]> Pileus Git - ~andy/linux/blob - fs/xfs/xfs_attr_leaf.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
[~andy/linux] / fs / xfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_attr_remote.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_cksum.h"
44 #include "xfs_dinode.h"
45 #include "xfs_dir2.h"
46
47
48 /*
49  * xfs_attr_leaf.c
50  *
51  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52  */
53
54 /*========================================================================
55  * Function prototypes for the kernel.
56  *========================================================================*/
57
58 /*
59  * Routines used for growing the Btree.
60  */
61 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
62                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
63 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
64                                    struct xfs_attr3_icleaf_hdr *ichdr,
65                                    struct xfs_da_args *args, int freemap_index);
66 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
67                                    struct xfs_attr3_icleaf_hdr *ichdr,
68                                    struct xfs_buf *leaf_buffer);
69 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
70                                                    xfs_da_state_blk_t *blk1,
71                                                    xfs_da_state_blk_t *blk2);
72 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
73                         xfs_da_state_blk_t *leaf_blk_1,
74                         struct xfs_attr3_icleaf_hdr *ichdr1,
75                         xfs_da_state_blk_t *leaf_blk_2,
76                         struct xfs_attr3_icleaf_hdr *ichdr2,
77                         int *number_entries_in_blk1,
78                         int *number_usedbytes_in_blk1);
79
80 /*
81  * Utility routines.
82  */
83 STATIC void xfs_attr3_leaf_moveents(struct xfs_attr_leafblock *src_leaf,
84                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
85                         struct xfs_attr_leafblock *dst_leaf,
86                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87                         int move_count, struct xfs_mount *mp);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
89
90 void
91 xfs_attr3_leaf_hdr_from_disk(
92         struct xfs_attr3_icleaf_hdr     *to,
93         struct xfs_attr_leafblock       *from)
94 {
95         int     i;
96
97         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
98                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
99
100         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
101                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
102
103                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
104                 to->back = be32_to_cpu(hdr3->info.hdr.back);
105                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
106                 to->count = be16_to_cpu(hdr3->count);
107                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
108                 to->firstused = be16_to_cpu(hdr3->firstused);
109                 to->holes = hdr3->holes;
110
111                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
112                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
113                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
114                 }
115                 return;
116         }
117         to->forw = be32_to_cpu(from->hdr.info.forw);
118         to->back = be32_to_cpu(from->hdr.info.back);
119         to->magic = be16_to_cpu(from->hdr.info.magic);
120         to->count = be16_to_cpu(from->hdr.count);
121         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
122         to->firstused = be16_to_cpu(from->hdr.firstused);
123         to->holes = from->hdr.holes;
124
125         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
126                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
127                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
128         }
129 }
130
131 void
132 xfs_attr3_leaf_hdr_to_disk(
133         struct xfs_attr_leafblock       *to,
134         struct xfs_attr3_icleaf_hdr     *from)
135 {
136         int     i;
137
138         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
139                from->magic == XFS_ATTR3_LEAF_MAGIC);
140
141         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
142                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
143
144                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
145                 hdr3->info.hdr.back = cpu_to_be32(from->back);
146                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
147                 hdr3->count = cpu_to_be16(from->count);
148                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
149                 hdr3->firstused = cpu_to_be16(from->firstused);
150                 hdr3->holes = from->holes;
151                 hdr3->pad1 = 0;
152
153                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
154                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
155                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
156                 }
157                 return;
158         }
159         to->hdr.info.forw = cpu_to_be32(from->forw);
160         to->hdr.info.back = cpu_to_be32(from->back);
161         to->hdr.info.magic = cpu_to_be16(from->magic);
162         to->hdr.count = cpu_to_be16(from->count);
163         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
164         to->hdr.firstused = cpu_to_be16(from->firstused);
165         to->hdr.holes = from->holes;
166         to->hdr.pad1 = 0;
167
168         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
169                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
170                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
171         }
172 }
173
174 static bool
175 xfs_attr3_leaf_verify(
176         struct xfs_buf          *bp)
177 {
178         struct xfs_mount        *mp = bp->b_target->bt_mount;
179         struct xfs_attr_leafblock *leaf = bp->b_addr;
180         struct xfs_attr3_icleaf_hdr ichdr;
181
182         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
183
184         if (xfs_sb_version_hascrc(&mp->m_sb)) {
185                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
186
187                 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
188                         return false;
189
190                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
191                         return false;
192                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
193                         return false;
194         } else {
195                 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
196                         return false;
197         }
198         if (ichdr.count == 0)
199                 return false;
200
201         /* XXX: need to range check rest of attr header values */
202         /* XXX: hash order check? */
203
204         return true;
205 }
206
207 static void
208 xfs_attr3_leaf_write_verify(
209         struct xfs_buf  *bp)
210 {
211         struct xfs_mount        *mp = bp->b_target->bt_mount;
212         struct xfs_buf_log_item *bip = bp->b_fspriv;
213         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
214
215         if (!xfs_attr3_leaf_verify(bp)) {
216                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
217                 xfs_buf_ioerror(bp, EFSCORRUPTED);
218                 return;
219         }
220
221         if (!xfs_sb_version_hascrc(&mp->m_sb))
222                 return;
223
224         if (bip)
225                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
226
227         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_ATTR3_LEAF_CRC_OFF);
228 }
229
230 /*
231  * leaf/node format detection on trees is sketchy, so a node read can be done on
232  * leaf level blocks when detection identifies the tree as a node format tree
233  * incorrectly. In this case, we need to swap the verifier to match the correct
234  * format of the block being read.
235  */
236 static void
237 xfs_attr3_leaf_read_verify(
238         struct xfs_buf          *bp)
239 {
240         struct xfs_mount        *mp = bp->b_target->bt_mount;
241
242         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
243              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
244                                           XFS_ATTR3_LEAF_CRC_OFF)) ||
245             !xfs_attr3_leaf_verify(bp)) {
246                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
247                 xfs_buf_ioerror(bp, EFSCORRUPTED);
248         }
249 }
250
251 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
252         .verify_read = xfs_attr3_leaf_read_verify,
253         .verify_write = xfs_attr3_leaf_write_verify,
254 };
255
256 int
257 xfs_attr3_leaf_read(
258         struct xfs_trans        *tp,
259         struct xfs_inode        *dp,
260         xfs_dablk_t             bno,
261         xfs_daddr_t             mappedbno,
262         struct xfs_buf          **bpp)
263 {
264         int                     err;
265
266         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
267                                 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
268         if (!err && tp)
269                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
270         return err;
271 }
272
273 /*========================================================================
274  * Namespace helper routines
275  *========================================================================*/
276
277 /*
278  * If namespace bits don't match return 0.
279  * If all match then return 1.
280  */
281 STATIC int
282 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
283 {
284         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
285 }
286
287
288 /*========================================================================
289  * External routines when attribute fork size < XFS_LITINO(mp).
290  *========================================================================*/
291
292 /*
293  * Query whether the requested number of additional bytes of extended
294  * attribute space will be able to fit inline.
295  *
296  * Returns zero if not, else the di_forkoff fork offset to be used in the
297  * literal area for attribute data once the new bytes have been added.
298  *
299  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
300  * special case for dev/uuid inodes, they have fixed size data forks.
301  */
302 int
303 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
304 {
305         int offset;
306         int minforkoff; /* lower limit on valid forkoff locations */
307         int maxforkoff; /* upper limit on valid forkoff locations */
308         int dsize;
309         xfs_mount_t *mp = dp->i_mount;
310
311         /* rounded down */
312         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
313
314         switch (dp->i_d.di_format) {
315         case XFS_DINODE_FMT_DEV:
316                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
317                 return (offset >= minforkoff) ? minforkoff : 0;
318         case XFS_DINODE_FMT_UUID:
319                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
320                 return (offset >= minforkoff) ? minforkoff : 0;
321         }
322
323         /*
324          * If the requested numbers of bytes is smaller or equal to the
325          * current attribute fork size we can always proceed.
326          *
327          * Note that if_bytes in the data fork might actually be larger than
328          * the current data fork size is due to delalloc extents. In that
329          * case either the extent count will go down when they are converted
330          * to real extents, or the delalloc conversion will take care of the
331          * literal area rebalancing.
332          */
333         if (bytes <= XFS_IFORK_ASIZE(dp))
334                 return dp->i_d.di_forkoff;
335
336         /*
337          * For attr2 we can try to move the forkoff if there is space in the
338          * literal area, but for the old format we are done if there is no
339          * space in the fixed attribute fork.
340          */
341         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
342                 return 0;
343
344         dsize = dp->i_df.if_bytes;
345
346         switch (dp->i_d.di_format) {
347         case XFS_DINODE_FMT_EXTENTS:
348                 /*
349                  * If there is no attr fork and the data fork is extents, 
350                  * determine if creating the default attr fork will result
351                  * in the extents form migrating to btree. If so, the
352                  * minimum offset only needs to be the space required for
353                  * the btree root.
354                  */
355                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
356                     xfs_default_attroffset(dp))
357                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
358                 break;
359         case XFS_DINODE_FMT_BTREE:
360                 /*
361                  * If we have a data btree then keep forkoff if we have one,
362                  * otherwise we are adding a new attr, so then we set
363                  * minforkoff to where the btree root can finish so we have
364                  * plenty of room for attrs
365                  */
366                 if (dp->i_d.di_forkoff) {
367                         if (offset < dp->i_d.di_forkoff)
368                                 return 0;
369                         return dp->i_d.di_forkoff;
370                 }
371                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
372                 break;
373         }
374
375         /*
376          * A data fork btree root must have space for at least
377          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
378          */
379         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
380         minforkoff = roundup(minforkoff, 8) >> 3;
381
382         /* attr fork btree root can have at least this many key/ptr pairs */
383         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
384                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
385         maxforkoff = maxforkoff >> 3;   /* rounded down */
386
387         if (offset >= maxforkoff)
388                 return maxforkoff;
389         if (offset >= minforkoff)
390                 return offset;
391         return 0;
392 }
393
394 /*
395  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
396  */
397 STATIC void
398 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
399 {
400         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
401             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
402                 spin_lock(&mp->m_sb_lock);
403                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
404                         xfs_sb_version_addattr2(&mp->m_sb);
405                         spin_unlock(&mp->m_sb_lock);
406                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
407                 } else
408                         spin_unlock(&mp->m_sb_lock);
409         }
410 }
411
412 /*
413  * Create the initial contents of a shortform attribute list.
