selftests/bpf: Make test_varlen work with 32-bit user-space arch
[platform/kernel/linux-starfive.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          * firstused is the block offset of the first name info structure.
313          * Make sure it doesn't go off the block or crash into the header.
314          */
315         if (ichdr.firstused > mp->m_attr_geo->blksize)
316                 return __this_address;
317         if (ichdr.firstused < xfs_attr3_leaf_hdr_size(leaf))
318                 return __this_address;
319
320         /* Make sure the entries array doesn't crash into the name info. */
321         entries = xfs_attr3_leaf_entryp(bp->b_addr);
322         if ((char *)&entries[ichdr.count] >
323             (char *)bp->b_addr + ichdr.firstused)
324                 return __this_address;
325
326         /*
327          * NOTE: This verifier historically failed empty leaf buffers because
328          * we expect the fork to be in another format. Empty attr fork format
329          * conversions are possible during xattr set, however, and format
330          * conversion is not atomic with the xattr set that triggers it. We
331          * cannot assume leaf blocks are non-empty until that is addressed.
332         */
333         buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
334         for (i = 0, ent = entries; i < ichdr.count; ent++, i++) {
335                 fa = xfs_attr3_leaf_verify_entry(mp, buf_end, leaf, &ichdr,
336                                 ent, i, &last_hashval);
337                 if (fa)
338                         return fa;
339         }
340
341         /*
342          * Quickly check the freemap information.  Attribute data has to be
343          * aligned to 4-byte boundaries, and likewise for the free space.
344          *
345          * Note that for 64k block size filesystems, the freemap entries cannot
346          * overflow as they are only be16 fields. However, when checking end
347          * pointer of the freemap, we have to be careful to detect overflows and
348          * so use uint32_t for those checks.
349          */
350         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
351                 if (ichdr.freemap[i].base > mp->m_attr_geo->blksize)
352                         return __this_address;
353                 if (ichdr.freemap[i].base & 0x3)
354                         return __this_address;
355                 if (ichdr.freemap[i].size > mp->m_attr_geo->blksize)
356                         return __this_address;
357                 if (ichdr.freemap[i].size & 0x3)
358                         return __this_address;
359
360                 /* be care of 16 bit overflows here */
361                 end = (uint32_t)ichdr.freemap[i].base + ichdr.freemap[i].size;
362                 if (end < ichdr.freemap[i].base)
363                         return __this_address;
364                 if (end > mp->m_attr_geo->blksize)
365                         return __this_address;
366         }
367
368         return NULL;
369 }
370
371 static void
372 xfs_attr3_leaf_write_verify(
373         struct xfs_buf  *bp)
374 {
375         struct xfs_mount        *mp = bp->b_mount;
376         struct xfs_buf_log_item *bip = bp->b_log_item;
377         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
378         xfs_failaddr_t          fa;
379
380         fa = xfs_attr3_leaf_verify(bp);
381         if (fa) {
382                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
383                 return;
384         }
385
386         if (!xfs_sb_version_hascrc(&mp->m_sb))
387                 return;
388
389         if (bip)
390                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
391
392         xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
393 }
394
395 /*
396  * leaf/node format detection on trees is sketchy, so a node read can be done on
397  * leaf level blocks when detection identifies the tree as a node format tree
398  * incorrectly. In this case, we need to swap the verifier to match the correct
399  * format of the block being read.
400  */
401 static void
402 xfs_attr3_leaf_read_verify(
403         struct xfs_buf          *bp)
404 {
405         struct xfs_mount        *mp = bp->b_mount;
406         xfs_failaddr_t          fa;
407
408         if (xfs_sb_version_hascrc(&mp->m_sb) &&
409              !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
410                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
411         else {
412                 fa = xfs_attr3_leaf_verify(bp);
413                 if (fa)
414                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
415         }
416 }
417
418 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
419         .name = "xfs_attr3_leaf",
420         .magic16 = { cpu_to_be16(XFS_ATTR_LEAF_MAGIC),
421                      cpu_to_be16(XFS_ATTR3_LEAF_MAGIC) },
422         .verify_read = xfs_attr3_leaf_read_verify,
423         .verify_write = xfs_attr3_leaf_write_verify,
424         .verify_struct = xfs_attr3_leaf_verify,
425 };
426
427 int
428 xfs_attr3_leaf_read(
429         struct xfs_trans        *tp,
430         struct xfs_inode        *dp,
431         xfs_dablk_t             bno,
432         struct xfs_buf          **bpp)
433 {
434         int                     err;
435
436         err = xfs_da_read_buf(tp, dp, bno, 0, bpp, XFS_ATTR_FORK,
437                         &xfs_attr3_leaf_buf_ops);
438         if (!err && tp && *bpp)
439                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
440         return err;
441 }
442
443 /*========================================================================
444  * Namespace helper routines
445  *========================================================================*/
446
447 static bool
448 xfs_attr_match(
449         struct xfs_da_args      *args,
450         uint8_t                 namelen,
451         unsigned char           *name,
452         int                     flags)
453 {
454         if (args->namelen != namelen)
455                 return false;
456         if (memcmp(args->name, name, namelen) != 0)
457                 return false;
458         /*
459          * If we are looking for incomplete entries, show only those, else only
460          * show complete entries.
461          */
462         if (args->attr_filter !=
463             (flags & (XFS_ATTR_NSP_ONDISK_MASK | XFS_ATTR_INCOMPLETE)))
464                 return false;
465         return true;
466 }
467
468 static int
469 xfs_attr_copy_value(
470         struct xfs_da_args      *args,
471         unsigned char           *value,
472         int                     valuelen)
473 {
474         /*
475          * No copy if all we have to do is get the length
476          */
477         if (!args->valuelen) {
478                 args->valuelen = valuelen;
479                 return 0;
480         }
481
482         /*
483          * No copy if the length of the existing buffer is too small
484          */
485         if (args->valuelen < valuelen) {
486                 args->valuelen = valuelen;
487                 return -ERANGE;
488         }
489
490         if (!args->value) {
491                 args->value = kmem_alloc_large(valuelen, KM_NOLOCKDEP);
492                 if (!args->value)
493                         return -ENOMEM;
494         }
495         args->valuelen = valuelen;
496
497         /* remote block xattr requires IO for copy-in */
498         if (args->rmtblkno)
499                 return xfs_attr_rmtval_get(args);
500
501         /*
502          * This is to prevent a GCC warning because the remote xattr case
503          * doesn't have a value to pass in. In that case, we never reach here,
504          * but GCC can't work that out and so throws a "passing NULL to
505          * memcpy" warning.
506          */
507         if (!value)
508                 return -EINVAL;
509         memcpy(args->value, value, valuelen);
510         return 0;
511 }
512
513 /*========================================================================
514  * External routines when attribute fork size < XFS_LITINO(mp).
515  *========================================================================*/
516
517 /*
518  * Query whether the requested number of additional bytes of extended
519  * attribute space will be able to fit inline.
520  *
521  * Returns zero if not, else the di_forkoff fork offset to be used in the
522  * literal area for attribute data once the new bytes have been added.
523  *
524  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
525  * special case for dev/uuid inodes, they have fixed size data forks.
526  */
527 int
528 xfs_attr_shortform_bytesfit(
529         struct xfs_inode        *dp,
530         int                     bytes)
531 {
532         struct xfs_mount        *mp = dp->i_mount;
533         int64_t                 dsize;
534         int                     minforkoff;
535         int                     maxforkoff;
536         int                     offset;
537
538         /* rounded down */
539         offset = (XFS_LITINO(mp) - bytes) >> 3;
540
541         if (dp->i_df.if_format == XFS_DINODE_FMT_DEV) {
542                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
543                 return (offset >= minforkoff) ? minforkoff : 0;
544         }
545
546         /*
547          * If the requested numbers of bytes is smaller or equal to the
548          * current attribute fork size we can always proceed.
549          *
550          * Note that if_bytes in the data fork might actually be larger than
551          * the current data fork size is due to delalloc extents. In that
552          * case either the extent count will go down when they are converted
553          * to real extents, or the delalloc conversion will take care of the
554          * literal area rebalancing.
555          */
556         if (bytes <= XFS_IFORK_ASIZE(dp))
557                 return dp->i_d.di_forkoff;
558
559         /*
560          * For attr2 we can try to move the forkoff if there is space in the
561          * literal area, but for the old format we are done if there is no
562          * space in the fixed attribute fork.
563          */
564         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
565                 return 0;
566
567         dsize = dp->i_df.if_bytes;
568
569         switch (dp->i_df.if_format) {
570         case XFS_DINODE_FMT_EXTENTS:
571                 /*
572                  * If there is no attr fork and the data fork is extents, 
573                  * determine if creating the default attr fork will result
574                  * in the extents form migrating to btree. If so, the
575                  * minimum offset only needs to be the space required for
576                  * the btree root.
577                  */
578                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
579                     xfs_default_attroffset(dp))
580                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
581                 break;
582         case XFS_DINODE_FMT_BTREE:
583                 /*
584                  * If we have a data btree then keep forkoff if we have one,
585                  * otherwise we are adding a new attr, so then we set
586                  * minforkoff to where the btree root can finish so we have
587                  * plenty of room for attrs
588                  */
589                 if (dp->i_d.di_forkoff) {
590                         if (offset < dp->i_d.di_forkoff)
591                                 return 0;
592                         return dp->i_d.di_forkoff;
593                 }
594                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
595                 break;
596         }
597
598         /*
599          * A data fork btree root must have space for at least
600          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
601          */
602         minforkoff = max_t(int64_t, dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
603         minforkoff = roundup(minforkoff, 8) >> 3;
604
605         /* attr fork btree root can have at least this many key/ptr pairs */
606         maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
607         maxforkoff = maxforkoff >> 3;   /* rounded down */
608
609         if (offset >= maxforkoff)
610                 return maxforkoff;
611         if (offset >= minforkoff)
612                 return offset;
613         return 0;
614 }
615
616 /*
617  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
618  */
619 STATIC void
620 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
621 {
622         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
623             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
624                 spin_lock(&mp->m_sb_lock);
625                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
626                         xfs_sb_version_addattr2(&mp->m_sb);
627                         spin_unlock(&mp->m_sb_lock);
628                         xfs_log_sb(tp);
629                 } else
630                         spin_unlock(&mp->m_sb_lock);
631         }
632 }
633
634 /*
635  * Create the initial contents of a shortform attribute list.
