]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/xfs/xfs_attr_leaf.c
Merge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge
[karo-tx-linux.git] / 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_types.h"
22 #include "xfs_bit.h"
23 #include "xfs_log.h"
24 #include "xfs_trans.h"
25 #include "xfs_trans_priv.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc.h"
34 #include "xfs_btree.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_attr_remote.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_bmap.h"
41 #include "xfs_attr.h"
42 #include "xfs_attr_leaf.h"
43 #include "xfs_error.h"
44 #include "xfs_trace.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_cksum.h"
47
48
49 /*
50  * xfs_attr_leaf.c
51  *
52  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
53  */
54
55 /*========================================================================
56  * Function prototypes for the kernel.
57  *========================================================================*/
58
59 /*
60  * Routines used for growing the Btree.
61  */
62 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
63                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
64 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
65                                    struct xfs_attr3_icleaf_hdr *ichdr,
66                                    struct xfs_da_args *args, int freemap_index);
67 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
68                                    struct xfs_attr3_icleaf_hdr *ichdr,
69                                    struct xfs_buf *leaf_buffer);
70 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
71                                                    xfs_da_state_blk_t *blk1,
72                                                    xfs_da_state_blk_t *blk2);
73 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
74                         xfs_da_state_blk_t *leaf_blk_1,
75                         struct xfs_attr3_icleaf_hdr *ichdr1,
76                         xfs_da_state_blk_t *leaf_blk_2,
77                         struct xfs_attr3_icleaf_hdr *ichdr2,
78                         int *number_entries_in_blk1,
79                         int *number_usedbytes_in_blk1);
80
81 /*
82  * Utility routines.
83  */
84 STATIC void xfs_attr3_leaf_moveents(struct xfs_attr_leafblock *src_leaf,
85                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
86                         struct xfs_attr_leafblock *dst_leaf,
87                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
88                         int move_count, struct xfs_mount *mp);
89 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
90
91 void
92 xfs_attr3_leaf_hdr_from_disk(
93         struct xfs_attr3_icleaf_hdr     *to,
94         struct xfs_attr_leafblock       *from)
95 {
96         int     i;
97
98         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
99                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
100
101         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
102                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
103
104                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
105                 to->back = be32_to_cpu(hdr3->info.hdr.back);
106                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
107                 to->count = be16_to_cpu(hdr3->count);
108                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
109                 to->firstused = be16_to_cpu(hdr3->firstused);
110                 to->holes = hdr3->holes;
111
112                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
113                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
114                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
115                 }
116                 return;
117         }
118         to->forw = be32_to_cpu(from->hdr.info.forw);
119         to->back = be32_to_cpu(from->hdr.info.back);
120         to->magic = be16_to_cpu(from->hdr.info.magic);
121         to->count = be16_to_cpu(from->hdr.count);
122         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
123         to->firstused = be16_to_cpu(from->hdr.firstused);
124         to->holes = from->hdr.holes;
125
126         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
127                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
128                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
129         }
130 }
131
132 void
133 xfs_attr3_leaf_hdr_to_disk(
134         struct xfs_attr_leafblock       *to,
135         struct xfs_attr3_icleaf_hdr     *from)
136 {
137         int     i;
138
139         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
140                from->magic == XFS_ATTR3_LEAF_MAGIC);
141
142         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
143                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
144
145                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
146                 hdr3->info.hdr.back = cpu_to_be32(from->back);
147                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
148                 hdr3->count = cpu_to_be16(from->count);
149                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
150                 hdr3->firstused = cpu_to_be16(from->firstused);
151                 hdr3->holes = from->holes;
152                 hdr3->pad1 = 0;
153
154                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
155                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
156                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
157                 }
158                 return;
159         }
160         to->hdr.info.forw = cpu_to_be32(from->forw);
161         to->hdr.info.back = cpu_to_be32(from->back);
162         to->hdr.info.magic = cpu_to_be16(from->magic);
163         to->hdr.count = cpu_to_be16(from->count);
164         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
165         to->hdr.firstused = cpu_to_be16(from->firstused);
166         to->hdr.holes = from->holes;
167         to->hdr.pad1 = 0;
168
169         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
170                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
171                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
172         }
173 }
174
175 static bool
176 xfs_attr3_leaf_verify(
177         struct xfs_buf          *bp)
178 {
179         struct xfs_mount        *mp = bp->b_target->bt_mount;
180         struct xfs_attr_leafblock *leaf = bp->b_addr;
181         struct xfs_attr3_icleaf_hdr ichdr;
182
183         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
184
185         if (xfs_sb_version_hascrc(&mp->m_sb)) {
186                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
187
188                 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
189                         return false;
190
191                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
192                         return false;
193                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
194                         return false;
195         } else {
196                 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
197                         return false;
198         }
199         if (ichdr.count == 0)
200                 return false;
201
202         /* XXX: need to range check rest of attr header values */
203         /* XXX: hash order check? */
204
205         return true;
206 }
207
208 static void
209 xfs_attr3_leaf_write_verify(
210         struct xfs_buf  *bp)
211 {
212         struct xfs_mount        *mp = bp->b_target->bt_mount;
213         struct xfs_buf_log_item *bip = bp->b_fspriv;
214         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
215
216         if (!xfs_attr3_leaf_verify(bp)) {
217                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
218                 xfs_buf_ioerror(bp, EFSCORRUPTED);
219                 return;
220         }
221
222         if (!xfs_sb_version_hascrc(&mp->m_sb))
223                 return;
224
225         if (bip)
226                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
227
228         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_ATTR3_LEAF_CRC_OFF);
229 }
230
231 /*
232  * leaf/node format detection on trees is sketchy, so a node read can be done on
233  * leaf level blocks when detection identifies the tree as a node format tree
234  * incorrectly. In this case, we need to swap the verifier to match the correct
235  * format of the block being read.
236  */
237 static void
238 xfs_attr3_leaf_read_verify(
239         struct xfs_buf          *bp)
240 {
241         struct xfs_mount        *mp = bp->b_target->bt_mount;
242
243         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
244              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
245                                           XFS_ATTR3_LEAF_CRC_OFF)) ||
246             !xfs_attr3_leaf_verify(bp)) {
247                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
248                 xfs_buf_ioerror(bp, EFSCORRUPTED);
249         }
250 }
251
252 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
253         .verify_read = xfs_attr3_leaf_read_verify,
254         .verify_write = xfs_attr3_leaf_write_verify,
255 };
256
257 int
258 xfs_attr3_leaf_read(
259         struct xfs_trans        *tp,
260         struct xfs_inode        *dp,
261         xfs_dablk_t             bno,
262         xfs_daddr_t             mappedbno,
263         struct xfs_buf          **bpp)
264 {
265         int                     err;
266
267         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
268                                 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
269         if (!err && tp)
270                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
271         return err;
272 }
273
274 /*========================================================================
275  * Namespace helper routines
276  *========================================================================*/
277
278 /*
279  * If namespace bits don't match return 0.
280  * If all match then return 1.
281  */
282 STATIC int
283 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
284 {
285         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
286 }
287
288
289 /*========================================================================
290  * External routines when attribute fork size < XFS_LITINO(mp).
291  *========================================================================*/
292
293 /*
294  * Query whether the requested number of additional bytes of extended
295  * attribute space will be able to fit inline.
296  *
297  * Returns zero if not, else the di_forkoff fork offset to be used in the
298  * literal area for attribute data once the new bytes have been added.
299  *
300  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
301  * special case for dev/uuid inodes, they have fixed size data forks.
302  */
303 int
304 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
305 {
306         int offset;
307         int minforkoff; /* lower limit on valid forkoff locations */
308         int maxforkoff; /* upper limit on valid forkoff locations */
309         int dsize;
310         xfs_mount_t *mp = dp->i_mount;
311
312         /* rounded down */
313         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
314
315         switch (dp->i_d.di_format) {
316         case XFS_DINODE_FMT_DEV:
317                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
318                 return (offset >= minforkoff) ? minforkoff : 0;
319         case XFS_DINODE_FMT_UUID:
320                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
321                 return (offset >= minforkoff) ? minforkoff : 0;
322         }
323
324         /*
325          * If the requested numbers of bytes is smaller or equal to the
326          * current attribute fork size we can always proceed.
