = args->blkno2 = blk2->blkno;
1782         } else if (blk1->index == ichdr1.count) {
1783                 if (state->inleaf) {
1784                         args->index = blk1->index;
1785                         args->blkno = blk1->blkno;
1786                         args->index2 = 0;
1787                         args->blkno2 = blk2->blkno;
1788                 } else {
1789                         /*
1790                          * On a double leaf split, the original attr location
1791                          * is already stored in blkno2/index2, so don't
1792                          * overwrite it overwise we corrupt the tree.
1793                          */
1794                         blk2->index = blk1->index - ichdr1.count;
1795                         args->index = blk2->index;
1796                         args->blkno = blk2->blkno;
1797                         if (!state->extravalid) {
1798                                 /*
1799                                  * set the new attr location to match the old
1800                                  * one and let the higher level split code
1801                                  * decide where in the leaf to place it.
1802                                  */
1803                                 args->index2 = blk2->index;
1804                                 args->blkno2 = blk2->blkno;
1805                         }
1806                 }
1807         } else {
1808                 ASSERT(state->inleaf == 1);
1809                 args->index = args->index2 = blk1->index;
1810                 args->blkno = args->blkno2 = blk1->blkno;
1811         }
1812 }
1813
1814 /*
1815  * Examine entries until we reduce the absolute difference in
1816  * byte usage between the two blocks to a minimum.
1817  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1818  * GROT: there will always be enough room in either block for a new entry.
1819  * GROT: Do a double-split for this case?
1820  */
1821 STATIC int
1822 xfs_attr3_leaf_figure_balance(
1823         struct xfs_da_state             *state,
1824         struct xfs_da_state_blk         *blk1,
1825         struct xfs_attr3_icleaf_hdr     *ichdr1,
1826         struct xfs_da_state_blk         *blk2,
1827         struct xfs_attr3_icleaf_hdr     *ichdr2,
1828         int                             *countarg,
1829         int                             *usedbytesarg)
1830 {
1831         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1832         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1833         struct xfs_attr_leaf_entry      *entry;
1834         int                             count;
1835         int                             max;
1836         int                             index;
1837         int                             totallen = 0;
1838         int                             half;
1839         int                             lastdelta;
1840         int                             foundit = 0;
1841         int                             tmp;
1842
1843         /*
1844          * Examine entries until we reduce the absolute difference in
1845          * byte usage between the two blocks to a minimum.
1846          */
1847         max = ichdr1->count + ichdr2->count;
1848         half = (max + 1) * sizeof(*entry);
1849         half += ichdr1->usedbytes + ichdr2->usedbytes +
1850                         xfs_attr_leaf_newentsize(state->args, NULL);
1851         half /= 2;
1852         lastdelta = state->args->geo->blksize;
1853         entry = xfs_attr3_leaf_entryp(leaf1);
1854         for (count = index = 0; count < max; entry++, index++, count++) {
1855
1856 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1857                 /*
1858                  * The new entry is in the first block, account for it.
1859                  */
1860                 if (count == blk1->index) {
1861                         tmp = totallen + sizeof(*entry) +
1862                                 xfs_attr_leaf_newentsize(state->args, NULL);
1863                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1864                                 break;
1865                         lastdelta = XFS_ATTR_ABS(half - tmp);
1866                         totallen = tmp;
1867                         foundit = 1;
1868                 }
1869
1870                 /*
1871                  * Wrap around into the second block if necessary.
1872                  */
1873                 if (count == ichdr1->count) {
1874                         leaf1 = leaf2;
1875                         entry = xfs_attr3_leaf_entryp(leaf1);
1876                         index = 0;
1877                 }
1878
1879                 /*
1880                  * Figure out if next leaf entry would be too much.
1881                  */
1882                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1883                                                                         index);
1884                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1885                         break;
1886                 lastdelta = XFS_ATTR_ABS(half - tmp);
1887                 totallen = tmp;
1888 #undef XFS_ATTR_ABS
1889         }
1890
1891         /*
1892          * Calculate the number of usedbytes that will end up in lower block.
1893          * If new entry not in lower block, fix up the count.
1894          */
1895         totallen -= count * sizeof(*entry);
1896         if (foundit) {
1897                 totallen -= sizeof(*entry) +
1898                                 xfs_attr_leaf_newentsize(state->args, NULL);
1899         }
1900
1901         *countarg = count;
1902         *usedbytesarg = totallen;
1903         return foundit;
1904 }
1905
1906 /*========================================================================
1907  * Routines used for shrinking the Btree.
1908  *========================================================================*/
1909
1910 /*
1911  * Check a leaf block and its neighbors to see if the block should be
1912  * collapsed into one or the other neighbor.  Always keep the block
1913  * with the smaller block number.