414  */
415 void
416 xfs_attr_shortform_create(xfs_da_args_t *args)
417 {
418         xfs_attr_sf_hdr_t *hdr;
419         xfs_inode_t *dp;
420         xfs_ifork_t *ifp;
421
422         trace_xfs_attr_sf_create(args);
423
424         dp = args->dp;
425         ASSERT(dp != NULL);
426         ifp = dp->i_afp;
427         ASSERT(ifp != NULL);
428         ASSERT(ifp->if_bytes == 0);
429         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
430                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
431                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
432                 ifp->if_flags |= XFS_IFINLINE;
433         } else {
434                 ASSERT(ifp->if_flags & XFS_IFINLINE);
435         }
436         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
437         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
438         hdr->count = 0;
439         hdr->totsize = cpu_to_be16(sizeof(*hdr));
440         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
441 }
442
443 /*
444  * Add a name/value pair to the shortform attribute list.
445  * Overflow from the inode has already been checked for.
446  */
447 void
448 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
449 {
450         xfs_attr_shortform_t *sf;
451         xfs_attr_sf_entry_t *sfe;
452         int i, offset, size;
453         xfs_mount_t *mp;
454         xfs_inode_t *dp;
455         xfs_ifork_t *ifp;
456
457         trace_xfs_attr_sf_add(args);
458
459         dp = args->dp;
460         mp = dp->i_mount;
461         dp->i_d.di_forkoff = forkoff;
462
463         ifp = dp->i_afp;
464         ASSERT(ifp->if_flags & XFS_IFINLINE);
465         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
466         sfe = &sf->list[0];
467         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
468 #ifdef DEBUG
469                 if (sfe->namelen != args->namelen)
470                         continue;
471                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
472                         continue;
473                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
474                         continue;
475                 ASSERT(0);
476 #endif
477         }
478
479         offset = (char *)sfe - (char *)sf;
480         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
481         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
482         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
483         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
484
485         sfe->namelen = args->namelen;
486         sfe->valuelen = args->valuelen;
487         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
488         memcpy(sfe->nameval, args->name, args->namelen);
489         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
490         sf->hdr.count++;
491         be16_add_cpu(&sf->hdr.totsize, size);
492         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
493
494         xfs_sbversion_add_attr2(mp, args->trans);
495 }
496
497 /*
498  * After the last attribute is removed revert to original inode format,
499  * making all literal area available to the data fork once more.
500  */
501 STATIC void
502 xfs_attr_fork_reset(
503         struct xfs_inode        *ip,
504         struct xfs_trans        *tp)
505 {
506         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
507         ip->i_d.di_forkoff = 0;
508         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
509
510         ASSERT(ip->i_d.di_anextents == 0);
511         ASSERT(ip->i_afp == NULL);
512
513         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
514 }
515
516 /*
517  * Remove an attribute from the shortform attribute list structure.
518  */
519 int
520 xfs_attr_shortform_remove(xfs_da_args_t *args)
521 {
522         xfs_attr_shortform_t *sf;
523         xfs_attr_sf_entry_t *sfe;
524         int base, size=0, end, totsize, i;
525         xfs_mount_t *mp;
526         xfs_inode_t *dp;
527
528         trace_xfs_attr_sf_remove(args);
529
530         dp = args->dp;
531         mp = dp->i_mount;
532         base = sizeof(xfs_attr_sf_hdr_t);
533         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
534         sfe = &sf->list[0];
535         end = sf->hdr.count;
536         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
537                                         base += size, i++) {
538                 size = XFS_ATTR_SF_ENTSIZE(sfe);
539                 if (sfe->namelen != args->namelen)
540                         continue;
541                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
542                         continue;
543                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
544                         continue;
545                 break;
546         }
547         if (i == end)
548                 return(XFS_ERROR(ENOATTR));
549
550         /*
551          * Fix up the attribute fork data, covering the hole
552          */
553         end = base + size;
554         totsize = be16_to_cpu(sf->hdr.totsize);
555         if (end != totsize)
556                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
557         sf->hdr.count--;
558         be16_add_cpu(&sf->hdr.totsize, -size);
559
560         /*
561          * Fix up the start offset of the attribute fork
562          */
563         totsize -= size;
564         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
565             (mp->m_flags & XFS_MOUNT_ATTR2) &&
566             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
567             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
568                 xfs_attr_fork_reset(dp, args->trans);
569         } else {
570                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
571                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
572                 ASSERT(dp->i_d.di_forkoff);
573                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
574                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
575                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
576                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
577                 xfs_trans_log_inode(args->trans, dp,
578                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
579         }
580
581         xfs_sbversion_add_attr2(mp, args->trans);
582
583         return(0);
584 }
585
586 /*
587  * Look up a name in a shortform attribute list structure.
588  */
589 /*ARGSUSED*/
590 int
591 xfs_attr_shortform_lookup(xfs_da_args_t *args)
592 {
593         xfs_attr_shortform_t *sf;
594         xfs_attr_sf_entry_t *sfe;
595         int i;
596         xfs_ifork_t *ifp;
597
598         trace_xfs_attr_sf_lookup(args);
599
600         ifp = args->dp->i_afp;
601         ASSERT(ifp->if_flags & XFS_IFINLINE);
602         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
603         sfe = &sf->list[0];
604         for (i = 0; i < sf->hdr.count;
605                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
606                 if (sfe->namelen != args->namelen)
607                         continue;
608                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
609                         continue;
610                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
611                         continue;
612                 return(XFS_ERROR(EEXIST));
613         }
614         return(XFS_ERROR(ENOATTR));
615 }
616
617 /*
618  * Look up a name in a shortform attribute list structure.
619  */
620 /*ARGSUSED*/
621 int
622 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
623 {
624         xfs_attr_shortform_t *sf;
625         xfs_attr_sf_entry_t *sfe;
626         int i;
627
628         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
629         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
630         sfe = &sf->list[0];
631         for (i = 0; i < sf->hdr.count;
632                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
633                 if (sfe->namelen != args->namelen)
634                         continue;
635                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
636                         continue;
637                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
638                         continue;
639                 if (args->flags & ATTR_KERNOVAL) {
640                         args->valuelen = sfe->valuelen;
641                         return(XFS_ERROR(EEXIST));
642                 }
643                 if (args->valuelen < sfe->valuelen) {
644                         args->valuelen = sfe->valuelen;
645                         return(XFS_ERROR(ERANGE));
646                 }
647                 args->valuelen = sfe->valuelen;
648                 memcpy(args->value, &sfe->nameval[args->namelen],
649                                                     args->valuelen);
650                 return(XFS_ERROR(EEXIST));
651         }
652         return(XFS_ERROR(ENOATTR));
653 }
654
655 /*
656  * Convert from using the shortform to the leaf.
657  */
658 int
659 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
660 {
661         xfs_inode_t *dp;
662         xfs_attr_shortform_t *sf;
663         xfs_attr_sf_entry_t *sfe;
664         xfs_da_args_t nargs;
665         char *tmpbuffer;
666         int error, i, size;
667         xfs_dablk_t blkno;
668         struct xfs_buf *bp;
669         xfs_ifork_t *ifp;
670
671         trace_xfs_attr_sf_to_leaf(args);
672
673         dp = args->dp;
674         ifp = dp->i_afp;
675         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
676         size = be16_to_cpu(sf->hdr.totsize);
677         tmpbuffer = kmem_alloc(size, KM_SLEEP);
678         ASSERT(tmpbuffer != NULL);
679         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
680         sf = (xfs_attr_shortform_t *)tmpbuffer;
681
682         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
683         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
684
685         bp = NULL;
686         error = xfs_da_grow_inode(args, &blkno);
687         if (error) {
688                 /*
689                  * If we hit an IO error middle of the transaction inside
690                  * grow_inode(), we may have inconsistent data. Bail out.
691                  */
692                 if (error == EIO)
693                         goto out;
694                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
695                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
696                 goto out;
697         }
698
699         ASSERT(blkno == 0);
700         error = xfs_attr3_leaf_create(args, blkno, &bp);
701         if (error) {
702                 error = xfs_da_shrink_inode(args, 0, bp);
703                 bp = NULL;
704                 if (error)
705                         goto out;
706                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
707                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
708                 goto out;
709         }
710
711         memset((char *)&nargs, 0, sizeof(nargs));
712         nargs.dp = dp;
713         nargs.firstblock = args->firstblock;
714         nargs.flist = args->flist;
715         nargs.total = args->total;
716         nargs.whichfork = XFS_ATTR_FORK;
717         nargs.trans = args->trans;
718         nargs.op_flags = XFS_DA_OP_OKNOENT;
719
720         sfe = &sf->list[0];
721         for (i = 0; i < sf->hdr.count; i++) {
722                 nargs.name = sfe->nameval;
723                 nargs.namelen = sfe->namelen;
724                 nargs.value = &sfe->nameval[nargs.namelen];
725                 nargs.valuelen = sfe->valuelen;
726                 nargs.hashval = xfs_da_hashname(sfe->nameval,
727                                                 sfe->namelen);
728                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
729                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
730                 ASSERT(error == ENOATTR);
731                 error = xfs_attr3_leaf_add(bp, &nargs);
732                 ASSERT(error != ENOSPC);
733                 if (error)
734                         goto out;
735                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
736         }
737         error = 0;
738
739 out:
740         kmem_free(tmpbuffer);
741         return(error);
742 }
743
744 /*
745  * Check a leaf attribute block to see if all the entries would fit into
746  * a shortform attribute list.