636  */
637 void
638 xfs_attr_shortform_create(
639         struct xfs_da_args      *args)
640 {
641         struct xfs_inode        *dp = args->dp;
642         struct xfs_ifork        *ifp = dp->i_afp;
643         struct xfs_attr_sf_hdr  *hdr;
644
645         trace_xfs_attr_sf_create(args);
646
647         ASSERT(ifp->if_bytes == 0);
648         if (ifp->if_format == XFS_DINODE_FMT_EXTENTS) {
649                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
650                 ifp->if_format = XFS_DINODE_FMT_LOCAL;
651                 ifp->if_flags |= XFS_IFINLINE;
652         } else {
653                 ASSERT(ifp->if_flags & XFS_IFINLINE);
654         }
655         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
656         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
657         hdr->count = 0;
658         hdr->totsize = cpu_to_be16(sizeof(*hdr));
659         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
660 }
661
662 /*
663  * Add a name/value pair to the shortform attribute list.
664  * Overflow from the inode has already been checked for.
665  */
666 void
667 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
668 {
669         xfs_attr_shortform_t *sf;
670         xfs_attr_sf_entry_t *sfe;
671         int i, offset, size;
672         xfs_mount_t *mp;
673         xfs_inode_t *dp;
674         struct xfs_ifork *ifp;
675
676         trace_xfs_attr_sf_add(args);
677
678         dp = args->dp;
679         mp = dp->i_mount;
680         dp->i_d.di_forkoff = forkoff;
681
682         ifp = dp->i_afp;
683         ASSERT(ifp->if_flags & XFS_IFINLINE);
684         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
685         sfe = &sf->list[0];
686         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
687                 ASSERT(!xfs_attr_match(args, sfe->namelen, sfe->nameval,
688                         sfe->flags));
689         }
690
691         offset = (char *)sfe - (char *)sf;
692         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
693         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
694         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
695         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
696
697         sfe->namelen = args->namelen;
698         sfe->valuelen = args->valuelen;
699         sfe->flags = args->attr_filter;
700         memcpy(sfe->nameval, args->name, args->namelen);
701         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
702         sf->hdr.count++;
703         be16_add_cpu(&sf->hdr.totsize, size);
704         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
705
706         xfs_sbversion_add_attr2(mp, args->trans);
707 }
708
709 /*
710  * After the last attribute is removed revert to original inode format,
711  * making all literal area available to the data fork once more.
712  */
713 void
714 xfs_attr_fork_remove(
715         struct xfs_inode        *ip,
716         struct xfs_trans        *tp)
717 {
718         ASSERT(ip->i_afp->if_nextents == 0);
719
720         xfs_idestroy_fork(ip->i_afp);
721         kmem_cache_free(xfs_ifork_zone, ip->i_afp);
722         ip->i_afp = NULL;
723         ip->i_d.di_forkoff = 0;
724         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
725 }
726
727 /*
728  * Remove an attribute from the shortform attribute list structure.
729  */
730 int
731 xfs_attr_shortform_remove(xfs_da_args_t *args)
732 {
733         xfs_attr_shortform_t *sf;
734         xfs_attr_sf_entry_t *sfe;
735         int base, size=0, end, totsize, i;
736         xfs_mount_t *mp;
737         xfs_inode_t *dp;
738
739         trace_xfs_attr_sf_remove(args);
740
741         dp = args->dp;
742         mp = dp->i_mount;
743         base = sizeof(xfs_attr_sf_hdr_t);
744         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
745         sfe = &sf->list[0];
746         end = sf->hdr.count;
747         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
748                                         base += size, i++) {
749                 size = XFS_ATTR_SF_ENTSIZE(sfe);
750                 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
751                                 sfe->flags))
752                         break;
753         }
754         if (i == end)
755                 return -ENOATTR;
756
757         /*
758          * Fix up the attribute fork data, covering the hole
759          */
760         end = base + size;
761         totsize = be16_to_cpu(sf->hdr.totsize);
762         if (end != totsize)
763                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
764         sf->hdr.count--;
765         be16_add_cpu(&sf->hdr.totsize, -size);
766
767         /*
768          * Fix up the start offset of the attribute fork
769          */
770         totsize -= size;
771         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
772             (mp->m_flags & XFS_MOUNT_ATTR2) &&
773             (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
774             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
775                 xfs_attr_fork_remove(dp, args->trans);
776         } else {
777                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
778                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
779                 ASSERT(dp->i_d.di_forkoff);
780                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
781                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
782                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
783                                 dp->i_df.if_format == XFS_DINODE_FMT_BTREE);
784                 xfs_trans_log_inode(args->trans, dp,
785                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
786         }
787
788         xfs_sbversion_add_attr2(mp, args->trans);
789
790         return 0;
791 }
792
793 /*
794  * Look up a name in a shortform attribute list structure.
795  */
796 /*ARGSUSED*/
797 int
798 xfs_attr_shortform_lookup(xfs_da_args_t *args)
799 {
800         xfs_attr_shortform_t *sf;
801         xfs_attr_sf_entry_t *sfe;
802         int i;
803         struct xfs_ifork *ifp;
804
805         trace_xfs_attr_sf_lookup(args);
806
807         ifp = args->dp->i_afp;
808         ASSERT(ifp->if_flags & XFS_IFINLINE);
809         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
810         sfe = &sf->list[0];
811         for (i = 0; i < sf->hdr.count;
812                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
813                 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
814                                 sfe->flags))
815                         return -EEXIST;
816         }
817         return -ENOATTR;
818 }
819
820 /*
821  * Retrieve the attribute value and length.
822  *
823  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
824  * lookup, we only return an error if the attribute does not exist or we can't
825  * retrieve the value.
826  */
827 int
828 xfs_attr_shortform_getvalue(
829         struct xfs_da_args      *args)
830 {
831         struct xfs_attr_shortform *sf;
832         struct xfs_attr_sf_entry *sfe;
833         int                     i;
834
835         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
836         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
837         sfe = &sf->list[0];
838         for (i = 0; i < sf->hdr.count;
839                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
840                 if (xfs_attr_match(args, sfe->namelen, sfe->nameval,
841                                 sfe->flags))
842                         return xfs_attr_copy_value(args,
843                                 &sfe->nameval[args->namelen], sfe->valuelen);
844         }
845         return -ENOATTR;
846 }
847
848 /*
849  * Convert from using the shortform to the leaf.  On success, return the
850  * buffer so that we can keep it locked until we're totally done with it.
851  */
852 int
853 xfs_attr_shortform_to_leaf(
854         struct xfs_da_args              *args,
855         struct xfs_buf                  **leaf_bp)
856 {
857         struct xfs_inode                *dp;
858         struct xfs_attr_shortform       *sf;
859         struct xfs_attr_sf_entry        *sfe;
860         struct xfs_da_args              nargs;
861         char                            *tmpbuffer;
862         int                             error, i, size;
863         xfs_dablk_t                     blkno;
864         struct xfs_buf                  *bp;
865         struct xfs_ifork                *ifp;
866
867         trace_xfs_attr_sf_to_leaf(args);
868
869         dp = args->dp;
870         ifp = dp->i_afp;
871         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
872         size = be16_to_cpu(sf->hdr.totsize);
873         tmpbuffer = kmem_alloc(size, 0);
874         ASSERT(tmpbuffer != NULL);
875         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
876         sf = (xfs_attr_shortform_t *)tmpbuffer;
877
878         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
879         xfs_bmap_local_to_extents_empty(args->trans, dp, XFS_ATTR_FORK);
880
881         bp = NULL;
882         error = xfs_da_grow_inode(args, &blkno);
883         if (error)
884                 goto out;
885
886         ASSERT(blkno == 0);
887         error = xfs_attr3_leaf_create(args, blkno, &bp);
888         if (error)
889                 goto out;
890
891         memset((char *)&nargs, 0, sizeof(nargs));
892         nargs.dp = dp;
893         nargs.geo = args->geo;
894         nargs.total = args->total;
895         nargs.whichfork = XFS_ATTR_FORK;
896         nargs.trans = args->trans;
897         nargs.op_flags = XFS_DA_OP_OKNOENT;
898
899         sfe = &sf->list[0];
900         for (i = 0; i < sf->hdr.count; i++) {
901                 nargs.name = sfe->nameval;
902                 nargs.namelen = sfe->namelen;
903                 nargs.value = &sfe->nameval[nargs.namelen];
904                 nargs.valuelen = sfe->valuelen;
905                 nargs.hashval = xfs_da_hashname(sfe->nameval,
906                                                 sfe->namelen);
907                 nargs.attr_filter = sfe->flags & XFS_ATTR_NSP_ONDISK_MASK;
908                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
909                 ASSERT(error == -ENOATTR);
910                 error = xfs_attr3_leaf_add(bp, &nargs);
911                 ASSERT(error != -ENOSPC);
912                 if (error)
913                         goto out;
914                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
915         }
916         error = 0;
917         *leaf_bp = bp;
918 out:
919         kmem_free(tmpbuffer);
920         return error;
921 }
922
923 /*
924  * Check a leaf attribute block to see if all the entries would fit into
925  * a shortform attribute list.
926  */
927 int
928 xfs_attr_shortform_allfit(
929         struct xfs_buf          *bp,
930         struct xfs_inode        *dp)
931 {
932         struct xfs_attr_leafblock *leaf;
933         struct xfs_attr_leaf_entry *entry;
934         xfs_attr_leaf_name_local_t *name_loc;
935         struct xfs_attr3_icleaf_hdr leafhdr;
936         int                     bytes;
937         int                     i;
938         struct xfs_mount        *mp = bp->b_mount;
939
940         leaf = bp->b_addr;
941         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
942         entry = xfs_attr3_leaf_entryp(leaf);
943
944         bytes = sizeof(struct xfs_attr_sf_hdr);
945         for (i = 0; i < leafhdr.count; entry++, i++) {
946                 if (entry->flags & XFS_ATTR_INCOMPLETE)
947                         continue;               /* don't copy partial entries */
948                 if (!(entry->flags & XFS_ATTR_LOCAL))
949                         return 0;
950                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
951                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
952                         return 0;
953                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
954                         return 0;
955                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
956                                 + name_loc->namelen
957                                 + be16_to_cpu(name_loc->valuelen);
958         }
959         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
960             (dp->i_df.if_format != XFS_DINODE_FMT_BTREE) &&
961             (bytes == sizeof(struct xfs_attr_sf_hdr)))
962                 return -1;
963         return xfs_attr_shortform_bytesfit(dp, bytes);
964 }
965
966 /* Verify the consistency of an inline attribute fork. */
967 xfs_failaddr_t
968 xfs_attr_shortform_verify(
969         struct xfs_inode                *ip)
970 {
971         struct xfs_attr_shortform       *sfp;
972         struct xfs_attr_sf_entry        *sfep;
973         struct xfs_attr_sf_entry        *next_sfep;
974         char                            *endp;
975         struct xfs_ifork                *ifp;
976         int                             i;
977         int64_t                         size;
978
979         ASSERT(ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL);
980         ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK);
981         sfp = (struct xfs_attr_shortform *)ifp->if_u1.if_data;
982         size = ifp->if_bytes;
983
984         /*
985          * Give up if the attribute is way too short.