327          *
328          * Note that if_bytes in the data fork might actually be larger than
329          * the current data fork size is due to delalloc extents. In that
330          * case either the extent count will go down when they are converted
331          * to real extents, or the delalloc conversion will take care of the
332          * literal area rebalancing.
333          */
334         if (bytes <= XFS_IFORK_ASIZE(dp))
335                 return dp->i_d.di_forkoff;
336
337         /*
338          * For attr2 we can try to move the forkoff if there is space in the
339          * literal area, but for the old format we are done if there is no
340          * space in the fixed attribute fork.
341          */
342         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
343                 return 0;
344
345         dsize = dp->i_df.if_bytes;
346
347         switch (dp->i_d.di_format) {
348         case XFS_DINODE_FMT_EXTENTS:
349                 /*
350                  * If there is no attr fork and the data fork is extents, 
351                  * determine if creating the default attr fork will result
352                  * in the extents form migrating to btree. If so, the
353                  * minimum offset only needs to be the space required for
354                  * the btree root.
355                  */
356                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
357                     xfs_default_attroffset(dp))
358                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
359                 break;
360         case XFS_DINODE_FMT_BTREE:
361                 /*
362                  * If we have a data btree then keep forkoff if we have one,
363                  * otherwise we are adding a new attr, so then we set
364                  * minforkoff to where the btree root can finish so we have
365                  * plenty of room for attrs
366                  */
367                 if (dp->i_d.di_forkoff) {
368                         if (offset < dp->i_d.di_forkoff)
369                                 return 0;
370                         return dp->i_d.di_forkoff;
371                 }
372                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
373                 break;
374         }
375
376         /*
377          * A data fork btree root must have space for at least
378          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
379          */
380         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
381         minforkoff = roundup(minforkoff, 8) >> 3;
382
383         /* attr fork btree root can have at least this many key/ptr pairs */
384         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
385                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
386         maxforkoff = maxforkoff >> 3;   /* rounded down */
387
388         if (offset >= maxforkoff)
389                 return maxforkoff;
390         if (offset >= minforkoff)
391                 return offset;
392         return 0;
393 }
394
395 /*
396  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
397  */
398 STATIC void
399 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
400 {
401         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
402             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
403                 spin_lock(&mp->m_sb_lock);
404                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
405                         xfs_sb_version_addattr2(&mp->m_sb);
406                         spin_unlock(&mp->m_sb_lock);
407                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
408                 } else
409                         spin_unlock(&mp->m_sb_lock);
410         }
411 }
412
413 /*
414  * Create the initial contents of a shortform attribute list.
415  */
416 void
417 xfs_attr_shortform_create(xfs_da_args_t *args)
418 {
419         xfs_attr_sf_hdr_t *hdr;
420         xfs_inode_t *dp;
421         xfs_ifork_t *ifp;
422
423         trace_xfs_attr_sf_create(args);
424
425         dp = args->dp;
426         ASSERT(dp != NULL);
427         ifp = dp->i_afp;
428         ASSERT(ifp != NULL);
429         ASSERT(ifp->if_bytes == 0);
430         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
431                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
432                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
433                 ifp->if_flags |= XFS_IFINLINE;
434         } else {
435                 ASSERT(ifp->if_flags & XFS_IFINLINE);
436         }
437         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
438         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
439         hdr->count = 0;
440         hdr->totsize = cpu_to_be16(sizeof(*hdr));
441         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
442 }
443
444 /*
445  * Add a name/value pair to the shortform attribute list.
446  * Overflow from the inode has already been checked for.
447  */
448 void
449 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
450 {
451         xfs_attr_shortform_t *sf;
452         xfs_attr_sf_entry_t *sfe;
453         int i, offset, size;
454         xfs_mount_t *mp;
455         xfs_inode_t *dp;
456         xfs_ifork_t *ifp;
457
458         trace_xfs_attr_sf_add(args);
459
460         dp = args->dp;
461         mp = dp->i_mount;
462         dp->i_d.di_forkoff = forkoff;
463
464         ifp = dp->i_afp;
465         ASSERT(ifp->if_flags & XFS_IFINLINE);
466         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
467         sfe = &sf->list[0];
468         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
469 #ifdef DEBUG
470                 if (sfe->namelen != args->namelen)
471                         continue;
472                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
473                         continue;
474                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
475                         continue;
476                 ASSERT(0);
477 #endif
478         }
479
480         offset = (char *)sfe - (char *)sf;
481         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
482         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
483         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
484         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
485
486         sfe->namelen = args->namelen;
487         sfe->valuelen = args->valuelen;
488         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
489         memcpy(sfe->nameval, args->name, args->namelen);
490         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
491         sf->hdr.count++;
492         be16_add_cpu(&sf->hdr.totsize, size);
493         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
494
495         xfs_sbversion_add_attr2(mp, args->trans);
496 }
497
498 /*
499  * After the last attribute is removed revert to original inode format,
500  * making all literal area available to the data fork once more.
501  */
502 STATIC void
503 xfs_attr_fork_reset(
504         struct xfs_inode        *ip,
505         struct xfs_trans        *tp)
506 {
507         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
508         ip->i_d.di_forkoff = 0;
509         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
510
511         ASSERT(ip->i_d.di_anextents == 0);
512         ASSERT(ip->i_afp == NULL);
513
514         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
515 }
516
517 /*
518  * Remove an attribute from the shortform attribute list structure.
519  */
520 int
521 xfs_attr_shortform_remove(xfs_da_args_t *args)
522 {
523         xfs_attr_shortform_t *sf;
524         xfs_attr_sf_entry_t *sfe;
525         int base, size=0, end, totsize, i;
526         xfs_mount_t *mp;
527         xfs_inode_t *dp;
528
529         trace_xfs_attr_sf_remove(args);
530
531         dp = args->dp;
532         mp = dp->i_mount;
533         base = sizeof(xfs_attr_sf_hdr_t);
534         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
535         sfe = &sf->list[0];
536         end = sf->hdr.count;
537         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
538                                         base += size, i++) {
539                 size = XFS_ATTR_SF_ENTSIZE(sfe);
540                 if (sfe->namelen != args->namelen)
541                         continue;
542                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
543                         continue;
544                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
545                         continue;
546                 break;
547         }
548         if (i == end)
549                 return(XFS_ERROR(ENOATTR));
550
551         /*
552          * Fix up the attribute fork data, covering the hole
553          */
554         end = base + size;
555         totsize = be16_to_cpu(sf->hdr.totsize);
556         if (end != totsize)
557                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
558         sf->hdr.count--;
559         be16_add_cpu(&sf->hdr.totsize, -size);
560
561         /*
562          * Fix up the start offset of the attribute fork
563          */
564         totsize -= size;
565         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
566             (mp->m_flags & XFS_MOUNT_ATTR2) &&
567             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
568             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
569                 xfs_attr_fork_reset(dp, args->trans);
570         } else {
571                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
572                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
573                 ASSERT(dp->i_d.di_forkoff);
574                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
575                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
576                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
577                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
578                 xfs_trans_log_inode(args->trans, dp,
579                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
580         }
581
582         xfs_sbversion_add_attr2(mp, args->trans);
583
584         return(0);
585 }
586
587 /*
588  * Look up a name in a shortform attribute list structure.
589  */
590 /*ARGSUSED*/
591 int
592 xfs_attr_shortform_lookup(xfs_da_args_t *args)
593 {
594         xfs_attr_shortform_t *sf;
595         xfs_attr_sf_entry_t *sfe;
596         int i;
597         xfs_ifork_t *ifp;
598
599         trace_xfs_attr_sf_lookup(args);
600
601         ifp = args->dp->i_afp;
602         ASSERT(ifp->if_flags & XFS_IFINLINE);
603         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
604         sfe = &sf->list[0];
605         for (i = 0; i < sf->hdr.count;
606                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
607                 if (sfe->namelen != args->namelen)
608                         continue;
609                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
610                         continue;
611                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
612                         continue;
613                 return(XFS_ERROR(EEXIST));
614         }
615         return(XFS_ERROR(ENOATTR));
616 }
617
618 /*
619  * Look up a name in a shortform attribute list structure.