747  */
748 int
749 xfs_attr_shortform_allfit(
750         struct xfs_buf          *bp,
751         struct xfs_inode        *dp)
752 {
753         struct xfs_attr_leafblock *leaf;
754         struct xfs_attr_leaf_entry *entry;
755         xfs_attr_leaf_name_local_t *name_loc;
756         struct xfs_attr3_icleaf_hdr leafhdr;
757         int                     bytes;
758         int                     i;
759
760         leaf = bp->b_addr;
761         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
762         entry = xfs_attr3_leaf_entryp(leaf);
763
764         bytes = sizeof(struct xfs_attr_sf_hdr);
765         for (i = 0; i < leafhdr.count; entry++, i++) {
766                 if (entry->flags & XFS_ATTR_INCOMPLETE)
767                         continue;               /* don't copy partial entries */
768                 if (!(entry->flags & XFS_ATTR_LOCAL))
769                         return(0);
770                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
771                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
772                         return(0);
773                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
774                         return(0);
775                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
776                                 + name_loc->namelen
777                                 + be16_to_cpu(name_loc->valuelen);
778         }
779         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
780             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
781             (bytes == sizeof(struct xfs_attr_sf_hdr)))
782                 return -1;
783         return xfs_attr_shortform_bytesfit(dp, bytes);
784 }
785
786 /*
787  * Convert a leaf attribute list to shortform attribute list
788  */
789 int
790 xfs_attr3_leaf_to_shortform(
791         struct xfs_buf          *bp,
792         struct xfs_da_args      *args,
793         int                     forkoff)
794 {
795         struct xfs_attr_leafblock *leaf;
796         struct xfs_attr3_icleaf_hdr ichdr;
797         struct xfs_attr_leaf_entry *entry;
798         struct xfs_attr_leaf_name_local *name_loc;
799         struct xfs_da_args      nargs;
800         struct xfs_inode        *dp = args->dp;
801         char                    *tmpbuffer;
802         int                     error;
803         int                     i;
804
805         trace_xfs_attr_leaf_to_sf(args);
806
807         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
808         if (!tmpbuffer)
809                 return ENOMEM;
810
811         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
812
813         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
814         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
815         entry = xfs_attr3_leaf_entryp(leaf);
816
817         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
818         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
819
820         /*
821          * Clean out the prior contents of the attribute list.
822          */
823         error = xfs_da_shrink_inode(args, 0, bp);
824         if (error)
825                 goto out;
826
827         if (forkoff == -1) {
828                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
829                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
830                 xfs_attr_fork_reset(dp, args->trans);
831                 goto out;
832         }
833
834         xfs_attr_shortform_create(args);
835
836         /*
837          * Copy the attributes
838          */
839         memset((char *)&nargs, 0, sizeof(nargs));
840         nargs.dp = dp;
841         nargs.firstblock = args->firstblock;
842         nargs.flist = args->flist;
843         nargs.total = args->total;
844         nargs.whichfork = XFS_ATTR_FORK;
845         nargs.trans = args->trans;
846         nargs.op_flags = XFS_DA_OP_OKNOENT;
847
848         for (i = 0; i < ichdr.count; entry++, i++) {
849                 if (entry->flags & XFS_ATTR_INCOMPLETE)
850                         continue;       /* don't copy partial entries */
851                 if (!entry->nameidx)
852                         continue;
853                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
854                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
855                 nargs.name = name_loc->nameval;
856                 nargs.namelen = name_loc->namelen;
857                 nargs.value = &name_loc->nameval[nargs.namelen];
858                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
859                 nargs.hashval = be32_to_cpu(entry->hashval);
860                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
861                 xfs_attr_shortform_add(&nargs, forkoff);
862         }
863         error = 0;
864
865 out:
866         kmem_free(tmpbuffer);
867         return error;
868 }
869
870 /*
871  * Convert from using a single leaf to a root node and a leaf.
872  */
873 int
874 xfs_attr3_leaf_to_node(
875         struct xfs_da_args      *args)
876 {
877         struct xfs_attr_leafblock *leaf;
878         struct xfs_attr3_icleaf_hdr icleafhdr;
879         struct xfs_attr_leaf_entry *entries;
880         struct xfs_da_node_entry *btree;
881         struct xfs_da3_icnode_hdr icnodehdr;
882         struct xfs_da_intnode   *node;
883         struct xfs_inode        *dp = args->dp;
884         struct xfs_mount        *mp = dp->i_mount;
885         struct xfs_buf          *bp1 = NULL;
886         struct xfs_buf          *bp2 = NULL;
887         xfs_dablk_t             blkno;
888         int                     error;
889
890         trace_xfs_attr_leaf_to_node(args);
891
892         error = xfs_da_grow_inode(args, &blkno);
893         if (error)
894                 goto out;
895         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
896         if (error)
897                 goto out;
898
899         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
900         if (error)
901                 goto out;
902
903         /* copy leaf to new buffer, update identifiers */
904         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
905         bp2->b_ops = bp1->b_ops;
906         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
907         if (xfs_sb_version_hascrc(&mp->m_sb)) {
908                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
909                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
910         }
911         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
912
913         /*
914          * Set up the new root node.
915          */
916         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
917         if (error)
918                 goto out;
919         node = bp1->b_addr;
920         dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
921         btree = dp->d_ops->node_tree_p(node);
922
923         leaf = bp2->b_addr;
924         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
925         entries = xfs_attr3_leaf_entryp(leaf);
926
927         /* both on-disk, don't endian-flip twice */
928         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
929         btree[0].before = cpu_to_be32(blkno);
930         icnodehdr.count = 1;
931         dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
932         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
933         error = 0;
934 out:
935         return error;
936 }
937
938 /*========================================================================
939  * Routines used for growing the Btree.
940  *========================================================================*/
941
942 /*
943  * Create the initial contents of a leaf attribute list
944  * or a leaf in a node attribute list.
945  */
946 STATIC int
947 xfs_attr3_leaf_create(
948         struct xfs_da_args      *args,
949         xfs_dablk_t             blkno,
950         struct xfs_buf          **bpp)
951 {
952         struct xfs_attr_leafblock *leaf;
953         struct xfs_attr3_icleaf_hdr ichdr;
954         struct xfs_inode        *dp = args->dp;
955         struct xfs_mount        *mp = dp->i_mount;
956         struct xfs_buf          *bp;
957         int                     error;
958
959         trace_xfs_attr_leaf_create(args);
960
961         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
962                                             XFS_ATTR_FORK);
963         if (error)
964                 return error;
965         bp->b_ops = &xfs_attr3_leaf_buf_ops;
966         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
967         leaf = bp->b_addr;
968         memset(leaf, 0, XFS_LBSIZE(mp));
969
970         memset(&ichdr, 0, sizeof(ichdr));
971         ichdr.firstused = XFS_LBSIZE(mp);
972
973         if (xfs_sb_version_hascrc(&mp->m_sb)) {
974                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
975
976                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
977
978                 hdr3->blkno = cpu_to_be64(bp->b_bn);
979                 hdr3->owner = cpu_to_be64(dp->i_ino);
980                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
981
982                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
983         } else {
984                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
985                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
986         }
987         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
988
989         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
990         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
991
992         *bpp = bp;
993         return 0;
994 }
995
996 /*
997  * Split the leaf node, rebalance, then add the new entry.
998  */
999 int
1000 xfs_attr3_leaf_split(
1001         struct xfs_da_state     *state,
1002         struct xfs_da_state_blk *oldblk,
1003         struct xfs_da_state_blk *newblk)
1004 {
1005         xfs_dablk_t blkno;
1006         int error;
1007
1008         trace_xfs_attr_leaf_split(state->args);
1009
1010         /*
1011          * Allocate space for a new leaf node.
1012          */
1013         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1014         error = xfs_da_grow_inode(state->args, &blkno);
1015         if (error)
1016                 return(error);
1017         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1018         if (error)
1019                 return(error);
1020         newblk->blkno = blkno;
1021         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1022
1023         /*
1024          * Rebalance the entries across the two leaves.
1025          * NOTE: rebalance() currently depends on the 2nd block being empty.
1026          */
1027         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1028         error = xfs_da3_blk_link(state, oldblk, newblk);
1029         if (error)
1030                 return(error);
1031
1032         /*
1033          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1034          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1035          * "new" attrs info.  Will need the "old" info to remove it later.
1036          *
1037          * Insert the "new" entry in the correct block.
1038          */
1039         if (state->inleaf) {
1040                 trace_xfs_attr_leaf_add_old(state->args);
1041                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1042         } else {
1043                 trace_xfs_attr_leaf_add_new(state->args);
1044                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1045         }
1046
1047         /*
1048          * Update last hashval in each block since we added the name.
1049          */
1050         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1051         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1052         return(error);
1053 }
1054
1055 /*
1056  * Add a name to the leaf attribute list structure.
1057  */
1058 int
1059 xfs_attr3_leaf_add(
1060         struct xfs_buf          *bp,
1061         struct xfs_da_args      *args)
1062 {
1063         struct xfs_attr_leafblock *leaf;
1064         struct xfs_attr3_icleaf_hdr ichdr;
1065         int                     tablesize;
1066         int                     entsize;
1067         int                     sum;
1068         int                     tmp;
1069         int                     i;
1070
1071         trace_xfs_attr_leaf_add(args);
1072
1073         leaf = bp->b_addr;
1074         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1075         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1076         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1077                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1078
1079         /*
1080          * Search through freemap for first-fit on new name length.
1081          * (may need to figure in size of entry struct too)
1082          */
1083         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1084                                         + xfs_attr3_leaf_hdr_size(leaf);
1085         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1086                 if (tablesize > ichdr.firstused) {
1087                         sum += ichdr.freemap[i].size;
1088                         continue;
1089                 }
1090                 if (!ichdr.freemap[i].size)
1091                         continue;       /* no space in this map */
1092                 tmp = entsize;
1093                 if (ichdr.freemap[i].base < ichdr.firstused)
1094                         tmp += sizeof(xfs_attr_leaf_entry_t);
1095                 if (ichdr.freemap[i].size >= tmp) {
1096                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1097                         goto out_log_hdr;
1098                 }
1099                 sum += ichdr.freemap[i].size;
1100         }
1101
1102         /*
1103          * If there are no holes in the address space of the block,
1104          * and we don't have enough freespace, then compaction will do us
1105          * no good and we should just give up.