986          */
987         if (size < sizeof(struct xfs_attr_sf_hdr))
988                 return __this_address;
989
990         endp = (char *)sfp + size;
991
992         /* Check all reported entries */
993         sfep = &sfp->list[0];
994         for (i = 0; i < sfp->hdr.count; i++) {
995                 /*
996                  * struct xfs_attr_sf_entry has a variable length.
997                  * Check the fixed-offset parts of the structure are
998                  * within the data buffer.
999                  */
1000                 if (((char *)sfep + sizeof(*sfep)) >= endp)
1001                         return __this_address;
1002
1003                 /* Don't allow names with known bad length. */
1004                 if (sfep->namelen == 0)
1005                         return __this_address;
1006
1007                 /*
1008                  * Check that the variable-length part of the structure is
1009                  * within the data buffer.  The next entry starts after the
1010                  * name component, so nextentry is an acceptable test.
1011                  */
1012                 next_sfep = XFS_ATTR_SF_NEXTENTRY(sfep);
1013                 if ((char *)next_sfep > endp)
1014                         return __this_address;
1015
1016                 /*
1017                  * Check for unknown flags.  Short form doesn't support
1018                  * the incomplete or local bits, so we can use the namespace
1019                  * mask here.
1020                  */
1021                 if (sfep->flags & ~XFS_ATTR_NSP_ONDISK_MASK)
1022                         return __this_address;
1023
1024                 /*
1025                  * Check for invalid namespace combinations.  We only allow
1026                  * one namespace flag per xattr, so we can just count the
1027                  * bits (i.e. hweight) here.
1028                  */
1029                 if (hweight8(sfep->flags & XFS_ATTR_NSP_ONDISK_MASK) > 1)
1030                         return __this_address;
1031
1032                 sfep = next_sfep;
1033         }
1034         if ((void *)sfep != (void *)endp)
1035                 return __this_address;
1036
1037         return NULL;
1038 }
1039
1040 /*
1041  * Convert a leaf attribute list to shortform attribute list
1042  */
1043 int
1044 xfs_attr3_leaf_to_shortform(
1045         struct xfs_buf          *bp,
1046         struct xfs_da_args      *args,
1047         int                     forkoff)
1048 {
1049         struct xfs_attr_leafblock *leaf;
1050         struct xfs_attr3_icleaf_hdr ichdr;
1051         struct xfs_attr_leaf_entry *entry;
1052         struct xfs_attr_leaf_name_local *name_loc;
1053         struct xfs_da_args      nargs;
1054         struct xfs_inode        *dp = args->dp;
1055         char                    *tmpbuffer;
1056         int                     error;
1057         int                     i;
1058
1059         trace_xfs_attr_leaf_to_sf(args);
1060
1061         tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1062         if (!tmpbuffer)
1063                 return -ENOMEM;
1064
1065         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1066
1067         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1068         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1069         entry = xfs_attr3_leaf_entryp(leaf);
1070
1071         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1072         memset(bp->b_addr, 0, args->geo->blksize);
1073
1074         /*
1075          * Clean out the prior contents of the attribute list.
1076          */
1077         error = xfs_da_shrink_inode(args, 0, bp);
1078         if (error)
1079                 goto out;
1080
1081         if (forkoff == -1) {
1082                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1083                 ASSERT(dp->i_df.if_format != XFS_DINODE_FMT_BTREE);
1084                 xfs_attr_fork_remove(dp, args->trans);
1085                 goto out;
1086         }
1087
1088         xfs_attr_shortform_create(args);
1089
1090         /*
1091          * Copy the attributes
1092          */
1093         memset((char *)&nargs, 0, sizeof(nargs));
1094         nargs.geo = args->geo;
1095         nargs.dp = dp;
1096         nargs.total = args->total;
1097         nargs.whichfork = XFS_ATTR_FORK;
1098         nargs.trans = args->trans;
1099         nargs.op_flags = XFS_DA_OP_OKNOENT;
1100
1101         for (i = 0; i < ichdr.count; entry++, i++) {
1102                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1103                         continue;       /* don't copy partial entries */
1104                 if (!entry->nameidx)
1105                         continue;
1106                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1107                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1108                 nargs.name = name_loc->nameval;
1109                 nargs.namelen = name_loc->namelen;
1110                 nargs.value = &name_loc->nameval[nargs.namelen];
1111                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1112                 nargs.hashval = be32_to_cpu(entry->hashval);
1113                 nargs.attr_filter = entry->flags & XFS_ATTR_NSP_ONDISK_MASK;
1114                 xfs_attr_shortform_add(&nargs, forkoff);
1115         }
1116         error = 0;
1117
1118 out:
1119         kmem_free(tmpbuffer);
1120         return error;
1121 }
1122
1123 /*
1124  * Convert from using a single leaf to a root node and a leaf.
1125  */
1126 int
1127 xfs_attr3_leaf_to_node(
1128         struct xfs_da_args      *args)
1129 {
1130         struct xfs_attr_leafblock *leaf;
1131         struct xfs_attr3_icleaf_hdr icleafhdr;
1132         struct xfs_attr_leaf_entry *entries;
1133         struct xfs_da3_icnode_hdr icnodehdr;
1134         struct xfs_da_intnode   *node;
1135         struct xfs_inode        *dp = args->dp;
1136         struct xfs_mount        *mp = dp->i_mount;
1137         struct xfs_buf          *bp1 = NULL;
1138         struct xfs_buf          *bp2 = NULL;
1139         xfs_dablk_t             blkno;
1140         int                     error;
1141
1142         trace_xfs_attr_leaf_to_node(args);
1143
1144         error = xfs_da_grow_inode(args, &blkno);
1145         if (error)
1146                 goto out;
1147         error = xfs_attr3_leaf_read(args->trans, dp, 0, &bp1);
1148         if (error)
1149                 goto out;
1150
1151         error = xfs_da_get_buf(args->trans, dp, blkno, &bp2, XFS_ATTR_FORK);
1152         if (error)
1153                 goto out;
1154
1155         /* copy leaf to new buffer, update identifiers */
1156         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1157         bp2->b_ops = bp1->b_ops;
1158         memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
1159         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1160                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1161                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1162         }
1163         xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1164
1165         /*
1166          * Set up the new root node.
1167          */
1168         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1169         if (error)
1170                 goto out;
1171         node = bp1->b_addr;
1172         xfs_da3_node_hdr_from_disk(mp, &icnodehdr, node);
1173
1174         leaf = bp2->b_addr;
1175         xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1176         entries = xfs_attr3_leaf_entryp(leaf);
1177
1178         /* both on-disk, don't endian-flip twice */
1179         icnodehdr.btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1180         icnodehdr.btree[0].before = cpu_to_be32(blkno);
1181         icnodehdr.count = 1;
1182         xfs_da3_node_hdr_to_disk(dp->i_mount, node, &icnodehdr);
1183         xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1184         error = 0;
1185 out:
1186         return error;
1187 }
1188
1189 /*========================================================================
1190  * Routines used for growing the Btree.
1191  *========================================================================*/
1192
1193 /*
1194  * Create the initial contents of a leaf attribute list
1195  * or a leaf in a node attribute list.
1196  */
1197 STATIC int
1198 xfs_attr3_leaf_create(
1199         struct xfs_da_args      *args,
1200         xfs_dablk_t             blkno,
1201         struct xfs_buf          **bpp)
1202 {
1203         struct xfs_attr_leafblock *leaf;
1204         struct xfs_attr3_icleaf_hdr ichdr;
1205         struct xfs_inode        *dp = args->dp;
1206         struct xfs_mount        *mp = dp->i_mount;
1207         struct xfs_buf          *bp;
1208         int                     error;
1209
1210         trace_xfs_attr_leaf_create(args);
1211
1212         error = xfs_da_get_buf(args->trans, args->dp, blkno, &bp,
1213                                             XFS_ATTR_FORK);
1214         if (error)
1215                 return error;
1216         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1217         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1218         leaf = bp->b_addr;
1219         memset(leaf, 0, args->geo->blksize);
1220
1221         memset(&ichdr, 0, sizeof(ichdr));
1222         ichdr.firstused = args->geo->blksize;
1223
1224         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1225                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1226
1227                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1228
1229                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1230                 hdr3->owner = cpu_to_be64(dp->i_ino);
1231                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1232
1233                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1234         } else {
1235                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1236                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1237         }
1238         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1239
1240         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1241         xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1242
1243         *bpp = bp;
1244         return 0;
1245 }
1246
1247 /*
1248  * Split the leaf node, rebalance, then add the new entry.
1249  */
1250 int
1251 xfs_attr3_leaf_split(
1252         struct xfs_da_state     *state,
1253         struct xfs_da_state_blk *oldblk,
1254         struct xfs_da_state_blk *newblk)
1255 {
1256         xfs_dablk_t blkno;
1257         int error;
1258
1259         trace_xfs_attr_leaf_split(state->args);
1260
1261         /*
1262          * Allocate space for a new leaf node.
1263          */
1264         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1265         error = xfs_da_grow_inode(state->args, &blkno);
1266         if (error)
1267                 return error;
1268         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1269         if (error)
1270                 return error;
1271         newblk->blkno = blkno;
1272         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1273
1274         /*
1275          * Rebalance the entries across the two leaves.
1276          * NOTE: rebalance() currently depends on the 2nd block being empty.
1277          */
1278         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1279         error = xfs_da3_blk_link(state, oldblk, newblk);
1280         if (error)
1281                 return error;
1282
1283         /*
1284          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1285          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1286          * "new" attrs info.  Will need the "old" info to remove it later.