1914  * If the current block is over 50% full, don't try to join it, return 0.
1915  * If the block is empty, fill in the state structure and return 2.
1916  * If it can be collapsed, fill in the state structure and return 1.
1917  * If nothing can be done, return 0.
1918  *
1919  * GROT: allow for INCOMPLETE entries in calculation.
1920  */
1921 int
1922 xfs_attr3_leaf_toosmall(
1923         struct xfs_da_state     *state,
1924         int                     *action)
1925 {
1926         struct xfs_attr_leafblock *leaf;
1927         struct xfs_da_state_blk *blk;
1928         struct xfs_attr3_icleaf_hdr ichdr;
1929         struct xfs_buf          *bp;
1930         xfs_dablk_t             blkno;
1931         int                     bytes;
1932         int                     forward;
1933         int                     error;
1934         int                     retval;
1935         int                     i;
1936
1937         trace_xfs_attr_leaf_toosmall(state->args);
1938
1939         /*
1940          * Check for the degenerate case of the block being over 50% full.
1941          * If so, it's not worth even looking to see if we might be able
1942          * to coalesce with a sibling.
1943          */
1944         blk = &state->path.blk[ state->path.active-1 ];
1945         leaf = blk->bp->b_addr;
1946         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1947         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1948                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1949                 ichdr.usedbytes;
1950         if (bytes > (state->args->geo->blksize >> 1)) {
1951                 *action = 0;    /* blk over 50%, don't try to join */
1952                 return 0;
1953         }
1954
1955         /*
1956          * Check for the degenerate case of the block being empty.
1957          * If the block is empty, we'll simply delete it, no need to
1958          * coalesce it with a sibling block.  We choose (arbitrarily)
1959          * to merge with the forward block unless it is NULL.
1960          */
1961         if (ichdr.count == 0) {
1962                 /*
1963                  * Make altpath point to the block we want to keep and
1964                  * path point to the block we want to drop (this one).
1965                  */
1966                 forward = (ichdr.forw != 0);
1967                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1968                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1969                                                  0, &retval);
1970                 if (error)
1971                         return error;
1972                 if (retval) {
1973                         *action = 0;
1974                 } else {
1975                         *action = 2;
1976                 }
1977                 return 0;
1978         }
1979
1980         /*
1981          * Examine each sibling block to see if we can coalesce with
1982          * at least 25% free space to spare.  We need to figure out
1983          * whether to merge with the forward or the backward block.
1984          * We prefer coalescing with the lower numbered sibling so as
1985          * to shrink an attribute list over time.
1986          */
1987         /* start with smaller blk num */
1988         forward = ichdr.forw < ichdr.back;
1989         for (i = 0; i < 2; forward = !forward, i++) {
1990                 struct xfs_attr3_icleaf_hdr ichdr2;
1991                 if (forward)
1992                         blkno = ichdr.forw;
1993                 else
1994                         blkno = ichdr.back;
1995                 if (blkno == 0)
1996                         continue;
1997                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1998                                         blkno, &bp);
1999                 if (error)
2000                         return error;
2001
2002                 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
2003
2004                 bytes = state->args->geo->blksize -
2005                         (state->args->geo->blksize >> 2) -
2006                         ichdr.usedbytes - ichdr2.usedbytes -
2007                         ((ichdr.count + ichdr2.count) *
2008                                         sizeof(xfs_attr_leaf_entry_t)) -
2009                         xfs_attr3_leaf_hdr_size(leaf);
2010
2011                 xfs_trans_brelse(state->args->trans, bp);
2012                 if (bytes >= 0)
2013                         break;  /* fits with at least 25% to spare */
2014         }
2015         if (i >= 2) {
2016                 *action = 0;
2017                 return 0;
2018         }
2019
2020         /*
2021          * Make altpath point to the block we want to keep (the lower
2022          * numbered block) and path point to the block we want to drop.
2023          */
2024         memcpy(&state->altpath, &state->path, sizeof(state->path));
2025         if (blkno < blk->blkno) {
2026                 error = xfs_da3_path_shift(state, &state->altpath, forward,
2027                                                  0, &retval);
2028         } else {
2029                 error = xfs_da3_path_shift(state, &state->path, forward,
2030                                                  0, &retval);
2031         }
2032         if (error)
2033                 return error;
2034         if (retval) {
2035                 *action = 0;
2036         } else {
2037                 *action = 1;
2038         }
2039         return 0;
2040 }
2041
2042 /*
2043  * Remove a name from the leaf attribute list structure.
2044  *
2045  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2046  * If two leaves are 37% full, when combined they will leave 25% free.