1106          */
1107         if (!ichdr.holes && sum < entsize)
1108                 return XFS_ERROR(ENOSPC);
1109
1110         /*
1111          * Compact the entries to coalesce free space.
1112          * This may change the hdr->count via dropping INCOMPLETE entries.
1113          */
1114         xfs_attr3_leaf_compact(args, &ichdr, bp);
1115
1116         /*
1117          * After compaction, the block is guaranteed to have only one
1118          * free region, in freemap[0].  If it is not big enough, give up.
1119          */
1120         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1121                 tmp = ENOSPC;
1122                 goto out_log_hdr;
1123         }
1124
1125         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1126
1127 out_log_hdr:
1128         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1129         xfs_trans_log_buf(args->trans, bp,
1130                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1131                                 xfs_attr3_leaf_hdr_size(leaf)));
1132         return tmp;
1133 }
1134
1135 /*
1136  * Add a name to a leaf attribute list structure.
1137  */
1138 STATIC int
1139 xfs_attr3_leaf_add_work(
1140         struct xfs_buf          *bp,
1141         struct xfs_attr3_icleaf_hdr *ichdr,
1142         struct xfs_da_args      *args,
1143         int                     mapindex)
1144 {
1145         struct xfs_attr_leafblock *leaf;
1146         struct xfs_attr_leaf_entry *entry;
1147         struct xfs_attr_leaf_name_local *name_loc;
1148         struct xfs_attr_leaf_name_remote *name_rmt;
1149         struct xfs_mount        *mp;
1150         int                     tmp;
1151         int                     i;
1152
1153         trace_xfs_attr_leaf_add_work(args);
1154
1155         leaf = bp->b_addr;
1156         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1157         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1158
1159         /*
1160          * Force open some space in the entry array and fill it in.
1161          */
1162         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1163         if (args->index < ichdr->count) {
1164                 tmp  = ichdr->count - args->index;
1165                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1166                 memmove(entry + 1, entry, tmp);
1167                 xfs_trans_log_buf(args->trans, bp,
1168                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1169         }
1170         ichdr->count++;
1171
1172         /*
1173          * Allocate space for the new string (at the end of the run).
1174          */
1175         mp = args->trans->t_mountp;
1176         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1177         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1178         ASSERT(ichdr->freemap[mapindex].size >=
1179                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1180                                          mp->m_sb.sb_blocksize, NULL));
1181         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1182         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1183
1184         ichdr->freemap[mapindex].size -=
1185                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1186                                                  mp->m_sb.sb_blocksize, &tmp);
1187
1188         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1189                                      ichdr->freemap[mapindex].size);
1190         entry->hashval = cpu_to_be32(args->hashval);
1191         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1192         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1193         if (args->op_flags & XFS_DA_OP_RENAME) {
1194                 entry->flags |= XFS_ATTR_INCOMPLETE;
1195                 if ((args->blkno2 == args->blkno) &&
1196                     (args->index2 <= args->index)) {
1197                         args->index2++;
1198                 }
1199         }
1200         xfs_trans_log_buf(args->trans, bp,
1201                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1202         ASSERT((args->index == 0) ||
1203                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1204         ASSERT((args->index == ichdr->count - 1) ||
1205                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1206
1207         /*
1208          * For "remote" attribute values, simply note that we need to
1209          * allocate space for the "remote" value.  We can't actually
1210          * allocate the extents in this transaction, and we can't decide
1211          * which blocks they should be as we might allocate more blocks
1212          * as part of this transaction (a split operation for example).
1213          */
1214         if (entry->flags & XFS_ATTR_LOCAL) {
1215                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1216                 name_loc->namelen = args->namelen;
1217                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1218                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1219                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1220                                    be16_to_cpu(name_loc->valuelen));
1221         } else {
1222                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1223                 name_rmt->namelen = args->namelen;
1224                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1225                 entry->flags |= XFS_ATTR_INCOMPLETE;
1226                 /* just in case */
1227                 name_rmt->valuelen = 0;
1228                 name_rmt->valueblk = 0;
1229                 args->rmtblkno = 1;
1230                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1231         }
1232         xfs_trans_log_buf(args->trans, bp,
1233              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1234                                    xfs_attr_leaf_entsize(leaf, args->index)));
1235
1236         /*
1237          * Update the control info for this leaf node
1238          */
1239         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1240                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1241
1242         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1243                                         + xfs_attr3_leaf_hdr_size(leaf));
1244         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1245                                         + xfs_attr3_leaf_hdr_size(leaf);
1246
1247         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1248                 if (ichdr->freemap[i].base == tmp) {
1249                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1250                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1251                 }
1252         }
1253         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1254         return 0;
1255 }
1256
1257 /*
1258  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1259  */
1260 STATIC void
1261 xfs_attr3_leaf_compact(
1262         struct xfs_da_args      *args,
1263         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1264         struct xfs_buf          *bp)
1265 {
1266         struct xfs_attr_leafblock *leaf_src;
1267         struct xfs_attr_leafblock *leaf_dst;
1268         struct xfs_attr3_icleaf_hdr ichdr_src;
1269         struct xfs_trans        *trans = args->trans;
1270         struct xfs_mount        *mp = trans->t_mountp;
1271         char                    *tmpbuffer;
1272
1273         trace_xfs_attr_leaf_compact(args);
1274
1275         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1276         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1277         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1278         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1279         leaf_dst = bp->b_addr;
1280
1281         /*
1282          * Copy the on-disk header back into the destination buffer to ensure
1283          * all the information in the header that is not part of the incore
1284          * header structure is preserved.
1285          */
1286         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1287
1288         /* Initialise the incore headers */
1289         ichdr_src = *ichdr_dst; /* struct copy */
1290         ichdr_dst->firstused = XFS_LBSIZE(mp);
1291         ichdr_dst->usedbytes = 0;
1292         ichdr_dst->count = 0;
1293         ichdr_dst->holes = 0;
1294         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1295         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1296                                                 ichdr_dst->freemap[0].base;
1297
1298         /* write the header back to initialise the underlying buffer */
1299         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1300
1301         /*
1302          * Copy all entry's in the same (sorted) order,
1303          * but allocate name/value pairs packed and in sequence.
1304          */
1305         xfs_attr3_leaf_moveents(leaf_src, &ichdr_src, 0, leaf_dst, ichdr_dst, 0,
1306                                 ichdr_src.count, mp);
1307         /*
1308          * this logs the entire buffer, but the caller must write the header
1309          * back to the buffer when it is finished modifying it.
1310          */
1311         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1312
1313         kmem_free(tmpbuffer);
1314 }
1315
1316 /*
1317  * Compare two leaf blocks "order".
1318  * Return 0 unless leaf2 should go before leaf1.
1319  */
1320 static int
1321 xfs_attr3_leaf_order(
1322         struct xfs_buf  *leaf1_bp,
1323         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1324         struct xfs_buf  *leaf2_bp,
1325         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1326 {
1327         struct xfs_attr_leaf_entry *entries1;
1328         struct xfs_attr_leaf_entry *entries2;
1329
1330         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1331         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1332         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1333             ((be32_to_cpu(entries2[0].hashval) <
1334               be32_to_cpu(entries1[0].hashval)) ||
1335              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1336               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1337                 return 1;
1338         }
1339         return 0;
1340 }
1341
1342 int
1343 xfs_attr_leaf_order(
1344         struct xfs_buf  *leaf1_bp,
1345         struct xfs_buf  *leaf2_bp)
1346 {
1347         struct xfs_attr3_icleaf_hdr ichdr1;
1348         struct xfs_attr3_icleaf_hdr ichdr2;
1349
1350         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1351         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1352         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1353 }
1354
1355 /*
1356  * Redistribute the attribute list entries between two leaf nodes,
1357  * taking into account the size of the new entry.
1358  *
1359  * NOTE: if new block is empty, then it will get the upper half of the
1360  * old block.  At present, all (one) callers pass in an empty second block.
1361  *
1362  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1363  * to match what it is doing in splitting the attribute leaf block.  Those
1364  * values are used in "atomic rename" operations on attributes.  Note that
1365  * the "new" and "old" values can end up in different blocks.
1366  */
1367 STATIC void
1368 xfs_attr3_leaf_rebalance(
1369         struct xfs_da_state     *state,
1370         struct xfs_da_state_blk *blk1,
1371         struct xfs_da_state_blk *blk2)
1372 {
1373         struct xfs_da_args      *args;
1374         struct xfs_attr_leafblock *leaf1;
1375         struct xfs_attr_leafblock *leaf2;
1376         struct xfs_attr3_icleaf_hdr ichdr1;
1377         struct xfs_attr3_icleaf_hdr ichdr2;
1378         struct xfs_attr_leaf_entry *entries1;
1379         struct xfs_attr_leaf_entry *entries2;
1380         int                     count;
1381         int                     totallen;
1382         int                     max;
1383         int                     space;
1384         int                     swap;
1385
1386         /*
1387          * Set up environment.
1388          */
1389         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1390         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1391         leaf1 = blk1->bp->b_addr;
1392         leaf2 = blk2->bp->b_addr;
1393         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1394         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1395         ASSERT(ichdr2.count == 0);
1396         args = state->args;
1397
1398         trace_xfs_attr_leaf_rebalance(args);
1399
1400         /*
1401          * Check ordering of blocks, reverse if it makes things simpler.
1402          *
1403          * NOTE: Given that all (current) callers pass in an empty
1404          * second block, this code should never set "swap".