1287          *
1288          * Insert the "new" entry in the correct block.
1289          */
1290         if (state->inleaf) {
1291                 trace_xfs_attr_leaf_add_old(state->args);
1292                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1293         } else {
1294                 trace_xfs_attr_leaf_add_new(state->args);
1295                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1296         }
1297
1298         /*
1299          * Update last hashval in each block since we added the name.
1300          */
1301         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1302         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1303         return error;
1304 }
1305
1306 /*
1307  * Add a name to the leaf attribute list structure.
1308  */
1309 int
1310 xfs_attr3_leaf_add(
1311         struct xfs_buf          *bp,
1312         struct xfs_da_args      *args)
1313 {
1314         struct xfs_attr_leafblock *leaf;
1315         struct xfs_attr3_icleaf_hdr ichdr;
1316         int                     tablesize;
1317         int                     entsize;
1318         int                     sum;
1319         int                     tmp;
1320         int                     i;
1321
1322         trace_xfs_attr_leaf_add(args);
1323
1324         leaf = bp->b_addr;
1325         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1326         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1327         entsize = xfs_attr_leaf_newentsize(args, NULL);
1328
1329         /*
1330          * Search through freemap for first-fit on new name length.
1331          * (may need to figure in size of entry struct too)
1332          */
1333         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1334                                         + xfs_attr3_leaf_hdr_size(leaf);
1335         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1336                 if (tablesize > ichdr.firstused) {
1337                         sum += ichdr.freemap[i].size;
1338                         continue;
1339                 }
1340                 if (!ichdr.freemap[i].size)
1341                         continue;       /* no space in this map */
1342                 tmp = entsize;
1343                 if (ichdr.freemap[i].base < ichdr.firstused)
1344                         tmp += sizeof(xfs_attr_leaf_entry_t);
1345                 if (ichdr.freemap[i].size >= tmp) {
1346                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1347                         goto out_log_hdr;
1348                 }
1349                 sum += ichdr.freemap[i].size;
1350         }
1351
1352         /*
1353          * If there are no holes in the address space of the block,
1354          * and we don't have enough freespace, then compaction will do us
1355          * no good and we should just give up.
1356          */
1357         if (!ichdr.holes && sum < entsize)
1358                 return -ENOSPC;
1359
1360         /*
1361          * Compact the entries to coalesce free space.
1362          * This may change the hdr->count via dropping INCOMPLETE entries.
1363          */
1364         xfs_attr3_leaf_compact(args, &ichdr, bp);
1365
1366         /*
1367          * After compaction, the block is guaranteed to have only one
1368          * free region, in freemap[0].  If it is not big enough, give up.
1369          */
1370         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1371                 tmp = -ENOSPC;
1372                 goto out_log_hdr;
1373         }
1374
1375         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1376
1377 out_log_hdr:
1378         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1379         xfs_trans_log_buf(args->trans, bp,
1380                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1381                                 xfs_attr3_leaf_hdr_size(leaf)));
1382         return tmp;
1383 }
1384
1385 /*
1386  * Add a name to a leaf attribute list structure.
1387  */
1388 STATIC int
1389 xfs_attr3_leaf_add_work(
1390         struct xfs_buf          *bp,
1391         struct xfs_attr3_icleaf_hdr *ichdr,
1392         struct xfs_da_args      *args,
1393         int                     mapindex)
1394 {
1395         struct xfs_attr_leafblock *leaf;
1396         struct xfs_attr_leaf_entry *entry;
1397         struct xfs_attr_leaf_name_local *name_loc;
1398         struct xfs_attr_leaf_name_remote *name_rmt;
1399         struct xfs_mount        *mp;
1400         int                     tmp;
1401         int                     i;
1402
1403         trace_xfs_attr_leaf_add_work(args);
1404
1405         leaf = bp->b_addr;
1406         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1407         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1408
1409         /*
1410          * Force open some space in the entry array and fill it in.
1411          */
1412         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1413         if (args->index < ichdr->count) {
1414                 tmp  = ichdr->count - args->index;
1415                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1416                 memmove(entry + 1, entry, tmp);
1417                 xfs_trans_log_buf(args->trans, bp,
1418                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1419         }
1420         ichdr->count++;
1421
1422         /*
1423          * Allocate space for the new string (at the end of the run).
1424          */
1425         mp = args->trans->t_mountp;
1426         ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1427         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1428         ASSERT(ichdr->freemap[mapindex].size >=
1429                 xfs_attr_leaf_newentsize(args, NULL));
1430         ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1431         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1432
1433         ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1434
1435         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1436                                      ichdr->freemap[mapindex].size);
1437         entry->hashval = cpu_to_be32(args->hashval);
1438         entry->flags = args->attr_filter;
1439         if (tmp)
1440                 entry->flags |= XFS_ATTR_LOCAL;
1441         if (args->op_flags & XFS_DA_OP_RENAME) {
1442                 entry->flags |= XFS_ATTR_INCOMPLETE;
1443                 if ((args->blkno2 == args->blkno) &&
1444                     (args->index2 <= args->index)) {
1445                         args->index2++;
1446                 }
1447         }
1448         xfs_trans_log_buf(args->trans, bp,
1449                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1450         ASSERT((args->index == 0) ||
1451                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1452         ASSERT((args->index == ichdr->count - 1) ||
1453                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1454
1455         /*
1456          * For "remote" attribute values, simply note that we need to
1457          * allocate space for the "remote" value.  We can't actually
1458          * allocate the extents in this transaction, and we can't decide
1459          * which blocks they should be as we might allocate more blocks
1460          * as part of this transaction (a split operation for example).
1461          */
1462         if (entry->flags & XFS_ATTR_LOCAL) {
1463                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1464                 name_loc->namelen = args->namelen;
1465                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1466                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1467                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1468                                    be16_to_cpu(name_loc->valuelen));
1469         } else {
1470                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1471                 name_rmt->namelen = args->namelen;
1472                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1473                 entry->flags |= XFS_ATTR_INCOMPLETE;
1474                 /* just in case */
1475                 name_rmt->valuelen = 0;
1476                 name_rmt->valueblk = 0;
1477                 args->rmtblkno = 1;
1478                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1479                 args->rmtvaluelen = args->valuelen;
1480         }
1481         xfs_trans_log_buf(args->trans, bp,
1482              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1483                                    xfs_attr_leaf_entsize(leaf, args->index)));
1484
1485         /*
1486          * Update the control info for this leaf node
1487          */
1488         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1489                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1490
1491         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1492                                         + xfs_attr3_leaf_hdr_size(leaf));
1493         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1494                                         + xfs_attr3_leaf_hdr_size(leaf);
1495
1496         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1497                 if (ichdr->freemap[i].base == tmp) {
1498                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1499                         ichdr->freemap[i].size -=
1500                                 min_t(uint16_t, ichdr->freemap[i].size,
1501                                                 sizeof(xfs_attr_leaf_entry_t));
1502                 }
1503         }
1504         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1505         return 0;
1506 }
1507
1508 /*
1509  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1510  */
1511 STATIC void
1512 xfs_attr3_leaf_compact(
1513         struct xfs_da_args      *args,
1514         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1515         struct xfs_buf          *bp)
1516 {
1517         struct xfs_attr_leafblock *leaf_src;
1518         struct xfs_attr_leafblock *leaf_dst;
1519         struct xfs_attr3_icleaf_hdr ichdr_src;
1520         struct xfs_trans        *trans = args->trans;
1521         char                    *tmpbuffer;
1522
1523         trace_xfs_attr_leaf_compact(args);
1524
1525         tmpbuffer = kmem_alloc(args->geo->blksize, 0);
1526         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1527         memset(bp->b_addr, 0, args->geo->blksize);
1528         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1529         leaf_dst = bp->b_addr;
1530
1531         /*
1532          * Copy the on-disk header back into the destination buffer to ensure
1533          * all the information in the header that is not part of the incore
1534          * header structure is preserved.
1535          */
1536         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1537
1538         /* Initialise the incore headers */
1539         ichdr_src = *ichdr_dst; /* struct copy */
1540         ichdr_dst->firstused = args->geo->blksize;
1541         ichdr_dst->usedbytes = 0;
1542         ichdr_dst->count = 0;
1543         ichdr_dst->holes = 0;
1544         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1545         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1546                                                 ichdr_dst->freemap[0].base;
1547
1548         /* write the header back to initialise the underlying buffer */
1549         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1550
1551         /*
1552          * Copy all entry's in the same (sorted) order,
1553          * but allocate name/value pairs packed and in sequence.
1554          */
1555         xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1556                                 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1557         /*
1558          * this logs the entire buffer, but the caller must write the header
1559          * back to the buffer when it is finished modifying it.
1560          */
1561         xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1562
1563         kmem_free(tmpbuffer);
1564 }
1565
1566 /*
1567  * Compare two leaf blocks "order".
1568  * Return 0 unless leaf2 should go before leaf1.
1569  */
1570 static int
1571 xfs_attr3_leaf_order(
1572         struct xfs_buf  *leaf1_bp,
1573         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1574         struct xfs_buf  *leaf2_bp,
1575         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1576 {
1577         struct xfs_attr_leaf_entry *entries1;
1578         struct xfs_attr_leaf_entry *entries2;
1579
1580         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1581         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1582         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1583             ((be32_to_cpu(entries2[0].hashval) <
1584               be32_to_cpu(entries1[0].hashval)) ||
1585              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1586               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1587                 return 1;
1588         }
1589         return 0;
1590 }
1591
1592 int
1593 xfs_attr_leaf_order(
1594         struct xfs_buf  *leaf1_bp,
1595         struct xfs_buf  *leaf2_bp)
1596 {
1597         struct xfs_attr3_icleaf_hdr ichdr1;
1598         struct xfs_attr3_icleaf_hdr ichdr2;
1599         struct xfs_mount *mp = leaf1_bp->b_mount;
1600
1601         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1602         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1603         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1604 }
1605
1606 /*
1607  * Redistribute the attribute list entries between two leaf nodes,
1608  * taking into account the size of the new entry.
1609  *
1610  * NOTE: if new block is empty, then it will get the upper half of the
1611  * old block.  At present, all (one) callers pass in an empty second block.
1612  *
1613  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1614  * to match what it is doing in splitting the attribute leaf block.  Those
1615  * values are used in "atomic rename" operations on attributes.  Note that
1616  * the "new" and "old" values can end up in different blocks.