2047  */
2048 int
2049 xfs_attr3_leaf_remove(
2050         struct xfs_buf          *bp,
2051         struct xfs_da_args      *args)
2052 {
2053         struct xfs_attr_leafblock *leaf;
2054         struct xfs_attr3_icleaf_hdr ichdr;
2055         struct xfs_attr_leaf_entry *entry;
2056         int                     before;
2057         int                     after;
2058         int                     smallest;
2059         int                     entsize;
2060         int                     tablesize;
2061         int                     tmp;
2062         int                     i;
2063
2064         trace_xfs_attr_leaf_remove(args);
2065
2066         leaf = bp->b_addr;
2067         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2068
2069         ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2070         ASSERT(args->index >= 0 && args->index < ichdr.count);
2071         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2072                                         xfs_attr3_leaf_hdr_size(leaf));
2073
2074         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2075
2076         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2077         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2078
2079         /*
2080          * Scan through free region table:
2081          *    check for adjacency of free'd entry with an existing one,
2082          *    find smallest free region in case we need to replace it,
2083          *    adjust any map that borders the entry table,
2084          */
2085         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2086                                         + xfs_attr3_leaf_hdr_size(leaf);
2087         tmp = ichdr.freemap[0].size;
2088         before = after = -1;
2089         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2090         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2091         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2092                 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2093                 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2094                 if (ichdr.freemap[i].base == tablesize) {
2095                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2096                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2097                 }
2098
2099                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2100                                 be16_to_cpu(entry->nameidx)) {
2101                         before = i;
2102                 } else if (ichdr.freemap[i].base ==
2103                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2104                         after = i;
2105                 } else if (ichdr.freemap[i].size < tmp) {
2106                         tmp = ichdr.freemap[i].size;
2107                         smallest = i;
2108                 }
2109         }
2110
2111         /*
2112          * Coalesce adjacent freemap regions,
2113          * or replace the smallest region.
2114          */
2115         if ((before >= 0) || (after >= 0)) {
2116                 if ((before >= 0) && (after >= 0)) {
2117                         ichdr.freemap[before].size += entsize;
2118                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2119                         ichdr.freemap[after].base = 0;
2120                         ichdr.freemap[after].size = 0;
2121                 } else if (before >= 0) {
2122                         ichdr.freemap[before].size += entsize;
2123                 } else {
2124                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2125                         ichdr.freemap[after].size += entsize;
2126                 }
2127         } else {
2128                 /*
2129                  * Replace smallest region (if it is smaller than free'd entry)
2130                  */
2131                 if (ichdr.freemap[smallest].size < entsize) {
2132                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2133                         ichdr.freemap[smallest].size = entsize;
2134                 }
2135         }
2136
2137         /*
2138          * Did we remove the first entry?
2139          */
2140         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2141                 smallest = 1;
2142         else
2143                 smallest = 0;
2144
2145         /*
2146          * Compress the remaining entries and zero out the removed stuff.
2147          */
2148         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2149         ichdr.usedbytes -= entsize;
2150         xfs_trans_log_buf(args->trans, bp,
2151              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2152                                    entsize));
2153
2154         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2155         memmove(entry, entry + 1, tmp);
2156         ichdr.count--;
2157         xfs_trans_log_buf(args->trans, bp,
2158             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2159
2160         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2161         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2162
2163         /*
2164          * If we removed the first entry, re-find the first used byte
2165          * in the name area.  Note that if the entry was the "firstused",
2166          * then we don't have a "hole" in our block resulting from
2167          * removing the name.
2168          */
2169         if (smallest) {
2170                 tmp = args->geo->blksize;
2171                 entry = xfs_attr3_leaf_entryp(leaf);
2172                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2173                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2174                         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2175
2176                         if (be16_to_cpu(entry->nameidx) < tmp)
2177                                 tmp = be16_to_cpu(entry->nameidx);
2178                 }
2179                 ichdr.firstused = tmp;
2180                 ASSERT(ichdr.firstused != 0);
2181         } else {
2182                 ichdr.holes = 1;        /* mark as needing compaction */
2183         }
2184         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2185         xfs_trans_log_buf(args->trans, bp,
2186                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2187                                           xfs_attr3_leaf_hdr_size(leaf)));
2188
2189         /*
2190          * Check if leaf is less than 50% full, caller may want to
2191          * "join" the leaf with a sibling if so.
2192          */
2193         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2194               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2195
2196         return tmp < args->geo->magicpct; /* leaf is < 37% full */
2197 }
2198
2199 /*
2200  * Move all the attribute list entries from drop_leaf into save_leaf.