1405          */
1406         swap = 0;
1407         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1408                 struct xfs_da_state_blk *tmp_blk;
1409                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1410
1411                 tmp_blk = blk1;
1412                 blk1 = blk2;
1413                 blk2 = tmp_blk;
1414
1415                 /* struct copies to swap them rather than reconverting */
1416                 tmp_ichdr = ichdr1;
1417                 ichdr1 = ichdr2;
1418                 ichdr2 = tmp_ichdr;
1419
1420                 leaf1 = blk1->bp->b_addr;
1421                 leaf2 = blk2->bp->b_addr;
1422                 swap = 1;
1423         }
1424
1425         /*
1426          * Examine entries until we reduce the absolute difference in
1427          * byte usage between the two blocks to a minimum.  Then get
1428          * the direction to copy and the number of elements to move.
1429          *
1430          * "inleaf" is true if the new entry should be inserted into blk1.
1431          * If "swap" is also true, then reverse the sense of "inleaf".
1432          */
1433         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1434                                                       blk2, &ichdr2,
1435                                                       &count, &totallen);
1436         if (swap)
1437                 state->inleaf = !state->inleaf;
1438
1439         /*
1440          * Move any entries required from leaf to leaf:
1441          */
1442         if (count < ichdr1.count) {
1443                 /*
1444                  * Figure the total bytes to be added to the destination leaf.
1445                  */
1446                 /* number entries being moved */
1447                 count = ichdr1.count - count;
1448                 space  = ichdr1.usedbytes - totallen;
1449                 space += count * sizeof(xfs_attr_leaf_entry_t);
1450
1451                 /*
1452                  * leaf2 is the destination, compact it if it looks tight.
1453                  */
1454                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1455                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1456                 if (space > max)
1457                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1458
1459                 /*
1460                  * Move high entries from leaf1 to low end of leaf2.
1461                  */
1462                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1463                                 leaf2, &ichdr2, 0, count, state->mp);
1464
1465         } else if (count > ichdr1.count) {
1466                 /*
1467                  * I assert that since all callers pass in an empty
1468                  * second buffer, this code should never execute.
1469                  */
1470                 ASSERT(0);
1471
1472                 /*
1473                  * Figure the total bytes to be added to the destination leaf.
1474                  */
1475                 /* number entries being moved */
1476                 count -= ichdr1.count;
1477                 space  = totallen - ichdr1.usedbytes;
1478                 space += count * sizeof(xfs_attr_leaf_entry_t);
1479
1480                 /*
1481                  * leaf1 is the destination, compact it if it looks tight.
1482                  */
1483                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1484                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1485                 if (space > max)
1486                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1487
1488                 /*
1489                  * Move low entries from leaf2 to high end of leaf1.
1490                  */
1491                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1492                                         ichdr1.count, count, state->mp);
1493         }
1494
1495         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1496         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1497         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1498         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1499
1500         /*
1501          * Copy out last hashval in each block for B-tree code.
1502          */
1503         entries1 = xfs_attr3_leaf_entryp(leaf1);
1504         entries2 = xfs_attr3_leaf_entryp(leaf2);
1505         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1506         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1507
1508         /*
1509          * Adjust the expected index for insertion.
1510          * NOTE: this code depends on the (current) situation that the
1511          * second block was originally empty.
1512          *
1513          * If the insertion point moved to the 2nd block, we must adjust
1514          * the index.  We must also track the entry just following the
1515          * new entry for use in an "atomic rename" operation, that entry
1516          * is always the "old" entry and the "new" entry is what we are
1517          * inserting.  The index/blkno fields refer to the "old" entry,
1518          * while the index2/blkno2 fields refer to the "new" entry.
1519          */
1520         if (blk1->index > ichdr1.count) {
1521                 ASSERT(state->inleaf == 0);
1522                 blk2->index = blk1->index - ichdr1.count;
1523                 args->index = args->index2 = blk2->index;
1524                 args->blkno = args->blkno2 = blk2->blkno;
1525         } else if (blk1->index == ichdr1.count) {
1526                 if (state->inleaf) {
1527                         args->index = blk1->index;
1528                         args->blkno = blk1->blkno;
1529                         args->index2 = 0;
1530                         args->blkno2 = blk2->blkno;
1531                 } else {
1532                         /*
1533                          * On a double leaf split, the original attr location
1534                          * is already stored in blkno2/index2, so don't
1535                          * overwrite it overwise we corrupt the tree.
1536                          */
1537                         blk2->index = blk1->index - ichdr1.count;
1538                         args->index = blk2->index;
1539                         args->blkno = blk2->blkno;
1540                         if (!state->extravalid) {
1541                                 /*
1542                                  * set the new attr location to match the old
1543                                  * one and let the higher level split code
1544                                  * decide where in the leaf to place it.
1545                                  */
1546                                 args->index2 = blk2->index;
1547                                 args->blkno2 = blk2->blkno;
1548                         }
1549                 }
1550         } else {
1551                 ASSERT(state->inleaf == 1);
1552                 args->index = args->index2 = blk1->index;
1553                 args->blkno = args->blkno2 = blk1->blkno;
1554         }
1555 }
1556
1557 /*
1558  * Examine entries until we reduce the absolute difference in
1559  * byte usage between the two blocks to a minimum.
1560  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1561  * GROT: there will always be enough room in either block for a new entry.
1562  * GROT: Do a double-split for this case?
1563  */
1564 STATIC int
1565 xfs_attr3_leaf_figure_balance(
1566         struct xfs_da_state             *state,
1567         struct xfs_da_state_blk         *blk1,
1568         struct xfs_attr3_icleaf_hdr     *ichdr1,
1569         struct xfs_da_state_blk         *blk2,
1570         struct xfs_attr3_icleaf_hdr     *ichdr2,
1571         int                             *countarg,
1572         int                             *usedbytesarg)
1573 {
1574         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1575         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1576         struct xfs_attr_leaf_entry      *entry;
1577         int                             count;
1578         int                             max;
1579         int                             index;
1580         int                             totallen = 0;
1581         int                             half;
1582         int                             lastdelta;
1583         int                             foundit = 0;
1584         int                             tmp;
1585
1586         /*
1587          * Examine entries until we reduce the absolute difference in
1588          * byte usage between the two blocks to a minimum.
1589          */
1590         max = ichdr1->count + ichdr2->count;
1591         half = (max + 1) * sizeof(*entry);
1592         half += ichdr1->usedbytes + ichdr2->usedbytes +
1593                         xfs_attr_leaf_newentsize(state->args->namelen,
1594                                                  state->args->valuelen,
1595                                                  state->blocksize, NULL);
1596         half /= 2;
1597         lastdelta = state->blocksize;
1598         entry = xfs_attr3_leaf_entryp(leaf1);
1599         for (count = index = 0; count < max; entry++, index++, count++) {
1600
1601 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1602                 /*
1603                  * The new entry is in the first block, account for it.
1604                  */
1605                 if (count == blk1->index) {
1606                         tmp = totallen + sizeof(*entry) +
1607                                 xfs_attr_leaf_newentsize(
1608                                                 state->args->namelen,
1609                                                 state->args->valuelen,
1610                                                 state->blocksize, NULL);
1611                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1612                                 break;
1613                         lastdelta = XFS_ATTR_ABS(half - tmp);
1614                         totallen = tmp;
1615                         foundit = 1;
1616                 }
1617
1618                 /*
1619                  * Wrap around into the second block if necessary.
1620                  */
1621                 if (count == ichdr1->count) {
1622                         leaf1 = leaf2;
1623                         entry = xfs_attr3_leaf_entryp(leaf1);
1624                         index = 0;
1625                 }
1626
1627                 /*
1628                  * Figure out if next leaf entry would be too much.
1629                  */
1630                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1631                                                                         index);
1632                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1633                         break;
1634                 lastdelta = XFS_ATTR_ABS(half - tmp);
1635                 totallen = tmp;
1636 #undef XFS_ATTR_ABS
1637         }
1638
1639         /*
1640          * Calculate the number of usedbytes that will end up in lower block.
1641          * If new entry not in lower block, fix up the count.
1642          */
1643         totallen -= count * sizeof(*entry);
1644         if (foundit) {
1645                 totallen -= sizeof(*entry) +
1646                                 xfs_attr_leaf_newentsize(
1647                                                 state->args->namelen,
1648                                                 state->args->valuelen,
1649                                                 state->blocksize, NULL);
1650         }
1651
1652         *countarg = count;
1653         *usedbytesarg = totallen;
1654         return foundit;
1655 }
1656
1657 /*========================================================================
1658  * Routines used for shrinking the Btree.
1659  *========================================================================*/
1660
1661 /*
1662  * Check a leaf block and its neighbors to see if the block should be
1663  * collapsed into one or the other neighbor.  Always keep the block
1664  * with the smaller block number.
1665  * If the current block is over 50% full, don't try to join it, return 0.
1666  * If the block is empty, fill in the state structure and return 2.
1667  * If it can be collapsed, fill in the state structure and return 1.
1668  * If nothing can be done, return 0.
1669  *
1670  * GROT: allow for INCOMPLETE entries in calculation.
1671  */
1672 int
1673 xfs_attr3_leaf_toosmall(
1674         struct xfs_da_state     *state,
1675         int                     *action)
1676 {
1677         struct xfs_attr_leafblock *leaf;
1678         struct xfs_da_state_blk *blk;
1679         struct xfs_attr3_icleaf_hdr ichdr;
1680         struct xfs_buf          *bp;
1681         xfs_dablk_t             blkno;
1682         int                     bytes;
1683         int                     forward;
1684         int                     error;
1685         int                     retval;
1686         int                     i;
1687
1688         trace_xfs_attr_leaf_toosmall(state->args);
1689
1690         /*
1691          * Check for the degenerate case of the block being over 50% full.
1692          * If so, it's not worth even looking to see if we might be able
1693          * to coalesce with a sibling.
1694          */
1695         blk = &state->path.blk[ state->path.active-1 ];
1696         leaf = blk->bp->b_addr;
1697         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1698         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1699                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1700                 ichdr.usedbytes;
1701         if (bytes > (state->blocksize >> 1)) {
1702                 *action = 0;    /* blk over 50%, don't try to join */
1703                 return(0);
1704         }
1705
1706         /*
1707          * Check for the degenerate case of the block being empty.