1617  */
1618 STATIC void
1619 xfs_attr3_leaf_rebalance(
1620         struct xfs_da_state     *state,
1621         struct xfs_da_state_blk *blk1,
1622         struct xfs_da_state_blk *blk2)
1623 {
1624         struct xfs_da_args      *args;
1625         struct xfs_attr_leafblock *leaf1;
1626         struct xfs_attr_leafblock *leaf2;
1627         struct xfs_attr3_icleaf_hdr ichdr1;
1628         struct xfs_attr3_icleaf_hdr ichdr2;
1629         struct xfs_attr_leaf_entry *entries1;
1630         struct xfs_attr_leaf_entry *entries2;
1631         int                     count;
1632         int                     totallen;
1633         int                     max;
1634         int                     space;
1635         int                     swap;
1636
1637         /*
1638          * Set up environment.
1639          */
1640         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1641         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1642         leaf1 = blk1->bp->b_addr;
1643         leaf2 = blk2->bp->b_addr;
1644         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1645         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1646         ASSERT(ichdr2.count == 0);
1647         args = state->args;
1648
1649         trace_xfs_attr_leaf_rebalance(args);
1650
1651         /*
1652          * Check ordering of blocks, reverse if it makes things simpler.
1653          *
1654          * NOTE: Given that all (current) callers pass in an empty
1655          * second block, this code should never set "swap".
1656          */
1657         swap = 0;
1658         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1659                 swap(blk1, blk2);
1660
1661                 /* swap structures rather than reconverting them */
1662                 swap(ichdr1, ichdr2);
1663
1664                 leaf1 = blk1->bp->b_addr;
1665                 leaf2 = blk2->bp->b_addr;
1666                 swap = 1;
1667         }
1668
1669         /*
1670          * Examine entries until we reduce the absolute difference in
1671          * byte usage between the two blocks to a minimum.  Then get
1672          * the direction to copy and the number of elements to move.
1673          *
1674          * "inleaf" is true if the new entry should be inserted into blk1.
1675          * If "swap" is also true, then reverse the sense of "inleaf".
1676          */
1677         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1678                                                       blk2, &ichdr2,
1679                                                       &count, &totallen);
1680         if (swap)
1681                 state->inleaf = !state->inleaf;
1682
1683         /*
1684          * Move any entries required from leaf to leaf:
1685          */
1686         if (count < ichdr1.count) {
1687                 /*
1688                  * Figure the total bytes to be added to the destination leaf.
1689                  */
1690                 /* number entries being moved */
1691                 count = ichdr1.count - count;
1692                 space  = ichdr1.usedbytes - totallen;
1693                 space += count * sizeof(xfs_attr_leaf_entry_t);
1694
1695                 /*
1696                  * leaf2 is the destination, compact it if it looks tight.
1697                  */
1698                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1699                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1700                 if (space > max)
1701                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1702
1703                 /*
1704                  * Move high entries from leaf1 to low end of leaf2.
1705                  */
1706                 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1707                                 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1708
1709         } else if (count > ichdr1.count) {
1710                 /*
1711                  * I assert that since all callers pass in an empty
1712                  * second buffer, this code should never execute.
1713                  */
1714                 ASSERT(0);
1715
1716                 /*
1717                  * Figure the total bytes to be added to the destination leaf.
1718                  */
1719                 /* number entries being moved */
1720                 count -= ichdr1.count;
1721                 space  = totallen - ichdr1.usedbytes;
1722                 space += count * sizeof(xfs_attr_leaf_entry_t);
1723
1724                 /*
1725                  * leaf1 is the destination, compact it if it looks tight.
1726                  */
1727                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1728                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1729                 if (space > max)
1730                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1731
1732                 /*
1733                  * Move low entries from leaf2 to high end of leaf1.
1734                  */
1735                 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1736                                         ichdr1.count, count);
1737         }
1738
1739         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1740         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1741         xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1742         xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1743
1744         /*
1745          * Copy out last hashval in each block for B-tree code.
1746          */
1747         entries1 = xfs_attr3_leaf_entryp(leaf1);
1748         entries2 = xfs_attr3_leaf_entryp(leaf2);
1749         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1750         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1751
1752         /*
1753          * Adjust the expected index for insertion.
1754          * NOTE: this code depends on the (current) situation that the
1755          * second block was originally empty.
1756          *
1757          * If the insertion point moved to the 2nd block, we must adjust
1758          * the index.  We must also track the entry just following the
1759          * new entry for use in an "atomic rename" operation, that entry
1760          * is always the "old" entry and the "new" entry is what we are
1761          * inserting.  The index/blkno fields refer to the "old" entry,
1762          * while the index2/blkno2 fields refer to the "new" entry.
1763          */
1764         if (blk1->index > ichdr1.count) {
1765                 ASSERT(state->inleaf == 0);
1766                 blk2->index = blk1->index - ichdr1.count;
1767                 args->index = args->index2 = blk2->index;
1768                 args->blkno = args->blkno2 = blk2->blkno;
1769         } else if (blk1->index == ichdr1.count) {
1770                 if (state->inleaf) {
1771                         args->index = blk1->index;
1772                         args->blkno = blk1->blkno;
1773                         args->index2 = 0;
1774                         args->blkno2 = blk2->blkno;
1775                 } else {
1776                         /*
1777                          * On a double leaf split, the original attr location
1778                          * is already stored in blkno2/index2, so don't
1779                          * overwrite it overwise we corrupt the tree.
1780                          */
1781                         blk2->index = blk1->index - ichdr1.count;
1782                         args->index = blk2->index;
1783                         args->blkno = blk2->blkno;
1784                         if (!state->extravalid) {
1785                                 /*
1786                                  * set the new attr location to match the old
1787                                  * one and let the higher level split code
1788                                  * decide where in the leaf to place it.
1789                                  */
1790                                 args->index2 = blk2->index;
1791                                 args->blkno2 = blk2->blkno;
1792                         }
1793                 }
1794         } else {
1795                 ASSERT(state->inleaf == 1);
1796                 args->index = args->index2 = blk1->index;
1797                 args->blkno = args->blkno2 = blk1->blkno;
1798         }
1799 }
1800
1801 /*
1802  * Examine entries until we reduce the absolute difference in
1803  * byte usage between the two blocks to a minimum.
1804  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1805  * GROT: there will always be enough room in either block for a new entry.
1806  * GROT: Do a double-split for this case?
1807  */
1808 STATIC int
1809 xfs_attr3_leaf_figure_balance(
1810         struct xfs_da_state             *state,
1811         struct xfs_da_state_blk         *blk1,
1812         struct xfs_attr3_icleaf_hdr     *ichdr1,
1813         struct xfs_da_state_blk         *blk2,
1814         struct xfs_attr3_icleaf_hdr     *ichdr2,
1815         int                             *countarg,
1816         int                             *usedbytesarg)
1817 {
1818         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1819         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1820         struct xfs_attr_leaf_entry      *entry;
1821         int                             count;
1822         int                             max;
1823         int                             index;
1824         int                             totallen = 0;
1825         int                             half;
1826         int                             lastdelta;
1827         int                             foundit = 0;
1828         int                             tmp;
1829
1830         /*
1831          * Examine entries until we reduce the absolute difference in
1832          * byte usage between the two blocks to a minimum.
1833          */
1834         max = ichdr1->count + ichdr2->count;
1835         half = (max + 1) * sizeof(*entry);
1836         half += ichdr1->usedbytes + ichdr2->usedbytes +
1837                         xfs_attr_leaf_newentsize(state->args, NULL);
1838         half /= 2;
1839         lastdelta = state->args->geo->blksize;
1840         entry = xfs_attr3_leaf_entryp(leaf1);
1841         for (count = index = 0; count < max; entry++, index++, count++) {
1842
1843 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1844                 /*
1845                  * The new entry is in the first block, account for it.
1846                  */
1847                 if (count == blk1->index) {
1848                         tmp = totallen + sizeof(*entry) +
1849                                 xfs_attr_leaf_newentsize(state->args, NULL);
1850                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1851                                 break;
1852                         lastdelta = XFS_ATTR_ABS(half - tmp);
1853                         totallen = tmp;
1854                         foundit = 1;
1855                 }
1856
1857                 /*
1858                  * Wrap around into the second block if necessary.
1859                  */
1860                 if (count == ichdr1->count) {
1861                         leaf1 = leaf2;
1862                         entry = xfs_attr3_leaf_entryp(leaf1);
1863                         index = 0;
1864                 }
1865
1866                 /*
1867                  * Figure out if next leaf entry would be too much.
1868                  */
1869                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1870                                                                         index);
1871                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1872                         break;
1873                 lastdelta = XFS_ATTR_ABS(half - tmp);
1874                 totallen = tmp;
1875 #undef XFS_ATTR_ABS
1876         }
1877
1878         /*
1879          * Calculate the number of usedbytes that will end up in lower block.
1880          * If new entry not in lower block, fix up the count.
1881          */
1882         totallen -= count * sizeof(*entry);
1883         if (foundit) {
1884                 totallen -= sizeof(*entry) +
1885                                 xfs_attr_leaf_newentsize(state->args, NULL);
1886         }
1887
1888         *countarg = count;
1889         *usedbytesarg = totallen;
1890         return foundit;
1891 }
1892
1893 /*========================================================================
1894  * Routines used for shrinking the Btree.
1895  *========================================================================*/
1896
1897 /*
1898  * Check a leaf block and its neighbors to see if the block should be
1899  * collapsed into one or the other neighbor.  Always keep the block
1900  * with the smaller block number.
1901  * If the current block is over 50% full, don't try to join it, return 0.
1902  * If the block is empty, fill in the state structure and return 2.
1903  * If it can be collapsed, fill in the state structure and return 1.
1904  * If nothing can be done, return 0.
1905  *
1906  * GROT: allow for INCOMPLETE entries in calculation.
1907  */
1908 int
1909 xfs_attr3_leaf_toosmall(
1910         struct xfs_da_state     *state,
1911         int                     *action)
1912 {
1913         struct xfs_attr_leafblock *leaf;
1914         struct xfs_da_state_blk *blk;
1915         struct xfs_attr3_icleaf_hdr ichdr;
1916         struct xfs_buf          *bp;
1917         xfs_dablk_t             blkno;
1918         int                     bytes;
1919         int                     forward;
1920         int                     error;
1921         int                     retval;
1922         int                     i;
1923
1924         trace_xfs_attr_leaf_toosmall(state->args);
1925
1926         /*
1927          * Check for the degenerate case of the block being over 50% full.