620  */
621 /*ARGSUSED*/
622 int
623 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
624 {
625         xfs_attr_shortform_t *sf;
626         xfs_attr_sf_entry_t *sfe;
627         int i;
628
629         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
630         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
631         sfe = &sf->list[0];
632         for (i = 0; i < sf->hdr.count;
633                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
634                 if (sfe->namelen != args->namelen)
635                         continue;
636                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
637                         continue;
638                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
639                         continue;
640                 if (args->flags & ATTR_KERNOVAL) {
641                         args->valuelen = sfe->valuelen;
642                         return(XFS_ERROR(EEXIST));
643                 }
644                 if (args->valuelen < sfe->valuelen) {
645                         args->valuelen = sfe->valuelen;
646                         return(XFS_ERROR(ERANGE));
647                 }
648                 args->valuelen = sfe->valuelen;
649                 memcpy(args->value, &sfe->nameval[args->namelen],
650                                                     args->valuelen);
651                 return(XFS_ERROR(EEXIST));
652         }
653         return(XFS_ERROR(ENOATTR));
654 }
655
656 /*
657  * Convert from using the shortform to the leaf.
658  */
659 int
660 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
661 {
662         xfs_inode_t *dp;
663         xfs_attr_shortform_t *sf;
664         xfs_attr_sf_entry_t *sfe;
665         xfs_da_args_t nargs;
666         char *tmpbuffer;
667         int error, i, size;
668         xfs_dablk_t blkno;
669         struct xfs_buf *bp;
670         xfs_ifork_t *ifp;
671
672         trace_xfs_attr_sf_to_leaf(args);
673
674         dp = args->dp;
675         ifp = dp->i_afp;
676         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
677         size = be16_to_cpu(sf->hdr.totsize);
678         tmpbuffer = kmem_alloc(size, KM_SLEEP);
679         ASSERT(tmpbuffer != NULL);
680         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
681         sf = (xfs_attr_shortform_t *)tmpbuffer;
682
683         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
684         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
685
686         bp = NULL;
687         error = xfs_da_grow_inode(args, &blkno);
688         if (error) {
689                 /*
690                  * If we hit an IO error middle of the transaction inside
691                  * grow_inode(), we may have inconsistent data. Bail out.
692                  */
693                 if (error == EIO)
694                         goto out;
695                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
696                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
697                 goto out;
698         }
699
700         ASSERT(blkno == 0);
701         error = xfs_attr3_leaf_create(args, blkno, &bp);
702         if (error) {
703                 error = xfs_da_shrink_inode(args, 0, bp);
704                 bp = NULL;
705                 if (error)
706                         goto out;
707                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
708                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
709                 goto out;
710         }
711
712         memset((char *)&nargs, 0, sizeof(nargs));
713         nargs.dp = dp;
714         nargs.firstblock = args->firstblock;
715         nargs.flist = args->flist;
716         nargs.total = args->total;
717         nargs.whichfork = XFS_ATTR_FORK;
718         nargs.trans = args->trans;
719         nargs.op_flags = XFS_DA_OP_OKNOENT;
720
721         sfe = &sf->list[0];
722         for (i = 0; i < sf->hdr.count; i++) {
723                 nargs.name = sfe->nameval;
724                 nargs.namelen = sfe->namelen;
725                 nargs.value = &sfe->nameval[nargs.namelen];
726                 nargs.valuelen = sfe->valuelen;
727                 nargs.hashval = xfs_da_hashname(sfe->nameval,
728                                                 sfe->namelen);
729                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
730                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
731                 ASSERT(error == ENOATTR);
732                 error = xfs_attr3_leaf_add(bp, &nargs);
733                 ASSERT(error != ENOSPC);
734                 if (error)
735                         goto out;
736                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
737         }
738         error = 0;
739
740 out:
741         kmem_free(tmpbuffer);
742         return(error);
743 }
744
745 /*
746  * Check a leaf attribute block to see if all the entries would fit into
747  * a shortform attribute list.
748  */
749 int
750 xfs_attr_shortform_allfit(
751         struct xfs_buf          *bp,
752         struct xfs_inode        *dp)
753 {
754         struct xfs_attr_leafblock *leaf;
755         struct xfs_attr_leaf_entry *entry;
756         xfs_attr_leaf_name_local_t *name_loc;
757         struct xfs_attr3_icleaf_hdr leafhdr;
758         int                     bytes;
759         int                     i;
760
761         leaf = bp->b_addr;
762         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
763         entry = xfs_attr3_leaf_entryp(leaf);
764
765         bytes = sizeof(struct xfs_attr_sf_hdr);
766         for (i = 0; i < leafhdr.count; entry++, i++) {
767                 if (entry->flags & XFS_ATTR_INCOMPLETE)
768                         continue;               /* don't copy partial entries */
769                 if (!(entry->flags & XFS_ATTR_LOCAL))
770                         return(0);
771                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
772                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
773                         return(0);
774                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
775                         return(0);
776                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
777                                 + name_loc->namelen
778                                 + be16_to_cpu(name_loc->valuelen);
779         }
780         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
781             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
782             (bytes == sizeof(struct xfs_attr_sf_hdr)))
783                 return -1;
784         return xfs_attr_shortform_bytesfit(dp, bytes);
785 }
786
787 /*
788  * Convert a leaf attribute list to shortform attribute list
789  */
790 int
791 xfs_attr3_leaf_to_shortform(
792         struct xfs_buf          *bp,
793         struct xfs_da_args      *args,
794         int                     forkoff)
795 {
796         struct xfs_attr_leafblock *leaf;
797         struct xfs_attr3_icleaf_hdr ichdr;
798         struct xfs_attr_leaf_entry *entry;
799         struct xfs_attr_leaf_name_local *name_loc;
800         struct xfs_da_args      nargs;
801         struct xfs_inode        *dp = args->dp;
802         char                    *tmpbuffer;
803         int                     error;
804         int                     i;
805
806         trace_xfs_attr_leaf_to_sf(args);
807
808         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
809         if (!tmpbuffer)
810                 return ENOMEM;
811
812         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
813
814         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
815         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
816         entry = xfs_attr3_leaf_entryp(leaf);
817
818         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
819         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
820
821         /*
822          * Clean out the prior contents of the attribute list.
823          */
824         error = xfs_da_shrink_inode(args, 0, bp);
825         if (error)
826                 goto out;
827
828         if (forkoff == -1) {
829                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
830                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
831                 xfs_attr_fork_reset(dp, args->trans);
832                 goto out;
833         }
834
835         xfs_attr_shortform_create(args);
836
837         /*
838          * Copy the attributes
839          */
840         memset((char *)&nargs, 0, sizeof(nargs));
841         nargs.dp = dp;
842         nargs.firstblock = args->firstblock;
843         nargs.flist = args->flist;
844         nargs.total = args->total;
845         nargs.whichfork = XFS_ATTR_FORK;
846         nargs.trans = args->trans;
847         nargs.op_flags = XFS_DA_OP_OKNOENT;
848
849         for (i = 0; i < ichdr.count; entry++, i++) {
850                 if (entry->flags & XFS_ATTR_INCOMPLETE)
851                         continue;       /* don't copy partial entries */
852                 if (!entry->nameidx)
853                         continue;
854                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
855                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
856                 nargs.name = name_loc->nameval;
857                 nargs.namelen = name_loc->namelen;
858                 nargs.value = &name_loc->nameval[nargs.namelen];
859                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
860                 nargs.hashval = be32_to_cpu(entry->hashval);
861                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
862                 xfs_attr_shortform_add(&nargs, forkoff);
863         }
864         error = 0;
865
866 out:
867         kmem_free(tmpbuffer);
868         return error;
869 }
870
871 /*
872  * Convert from using a single leaf to a root node and a leaf.
873  */
874 int
875 xfs_attr3_leaf_to_node(
876         struct xfs_da_args      *args)
877 {
878         struct xfs_attr_leafblock *leaf;
879         struct xfs_attr3_icleaf_hdr icleafhdr;
880         struct xfs_attr_leaf_entry *entries;
881         struct xfs_da_node_entry *btree;
882         struct xfs_da3_icnode_hdr icnodehdr;
883         struct xfs_da_intnode   *node;
884         struct xfs_inode        *dp = args->dp;
885         struct xfs_mount        *mp = dp->i_mount;
886         struct xfs_buf          *bp1 = NULL;
887         struct xfs_buf          *bp2 = NULL;
888         xfs_dablk_t             blkno;
889         int                     error;
890
891         trace_xfs_attr_leaf_to_node(args);
892
893         error = xfs_da_grow_inode(args, &blkno);
894         if (error)
895                 goto out;
896         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
897         if (error)
898                 goto out;
899
900         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
901         if (error)
902                 goto out;
903
904         /* copy leaf to new buffer, update identifiers */
905         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
906         bp2->b_ops = bp1->b_ops;
907         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
908         if (xfs_sb_version_hascrc(&mp->m_sb)) {
909                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
910                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
911         }
912         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
913
914         /*
915          * Set up the new root node.