1708          * If the block is empty, we'll simply delete it, no need to
1709          * coalesce it with a sibling block.  We choose (arbitrarily)
1710          * to merge with the forward block unless it is NULL.
1711          */
1712         if (ichdr.count == 0) {
1713                 /*
1714                  * Make altpath point to the block we want to keep and
1715                  * path point to the block we want to drop (this one).
1716                  */
1717                 forward = (ichdr.forw != 0);
1718                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1719                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1720                                                  0, &retval);
1721                 if (error)
1722                         return(error);
1723                 if (retval) {
1724                         *action = 0;
1725                 } else {
1726                         *action = 2;
1727                 }
1728                 return 0;
1729         }
1730
1731         /*
1732          * Examine each sibling block to see if we can coalesce with
1733          * at least 25% free space to spare.  We need to figure out
1734          * whether to merge with the forward or the backward block.
1735          * We prefer coalescing with the lower numbered sibling so as
1736          * to shrink an attribute list over time.
1737          */
1738         /* start with smaller blk num */
1739         forward = ichdr.forw < ichdr.back;
1740         for (i = 0; i < 2; forward = !forward, i++) {
1741                 struct xfs_attr3_icleaf_hdr ichdr2;
1742                 if (forward)
1743                         blkno = ichdr.forw;
1744                 else
1745                         blkno = ichdr.back;
1746                 if (blkno == 0)
1747                         continue;
1748                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1749                                         blkno, -1, &bp);
1750                 if (error)
1751                         return(error);
1752
1753                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1754
1755                 bytes = state->blocksize - (state->blocksize >> 2) -
1756                         ichdr.usedbytes - ichdr2.usedbytes -
1757                         ((ichdr.count + ichdr2.count) *
1758                                         sizeof(xfs_attr_leaf_entry_t)) -
1759                         xfs_attr3_leaf_hdr_size(leaf);
1760
1761                 xfs_trans_brelse(state->args->trans, bp);
1762                 if (bytes >= 0)
1763                         break;  /* fits with at least 25% to spare */
1764         }
1765         if (i >= 2) {
1766                 *action = 0;
1767                 return(0);
1768         }
1769
1770         /*
1771          * Make altpath point to the block we want to keep (the lower
1772          * numbered block) and path point to the block we want to drop.
1773          */
1774         memcpy(&state->altpath, &state->path, sizeof(state->path));
1775         if (blkno < blk->blkno) {
1776                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1777                                                  0, &retval);
1778         } else {
1779                 error = xfs_da3_path_shift(state, &state->path, forward,
1780                                                  0, &retval);
1781         }
1782         if (error)
1783                 return(error);
1784         if (retval) {
1785                 *action = 0;
1786         } else {
1787                 *action = 1;
1788         }
1789         return(0);
1790 }
1791
1792 /*
1793  * Remove a name from the leaf attribute list structure.
1794  *
1795  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1796  * If two leaves are 37% full, when combined they will leave 25% free.
1797  */
1798 int
1799 xfs_attr3_leaf_remove(
1800         struct xfs_buf          *bp,
1801         struct xfs_da_args      *args)
1802 {
1803         struct xfs_attr_leafblock *leaf;
1804         struct xfs_attr3_icleaf_hdr ichdr;
1805         struct xfs_attr_leaf_entry *entry;
1806         struct xfs_mount        *mp = args->trans->t_mountp;
1807         int                     before;
1808         int                     after;
1809         int                     smallest;
1810         int                     entsize;
1811         int                     tablesize;
1812         int                     tmp;
1813         int                     i;
1814
1815         trace_xfs_attr_leaf_remove(args);
1816
1817         leaf = bp->b_addr;
1818         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1819
1820         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
1821         ASSERT(args->index >= 0 && args->index < ichdr.count);
1822         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1823                                         xfs_attr3_leaf_hdr_size(leaf));
1824
1825         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1826
1827         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1828         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1829
1830         /*
1831          * Scan through free region table:
1832          *    check for adjacency of free'd entry with an existing one,
1833          *    find smallest free region in case we need to replace it,
1834          *    adjust any map that borders the entry table,
1835          */
1836         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1837                                         + xfs_attr3_leaf_hdr_size(leaf);
1838         tmp = ichdr.freemap[0].size;
1839         before = after = -1;
1840         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1841         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1842         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1843                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
1844                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
1845                 if (ichdr.freemap[i].base == tablesize) {
1846                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1847                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1848                 }
1849
1850                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1851                                 be16_to_cpu(entry->nameidx)) {
1852                         before = i;
1853                 } else if (ichdr.freemap[i].base ==
1854                                 (be16_to_cpu(entry->nameidx) + entsize)) {
1855                         after = i;
1856                 } else if (ichdr.freemap[i].size < tmp) {
1857                         tmp = ichdr.freemap[i].size;
1858                         smallest = i;
1859                 }
1860         }
1861
1862         /*
1863          * Coalesce adjacent freemap regions,
1864          * or replace the smallest region.
1865          */
1866         if ((before >= 0) || (after >= 0)) {
1867                 if ((before >= 0) && (after >= 0)) {
1868                         ichdr.freemap[before].size += entsize;
1869                         ichdr.freemap[before].size += ichdr.freemap[after].size;
1870                         ichdr.freemap[after].base = 0;
1871                         ichdr.freemap[after].size = 0;
1872                 } else if (before >= 0) {
1873                         ichdr.freemap[before].size += entsize;
1874                 } else {
1875                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1876                         ichdr.freemap[after].size += entsize;
1877                 }
1878         } else {
1879                 /*
1880                  * Replace smallest region (if it is smaller than free'd entry)
1881                  */
1882                 if (ichdr.freemap[smallest].size < entsize) {
1883                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1884                         ichdr.freemap[smallest].size = entsize;
1885                 }
1886         }
1887
1888         /*
1889          * Did we remove the first entry?
1890          */
1891         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1892                 smallest = 1;
1893         else
1894                 smallest = 0;
1895
1896         /*
1897          * Compress the remaining entries and zero out the removed stuff.
1898          */
1899         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1900         ichdr.usedbytes -= entsize;
1901         xfs_trans_log_buf(args->trans, bp,
1902              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1903                                    entsize));
1904
1905         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1906         memmove(entry, entry + 1, tmp);
1907         ichdr.count--;
1908         xfs_trans_log_buf(args->trans, bp,
1909             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1910
1911         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1912         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1913
1914         /*
1915          * If we removed the first entry, re-find the first used byte
1916          * in the name area.  Note that if the entry was the "firstused",
1917          * then we don't have a "hole" in our block resulting from
1918          * removing the name.
1919          */
1920         if (smallest) {
1921                 tmp = XFS_LBSIZE(mp);
1922                 entry = xfs_attr3_leaf_entryp(leaf);
1923                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1924                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1925                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1926
1927                         if (be16_to_cpu(entry->nameidx) < tmp)
1928                                 tmp = be16_to_cpu(entry->nameidx);
1929                 }
1930                 ichdr.firstused = tmp;
1931                 if (!ichdr.firstused)
1932                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
1933         } else {
1934                 ichdr.holes = 1;        /* mark as needing compaction */
1935         }
1936         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1937         xfs_trans_log_buf(args->trans, bp,
1938                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1939                                           xfs_attr3_leaf_hdr_size(leaf)));
1940
1941         /*
1942          * Check if leaf is less than 50% full, caller may want to
1943          * "join" the leaf with a sibling if so.
1944          */
1945         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
1946               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
1947
1948         return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
1949 }
1950
1951 /*
1952  * Move all the attribute list entries from drop_leaf into save_leaf.
1953  */
1954 void
1955 xfs_attr3_leaf_unbalance(
1956         struct xfs_da_state     *state,
1957         struct xfs_da_state_blk *drop_blk,
1958         struct xfs_da_state_blk *save_blk)
1959 {
1960         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
1961         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
1962         struct xfs_attr3_icleaf_hdr drophdr;
1963         struct xfs_attr3_icleaf_hdr savehdr;
1964         struct xfs_attr_leaf_entry *entry;
1965         struct xfs_mount        *mp = state->mp;
1966
1967         trace_xfs_attr_leaf_unbalance(state->args);
1968
1969         drop_leaf = drop_blk->bp->b_addr;
1970         save_leaf = save_blk->bp->b_addr;
1971         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1972         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
1973         entry = xfs_attr3_leaf_entryp(drop_leaf);
1974
1975         /*
1976          * Save last hashval from dying block for later Btree fixup.
1977          */
1978         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
1979
1980         /*
1981          * Check if we need a temp buffer, or can we do it in place.
1982          * Note that we don't check "leaf" for holes because we will
1983          * always be dropping it, toosmall() decided that for us already.
1984          */
1985         if (savehdr.holes == 0) {
1986                 /*
1987                  * dest leaf has no holes, so we add there.  May need
1988                  * to make some room in the entry array.
1989                  */
1990                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
1991                                          drop_blk->bp, &drophdr)) {
1992                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1993                                                 save_leaf, &savehdr, 0,
1994                                                 drophdr.count, mp);
1995                 } else {
1996                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1997                                                 save_leaf, &savehdr,
1998                                                 savehdr.count, drophdr.count, mp);
1999                 }
2000         } else {
2001                 /*
2002                  * Destination has holes, so we make a temporary copy
2003                  * of the leaf and add them both to that.
2004                  */
2005                 struct xfs_attr_leafblock *tmp_leaf;
2006                 struct xfs_attr3_icleaf_hdr tmphdr;
2007
2008                 tmp_leaf = kmem_zalloc(state->blocksize, KM_SLEEP);
2009
2010                 /*
2011                  * Copy the header into the temp leaf so that all the stuff
2012                  * not in the incore header is present and gets copied back in
2013                  * once we've moved all the entries.