1928          * If so, it's not worth even looking to see if we might be able
1929          * to coalesce with a sibling.
1930          */
1931         blk = &state->path.blk[ state->path.active-1 ];
1932         leaf = blk->bp->b_addr;
1933         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1934         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1935                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1936                 ichdr.usedbytes;
1937         if (bytes > (state->args->geo->blksize >> 1)) {
1938                 *action = 0;    /* blk over 50%, don't try to join */
1939                 return 0;
1940         }
1941
1942         /*
1943          * Check for the degenerate case of the block being empty.
1944          * If the block is empty, we'll simply delete it, no need to
1945          * coalesce it with a sibling block.  We choose (arbitrarily)
1946          * to merge with the forward block unless it is NULL.
1947          */
1948         if (ichdr.count == 0) {
1949                 /*
1950                  * Make altpath point to the block we want to keep and
1951                  * path point to the block we want to drop (this one).
1952                  */
1953                 forward = (ichdr.forw != 0);
1954                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1955                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1956                                                  0, &retval);
1957                 if (error)
1958                         return error;
1959                 if (retval) {
1960                         *action = 0;
1961                 } else {
1962                         *action = 2;
1963                 }
1964                 return 0;
1965         }
1966
1967         /*
1968          * Examine each sibling block to see if we can coalesce with
1969          * at least 25% free space to spare.  We need to figure out
1970          * whether to merge with the forward or the backward block.
1971          * We prefer coalescing with the lower numbered sibling so as
1972          * to shrink an attribute list over time.
1973          */
1974         /* start with smaller blk num */
1975         forward = ichdr.forw < ichdr.back;
1976         for (i = 0; i < 2; forward = !forward, i++) {
1977                 struct xfs_attr3_icleaf_hdr ichdr2;
1978                 if (forward)
1979                         blkno = ichdr.forw;
1980                 else
1981                         blkno = ichdr.back;
1982                 if (blkno == 0)
1983                         continue;
1984                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1985                                         blkno, &bp);
1986                 if (error)
1987                         return error;
1988
1989                 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1990
1991                 bytes = state->args->geo->blksize -
1992                         (state->args->geo->blksize >> 2) -
1993                         ichdr.usedbytes - ichdr2.usedbytes -
1994                         ((ichdr.count + ichdr2.count) *
1995                                         sizeof(xfs_attr_leaf_entry_t)) -
1996                         xfs_attr3_leaf_hdr_size(leaf);
1997
1998                 xfs_trans_brelse(state->args->trans, bp);
1999                 if (bytes >= 0)
2000                         break;  /* fits with at least 25% to spare */
2001         }
2002         if (i >= 2) {
2003                 *action = 0;
2004                 return 0;
2005         }
2006
2007         /*
2008          * Make altpath point to the block we want to keep (the lower
2009          * numbered block) and path point to the block we want to drop.
2010          */
2011         memcpy(&state->altpath, &state->path, sizeof(state->path));
2012         if (blkno < blk->blkno) {
2013                 error = xfs_da3_path_shift(state, &state->altpath, forward,
2014                                                  0, &retval);
2015         } else {
2016                 error = xfs_da3_path_shift(state, &state->path, forward,
2017                                                  0, &retval);
2018         }
2019         if (error)
2020                 return error;
2021         if (retval) {
2022                 *action = 0;
2023         } else {
2024                 *action = 1;
2025         }
2026         return 0;
2027 }
2028
2029 /*
2030  * Remove a name from the leaf attribute list structure.
2031  *
2032  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
2033  * If two leaves are 37% full, when combined they will leave 25% free.
2034  */
2035 int
2036 xfs_attr3_leaf_remove(
2037         struct xfs_buf          *bp,
2038         struct xfs_da_args      *args)
2039 {
2040         struct xfs_attr_leafblock *leaf;
2041         struct xfs_attr3_icleaf_hdr ichdr;
2042         struct xfs_attr_leaf_entry *entry;
2043         int                     before;
2044         int                     after;
2045         int                     smallest;
2046         int                     entsize;
2047         int                     tablesize;
2048         int                     tmp;
2049         int                     i;
2050
2051         trace_xfs_attr_leaf_remove(args);
2052
2053         leaf = bp->b_addr;
2054         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2055
2056         ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
2057         ASSERT(args->index >= 0 && args->index < ichdr.count);
2058         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2059                                         xfs_attr3_leaf_hdr_size(leaf));
2060
2061         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2062
2063         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2064         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2065
2066         /*
2067          * Scan through free region table:
2068          *    check for adjacency of free'd entry with an existing one,
2069          *    find smallest free region in case we need to replace it,
2070          *    adjust any map that borders the entry table,
2071          */
2072         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2073                                         + xfs_attr3_leaf_hdr_size(leaf);
2074         tmp = ichdr.freemap[0].size;
2075         before = after = -1;
2076         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2077         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2078         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2079                 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
2080                 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
2081                 if (ichdr.freemap[i].base == tablesize) {
2082                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2083                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2084                 }
2085
2086                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2087                                 be16_to_cpu(entry->nameidx)) {
2088                         before = i;
2089                 } else if (ichdr.freemap[i].base ==
2090                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2091                         after = i;
2092                 } else if (ichdr.freemap[i].size < tmp) {
2093                         tmp = ichdr.freemap[i].size;
2094                         smallest = i;
2095                 }
2096         }
2097
2098         /*
2099          * Coalesce adjacent freemap regions,
2100          * or replace the smallest region.
2101          */
2102         if ((before >= 0) || (after >= 0)) {
2103                 if ((before >= 0) && (after >= 0)) {
2104                         ichdr.freemap[before].size += entsize;
2105                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2106                         ichdr.freemap[after].base = 0;
2107                         ichdr.freemap[after].size = 0;
2108                 } else if (before >= 0) {
2109                         ichdr.freemap[before].size += entsize;
2110                 } else {
2111                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2112                         ichdr.freemap[after].size += entsize;
2113                 }
2114         } else {
2115                 /*
2116                  * Replace smallest region (if it is smaller than free'd entry)
2117                  */
2118                 if (ichdr.freemap[smallest].size < entsize) {
2119                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2120                         ichdr.freemap[smallest].size = entsize;
2121                 }
2122         }
2123
2124         /*
2125          * Did we remove the first entry?
2126          */
2127         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2128                 smallest = 1;
2129         else
2130                 smallest = 0;
2131
2132         /*
2133          * Compress the remaining entries and zero out the removed stuff.
2134          */
2135         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2136         ichdr.usedbytes -= entsize;
2137         xfs_trans_log_buf(args->trans, bp,
2138              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2139                                    entsize));
2140
2141         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2142         memmove(entry, entry + 1, tmp);
2143         ichdr.count--;
2144         xfs_trans_log_buf(args->trans, bp,
2145             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2146
2147         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2148         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2149
2150         /*
2151          * If we removed the first entry, re-find the first used byte
2152          * in the name area.  Note that if the entry was the "firstused",
2153          * then we don't have a "hole" in our block resulting from
2154          * removing the name.
2155          */
2156         if (smallest) {
2157                 tmp = args->geo->blksize;
2158                 entry = xfs_attr3_leaf_entryp(leaf);
2159                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2160                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2161                         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2162
2163                         if (be16_to_cpu(entry->nameidx) < tmp)
2164                                 tmp = be16_to_cpu(entry->nameidx);
2165                 }
2166                 ichdr.firstused = tmp;
2167                 ASSERT(ichdr.firstused != 0);
2168         } else {
2169                 ichdr.holes = 1;        /* mark as needing compaction */
2170         }
2171         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2172         xfs_trans_log_buf(args->trans, bp,
2173                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2174                                           xfs_attr3_leaf_hdr_size(leaf)));
2175
2176         /*
2177          * Check if leaf is less than 50% full, caller may want to
2178          * "join" the leaf with a sibling if so.
2179          */
2180         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2181               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2182
2183         return tmp < args->geo->magicpct; /* leaf is < 37% full */
2184 }
2185
2186 /*
2187  * Move all the attribute list entries from drop_leaf into save_leaf.
2188  */
2189 void
2190 xfs_attr3_leaf_unbalance(
2191         struct xfs_da_state     *state,
2192         struct xfs_da_state_blk *drop_blk,
2193         struct xfs_da_state_blk *save_blk)
2194 {
2195         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2196         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2197         struct xfs_attr3_icleaf_hdr drophdr;
2198         struct xfs_attr3_icleaf_hdr savehdr;
2199         struct xfs_attr_leaf_entry *entry;
2200
2201         trace_xfs_attr_leaf_unbalance(state->args);
2202
2203         drop_leaf = drop_blk->bp->b_addr;
2204         save_leaf = save_blk->bp->b_addr;
2205         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2206         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2207         entry = xfs_attr3_leaf_entryp(drop_leaf);
2208
2209         /*
2210          * Save last hashval from dying block for later Btree fixup.
2211          */
2212         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2213
2214         /*
2215          * Check if we need a temp buffer, or can we do it in place.
2216          * Note that we don't check "leaf" for holes because we will
2217          * always be dropping it, toosmall() decided that for us already.
2218          */
2219         if (savehdr.holes == 0) {
2220                 /*
2221                  * dest leaf has no holes, so we add there.  May need
2222                  * to make some room in the entry array.
2223                  */
2224                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2225                                          drop_blk->bp, &drophdr)) {
2226                         xfs_attr3_leaf_moveents(state->args,
2227                                                 drop_leaf, &drophdr, 0,
2228                                                 save_leaf, &savehdr, 0,
2229                                                 drophdr.count);
2230                 } else {
2231                         xfs_attr3_leaf_moveents(state->args,
2232                                                 drop_leaf, &drophdr, 0,
2233                                                 save_leaf, &savehdr,
2234                                                 savehdr.count, drophdr.count);
2235                 }
2236         } else {
2237                 /*
2238                  * Destination has holes, so we make a temporary copy
2239                  * of the leaf and add them both to that.
2240                  */
2241                 struct xfs_attr_leafblock *tmp_leaf;
2242                 struct xfs_attr3_icleaf_hdr tmphdr;
2243
2244                 tmp_leaf = kmem_zalloc(state->args->geo->blksize, 0);
2245
2246                 /*
2247                  * Copy the header into the temp leaf so that all the stuff
2248                  * not in the incore header is present and gets copied back in
2249                  * once we've moved all the entries.