916          */
917         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
918         if (error)
919                 goto out;
920         node = bp1->b_addr;
921         xfs_da3_node_hdr_from_disk(&icnodehdr, node);
922         btree = xfs_da3_node_tree_p(node);
923
924         leaf = bp2->b_addr;
925         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
926         entries = xfs_attr3_leaf_entryp(leaf);
927
928         /* both on-disk, don't endian-flip twice */
929         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
930         btree[0].before = cpu_to_be32(blkno);
931         icnodehdr.count = 1;
932         xfs_da3_node_hdr_to_disk(node, &icnodehdr);
933         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
934         error = 0;
935 out:
936         return error;
937 }
938
939 /*========================================================================
940  * Routines used for growing the Btree.
941  *========================================================================*/
942
943 /*
944  * Create the initial contents of a leaf attribute list
945  * or a leaf in a node attribute list.
946  */
947 STATIC int
948 xfs_attr3_leaf_create(
949         struct xfs_da_args      *args,
950         xfs_dablk_t             blkno,
951         struct xfs_buf          **bpp)
952 {
953         struct xfs_attr_leafblock *leaf;
954         struct xfs_attr3_icleaf_hdr ichdr;
955         struct xfs_inode        *dp = args->dp;
956         struct xfs_mount        *mp = dp->i_mount;
957         struct xfs_buf          *bp;
958         int                     error;
959
960         trace_xfs_attr_leaf_create(args);
961
962         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
963                                             XFS_ATTR_FORK);
964         if (error)
965                 return error;
966         bp->b_ops = &xfs_attr3_leaf_buf_ops;
967         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
968         leaf = bp->b_addr;
969         memset(leaf, 0, XFS_LBSIZE(mp));
970
971         memset(&ichdr, 0, sizeof(ichdr));
972         ichdr.firstused = XFS_LBSIZE(mp);
973
974         if (xfs_sb_version_hascrc(&mp->m_sb)) {
975                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
976
977                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
978
979                 hdr3->blkno = cpu_to_be64(bp->b_bn);
980                 hdr3->owner = cpu_to_be64(dp->i_ino);
981                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
982
983                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
984         } else {
985                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
986                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
987         }
988         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
989
990         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
991         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
992
993         *bpp = bp;
994         return 0;
995 }
996
997 /*
998  * Split the leaf node, rebalance, then add the new entry.
999  */
1000 int
1001 xfs_attr3_leaf_split(
1002         struct xfs_da_state     *state,
1003         struct xfs_da_state_blk *oldblk,
1004         struct xfs_da_state_blk *newblk)
1005 {
1006         xfs_dablk_t blkno;
1007         int error;
1008
1009         trace_xfs_attr_leaf_split(state->args);
1010
1011         /*
1012          * Allocate space for a new leaf node.
1013          */
1014         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1015         error = xfs_da_grow_inode(state->args, &blkno);
1016         if (error)
1017                 return(error);
1018         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1019         if (error)
1020                 return(error);
1021         newblk->blkno = blkno;
1022         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1023
1024         /*
1025          * Rebalance the entries across the two leaves.
1026          * NOTE: rebalance() currently depends on the 2nd block being empty.
1027          */
1028         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1029         error = xfs_da3_blk_link(state, oldblk, newblk);
1030         if (error)
1031                 return(error);
1032
1033         /*
1034          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1035          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1036          * "new" attrs info.  Will need the "old" info to remove it later.
1037          *
1038          * Insert the "new" entry in the correct block.
1039          */
1040         if (state->inleaf) {
1041                 trace_xfs_attr_leaf_add_old(state->args);
1042                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1043         } else {
1044                 trace_xfs_attr_leaf_add_new(state->args);
1045                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1046         }
1047
1048         /*
1049          * Update last hashval in each block since we added the name.
1050          */
1051         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1052         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1053         return(error);
1054 }
1055
1056 /*
1057  * Add a name to the leaf attribute list structure.
1058  */
1059 int
1060 xfs_attr3_leaf_add(
1061         struct xfs_buf          *bp,
1062         struct xfs_da_args      *args)
1063 {
1064         struct xfs_attr_leafblock *leaf;
1065         struct xfs_attr3_icleaf_hdr ichdr;
1066         int                     tablesize;
1067         int                     entsize;
1068         int                     sum;
1069         int                     tmp;
1070         int                     i;
1071
1072         trace_xfs_attr_leaf_add(args);
1073
1074         leaf = bp->b_addr;
1075         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1076         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1077         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1078                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1079
1080         /*
1081          * Search through freemap for first-fit on new name length.
1082          * (may need to figure in size of entry struct too)
1083          */
1084         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1085                                         + xfs_attr3_leaf_hdr_size(leaf);
1086         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1087                 if (tablesize > ichdr.firstused) {
1088                         sum += ichdr.freemap[i].size;
1089                         continue;
1090                 }
1091                 if (!ichdr.freemap[i].size)
1092                         continue;       /* no space in this map */
1093                 tmp = entsize;
1094                 if (ichdr.freemap[i].base < ichdr.firstused)
1095                         tmp += sizeof(xfs_attr_leaf_entry_t);
1096                 if (ichdr.freemap[i].size >= tmp) {
1097                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1098                         goto out_log_hdr;
1099                 }
1100                 sum += ichdr.freemap[i].size;
1101         }
1102
1103         /*
1104          * If there are no holes in the address space of the block,
1105          * and we don't have enough freespace, then compaction will do us
1106          * no good and we should just give up.
1107          */
1108         if (!ichdr.holes && sum < entsize)
1109                 return XFS_ERROR(ENOSPC);
1110
1111         /*
1112          * Compact the entries to coalesce free space.
1113          * This may change the hdr->count via dropping INCOMPLETE entries.
1114          */
1115         xfs_attr3_leaf_compact(args, &ichdr, bp);
1116
1117         /*
1118          * After compaction, the block is guaranteed to have only one
1119          * free region, in freemap[0].  If it is not big enough, give up.
1120          */
1121         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1122                 tmp = ENOSPC;
1123                 goto out_log_hdr;
1124         }
1125
1126         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1127
1128 out_log_hdr:
1129         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1130         xfs_trans_log_buf(args->trans, bp,
1131                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1132                                 xfs_attr3_leaf_hdr_size(leaf)));
1133         return tmp;
1134 }
1135
1136 /*
1137  * Add a name to a leaf attribute list structure.
1138  */
1139 STATIC int
1140 xfs_attr3_leaf_add_work(
1141         struct xfs_buf          *bp,
1142         struct xfs_attr3_icleaf_hdr *ichdr,
1143         struct xfs_da_args      *args,
1144         int                     mapindex)
1145 {
1146         struct xfs_attr_leafblock *leaf;
1147         struct xfs_attr_leaf_entry *entry;
1148         struct xfs_attr_leaf_name_local *name_loc;
1149         struct xfs_attr_leaf_name_remote *name_rmt;
1150         struct xfs_mount        *mp;
1151         int                     tmp;
1152         int                     i;
1153
1154         trace_xfs_attr_leaf_add_work(args);
1155
1156         leaf = bp->b_addr;
1157         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1158         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1159
1160         /*
1161          * Force open some space in the entry array and fill it in.
1162          */
1163         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1164         if (args->index < ichdr->count) {
1165                 tmp  = ichdr->count - args->index;
1166                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1167                 memmove(entry + 1, entry, tmp);
1168                 xfs_trans_log_buf(args->trans, bp,
1169                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1170         }
1171         ichdr->count++;
1172
1173         /*
1174          * Allocate space for the new string (at the end of the run).