2014                  */
2015                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2016
2017                 memset(&tmphdr, 0, sizeof(tmphdr));
2018                 tmphdr.magic = savehdr.magic;
2019                 tmphdr.forw = savehdr.forw;
2020                 tmphdr.back = savehdr.back;
2021                 tmphdr.firstused = state->blocksize;
2022
2023                 /* write the header to the temp buffer to initialise it */
2024                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2025
2026                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2027                                          drop_blk->bp, &drophdr)) {
2028                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2029                                                 tmp_leaf, &tmphdr, 0,
2030                                                 drophdr.count, mp);
2031                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2032                                                 tmp_leaf, &tmphdr, tmphdr.count,
2033                                                 savehdr.count, mp);
2034                 } else {
2035                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2036                                                 tmp_leaf, &tmphdr, 0,
2037                                                 savehdr.count, mp);
2038                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2039                                                 tmp_leaf, &tmphdr, tmphdr.count,
2040                                                 drophdr.count, mp);
2041                 }
2042                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2043                 savehdr = tmphdr; /* struct copy */
2044                 kmem_free(tmp_leaf);
2045         }
2046
2047         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2048         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2049                                            state->blocksize - 1);
2050
2051         /*
2052          * Copy out last hashval in each block for B-tree code.
2053          */
2054         entry = xfs_attr3_leaf_entryp(save_leaf);
2055         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2056 }
2057
2058 /*========================================================================
2059  * Routines used for finding things in the Btree.
2060  *========================================================================*/
2061
2062 /*
2063  * Look up a name in a leaf attribute list structure.
2064  * This is the internal routine, it uses the caller's buffer.
2065  *
2066  * Note that duplicate keys are allowed, but only check within the
2067  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2068  *
2069  * Return in args->index the index into the entry[] array of either
2070  * the found entry, or where the entry should have been (insert before
2071  * that entry).
2072  *
2073  * Don't change the args->value unless we find the attribute.
2074  */
2075 int
2076 xfs_attr3_leaf_lookup_int(
2077         struct xfs_buf          *bp,
2078         struct xfs_da_args      *args)
2079 {
2080         struct xfs_attr_leafblock *leaf;
2081         struct xfs_attr3_icleaf_hdr ichdr;
2082         struct xfs_attr_leaf_entry *entry;
2083         struct xfs_attr_leaf_entry *entries;
2084         struct xfs_attr_leaf_name_local *name_loc;
2085         struct xfs_attr_leaf_name_remote *name_rmt;
2086         xfs_dahash_t            hashval;
2087         int                     probe;
2088         int                     span;
2089
2090         trace_xfs_attr_leaf_lookup(args);
2091
2092         leaf = bp->b_addr;
2093         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2094         entries = xfs_attr3_leaf_entryp(leaf);
2095         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2096
2097         /*
2098          * Binary search.  (note: small blocks will skip this loop)
2099          */
2100         hashval = args->hashval;
2101         probe = span = ichdr.count / 2;
2102         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2103                 span /= 2;
2104                 if (be32_to_cpu(entry->hashval) < hashval)
2105                         probe += span;
2106                 else if (be32_to_cpu(entry->hashval) > hashval)
2107                         probe -= span;
2108                 else
2109                         break;
2110         }
2111         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2112         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2113
2114         /*
2115          * Since we may have duplicate hashval's, find the first matching
2116          * hashval in the leaf.
2117          */
2118         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2119                 entry--;
2120                 probe--;
2121         }
2122         while (probe < ichdr.count &&
2123                be32_to_cpu(entry->hashval) < hashval) {
2124                 entry++;
2125                 probe++;
2126         }
2127         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2128                 args->index = probe;
2129                 return XFS_ERROR(ENOATTR);
2130         }
2131
2132         /*
2133          * Duplicate keys may be present, so search all of them for a match.
2134          */
2135         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2136                         entry++, probe++) {
2137 /*
2138  * GROT: Add code to remove incomplete entries.
2139  */
2140                 /*
2141                  * If we are looking for INCOMPLETE entries, show only those.
2142                  * If we are looking for complete entries, show only those.
2143                  */
2144                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2145                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2146                         continue;
2147                 }
2148                 if (entry->flags & XFS_ATTR_LOCAL) {
2149                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2150                         if (name_loc->namelen != args->namelen)
2151                                 continue;
2152                         if (memcmp(args->name, name_loc->nameval,
2153                                                         args->namelen) != 0)
2154                                 continue;
2155                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2156                                 continue;
2157                         args->index = probe;
2158                         return XFS_ERROR(EEXIST);
2159                 } else {
2160                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2161                         if (name_rmt->namelen != args->namelen)
2162                                 continue;
2163                         if (memcmp(args->name, name_rmt->name,
2164                                                         args->namelen) != 0)
2165                                 continue;
2166                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2167                                 continue;
2168                         args->index = probe;
2169                         args->valuelen = be32_to_cpu(name_rmt->valuelen);
2170                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2171                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2172                                                         args->dp->i_mount,
2173                                                         args->valuelen);
2174                         return XFS_ERROR(EEXIST);
2175                 }
2176         }
2177         args->index = probe;
2178         return XFS_ERROR(ENOATTR);
2179 }
2180
2181 /*
2182  * Get the value associated with an attribute name from a leaf attribute
2183  * list structure.
2184  */
2185 int
2186 xfs_attr3_leaf_getvalue(
2187         struct xfs_buf          *bp,
2188         struct xfs_da_args      *args)
2189 {
2190         struct xfs_attr_leafblock *leaf;
2191         struct xfs_attr3_icleaf_hdr ichdr;
2192         struct xfs_attr_leaf_entry *entry;
2193         struct xfs_attr_leaf_name_local *name_loc;
2194         struct xfs_attr_leaf_name_remote *name_rmt;
2195         int                     valuelen;
2196
2197         leaf = bp->b_addr;
2198         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2199         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2200         ASSERT(args->index < ichdr.count);
2201
2202         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2203         if (entry->flags & XFS_ATTR_LOCAL) {
2204                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2205                 ASSERT(name_loc->namelen == args->namelen);
2206                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2207                 valuelen = be16_to_cpu(name_loc->valuelen);
2208                 if (args->flags & ATTR_KERNOVAL) {
2209                         args->valuelen = valuelen;
2210                         return 0;
2211                 }
2212                 if (args->valuelen < valuelen) {
2213                         args->valuelen = valuelen;
2214                         return XFS_ERROR(ERANGE);
2215                 }
2216                 args->valuelen = valuelen;
2217                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2218         } else {
2219                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2220                 ASSERT(name_rmt->namelen == args->namelen);
2221                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2222                 valuelen = be32_to_cpu(name_rmt->valuelen);
2223                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2224                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2225                                                        valuelen);
2226                 if (args->flags & ATTR_KERNOVAL) {
2227                         args->valuelen = valuelen;
2228                         return 0;
2229                 }
2230                 if (args->valuelen < valuelen) {
2231                         args->valuelen = valuelen;
2232                         return XFS_ERROR(ERANGE);
2233                 }
2234                 args->valuelen = valuelen;
2235         }
2236         return 0;
2237 }
2238
2239 /*========================================================================
2240  * Utility routines.
2241  *========================================================================*/
2242
2243 /*
2244  * Move the indicated entries from one leaf to another.
2245  * NOTE: this routine modifies both source and destination leaves.
2246  */
2247 /*ARGSUSED*/
2248 STATIC void
2249 xfs_attr3_leaf_moveents(
2250         struct xfs_attr_leafblock       *leaf_s,
2251         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2252         int                             start_s,
2253         struct xfs_attr_leafblock       *leaf_d,
2254         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2255         int                             start_d,
2256         int                             count,
2257         struct xfs_mount                *mp)
2258 {
2259         struct xfs_attr_leaf_entry      *entry_s;
2260         struct xfs_attr_leaf_entry      *entry_d;
2261         int                             desti;
2262         int                             tmp;
2263         int                             i;
2264
2265         /*
2266          * Check for nothing to do.
2267          */
2268         if (count == 0)
2269                 return;
2270
2271         /*
2272          * Set up environment.
2273          */
2274         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2275                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2276         ASSERT(ichdr_s->magic == ichdr_d->magic);
2277         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2278         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2279                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2280         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2281         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2282                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2283
2284         ASSERT(start_s < ichdr_s->count);
2285         ASSERT(start_d <= ichdr_d->count);
2286         ASSERT(count <= ichdr_s->count);
2287
2288
2289         /*
2290          * Move the entries in the destination leaf up to make a hole?
2291          */
2292         if (start_d < ichdr_d->count) {
2293                 tmp  = ichdr_d->count - start_d;
2294                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2295                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2296                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2297                 memmove(entry_d, entry_s, tmp);
2298         }
2299
2300         /*
2301          * Copy all entry's in the same (sorted) order,
2302          * but allocate attribute info packed and in sequence.
2303          */
2304         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2305         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2306         desti = start_d;
2307         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2308                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2309                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2310 #ifdef GROT
2311                 /*
2312                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2313                  * may also need to change the insertion index.  Code turned
2314                  * off for 6.2, should be revisited later.
2315                  */
2316                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2317                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2318                         ichdr_s->usedbytes -= tmp;
2319                         ichdr_s->count -= 1;
2320                         entry_d--;      /* to compensate for ++ in loop hdr */
2321                         desti--;
2322                         if ((start_s + i) < offset)
2323                                 result++;       /* insertion index adjustment */
2324                 } else {
2325 #endif /* GROT */
2326                         ichdr_d->firstused -= tmp;
2327                         /* both on-disk, don't endian flip twice */
2328                         entry_d->hashval = entry_s->hashval;
2329                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2330                         entry_d->flags = entry_s->flags;
2331                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2332                                                         <= XFS_LBSIZE(mp));
2333                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2334                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2335                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2336                                                         <= XFS_LBSIZE(mp));
2337                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2338                         ichdr_s->usedbytes -= tmp;
2339                         ichdr_d->usedbytes += tmp;
2340                         ichdr_s->count -= 1;
2341                         ichdr_d->count += 1;
2342                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2343                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2344                         ASSERT(ichdr_d->firstused >= tmp);
2345 #ifdef GROT
2346                 }
2347 #endif /* GROT */
2348         }
2349
2350         /*
2351          * Zero out the entries we just copied.