2250                  */
2251                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2252
2253                 memset(&tmphdr, 0, sizeof(tmphdr));
2254                 tmphdr.magic = savehdr.magic;
2255                 tmphdr.forw = savehdr.forw;
2256                 tmphdr.back = savehdr.back;
2257                 tmphdr.firstused = state->args->geo->blksize;
2258
2259                 /* write the header to the temp buffer to initialise it */
2260                 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2261
2262                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2263                                          drop_blk->bp, &drophdr)) {
2264                         xfs_attr3_leaf_moveents(state->args,
2265                                                 drop_leaf, &drophdr, 0,
2266                                                 tmp_leaf, &tmphdr, 0,
2267                                                 drophdr.count);
2268                         xfs_attr3_leaf_moveents(state->args,
2269                                                 save_leaf, &savehdr, 0,
2270                                                 tmp_leaf, &tmphdr, tmphdr.count,
2271                                                 savehdr.count);
2272                 } else {
2273                         xfs_attr3_leaf_moveents(state->args,
2274                                                 save_leaf, &savehdr, 0,
2275                                                 tmp_leaf, &tmphdr, 0,
2276                                                 savehdr.count);
2277                         xfs_attr3_leaf_moveents(state->args,
2278                                                 drop_leaf, &drophdr, 0,
2279                                                 tmp_leaf, &tmphdr, tmphdr.count,
2280                                                 drophdr.count);
2281                 }
2282                 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2283                 savehdr = tmphdr; /* struct copy */
2284                 kmem_free(tmp_leaf);
2285         }
2286
2287         xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2288         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2289                                            state->args->geo->blksize - 1);
2290
2291         /*
2292          * Copy out last hashval in each block for B-tree code.
2293          */
2294         entry = xfs_attr3_leaf_entryp(save_leaf);
2295         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2296 }
2297
2298 /*========================================================================
2299  * Routines used for finding things in the Btree.
2300  *========================================================================*/
2301
2302 /*
2303  * Look up a name in a leaf attribute list structure.
2304  * This is the internal routine, it uses the caller's buffer.
2305  *
2306  * Note that duplicate keys are allowed, but only check within the
2307  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2308  *
2309  * Return in args->index the index into the entry[] array of either
2310  * the found entry, or where the entry should have been (insert before
2311  * that entry).
2312  *
2313  * Don't change the args->value unless we find the attribute.
2314  */
2315 int
2316 xfs_attr3_leaf_lookup_int(
2317         struct xfs_buf          *bp,
2318         struct xfs_da_args      *args)
2319 {
2320         struct xfs_attr_leafblock *leaf;
2321         struct xfs_attr3_icleaf_hdr ichdr;
2322         struct xfs_attr_leaf_entry *entry;
2323         struct xfs_attr_leaf_entry *entries;
2324         struct xfs_attr_leaf_name_local *name_loc;
2325         struct xfs_attr_leaf_name_remote *name_rmt;
2326         xfs_dahash_t            hashval;
2327         int                     probe;
2328         int                     span;
2329
2330         trace_xfs_attr_leaf_lookup(args);
2331
2332         leaf = bp->b_addr;
2333         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2334         entries = xfs_attr3_leaf_entryp(leaf);
2335         if (ichdr.count >= args->geo->blksize / 8) {
2336                 xfs_buf_mark_corrupt(bp);
2337                 return -EFSCORRUPTED;
2338         }
2339
2340         /*
2341          * Binary search.  (note: small blocks will skip this loop)
2342          */
2343         hashval = args->hashval;
2344         probe = span = ichdr.count / 2;
2345         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2346                 span /= 2;
2347                 if (be32_to_cpu(entry->hashval) < hashval)
2348                         probe += span;
2349                 else if (be32_to_cpu(entry->hashval) > hashval)
2350                         probe -= span;
2351                 else
2352                         break;
2353         }
2354         if (!(probe >= 0 && (!ichdr.count || probe < ichdr.count))) {
2355                 xfs_buf_mark_corrupt(bp);
2356                 return -EFSCORRUPTED;
2357         }
2358         if (!(span <= 4 || be32_to_cpu(entry->hashval) == hashval)) {
2359                 xfs_buf_mark_corrupt(bp);
2360                 return -EFSCORRUPTED;
2361         }
2362
2363         /*
2364          * Since we may have duplicate hashval's, find the first matching
2365          * hashval in the leaf.
2366          */
2367         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2368                 entry--;
2369                 probe--;
2370         }
2371         while (probe < ichdr.count &&
2372                be32_to_cpu(entry->hashval) < hashval) {
2373                 entry++;
2374                 probe++;
2375         }
2376         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2377                 args->index = probe;
2378                 return -ENOATTR;
2379         }
2380
2381         /*
2382          * Duplicate keys may be present, so search all of them for a match.
2383          */
2384         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2385                         entry++, probe++) {
2386 /*
2387  * GROT: Add code to remove incomplete entries.
2388  */
2389                 if (entry->flags & XFS_ATTR_LOCAL) {
2390                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2391                         if (!xfs_attr_match(args, name_loc->namelen,
2392                                         name_loc->nameval, entry->flags))
2393                                 continue;
2394                         args->index = probe;
2395                         return -EEXIST;
2396                 } else {
2397                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2398                         if (!xfs_attr_match(args, name_rmt->namelen,
2399                                         name_rmt->name, entry->flags))
2400                                 continue;
2401                         args->index = probe;
2402                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2403                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2404                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2405                                                         args->dp->i_mount,
2406                                                         args->rmtvaluelen);
2407                         return -EEXIST;
2408                 }
2409         }
2410         args->index = probe;
2411         return -ENOATTR;
2412 }
2413
2414 /*
2415  * Get the value associated with an attribute name from a leaf attribute
2416  * list structure.
2417  *
2418  * If args->valuelen is zero, only the length needs to be returned.  Unlike a
2419  * lookup, we only return an error if the attribute does not exist or we can't
2420  * retrieve the value.
2421  */
2422 int
2423 xfs_attr3_leaf_getvalue(
2424         struct xfs_buf          *bp,
2425         struct xfs_da_args      *args)
2426 {
2427         struct xfs_attr_leafblock *leaf;
2428         struct xfs_attr3_icleaf_hdr ichdr;
2429         struct xfs_attr_leaf_entry *entry;
2430         struct xfs_attr_leaf_name_local *name_loc;
2431         struct xfs_attr_leaf_name_remote *name_rmt;
2432
2433         leaf = bp->b_addr;
2434         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2435         ASSERT(ichdr.count < args->geo->blksize / 8);
2436         ASSERT(args->index < ichdr.count);
2437
2438         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2439         if (entry->flags & XFS_ATTR_LOCAL) {
2440                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2441                 ASSERT(name_loc->namelen == args->namelen);
2442                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2443                 return xfs_attr_copy_value(args,
2444                                         &name_loc->nameval[args->namelen],
2445                                         be16_to_cpu(name_loc->valuelen));
2446         }
2447
2448         name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2449         ASSERT(name_rmt->namelen == args->namelen);
2450         ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2451         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2452         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2453         args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2454                                                args->rmtvaluelen);
2455         return xfs_attr_copy_value(args, NULL, args->rmtvaluelen);
2456 }
2457
2458 /*========================================================================
2459  * Utility routines.
2460  *========================================================================*/
2461
2462 /*
2463  * Move the indicated entries from one leaf to another.
2464  * NOTE: this routine modifies both source and destination leaves.
2465  */
2466 /*ARGSUSED*/
2467 STATIC void
2468 xfs_attr3_leaf_moveents(
2469         struct xfs_da_args              *args,
2470         struct xfs_attr_leafblock       *leaf_s,
2471         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2472         int                             start_s,
2473         struct xfs_attr_leafblock       *leaf_d,
2474         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2475         int                             start_d,
2476         int                             count)
2477 {
2478         struct xfs_attr_leaf_entry      *entry_s;
2479         struct xfs_attr_leaf_entry      *entry_d;
2480         int                             desti;
2481         int                             tmp;
2482         int                             i;
2483
2484         /*
2485          * Check for nothing to do.
2486          */
2487         if (count == 0)
2488                 return;
2489
2490         /*
2491          * Set up environment.
2492          */
2493         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2494                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2495         ASSERT(ichdr_s->magic == ichdr_d->magic);
2496         ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2497         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2498                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2499         ASSERT(ichdr_d->count < args->geo->blksize / 8);
2500         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2501                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2502
2503         ASSERT(start_s < ichdr_s->count);
2504         ASSERT(start_d <= ichdr_d->count);
2505         ASSERT(count <= ichdr_s->count);
2506
2507
2508         /*
2509          * Move the entries in the destination leaf up to make a hole?
2510          */
2511         if (start_d < ichdr_d->count) {
2512                 tmp  = ichdr_d->count - start_d;
2513                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2514                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2515                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2516                 memmove(entry_d, entry_s, tmp);
2517         }
2518
2519         /*
2520          * Copy all entry's in the same (sorted) order,
2521          * but allocate attribute info packed and in sequence.
2522          */
2523         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2524         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2525         desti = start_d;
2526         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2527                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2528                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2529 #ifdef GROT
2530                 /*
2531                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2532                  * may also need to change the insertion index.  Code turned
2533                  * off for 6.2, should be revisited later.
2534                  */
2535                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2536                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2537                         ichdr_s->usedbytes -= tmp;
2538                         ichdr_s->count -= 1;
2539                         entry_d--;      /* to compensate for ++ in loop hdr */
2540                         desti--;
2541                         if ((start_s + i) < offset)
2542                                 result++;       /* insertion index adjustment */
2543                 } else {
2544 #endif /* GROT */
2545                         ichdr_d->firstused -= tmp;
2546                         /* both on-disk, don't endian flip twice */
2547                         entry_d->hashval = entry_s->hashval;
2548                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2549                         entry_d->flags = entry_s->flags;
2550                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2551                                                         <= args->geo->blksize);
2552                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2553                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2554                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2555                                                         <= args->geo->blksize);
2556                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2557                         ichdr_s->usedbytes -= tmp;
2558                         ichdr_d->usedbytes += tmp;
2559                         ichdr_s->count -= 1;
2560                         ichdr_d->count += 1;
2561                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2562                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2563                         ASSERT(ichdr_d->firstused >= tmp);
2564 #ifdef GROT
2565                 }
2566 #endif /* GROT */
2567         }
2568
2569         /*
2570          * Zero out the entries we just copied.