1175          */
1176         mp = args->trans->t_mountp;
1177         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1178         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1179         ASSERT(ichdr->freemap[mapindex].size >=
1180                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1181                                          mp->m_sb.sb_blocksize, NULL));
1182         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1183         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1184
1185         ichdr->freemap[mapindex].size -=
1186                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1187                                                  mp->m_sb.sb_blocksize, &tmp);
1188
1189         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1190                                      ichdr->freemap[mapindex].size);
1191         entry->hashval = cpu_to_be32(args->hashval);
1192         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1193         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1194         if (args->op_flags & XFS_DA_OP_RENAME) {
1195                 entry->flags |= XFS_ATTR_INCOMPLETE;
1196                 if ((args->blkno2 == args->blkno) &&
1197                     (args->index2 <= args->index)) {
1198                         args->index2++;
1199                 }
1200         }
1201         xfs_trans_log_buf(args->trans, bp,
1202                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1203         ASSERT((args->index == 0) ||
1204                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1205         ASSERT((args->index == ichdr->count - 1) ||
1206                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1207
1208         /*
1209          * For "remote" attribute values, simply note that we need to
1210          * allocate space for the "remote" value.  We can't actually
1211          * allocate the extents in this transaction, and we can't decide
1212          * which blocks they should be as we might allocate more blocks
1213          * as part of this transaction (a split operation for example).
1214          */
1215         if (entry->flags & XFS_ATTR_LOCAL) {
1216                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1217                 name_loc->namelen = args->namelen;
1218                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1219                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1220                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1221                                    be16_to_cpu(name_loc->valuelen));
1222         } else {
1223                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1224                 name_rmt->namelen = args->namelen;
1225                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1226                 entry->flags |= XFS_ATTR_INCOMPLETE;
1227                 /* just in case */
1228                 name_rmt->valuelen = 0;
1229                 name_rmt->valueblk = 0;
1230                 args->rmtblkno = 1;
1231                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1232         }
1233         xfs_trans_log_buf(args->trans, bp,
1234              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1235                                    xfs_attr_leaf_entsize(leaf, args->index)));
1236
1237         /*
1238          * Update the control info for this leaf node
1239          */
1240         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1241                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1242
1243         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1244                                         + xfs_attr3_leaf_hdr_size(leaf));
1245         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1246                                         + xfs_attr3_leaf_hdr_size(leaf);
1247
1248         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1249                 if (ichdr->freemap[i].base == tmp) {
1250                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1251                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1252                 }
1253         }
1254         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1255         return 0;
1256 }
1257
1258 /*
1259  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1260  */
1261 STATIC void
1262 xfs_attr3_leaf_compact(
1263         struct xfs_da_args      *args,
1264         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1265         struct xfs_buf          *bp)
1266 {
1267         struct xfs_attr_leafblock *leaf_src;
1268         struct xfs_attr_leafblock *leaf_dst;
1269         struct xfs_attr3_icleaf_hdr ichdr_src;
1270         struct xfs_trans        *trans = args->trans;
1271         struct xfs_mount        *mp = trans->t_mountp;
1272         char                    *tmpbuffer;
1273
1274         trace_xfs_attr_leaf_compact(args);
1275
1276         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1277         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1278         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1279         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1280         leaf_dst = bp->b_addr;
1281
1282         /*
1283          * Copy the on-disk header back into the destination buffer to ensure
1284          * all the information in the header that is not part of the incore
1285          * header structure is preserved.
1286          */
1287         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1288
1289         /* Initialise the incore headers */
1290         ichdr_src = *ichdr_dst; /* struct copy */
1291         ichdr_dst->firstused = XFS_LBSIZE(mp);
1292         ichdr_dst->usedbytes = 0;
1293         ichdr_dst->count = 0;
1294         ichdr_dst->holes = 0;
1295         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1296         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1297                                                 ichdr_dst->freemap[0].base;
1298
1299         /* write the header back to initialise the underlying buffer */
1300         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1301
1302         /*
1303          * Copy all entry's in the same (sorted) order,
1304          * but allocate name/value pairs packed and in sequence.
1305          */
1306         xfs_attr3_leaf_moveents(leaf_src, &ichdr_src, 0, leaf_dst, ichdr_dst, 0,
1307                                 ichdr_src.count, mp);
1308         /*
1309          * this logs the entire buffer, but the caller must write the header
1310          * back to the buffer when it is finished modifying it.
1311          */
1312         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1313
1314         kmem_free(tmpbuffer);
1315 }
1316
1317 /*
1318  * Compare two leaf blocks "order".
1319  * Return 0 unless leaf2 should go before leaf1.
1320  */
1321 static int
1322 xfs_attr3_leaf_order(
1323         struct xfs_buf  *leaf1_bp,
1324         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1325         struct xfs_buf  *leaf2_bp,
1326         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1327 {
1328         struct xfs_attr_leaf_entry *entries1;
1329         struct xfs_attr_leaf_entry *entries2;
1330
1331         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1332         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1333         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1334             ((be32_to_cpu(entries2[0].hashval) <
1335               be32_to_cpu(entries1[0].hashval)) ||
1336              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1337               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1338                 return 1;
1339         }
1340         return 0;
1341 }
1342
1343 int
1344 xfs_attr_leaf_order(
1345         struct xfs_buf  *leaf1_bp,
1346         struct xfs_buf  *leaf2_bp)
1347 {
1348         struct xfs_attr3_icleaf_hdr ichdr1;
1349         struct xfs_attr3_icleaf_hdr ichdr2;
1350
1351         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1352         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1353         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1354 }
1355
1356 /*
1357  * Redistribute the attribute list entries between two leaf nodes,
1358  * taking into account the size of the new entry.
1359  *
1360  * NOTE: if new block is empty, then it will get the upper half of the
1361  * old block.  At present, all (one) callers pass in an empty second block.
1362  *
1363  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1364  * to match what it is doing in splitting the attribute leaf block.  Those
1365  * values are used in "atomic rename" operations on attributes.  Note that
1366  * the "new" and "old" values can end up in different blocks.
1367  */
1368 STATIC void
1369 xfs_attr3_leaf_rebalance(
1370         struct xfs_da_state     *state,
1371         struct xfs_da_state_blk *blk1,
1372         struct xfs_da_state_blk *blk2)
1373 {
1374         struct xfs_da_args      *args;
1375         struct xfs_attr_leafblock *leaf1;
1376         struct xfs_attr_leafblock *leaf2;
1377         struct xfs_attr3_icleaf_hdr ichdr1;
1378         struct xfs_attr3_icleaf_hdr ichdr2;
1379         struct xfs_attr_leaf_entry *entries1;
1380         struct xfs_attr_leaf_entry *entries2;
1381         int                     count;
1382         int                     totallen;
1383         int                     max;
1384         int                     space;
1385         int                     swap;
1386
1387         /*
1388          * Set up environment.
1389          */
1390         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1391         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1392         leaf1 = blk1->bp->b_addr;
1393         leaf2 = blk2->bp->b_addr;
1394         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1395         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1396         ASSERT(ichdr2.count == 0);
1397         args = state->args;
1398
1399         trace_xfs_attr_leaf_rebalance(args);
1400
1401         /*
1402          * Check ordering of blocks, reverse if it makes things simpler.
1403          *
1404          * NOTE: Given that all (current) callers pass in an empty
1405          * second block, this code should never set "swap".
1406          */
1407         swap = 0;
1408         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1409                 struct xfs_da_state_blk *tmp_blk;
1410                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1411
1412                 tmp_blk = blk1;
1413                 blk1 = blk2;
1414                 blk2 = tmp_blk;
1415
1416                 /* struct copies to swap them rather than reconverting */
1417                 tmp_ichdr = ichdr1;
1418                 ichdr1 = ichdr2;
1419                 ichdr2 = tmp_ichdr;
1420
1421                 leaf1 = blk1->bp->b_addr;
1422                 leaf2 = blk2->bp->b_addr;
1423                 swap = 1;
1424         }
1425
1426         /*
1427          * Examine entries until we reduce the absolute difference in
1428          * byte usage between the two blocks to a minimum.  Then get
1429          * the direction to copy and the number of elements to move.
1430          *
1431          * "inleaf" is true if the new entry should be inserted into blk1.
1432          * If "swap" is also true, then reverse the sense of "inleaf".
1433          */
1434         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1435                                                       blk2, &ichdr2,
1436                                                       &count, &totallen);
1437         if (swap)
1438                 state->inleaf = !state->inleaf;
1439
1440         /*
1441          * Move any entries required from leaf to leaf:
1442          */
1443         if (count < ichdr1.count) {
1444                 /*
1445                  * Figure the total bytes to be added to the destination leaf.