2352          */
2353         if (start_s == ichdr_s->count) {
2354                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2355                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2356                 ASSERT(((char *)entry_s + tmp) <=
2357                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2358                 memset(entry_s, 0, tmp);
2359         } else {
2360                 /*
2361                  * Move the remaining entries down to fill the hole,
2362                  * then zero the entries at the top.
2363                  */
2364                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2365                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2366                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2367                 memmove(entry_d, entry_s, tmp);
2368
2369                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2370                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2371                 ASSERT(((char *)entry_s + tmp) <=
2372                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2373                 memset(entry_s, 0, tmp);
2374         }
2375
2376         /*
2377          * Fill in the freemap information
2378          */
2379         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2380         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2381         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2382         ichdr_d->freemap[1].base = 0;
2383         ichdr_d->freemap[2].base = 0;
2384         ichdr_d->freemap[1].size = 0;
2385         ichdr_d->freemap[2].size = 0;
2386         ichdr_s->holes = 1;     /* leaf may not be compact */
2387 }
2388
2389 /*
2390  * Pick up the last hashvalue from a leaf block.
2391  */
2392 xfs_dahash_t
2393 xfs_attr_leaf_lasthash(
2394         struct xfs_buf  *bp,
2395         int             *count)
2396 {
2397         struct xfs_attr3_icleaf_hdr ichdr;
2398         struct xfs_attr_leaf_entry *entries;
2399
2400         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2401         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2402         if (count)
2403                 *count = ichdr.count;
2404         if (!ichdr.count)
2405                 return 0;
2406         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2407 }
2408
2409 /*
2410  * Calculate the number of bytes used to store the indicated attribute
2411  * (whether local or remote only calculate bytes in this block).
2412  */
2413 STATIC int
2414 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2415 {
2416         struct xfs_attr_leaf_entry *entries;
2417         xfs_attr_leaf_name_local_t *name_loc;
2418         xfs_attr_leaf_name_remote_t *name_rmt;
2419         int size;
2420
2421         entries = xfs_attr3_leaf_entryp(leaf);
2422         if (entries[index].flags & XFS_ATTR_LOCAL) {
2423                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2424                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2425                                                    be16_to_cpu(name_loc->valuelen));
2426         } else {
2427                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2428                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2429         }
2430         return size;
2431 }
2432
2433 /*
2434  * Calculate the number of bytes that would be required to store the new
2435  * attribute (whether local or remote only calculate bytes in this block).
2436  * This routine decides as a side effect whether the attribute will be
2437  * a "local" or a "remote" attribute.
2438  */
2439 int
2440 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2441 {
2442         int size;
2443
2444         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2445         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2446                 if (local) {
2447                         *local = 1;
2448                 }
2449         } else {
2450                 size = xfs_attr_leaf_entsize_remote(namelen);
2451                 if (local) {
2452                         *local = 0;
2453                 }
2454         }
2455         return size;
2456 }
2457
2458
2459 /*========================================================================
2460  * Manage the INCOMPLETE flag in a leaf entry
2461  *========================================================================*/
2462
2463 /*
2464  * Clear the INCOMPLETE flag on an entry in a leaf block.
2465  */
2466 int
2467 xfs_attr3_leaf_clearflag(
2468         struct xfs_da_args      *args)
2469 {
2470         struct xfs_attr_leafblock *leaf;
2471         struct xfs_attr_leaf_entry *entry;
2472         struct xfs_attr_leaf_name_remote *name_rmt;
2473         struct xfs_buf          *bp;
2474         int                     error;
2475 #ifdef DEBUG
2476         struct xfs_attr3_icleaf_hdr ichdr;
2477         xfs_attr_leaf_name_local_t *name_loc;
2478         int namelen;
2479         char *name;
2480 #endif /* DEBUG */
2481
2482         trace_xfs_attr_leaf_clearflag(args);
2483         /*
2484          * Set up the operation.
2485          */
2486         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2487         if (error)
2488                 return(error);
2489
2490         leaf = bp->b_addr;
2491         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2492         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2493
2494 #ifdef DEBUG
2495         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2496         ASSERT(args->index < ichdr.count);
2497         ASSERT(args->index >= 0);
2498
2499         if (entry->flags & XFS_ATTR_LOCAL) {
2500                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2501                 namelen = name_loc->namelen;
2502                 name = (char *)name_loc->nameval;
2503         } else {
2504                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2505                 namelen = name_rmt->namelen;
2506                 name = (char *)name_rmt->name;
2507         }
2508         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2509         ASSERT(namelen == args->namelen);
2510         ASSERT(memcmp(name, args->name, namelen) == 0);
2511 #endif /* DEBUG */
2512
2513         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2514         xfs_trans_log_buf(args->trans, bp,
2515                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2516
2517         if (args->rmtblkno) {
2518                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2519                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2520                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2521                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2522                 xfs_trans_log_buf(args->trans, bp,
2523                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2524         }
2525
2526         /*
2527          * Commit the flag value change and start the next trans in series.
2528          */
2529         return xfs_trans_roll(&args->trans, args->dp);
2530 }
2531
2532 /*
2533  * Set the INCOMPLETE flag on an entry in a leaf block.
2534  */
2535 int
2536 xfs_attr3_leaf_setflag(
2537         struct xfs_da_args      *args)
2538 {
2539         struct xfs_attr_leafblock *leaf;
2540         struct xfs_attr_leaf_entry *entry;
2541         struct xfs_attr_leaf_name_remote *name_rmt;
2542         struct xfs_buf          *bp;
2543         int error;
2544 #ifdef DEBUG
2545         struct xfs_attr3_icleaf_hdr ichdr;
2546 #endif
2547
2548         trace_xfs_attr_leaf_setflag(args);
2549
2550         /*
2551          * Set up the operation.
2552          */
2553         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2554         if (error)
2555                 return(error);
2556
2557         leaf = bp->b_addr;
2558 #ifdef DEBUG
2559         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2560         ASSERT(args->index < ichdr.count);
2561         ASSERT(args->index >= 0);
2562 #endif
2563         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2564
2565         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2566         entry->flags |= XFS_ATTR_INCOMPLETE;
2567         xfs_trans_log_buf(args->trans, bp,
2568                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2569         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2570                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2571                 name_rmt->valueblk = 0;
2572                 name_rmt->valuelen = 0;
2573                 xfs_trans_log_buf(args->trans, bp,
2574                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2575         }
2576
2577         /*
2578          * Commit the flag value change and start the next trans in series.
2579          */
2580         return xfs_trans_roll(&args->trans, args->dp);
2581 }
2582
2583 /*
2584  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2585  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2586  * entry given by args->blkno2/index2.
2587  *
2588  * Note that they could be in different blocks, or in the same block.
2589  */
2590 int
2591 xfs_attr3_leaf_flipflags(
2592         struct xfs_da_args      *args)
2593 {
2594         struct xfs_attr_leafblock *leaf1;
2595         struct xfs_attr_leafblock *leaf2;
2596         struct xfs_attr_leaf_entry *entry1;
2597         struct xfs_attr_leaf_entry *entry2;
2598         struct xfs_attr_leaf_name_remote *name_rmt;
2599         struct xfs_buf          *bp1;
2600         struct xfs_buf          *bp2;
2601         int error;
2602 #ifdef DEBUG
2603         struct xfs_attr3_icleaf_hdr ichdr1;
2604         struct xfs_attr3_icleaf_hdr ichdr2;
2605         xfs_attr_leaf_name_local_t *name_loc;
2606         int namelen1, namelen2;
2607         char *name1, *name2;
2608 #endif /* DEBUG */
2609
2610         trace_xfs_attr_leaf_flipflags(args);
2611
2612         /*
2613          * Read the block containing the "old" attr
2614          */
2615         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2616         if (error)
2617                 return error;
2618
2619         /*
2620          * Read the block containing the "new" attr, if it is different
2621          */
2622         if (args->blkno2 != args->blkno) {
2623                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2624                                            -1, &bp2);
2625                 if (error)
2626                         return error;
2627         } else {
2628                 bp2 = bp1;
2629         }
2630
2631         leaf1 = bp1->b_addr;
2632         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2633
2634         leaf2 = bp2->b_addr;
2635         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2636
2637 #ifdef DEBUG
2638         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2639         ASSERT(args->index < ichdr1.count);
2640         ASSERT(args->index >= 0);
2641
2642         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2643         ASSERT(args->index2 < ichdr2.count);
2644         ASSERT(args->index2 >= 0);
2645
2646         if (entry1->flags & XFS_ATTR_LOCAL) {
2647                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2648                 namelen1 = name_loc->namelen;
2649                 name1 = (char *)name_loc->nameval;
2650         } else {
2651                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2652                 namelen1 = name_rmt->namelen;
2653                 name1 = (char *)name_rmt->name;
2654         }
2655         if (entry2->flags & XFS_ATTR_LOCAL) {
2656                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2657                 namelen2 = name_loc->namelen;
2658                 name2 = (char *)name_loc->nameval;
2659         } else {
2660                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2661                 namelen2 = name_rmt->namelen;
2662                 name2 = (char *)name_rmt->name;
2663         }
2664         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2665         ASSERT(namelen1 == namelen2);
2666         ASSERT(memcmp(name1, name2, namelen1) == 0);
2667 #endif /* DEBUG */
2668
2669         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2670         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2671
2672         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2673         xfs_trans_log_buf(args->trans, bp1,
2674                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2675         if (args->rmtblkno) {
2676                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2677                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2678                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2679                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2680                 xfs_trans_log_buf(args->trans, bp1,
2681                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2682         }
2683
2684         entry2->flags |= XFS_ATTR_INCOMPLETE;
2685         xfs_trans_log_buf(args->trans, bp2,
2686                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2687         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2688                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2689                 name_rmt->valueblk = 0;
2690                 name_rmt->valuelen = 0;
2691                 xfs_trans_log_buf(args->trans, bp2,
2692                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2693         }
2694
2695         /*
2696          * Commit the flag value change and start the next trans in series.
2697          */
2698         error = xfs_trans_roll(&args->trans, args->dp);
2699
2700         return error;
2701 }