2201  */
2202 void
2203 xfs_attr3_leaf_unbalance(
2204         struct xfs_da_state     *state,
2205         struct xfs_da_state_blk *drop_blk,
2206         struct xfs_da_state_blk *save_blk)
2207 {
2208         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2209         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2210         struct xfs_attr3_icleaf_hdr drophdr;
2211         struct xfs_attr3_icleaf_hdr savehdr;
2212         struct xfs_attr_leaf_entry *entry;
2213
2214         trace_xfs_attr_leaf_unbalance(state->args);
2215
2216         drop_leaf = drop_blk->bp->b_addr;
2217         save_leaf = save_blk->bp->b_addr;
2218         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2219         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2220         entry = xfs_attr3_leaf_entryp(drop_leaf);
2221
2222         /*
2223          * Save last hashval from dying block for later Btree fixup.
2224          */
2225         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2226
2227         /*
2228          * Check if we need a temp buffer, or can we do it in place.
2229          * Note that we don't check "leaf" for holes because we will
2230          * always be dropping it, toosmall() decided that for us already.
2231          */
2232         if (savehdr.holes == 0) {
2233                 /*
2234                  * dest leaf has no holes, so we add there.  May need
2235                  * to make some room in the entry array.
2236                  */
2237                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2238                                          drop_blk->bp, &drophdr)) {
2239                         xfs_attr3_leaf_moveents(state->args,
2240                                                 drop_leaf, &drophdr, 0,
2241                                                 save_leaf, &savehdr, 0,
2242                                                 drophdr.count);
2243                 } else {
2244                         xfs_attr3_leaf_moveents(state->args,
2245                                                 drop_leaf, &drophdr, 0,
2246                                                 save_leaf, &savehdr,
2247                                                 savehdr.count, drophdr.count);
2248                 }
2249         } else {
2250                 /*
2251                  * Destination has holes, so we make a temporary copy
2252                  * of the leaf and add them both to that.
2253                  */
2254                 struct xfs_attr_leafblock *tmp_leaf;
2255                 struct xfs_attr3_icleaf_hdr tmphdr;
2256
2257                 tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2258
2259                 /*
2260                  * Copy the header into the temp leaf so that all the stuff
2261                  * not in the incore header is present and gets copied back in
2262                  * once we've moved all the entries.
2263                  */
2264                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2265
2266                 memset(&tmphdr, 0, sizeof(tmphdr));
2267                 tmphdr.magic = savehdr.magic;
2268                 tmphdr.forw = savehdr.forw;
2269                 tmphdr.back = savehdr.back;
2270                 tmphdr.firstused = state->args->geo->blksize;
2271
2272                 /* write the header to the temp buffer to initialise it */
2273                 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2274
2275                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2276                                          drop_blk->bp, &drophdr)) {
2277                         xfs_attr3_leaf_moveents(state->args,
2278                                                 drop_leaf, &drophdr, 0,
2279                                                 tmp_leaf, &tmphdr, 0,
2280                                                 drophdr.count);
2281                         xfs_attr3_leaf_moveents(state->args,
2282                                                 save_leaf, &savehdr, 0,
2283                                                 tmp_leaf, &tmphdr, tmphdr.count,
2284                                                 savehdr.count);
2285                 } else {
2286                         xfs_attr3_leaf_moveents(state->args,
2287                                                 save_leaf, &savehdr, 0,
2288                                                 tmp_leaf, &tmphdr, 0,
2289                                                 savehdr.count);
2290                         xfs_attr3_leaf_moveents(state->args,
2291                                                 drop_leaf, &drophdr, 0,
2292                                                 tmp_leaf, &tmphdr, tmphdr.count,
2293                                                 drophdr.count);
2294                 }
2295                 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2296                 savehdr = tmphdr; /* struct copy */
2297                 kmem_free(tmp_leaf);
2298         }
2299
2300         xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2301         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2302                                            state->args->geo->blksize - 1);
2303
2304         /*
2305          * Copy out last hashval in each block for B-tree code.
2306          */
2307         entry = xfs_attr3_leaf_entryp(save_leaf);
2308         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2309 }
2310
2311 /*========================================================================
2312  * Routines used for finding things in the Btree.
2313  *========================================================================*/
2314
2315 /*
2316  * Look up a name in a leaf attribute list structure.
2317  * This is the internal routine, it uses the caller's buffer.
2318  *
2319  * Note that duplicate keys are allowed, but only check within the
2320  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2321  *
2322  * Return in args->index the index into the entry[] array of either
2323  * the found entry, or where the entry should have been (insert before
2324  * that entry).
2325  *
2326  * Don't change the args->value unless we find the attribute.