1446                  */
1447                 /* number entries being moved */
1448                 count = ichdr1.count - count;
1449                 space  = ichdr1.usedbytes - totallen;
1450                 space += count * sizeof(xfs_attr_leaf_entry_t);
1451
1452                 /*
1453                  * leaf2 is the destination, compact it if it looks tight.
1454                  */
1455                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1456                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1457                 if (space > max)
1458                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1459
1460                 /*
1461                  * Move high entries from leaf1 to low end of leaf2.
1462                  */
1463                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1464                                 leaf2, &ichdr2, 0, count, state->mp);
1465
1466         } else if (count > ichdr1.count) {
1467                 /*
1468                  * I assert that since all callers pass in an empty
1469                  * second buffer, this code should never execute.
1470                  */
1471                 ASSERT(0);
1472
1473                 /*
1474                  * Figure the total bytes to be added to the destination leaf.
1475                  */
1476                 /* number entries being moved */
1477                 count -= ichdr1.count;
1478                 space  = totallen - ichdr1.usedbytes;
1479                 space += count * sizeof(xfs_attr_leaf_entry_t);
1480
1481                 /*
1482                  * leaf1 is the destination, compact it if it looks tight.
1483                  */
1484                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1485                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1486                 if (space > max)
1487                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1488
1489                 /*
1490                  * Move low entries from leaf2 to high end of leaf1.
1491                  */
1492                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1493                                         ichdr1.count, count, state->mp);
1494         }
1495
1496         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1497         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1498         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1499         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1500
1501         /*
1502          * Copy out last hashval in each block for B-tree code.
1503          */
1504         entries1 = xfs_attr3_leaf_entryp(leaf1);
1505         entries2 = xfs_attr3_leaf_entryp(leaf2);
1506         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1507         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1508
1509         /*
1510          * Adjust the expected index for insertion.
1511          * NOTE: this code depends on the (current) situation that the
1512          * second block was originally empty.
1513          *
1514          * If the insertion point moved to the 2nd block, we must adjust
1515          * the index.  We must also track the entry just following the
1516          * new entry for use in an "atomic rename" operation, that entry
1517          * is always the "old" entry and the "new" entry is what we are
1518          * inserting.  The index/blkno fields refer to the "old" entry,
1519          * while the index2/blkno2 fields refer to the "new" entry.
1520          */
1521         if (blk1->index > ichdr1.count) {
1522                 ASSERT(state->inleaf == 0);
1523                 blk2->index = blk1->index - ichdr1.count;
1524                 args->index = args->index2 = blk2->index;
1525                 args->blkno = args->blkno2 = blk2->blkno;
1526         } else if (blk1->index == ichdr1.count) {
1527                 if (state->inleaf) {
1528                         args->index = blk1->index;
1529                         args->blkno = blk1->blkno;
1530                         args->index2 = 0;
1531                         args->blkno2 = blk2->blkno;
1532                 } else {
1533                         /*
1534                          * On a double leaf split, the original attr location
1535                          * is already stored in blkno2/index2, so don't
1536                          * overwrite it overwise we corrupt the tree.
1537                          */
1538                         blk2->index = blk1->index - ichdr1.count;
1539                         args->index = blk2->index;
1540                         args->blkno = blk2->blkno;
1541                         if (!state->extravalid) {
1542                                 /*
1543                                  * set the new attr location to match the old
1544                                  * one and let the higher level split code
1545                                  * decide where in the leaf to place it.
1546                                  */
1547                                 args->index2 = blk2->index;
1548                                 args->blkno2 = blk2->blkno;
1549                         }
1550                 }
1551         } else {
1552                 ASSERT(state->inleaf == 1);
1553                 args->index = args->index2 = blk1->index;
1554                 args->blkno = args->blkno2 = blk1->blkno;
1555         }
1556 }
1557
1558 /*
1559  * Examine entries until we reduce the absolute difference in
1560  * byte usage between the two blocks to a minimum.
1561  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1562  * GROT: there will always be enough room in either block for a new entry.
1563  * GROT: Do a double-split for this case?
1564  */
1565 STATIC int
1566 xfs_attr3_leaf_figure_balance(
1567         struct xfs_da_state             *state,
1568         struct xfs_da_state_blk         *blk1,
1569         struct xfs_attr3_icleaf_hdr     *ichdr1,
1570         struct xfs_da_state_blk         *blk2,
1571         struct xfs_attr3_icleaf_hdr     *ichdr2,
1572         int                             *countarg,
1573         int                             *usedbytesarg)
1574 {
1575         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1576         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1577         struct xfs_attr_leaf_entry      *entry;
1578         int                             count;
1579         int                             max;
1580         int                             index;
1581         int                             totallen = 0;
1582         int                             half;
1583         int                             lastdelta;
1584         int                             foundit = 0;
1585         int                             tmp;
1586
1587         /*
1588          * Examine entries until we reduce the absolute difference in
1589          * byte usage between the two blocks to a minimum.
1590          */
1591         max = ichdr1->count + ichdr2->count;
1592         half = (max + 1) * sizeof(*entry);
1593         half += ichdr1->usedbytes + ichdr2->usedbytes +
1594                         xfs_attr_leaf_newentsize(state->args->namelen,
1595                                                  state->args->valuelen,
1596                                                  state->blocksize, NULL);
1597         half /= 2;
1598         lastdelta = state->blocksize;
1599         entry = xfs_attr3_leaf_entryp(leaf1);
1600         for (count = index = 0; count < max; entry++, index++, count++) {
1601
1602 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1603                 /*
1604                  * The new entry is in the first block, account for it.
1605                  */
1606                 if (count == blk1->index) {
1607                         tmp = totallen + sizeof(*entry) +
1608                                 xfs_attr_leaf_newentsize(
1609                                                 state->args->namelen,
1610                                                 state->args->valuelen,
1611                                                 state->blocksize, NULL);
1612                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1613                                 break;
1614                         lastdelta = XFS_ATTR_ABS(half - tmp);
1615                         totallen = tmp;
1616                         foundit = 1;
1617                 }
1618
1619                 /*
1620                  * Wrap around into the second block if necessary.
1621                  */
1622                 if (count == ichdr1->count) {
1623                         leaf1 = leaf2;
1624                         entry = xfs_attr3_leaf_entryp(leaf1);
1625                         index = 0;
1626                 }
1627
1628                 /*
1629                  * Figure out if next leaf entry would be too much.
1630                  */
1631                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1632                                                                         index);
1633                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1634                         break;
1635                 lastdelta = XFS_ATTR_ABS(half - tmp);
1636                 totallen = tmp;
1637 #undef XFS_ATTR_ABS
1638         }
1639
1640         /*
1641          * Calculate the number of usedbytes that will end up in lower block.
1642          * If new entry not in lower block, fix up the count.
1643          */
1644         totallen -= count * sizeof(*entry);
1645         if (foundit) {
1646                 totallen -= sizeof(*entry) +
1647                                 xfs_attr_leaf_newentsize(
1648                                                 state->args->namelen,
1649                                                 state->args->valuelen,
1650                                                 state->blocksize, NULL);
1651         }
1652
1653         *countarg = count;
1654         *usedbytesarg = totallen;
1655         return foundit;
1656 }
1657
1658 /*========================================================================
1659  * Routines used for shrinking the Btree.
1660  *========================================================================*/
1661
1662 /*
1663  * Check a leaf block and its neighbors to see if the block should be
1664  * collapsed into one or the other neighbor.  Always keep the block
1665  * with the smaller block number.
1666  * If the current block is over 50% full, don't try to join it, return 0.
1667  * If the block is empty, fill in the state structure and return 2.
1668  * If it can be collapsed, fill in the state structure and return 1.
1669  * If nothing can be done, return 0.
1670  *
1671  * GROT: allow for INCOMPLETE entries in calculation.
1672  */
1673 int
1674 xfs_attr3_leaf_toosmall(
1675         struct xfs_da_state     *state,
1676         int                     *action)
1677 {
1678         struct xfs_attr_leafblock *leaf;
1679         struct xfs_da_state_blk *blk;
1680         struct xfs_attr3_icleaf_hdr ichdr;
1681         struct xfs_buf          *bp;
1682         xfs_dablk_t             blkno;
1683         int                     bytes;
1684         int                     forward;
1685         int                     error;
1686         int                     retval;
1687         int                     i;
1688
1689         trace_xfs_attr_leaf_toosmall(state->args);
1690
1691         /*
1692          * Check for the degenerate case of the block being over 50% full.