2571          */
2572         if (start_s == ichdr_s->count) {
2573                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2574                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2575                 ASSERT(((char *)entry_s + tmp) <=
2576                        ((char *)leaf_s + args->geo->blksize));
2577                 memset(entry_s, 0, tmp);
2578         } else {
2579                 /*
2580                  * Move the remaining entries down to fill the hole,
2581                  * then zero the entries at the top.
2582                  */
2583                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2584                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2585                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2586                 memmove(entry_d, entry_s, tmp);
2587
2588                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2589                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2590                 ASSERT(((char *)entry_s + tmp) <=
2591                        ((char *)leaf_s + args->geo->blksize));
2592                 memset(entry_s, 0, tmp);
2593         }
2594
2595         /*
2596          * Fill in the freemap information
2597          */
2598         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2599         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2600         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2601         ichdr_d->freemap[1].base = 0;
2602         ichdr_d->freemap[2].base = 0;
2603         ichdr_d->freemap[1].size = 0;
2604         ichdr_d->freemap[2].size = 0;
2605         ichdr_s->holes = 1;     /* leaf may not be compact */
2606 }
2607
2608 /*
2609  * Pick up the last hashvalue from a leaf block.
2610  */
2611 xfs_dahash_t
2612 xfs_attr_leaf_lasthash(
2613         struct xfs_buf  *bp,
2614         int             *count)
2615 {
2616         struct xfs_attr3_icleaf_hdr ichdr;
2617         struct xfs_attr_leaf_entry *entries;
2618         struct xfs_mount *mp = bp->b_mount;
2619
2620         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2621         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2622         if (count)
2623                 *count = ichdr.count;
2624         if (!ichdr.count)
2625                 return 0;
2626         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2627 }
2628
2629 /*
2630  * Calculate the number of bytes used to store the indicated attribute
2631  * (whether local or remote only calculate bytes in this block).
2632  */
2633 STATIC int
2634 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2635 {
2636         struct xfs_attr_leaf_entry *entries;
2637         xfs_attr_leaf_name_local_t *name_loc;
2638         xfs_attr_leaf_name_remote_t *name_rmt;
2639         int size;
2640
2641         entries = xfs_attr3_leaf_entryp(leaf);
2642         if (entries[index].flags & XFS_ATTR_LOCAL) {
2643                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2644                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2645                                                    be16_to_cpu(name_loc->valuelen));
2646         } else {
2647                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2648                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2649         }
2650         return size;
2651 }
2652
2653 /*
2654  * Calculate the number of bytes that would be required to store the new
2655  * attribute (whether local or remote only calculate bytes in this block).
2656  * This routine decides as a side effect whether the attribute will be
2657  * a "local" or a "remote" attribute.
2658  */
2659 int
2660 xfs_attr_leaf_newentsize(
2661         struct xfs_da_args      *args,
2662         int                     *local)
2663 {
2664         int                     size;
2665
2666         size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2667         if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2668                 if (local)
2669                         *local = 1;
2670                 return size;
2671         }
2672         if (local)
2673                 *local = 0;
2674         return xfs_attr_leaf_entsize_remote(args->namelen);
2675 }
2676
2677
2678 /*========================================================================
2679  * Manage the INCOMPLETE flag in a leaf entry
2680  *========================================================================*/
2681
2682 /*
2683  * Clear the INCOMPLETE flag on an entry in a leaf block.
2684  */
2685 int
2686 xfs_attr3_leaf_clearflag(
2687         struct xfs_da_args      *args)
2688 {
2689         struct xfs_attr_leafblock *leaf;
2690         struct xfs_attr_leaf_entry *entry;
2691         struct xfs_attr_leaf_name_remote *name_rmt;
2692         struct xfs_buf          *bp;
2693         int                     error;
2694 #ifdef DEBUG
2695         struct xfs_attr3_icleaf_hdr ichdr;
2696         xfs_attr_leaf_name_local_t *name_loc;
2697         int namelen;
2698         char *name;
2699 #endif /* DEBUG */
2700
2701         trace_xfs_attr_leaf_clearflag(args);
2702         /*
2703          * Set up the operation.
2704          */
2705         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2706         if (error)
2707                 return error;
2708
2709         leaf = bp->b_addr;
2710         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2711         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2712
2713 #ifdef DEBUG
2714         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2715         ASSERT(args->index < ichdr.count);
2716         ASSERT(args->index >= 0);
2717
2718         if (entry->flags & XFS_ATTR_LOCAL) {
2719                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2720                 namelen = name_loc->namelen;
2721                 name = (char *)name_loc->nameval;
2722         } else {
2723                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2724                 namelen = name_rmt->namelen;
2725                 name = (char *)name_rmt->name;
2726         }
2727         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2728         ASSERT(namelen == args->namelen);
2729         ASSERT(memcmp(name, args->name, namelen) == 0);
2730 #endif /* DEBUG */
2731
2732         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2733         xfs_trans_log_buf(args->trans, bp,
2734                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2735
2736         if (args->rmtblkno) {
2737                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2738                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2739                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2740                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2741                 xfs_trans_log_buf(args->trans, bp,
2742                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2743         }
2744
2745         /*
2746          * Commit the flag value change and start the next trans in series.
2747          */
2748         return xfs_trans_roll_inode(&args->trans, args->dp);
2749 }
2750
2751 /*
2752  * Set the INCOMPLETE flag on an entry in a leaf block.
2753  */
2754 int
2755 xfs_attr3_leaf_setflag(
2756         struct xfs_da_args      *args)
2757 {
2758         struct xfs_attr_leafblock *leaf;
2759         struct xfs_attr_leaf_entry *entry;
2760         struct xfs_attr_leaf_name_remote *name_rmt;
2761         struct xfs_buf          *bp;
2762         int error;
2763 #ifdef DEBUG
2764         struct xfs_attr3_icleaf_hdr ichdr;
2765 #endif
2766
2767         trace_xfs_attr_leaf_setflag(args);
2768
2769         /*
2770          * Set up the operation.
2771          */
2772         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp);
2773         if (error)
2774                 return error;
2775
2776         leaf = bp->b_addr;
2777 #ifdef DEBUG
2778         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2779         ASSERT(args->index < ichdr.count);
2780         ASSERT(args->index >= 0);
2781 #endif
2782         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2783
2784         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2785         entry->flags |= XFS_ATTR_INCOMPLETE;
2786         xfs_trans_log_buf(args->trans, bp,
2787                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2788         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2789                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2790                 name_rmt->valueblk = 0;
2791                 name_rmt->valuelen = 0;
2792                 xfs_trans_log_buf(args->trans, bp,
2793                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2794         }
2795
2796         /*
2797          * Commit the flag value change and start the next trans in series.
2798          */
2799         return xfs_trans_roll_inode(&args->trans, args->dp);
2800 }
2801
2802 /*
2803  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2804  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2805  * entry given by args->blkno2/index2.
2806  *
2807  * Note that they could be in different blocks, or in the same block.
2808  */
2809 int
2810 xfs_attr3_leaf_flipflags(
2811         struct xfs_da_args      *args)
2812 {
2813         struct xfs_attr_leafblock *leaf1;
2814         struct xfs_attr_leafblock *leaf2;
2815         struct xfs_attr_leaf_entry *entry1;
2816         struct xfs_attr_leaf_entry *entry2;
2817         struct xfs_attr_leaf_name_remote *name_rmt;
2818         struct xfs_buf          *bp1;
2819         struct xfs_buf          *bp2;
2820         int error;
2821 #ifdef DEBUG
2822         struct xfs_attr3_icleaf_hdr ichdr1;
2823         struct xfs_attr3_icleaf_hdr ichdr2;
2824         xfs_attr_leaf_name_local_t *name_loc;
2825         int namelen1, namelen2;
2826         char *name1, *name2;
2827 #endif /* DEBUG */
2828
2829         trace_xfs_attr_leaf_flipflags(args);
2830
2831         /*
2832          * Read the block containing the "old" attr
2833          */
2834         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, &bp1);
2835         if (error)
2836                 return error;
2837
2838         /*
2839          * Read the block containing the "new" attr, if it is different
2840          */
2841         if (args->blkno2 != args->blkno) {
2842                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2843                                            &bp2);
2844                 if (error)
2845                         return error;
2846         } else {
2847                 bp2 = bp1;
2848         }
2849
2850         leaf1 = bp1->b_addr;
2851         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2852
2853         leaf2 = bp2->b_addr;
2854         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2855
2856 #ifdef DEBUG
2857         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2858         ASSERT(args->index < ichdr1.count);
2859         ASSERT(args->index >= 0);
2860
2861         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2862         ASSERT(args->index2 < ichdr2.count);
2863         ASSERT(args->index2 >= 0);
2864
2865         if (entry1->flags & XFS_ATTR_LOCAL) {
2866                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2867                 namelen1 = name_loc->namelen;
2868                 name1 = (char *)name_loc->nameval;
2869         } else {
2870                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2871                 namelen1 = name_rmt->namelen;
2872                 name1 = (char *)name_rmt->name;
2873         }
2874         if (entry2->flags & XFS_ATTR_LOCAL) {
2875                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2876                 namelen2 = name_loc->namelen;
2877                 name2 = (char *)name_loc->nameval;
2878         } else {
2879                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2880                 namelen2 = name_rmt->namelen;
2881                 name2 = (char *)name_rmt->name;
2882         }
2883         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2884         ASSERT(namelen1 == namelen2);
2885         ASSERT(memcmp(name1, name2, namelen1) == 0);
2886 #endif /* DEBUG */
2887
2888         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2889         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2890
2891         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2892         xfs_trans_log_buf(args->trans, bp1,
2893                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2894         if (args->rmtblkno) {
2895                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2896                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2897                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2898                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2899                 xfs_trans_log_buf(args->trans, bp1,
2900                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2901         }
2902
2903         entry2->flags |= XFS_ATTR_INCOMPLETE;
2904         xfs_trans_log_buf(args->trans, bp2,
2905                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2906         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2907                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2908                 name_rmt->valueblk = 0;
2909                 name_rmt->valuelen = 0;
2910                 xfs_trans_log_buf(args->trans, bp2,
2911                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2912         }
2913
2914         /*
2915          * Commit the flag value change and start the next trans in series.
2916          */
2917         error = xfs_trans_roll_inode(&args->trans, args->dp);
2918
2919         return error;
2920 }