2327  */
2328 int
2329 xfs_attr3_leaf_lookup_int(
2330         struct xfs_buf          *bp,
2331         struct xfs_da_args      *args)
2332 {
2333         struct xfs_attr_leafblock *leaf;
2334         struct xfs_attr3_icleaf_hdr ichdr;
2335         struct xfs_attr_leaf_entry *entry;
2336         struct xfs_attr_leaf_entry *entries;
2337         struct xfs_attr_leaf_name_local *name_loc;
2338         struct xfs_attr_leaf_name_remote *name_rmt;
2339         xfs_dahash_t            hashval;
2340         int                     probe;
2341         int                     span;
2342
2343         trace_xfs_attr_leaf_lookup(args);
2344
2345         leaf = bp->b_addr;
2346         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2347         entries = xfs_attr3_leaf_entryp(leaf);
2348         if (ichdr.count >= args->geo->blksize / 8) {
2349                 xfs_buf_corruption_error(bp);
2350                 return -EFSCORRUPTED;
2351         }
2352
2353         /*
2354          * Binary search.  (note: small blocks will skip this loop)
2355          */
2356         hashval = args->hashval;
2357         probe = span = ichdr.count / 2;
2358         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2359                 span /= 2;
2360                 if (be32_to_cpu(entry->hashval) < hashval)
2361                         probe += span;
2362                 else if (be32_to_cpu(entry->hashval) > hashval)
2363                         probe -= span;
2364                 else
2365                         break;
2366         }
2367         if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2368                 xfs_buf_corruption_error(bp);
2369                 return -EFSCORRUPTED;
2370         }
2371         if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2372                 xfs_buf_corruption_error(bp);
2373                 return -EFSCORRUPTED;
2374         }
2375
2376         /*
2377          * Since we may have duplicate hashval's, find the first matching
2378          * hashval in the leaf.
2379          */
2380         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2381                 entry--;
2382                 probe--;
2383         }
2384         while (probe < ichdr.count &&
2385                be32_to_cpu(entry->hashval) < hashval) {
2386                 entry++;
2387                 probe++;
2388         }
2389         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2390                 args->index = probe;
2391                 return -ENOATTR;
2392         }
2393
2394         /*
2395          * Duplicate keys may be present, so search all of them for a match.
2396          */
2397         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2398                         entry++, probe++) {
2399 /*
2400  * GROT: Add code to remove incomplete entries.
2401  */
2402                 /*
2403                  * If we are looking for INCOMPLETE entries, show only those.
2404                  * If we are looking for complete entries, show only those.
2405                  */
2406                 if (!!(args->op_flags & XFS_DA_OP_INCOMPLETE) !=
2407                     !!(entry->flags & XFS_ATTR_INCOMPLETE)) {
2408                         continue;
2409                 }
2410                 if (entry->flags & XFS_ATTR_LOCAL) {
2411                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2412                         if (name_loc->namelen != args->namelen)
2413                                 continue;
2414                         if (memcmp(args->name, name_loc->nameval,
2415                                                         args->namelen) != 0)
2416                                 continue;
2417                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2418                                 continue;
2419                         args->index = probe;
2420                         return -EEXIST;
2421                 } else {
2422                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2423                         if (name_rmt->namelen != args->namelen)
2424                                 continue;
2425                         if (memcmp(args->name, name_rmt->name,
2426                                                         args->namelen) != 0)
2427                                 continue;
2428                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2429                                 continue;
2430                         args->index = probe;
2431                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2432                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2433                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2434                                                         args->dp->i_mount,
2435                                                         args->rmtvaluelen);
2436                         return -EEXIST;
2437                 }
2438         }
2439         args->index = probe;
2440         return -ENOATTR;
2441 }
2442
2443 /*
2444  * Get the value associated with an attribute name from a leaf attribute
2445  * list structure.
2446  *
2447  * If ATTR_KERNOVAL is specified, only the length needs to be returned.
2448  * Unlike a lookup, we only return an error if the attribute does not
2449  * exist or we can't retrieve the value.
2450  */
2451 int
2452 xfs_attr3_leaf_getvalue(
2453         struct xfs_buf          *bp,
2454         struct xfs_da_args      *args)
2455 {
2456         struct xfs_attr_leafblock *leaf;
2457         struct xfs_attr3_icleaf_hdr ichdr;
2458         struct xfs_attr_leaf_entry *entry;
2459         struct xfs_attr_leaf_name_local *name_loc;
2460         struct xfs_attr_leaf_name_remote *name_rmt;
2461
2462         leaf = bp->b_addr;
2463         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2464         ASSERT(ichdr.count < args->geo->blksize / 8);
2465         ASSERT(args->index < ichdr.count);
2466
2467         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2468         if (entry->flags & XFS_ATTR_LOCAL) {
2469                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2470                 ASSERT(name_loc->namelen == args->namelen);
2471                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2472                 return xfs_attr_copy_value(args,
2473                                         &name_loc->nameval[args->namelen],
2474                                         be16_to_cpu(name_loc->valuelen));
2475         }
2476
2477         name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2478         ASSERT(name_rmt->namelen == args->namelen);
2479         ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2480         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2481         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2482         args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2483                                                args->rmtvaluelen);
2484         return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2485 }
2486
2487 /*========================================================================
2488  * Utility routines.