1693          * If so, it's not worth even looking to see if we might be able
1694          * to coalesce with a sibling.
1695          */
1696         blk = &state->path.blk[ state->path.active-1 ];
1697         leaf = blk->bp->b_addr;
1698         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1699         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1700                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1701                 ichdr.usedbytes;
1702         if (bytes > (state->blocksize >> 1)) {
1703                 *action = 0;    /* blk over 50%, don't try to join */
1704                 return(0);
1705         }
1706
1707         /*
1708          * Check for the degenerate case of the block being empty.
1709          * If the block is empty, we'll simply delete it, no need to
1710          * coalesce it with a sibling block.  We choose (arbitrarily)
1711          * to merge with the forward block unless it is NULL.
1712          */
1713         if (ichdr.count == 0) {
1714                 /*
1715                  * Make altpath point to the block we want to keep and
1716                  * path point to the block we want to drop (this one).
1717                  */
1718                 forward = (ichdr.forw != 0);
1719                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1720                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1721                                                  0, &retval);
1722                 if (error)
1723                         return(error);
1724                 if (retval) {
1725                         *action = 0;
1726                 } else {
1727                         *action = 2;
1728                 }
1729                 return 0;
1730         }
1731
1732         /*
1733          * Examine each sibling block to see if we can coalesce with
1734          * at least 25% free space to spare.  We need to figure out
1735          * whether to merge with the forward or the backward block.
1736          * We prefer coalescing with the lower numbered sibling so as
1737          * to shrink an attribute list over time.
1738          */
1739         /* start with smaller blk num */
1740         forward = ichdr.forw < ichdr.back;
1741         for (i = 0; i < 2; forward = !forward, i++) {
1742                 struct xfs_attr3_icleaf_hdr ichdr2;
1743                 if (forward)
1744                         blkno = ichdr.forw;
1745                 else
1746                         blkno = ichdr.back;
1747                 if (blkno == 0)
1748                         continue;
1749                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1750                                         blkno, -1, &bp);
1751                 if (error)
1752                         return(error);
1753
1754                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1755
1756                 bytes = state->blocksize - (state->blocksize >> 2) -
1757                         ichdr.usedbytes - ichdr2.usedbytes -
1758                         ((ichdr.count + ichdr2.count) *
1759                                         sizeof(xfs_attr_leaf_entry_t)) -
1760                         xfs_attr3_leaf_hdr_size(leaf);
1761
1762                 xfs_trans_brelse(state->args->trans, bp);
1763                 if (bytes >= 0)
1764                         break;  /* fits with at least 25% to spare */
1765         }
1766         if (i >= 2) {
1767                 *action = 0;
1768                 return(0);
1769         }
1770
1771         /*
1772          * Make altpath point to the block we want to keep (the lower
1773          * numbered block) and path point to the block we want to drop.
1774          */
1775         memcpy(&state->altpath, &state->path, sizeof(state->path));
1776         if (blkno < blk->blkno) {
1777                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1778                                                  0, &retval);
1779         } else {
1780                 error = xfs_da3_path_shift(state, &state->path, forward,
1781                                                  0, &retval);
1782         }
1783         if (error)
1784                 return(error);
1785         if (retval) {
1786                 *action = 0;
1787         } else {
1788                 *action = 1;
1789         }
1790         return(0);
1791 }
1792
1793 /*
1794  * Remove a name from the leaf attribute list structure.
1795  *
1796  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1797  * If two leaves are 37% full, when combined they will leave 25% free.
1798  */
1799 int
1800 xfs_attr3_leaf_remove(
1801         struct xfs_buf          *bp,
1802         struct xfs_da_args      *args)
1803 {
1804         struct xfs_attr_leafblock *leaf;
1805         struct xfs_attr3_icleaf_hdr ichdr;
1806         struct xfs_attr_leaf_entry *entry;
1807         struct xfs_mount        *mp = args->trans->t_mountp;
1808         int                     before;
1809         int                     after;
1810         int                     smallest;
1811         int                     entsize;
1812         int                     tablesize;
1813         int                     tmp;
1814         int                     i;
1815
1816         trace_xfs_attr_leaf_remove(args);
1817
1818         leaf = bp->b_addr;
1819         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1820
1821         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
1822         ASSERT(args->index >= 0 && args->index < ichdr.count);
1823         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1824                                         xfs_attr3_leaf_hdr_size(leaf));
1825
1826         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1827
1828         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1829         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1830
1831         /*
1832          * Scan through free region table:
1833          *    check for adjacency of free'd entry with an existing one,
1834          *    find smallest free region in case we need to replace it,
1835          *    adjust any map that borders the entry table,
1836          */
1837         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1838                                         + xfs_attr3_leaf_hdr_size(leaf);
1839         tmp = ichdr.freemap[0].size;
1840         before = after = -1;
1841         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1842         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1843         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1844                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
1845                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
1846                 if (ichdr.freemap[i].base == tablesize) {
1847                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1848                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1849                 }
1850
1851                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1852                                 be16_to_cpu(entry->nameidx)) {
1853                         before = i;
1854                 } else if (ichdr.freemap[i].base ==
1855                                 (be16_to_cpu(entry->nameidx) + entsize)) {
1856                         after = i;
1857                 } else if (ichdr.freemap[i].size < tmp) {
1858                         tmp = ichdr.freemap[i].size;
1859                         smallest = i;
1860                 }
1861         }
1862
1863         /*
1864          * Coalesce adjacent freemap regions,
1865          * or replace the smallest region.
1866          */
1867         if ((before >= 0) || (after >= 0)) {
1868                 if ((before >= 0) && (after >= 0)) {
1869                         ichdr.freemap[before].size += entsize;
1870                         ichdr.freemap[before].size += ichdr.freemap[after].size;
1871                         ichdr.freemap[after].base = 0;
1872                         ichdr.freemap[after].size = 0;
1873                 } else if (before >= 0) {
1874                         ichdr.freemap[before].size += entsize;
1875                 } else {
1876                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1877                         ichdr.freemap[after].size += entsize;
1878                 }
1879         } else {
1880                 /*
1881                  * Replace smallest region (if it is smaller than free'd entry)
1882                  */
1883                 if (ichdr.freemap[smallest].size < entsize) {
1884                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1885                         ichdr.freemap[smallest].size = entsize;
1886                 }
1887         }
1888
1889         /*
1890          * Did we remove the first entry?
1891          */
1892         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1893                 smallest = 1;
1894         else
1895                 smallest = 0;
1896
1897         /*
1898          * Compress the remaining entries and zero out the removed stuff.
1899          */
1900         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1901         ichdr.usedbytes -= entsize;
1902         xfs_trans_log_buf(args->trans, bp,
1903              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1904                                    entsize));
1905
1906         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1907         memmove(entry, entry + 1, tmp);
1908         ichdr.count--;
1909         xfs_trans_log_buf(args->trans, bp,
1910             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1911
1912         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1913         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1914
1915         /*
1916          * If we removed the first entry, re-find the first used byte
1917          * in the name area.  Note that if the entry was the "firstused",
1918          * then we don't have a "hole" in our block resulting from
1919          * removing the name.
1920          */
1921         if (smallest) {
1922                 tmp = XFS_LBSIZE(mp);
1923                 entry = xfs_attr3_leaf_entryp(leaf);
1924                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1925                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1926                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1927
1928                         if (be16_to_cpu(entry->nameidx) < tmp)
1929                                 tmp = be16_to_cpu(entry->nameidx);
1930                 }
1931                 ichdr.firstused = tmp;
1932                 if (!ichdr.firstused)
1933                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
1934         } else {
1935                 ichdr.holes = 1;        /* mark as needing compaction */
1936         }
1937         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1938         xfs_trans_log_buf(args->trans, bp,
1939                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1940                                           xfs_attr3_leaf_hdr_size(leaf)));
1941
1942         /*
1943          * Check if leaf is less than 50% full, caller may want to
1944          * "join" the leaf with a sibling if so.
1945          */
1946         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
1947               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
1948
1949         return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
1950 }
1951
1952 /*
1953  * Move all the attribute list entries from drop_leaf into save_leaf.