2489  *========================================================================*/
2490
2491 /*
2492  * Move the indicated entries from one leaf to another.
2493  * NOTE: this routine modifies both source and destination leaves.
2494  */
2495 /*ARGSUSED*/
2496 STATIC void
2497 xfs_attr3_leaf_moveents(
2498         struct xfs_da_args              *args,
2499         struct xfs_attr_leafblock       *leaf_s,
2500         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2501         int                             start_s,
2502         struct xfs_attr_leafblock       *leaf_d,
2503         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2504         int                             start_d,
2505         int                             count)
2506 {
2507         struct xfs_attr_leaf_entry      *entry_s;
2508         struct xfs_attr_leaf_entry      *entry_d;
2509         int                             desti;
2510         int                             tmp;
2511         int                             i;
2512
2513         /*
2514          * Check for nothing to do.
2515          */
2516         if (count == 0)
2517                 return;
2518
2519         /*
2520          * Set up environment.
2521          */
2522         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2523                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2524         ASSERT(ichdr_s->magic == ichdr_d->magic);
2525         ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2526         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2527                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2528         ASSERT(ichdr_d->count < args->geo->blksize / 8);
2529         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2530                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2531
2532         ASSERT(start_s < ichdr_s->count);
2533         ASSERT(start_d <= ichdr_d->count);
2534         ASSERT(count <= ichdr_s->count);
2535
2536
2537         /*
2538          * Move the entries in the destination leaf up to make a hole?
2539          */
2540         if (start_d < ichdr_d->count) {
2541                 tmp  = ichdr_d->count - start_d;
2542                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2543                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2544                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2545                 memmove(entry_d, entry_s, tmp);
2546         }
2547
2548         /*
2549          * Copy all entry's in the same (sorted) order,
2550          * but allocate attribute info packed and in sequence.
2551          */
2552         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2553         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2554         desti = start_d;
2555         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2556                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2557                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2558 #ifdef GROT
2559                 /*
2560                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2561                  * may also need to change the insertion index.  Code turned
2562                  * off for 6.2, should be revisited later.
2563                  */
2564                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2565                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2566                         ichdr_s->usedbytes -= tmp;
2567                         ichdr_s->count -= 1;
2568                         entry_d--;      /* to compensate for ++ in loop hdr */
2569                         desti--;
2570                         if ((start_s + i) < offset)
2571                                 result++;       /* insertion index adjustment */
2572                 } else {
2573 #endif /* GROT */
2574                         ichdr_d->firstused -= tmp;
2575                         /* both on-disk, don't endian flip twice */
2576                         entry_d->hashval = entry_s->hashval;
2577                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2578                         entry_d->flags = entry_s->flags;
2579                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2580                                                         <= args->geo->blksize);
2581                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2582                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2583                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2584                                                         <= args->geo->blksize);
2585                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2586                         ichdr_s->usedbytes -= tmp;
2587                         ichdr_d->usedbytes += tmp;
2588                         ichdr_s->count -= 1;
2589                         ichdr_d->count += 1;
2590                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2591                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2592                         ASSERT(ichdr_d->firstused >= tmp);
2593 #ifdef GROT
2594                 }
2595 #endif /* GROT */
2596         }
2597
2598         /*
2599          * Zero out the entries we just copied.
2600          */
2601         if (start_s == ichdr_s->count) {
2602                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2603                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2604                 ASSERT(((char *)entry_s + tmp) <=
2605                        ((char *)leaf_s + args->geo->blksize));
2606                 memset(entry_s, 0, tmp);
2607         } else {
2608                 /*
2609                  * Move the remaining entries down to fill the hole,
2610                  * then zero the entries at the top.
2611                  */
2612                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2613                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2614                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2615                 memmove(entry_d, entry_s, tmp);
2616
2617                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2618                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2619                 ASSERT(((char *)entry_s + tmp) <=
2620                        ((char *)leaf_s + args->geo->blksize));
2621                 memset(entry_s, 0, tmp);
2622         }
2623
2624         /*
2625          * Fill in the freemap information
2626          */
2627         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2628         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2629         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2630         ichdr_d->freemap[1].base = 0;
2631         ichdr_d->freemap[2].base = 0;
2632         ichdr_d->freemap[1].size = 0;
2633         ichdr_d->freemap[2].size = 0;
2634         ichdr_s->holes = 1;     /* leaf may not be compact */
2635 }
2636
2637 /*
2638  * Pick up the last hashvalue from a leaf block.
2639  */
2640 xfs_dahash_t
2641 xfs_attr_leaf_lasthash(
2642         struct xfs_buf  *bp,
2643         int             *count)
2644 {
2645         struct xfs_attr3_icleaf_hdr ichdr;
2646         struct xfs_attr_leaf_entry *entries;
2647         struct xfs_mount *mp = bp->b_mount;
2648
2649         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2650         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2651         if (count)
2652                 *count = ichdr.count;
2653         if (!ichdr.count)
2654                 return 0;
2655         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2656 }
2657
2658 /*
2659  * Calculate the number of bytes used to store the indicated attribute
2660  * (whether local or remote only calculate bytes in this block).
2661  */
2662 STATIC int
2663 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2664 {
2665         struct xfs_attr_leaf_entry *entries;
2666         xfs_attr_leaf_name_local_t *name_loc;
2667         xfs_attr_leaf_name_remote_t *name_rmt;
2668         int size;
2669
2670         entries = xfs_attr3_leaf_entryp(leaf);
2671         if (entries[index].flags & XFS_ATTR_LOCAL) {
2672                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2673                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2674                                                    be16_to_cpu(name_loc->valuelen));
2675         } else {
2676                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2677                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2678         }
2679         return size;
2680 }
2681
2682 /*
2683  * Calculate the number of bytes that would be required to store the new
2684  * attribute (whether local or remote only calculate bytes in this block).
2685  * This routine decides as a side effect whether the attribute will be
2686  * a "local" or a "remote" attribute.
2687  */
2688 int
2689 xfs_attr_leaf_newentsize(
2690         struct xfs_da_args      *args,
2691         int                     *local)
2692 {
2693         int                     size;
2694
2695         size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2696         if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2697                 if (local)
2698                         *local = 1;
2699                 return size;
2700         }
2701         if (local)
2702                 *local = 0;
2703         return xfs_attr_leaf_entsize_remote(args->namelen);
2704 }
2705
2706
2707 /*========================================================================
2708  * Manage the INCOMPLETE flag in a leaf entry
2709  *========================================================================*/
2710
2711 /*
2712  * Clear the INCOMPLETE flag on an entry in a leaf block.
2713  */
2714 int
2715 xfs_attr3_leaf_clearflag(
2716         struct xfs_da_args      *args)
2717 {
2718         struct xfs_attr_leafblock *leaf;
2719         struct xfs_attr_leaf_entry *entry;
2720         struct xfs_attr_leaf_name_remote *name_rmt;
2721         struct xfs_buf          *bp;
2722         int                     error;
2723 #ifdef DEBUG
2724         struct xfs_attr3_icleaf_hdr ichdr;
2725         xfs_attr_leaf_name_local_t *name_loc;
2726         int namelen;
2727         char *name;
2728 #endif /* DEBUG */
2729
2730         trace_xfs_attr_leaf_clearflag(args);
2731         /*
2732          * Set up the operation.
2733          */
2734         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2735         if (error)
2736                 return error;
2737
2738         leaf = bp->b_addr;
2739         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2740         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2741
2742 #ifdef DEBUG
2743         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2744         ASSERT(args->index < ichdr.count);
2745         ASSERT(args->index >= 0);
2746
2747         if (entry->flags & XFS_ATTR_LOCAL) {
2748                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2749                 namelen = name_loc->namelen;
2750                 name = (char *)name_loc->nameval;
2751         } else {
2752                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2753                 namelen = name_rmt->namelen;
2754                 name = (char *)name_rmt->name;
2755         }
2756         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2757         ASSERT(namelen == args->namelen);
2758         ASSERT(memcmp(name, args->name, namelen) == 0);
2759 #endif /* DEBUG */
2760
2761         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2762         xfs_trans_log_buf(args->trans, bp,
2763                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2764
2765         if (args->rmtblkno) {
2766                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2767                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2768                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2769                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2770                 xfs_trans_log_buf(args->trans, bp,
2771                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2772         }
2773
2774         /*
2775          * Commit the flag value change and start the next trans in series.
2776          */
2777         return xfs_trans_roll_inode(&args->trans, args->dp);
2778 }
2779
2780 /*
2781  * Set the INCOMPLETE flag on an entry in a leaf block.
2782  */
2783 int
2784 xfs_attr3_leaf_setflag(
2785         struct xfs_da_args      *args)
2786 {
2787         struct xfs_attr_leafblock *leaf;
2788         struct xfs_attr_leaf_entry *entry;
2789         struct xfs_attr_leaf_name_remote *name_rmt;
2790         struct xfs_buf          *bp;
2791         int error;
2792 #ifdef DEBUG
2793         struct xfs_attr3_icleaf_hdr ichdr;
2794 #endif
2795
2796         trace_xfs_attr_leaf_setflag(args);
2797
2798         /*
2799          * Set up the operation.
2800          */
2801         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2802         if (error)
2803                 return error;
2804
2805         leaf = bp->b_addr;
2806 #ifdef DEBUG
2807         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2808         ASSERT(args->index < ichdr.count);
2809         ASSERT(args->index >= 0);
2810 #endif
2811         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2812
2813         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2814         entry->flags |= XFS_ATTR_INCOMPLETE;
2815         xfs_trans_log_buf(args->trans, bp,
2816                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2817         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2818                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2819                 name_rmt->valueblk = 0;
2820                 name_rmt->valuelen = 0;
2821                 xfs_trans_log_buf(args->trans, bp,
2822                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2823         }
2824
2825         /*
2826          * Commit the flag value change and start the next trans in series.
2827          */
2828         return xfs_trans_roll_inode(&args->trans, args->dp);
2829 }
2830
2831 /*
2832  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2833  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2834  * entry given by args->blkno2/index2.
2835  *
2836  * Note that they could be in different blocks, or in the same block.
2837  */
2838 int
2839 xfs_attr3_leaf_flipflags(
2840         struct xfs_da_args      *args)
2841 {
2842         struct xfs_attr_leafblock *leaf1;
2843         struct xfs_attr_leafblock *leaf2;
2844         struct xfs_attr_leaf_entry *entry1;
2845         struct xfs_attr_leaf_entry *entry2;
2846         struct xfs_attr_leaf_name_remote *name_rmt;
2847         struct xfs_buf          *bp1;
2848         struct xfs_buf          *bp2;
2849         int error;
2850 #ifdef DEBUG
2851         struct xfs_attr3_icleaf_hdr ichdr1;
2852         struct xfs_attr3_icleaf_hdr ichdr2;
2853         xfs_attr_leaf_name_local_t *name_loc;
2854         int namelen1, namelen2;
2855         char *name1, *name2;
2856 #endif /* DEBUG */
2857
2858         trace_xfs_attr_leaf_flipflags(args);
2859
2860         /*
2861          * Read the block containing the "old" attr
2862          */
2863         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2864         if (error)
2865                 return error;
2866
2867         /*
2868          * Read the block containing the "new" attr, if it is different
2869          */
2870         if (args->blkno2 != args->blkno) {
2871                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2872                                            &bp2);
2873                 if (error)
2874                         return error;
2875         } else {
2876                 bp2 = bp1;
2877         }
2878
2879         leaf1 = bp1->b_addr;
2880         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2881
2882         leaf2 = bp2->b_addr;
2883         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2884
2885 #ifdef DEBUG
2886         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2887         ASSERT(args->index < ichdr1.count);
2888         ASSERT(args->index >= 0);
2889
2890         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2891         ASSERT(args->index2 < ichdr2.count);
2892         ASSERT(args->index2 >= 0);
2893
2894         if (entry1->flags & XFS_ATTR_LOCAL) {
2895                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2896                 namelen1 = name_loc->namelen;
2897                 name1 = (char *)name_loc->nameval;
2898         } else {
2899                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2900                 namelen1 = name_rmt->namelen;
2901                 name1 = (char *)name_rmt->name;
2902         }
2903         if (entry2->flags & XFS_ATTR_LOCAL) {
2904                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2905                 namelen2 = name_loc->namelen;
2906                 name2 = (char *)name_loc->nameval;
2907         } else {
2908                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2909                 namelen2 = name_rmt->namelen;
2910                 name2 = (char *)name_rmt->name;
2911         }
2912         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2913         ASSERT(namelen1 == namelen2);
2914         ASSERT(memcmp(name1, name2, namelen1) == 0);
2915 #endif /* DEBUG */
2916
2917         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2918         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2919
2920         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2921         xfs_trans_log_buf(args->trans, bp1,
2922                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2923         if (args->rmtblkno) {
2924                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2925                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2926                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2927                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2928                 xfs_trans_log_buf(args->trans, bp1,
2929                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2930         }
2931
2932         entry2->flags |= XFS_ATTR_INCOMPLETE;
2933         xfs_trans_log_buf(args->trans, bp2,
2934                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2935         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2936                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2937                 name_rmt->valueblk = 0;
2938                 name_rmt->valuelen = 0;
2939                 xfs_trans_log_buf(args->trans, bp2,
2940                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2941         }
2942
2943         /*
2944          * Commit the flag value change and start the next trans in series.
2945          */
2946         error = xfs_trans_roll_inode(&args->trans, args->dp);
2947
2948         return error;
2949 }