1954  */
1955 void
1956 xfs_attr3_leaf_unbalance(
1957         struct xfs_da_state     *state,
1958         struct xfs_da_state_blk *drop_blk,
1959         struct xfs_da_state_blk *save_blk)
1960 {
1961         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
1962         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
1963         struct xfs_attr3_icleaf_hdr drophdr;
1964         struct xfs_attr3_icleaf_hdr savehdr;
1965         struct xfs_attr_leaf_entry *entry;
1966         struct xfs_mount        *mp = state->mp;
1967
1968         trace_xfs_attr_leaf_unbalance(state->args);
1969
1970         drop_leaf = drop_blk->bp->b_addr;
1971         save_leaf = save_blk->bp->b_addr;
1972         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1973         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
1974         entry = xfs_attr3_leaf_entryp(drop_leaf);
1975
1976         /*
1977          * Save last hashval from dying block for later Btree fixup.
1978          */
1979         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
1980
1981         /*
1982          * Check if we need a temp buffer, or can we do it in place.
1983          * Note that we don't check "leaf" for holes because we will
1984          * always be dropping it, toosmall() decided that for us already.
1985          */
1986         if (savehdr.holes == 0) {
1987                 /*
1988                  * dest leaf has no holes, so we add there.  May need
1989                  * to make some room in the entry array.
1990                  */
1991                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
1992                                          drop_blk->bp, &drophdr)) {
1993                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1994                                                 save_leaf, &savehdr, 0,
1995                                                 drophdr.count, mp);
1996                 } else {
1997                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1998                                                 save_leaf, &savehdr,
1999                                                 savehdr.count, drophdr.count, mp);
2000                 }
2001         } else {
2002                 /*
2003                  * Destination has holes, so we make a temporary copy
2004                  * of the leaf and add them both to that.
2005                  */
2006                 struct xfs_attr_leafblock *tmp_leaf;
2007                 struct xfs_attr3_icleaf_hdr tmphdr;
2008
2009                 tmp_leaf = kmem_zalloc(state->blocksize, KM_SLEEP);
2010
2011                 /*
2012                  * Copy the header into the temp leaf so that all the stuff
2013                  * not in the incore header is present and gets copied back in
2014                  * once we've moved all the entries.
2015                  */
2016                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2017
2018                 memset(&tmphdr, 0, sizeof(tmphdr));
2019                 tmphdr.magic = savehdr.magic;
2020                 tmphdr.forw = savehdr.forw;
2021                 tmphdr.back = savehdr.back;
2022                 tmphdr.firstused = state->blocksize;
2023
2024                 /* write the header to the temp buffer to initialise it */
2025                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2026
2027                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2028                                          drop_blk->bp, &drophdr)) {
2029                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2030                                                 tmp_leaf, &tmphdr, 0,
2031                                                 drophdr.count, mp);
2032                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2033                                                 tmp_leaf, &tmphdr, tmphdr.count,
2034                                                 savehdr.count, mp);
2035                 } else {
2036                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2037                                                 tmp_leaf, &tmphdr, 0,
2038                                                 savehdr.count, mp);
2039                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2040                                                 tmp_leaf, &tmphdr, tmphdr.count,
2041                                                 drophdr.count, mp);
2042                 }
2043                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2044                 savehdr = tmphdr; /* struct copy */
2045                 kmem_free(tmp_leaf);
2046         }
2047
2048         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2049         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2050                                            state->blocksize - 1);
2051
2052         /*
2053          * Copy out last hashval in each block for B-tree code.
2054          */
2055         entry = xfs_attr3_leaf_entryp(save_leaf);
2056         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2057 }
2058
2059 /*========================================================================
2060  * Routines used for finding things in the Btree.
2061  *========================================================================*/
2062
2063 /*
2064  * Look up a name in a leaf attribute list structure.
2065  * This is the internal routine, it uses the caller's buffer.
2066  *
2067  * Note that duplicate keys are allowed, but only check within the
2068  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2069  *
2070  * Return in args->index the index into the entry[] array of either
2071  * the found entry, or where the entry should have been (insert before
2072  * that entry).
2073  *
2074  * Don't change the args->value unless we find the attribute.
2075  */
2076 int
2077 xfs_attr3_leaf_lookup_int(
2078         struct xfs_buf          *bp,
2079         struct xfs_da_args      *args)
2080 {
2081         struct xfs_attr_leafblock *leaf;
2082         struct xfs_attr3_icleaf_hdr ichdr;
2083         struct xfs_attr_leaf_entry *entry;
2084         struct xfs_attr_leaf_entry *entries;
2085         struct xfs_attr_leaf_name_local *name_loc;
2086         struct xfs_attr_leaf_name_remote *name_rmt;
2087         xfs_dahash_t            hashval;
2088         int                     probe;
2089         int                     span;
2090
2091         trace_xfs_attr_leaf_lookup(args);
2092
2093         leaf = bp->b_addr;
2094         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2095         entries = xfs_attr3_leaf_entryp(leaf);
2096         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2097
2098         /*
2099          * Binary search.  (note: small blocks will skip this loop)
2100          */
2101         hashval = args->hashval;
2102         probe = span = ichdr.count / 2;
2103         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2104                 span /= 2;
2105                 if (be32_to_cpu(entry->hashval) < hashval)
2106                         probe += span;
2107                 else if (be32_to_cpu(entry->hashval) > hashval)
2108                         probe -= span;
2109                 else
2110                         break;
2111         }
2112         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2113         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2114
2115         /*
2116          * Since we may have duplicate hashval's, find the first matching
2117          * hashval in the leaf.
2118          */
2119         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2120                 entry--;
2121                 probe--;
2122         }
2123         while (probe < ichdr.count &&
2124                be32_to_cpu(entry->hashval) < hashval) {
2125                 entry++;
2126                 probe++;
2127         }
2128         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2129                 args->index = probe;
2130                 return XFS_ERROR(ENOATTR);
2131         }
2132
2133         /*
2134          * Duplicate keys may be present, so search all of them for a match.
2135          */
2136         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2137                         entry++, probe++) {
2138 /*
2139  * GROT: Add code to remove incomplete entries.
2140  */
2141                 /*
2142                  * If we are looking for INCOMPLETE entries, show only those.
2143                  * If we are looking for complete entries, show only those.
2144                  */
2145                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2146                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2147                         continue;
2148                 }
2149                 if (entry->flags & XFS_ATTR_LOCAL) {
2150                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2151                         if (name_loc->namelen != args->namelen)
2152                                 continue;
2153                         if (memcmp(args->name, name_loc->nameval,
2154                                                         args->namelen) != 0)
2155                                 continue;
2156                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2157                                 continue;
2158                         args->index = probe;
2159                         return XFS_ERROR(EEXIST);
2160                 } else {
2161                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2162                         if (name_rmt->namelen != args->namelen)
2163                                 continue;
2164                         if (memcmp(args->name, name_rmt->name,
2165                                                         args->namelen) != 0)
2166                                 continue;
2167                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2168                                 continue;
2169                         args->index = probe;
2170                         args->valuelen = be32_to_cpu(name_rmt->valuelen);
2171                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2172                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2173                                                         args->dp->i_mount,
2174                                                         args->valuelen);
2175                         return XFS_ERROR(EEXIST);
2176                 }
2177         }
2178         args->index = probe;
2179         return XFS_ERROR(ENOATTR);
2180 }
2181
2182 /*
2183  * Get the value associated with an attribute name from a leaf attribute
2184  * list structure.
2185  */
2186 int
2187 xfs_attr3_leaf_getvalue(
2188         struct xfs_buf          *bp,
2189         struct xfs_da_args      *args)
2190 {
2191         struct xfs_attr_leafblock *leaf;
2192         struct xfs_attr3_icleaf_hdr ichdr;
2193         struct xfs_attr_leaf_entry *entry;
2194         struct xfs_attr_leaf_name_local *name_loc;
2195         struct xfs_attr_leaf_name_remote *name_rmt;
2196         int                     valuelen;
2197
2198         leaf = bp->b_addr;
2199         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2200         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2201         ASSERT(args->index < ichdr.count);
2202
2203         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2204         if (entry->flags & XFS_ATTR_LOCAL) {
2205                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2206                 ASSERT(name_loc->namelen == args->namelen);
2207                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2208                 valuelen = be16_to_cpu(name_loc->valuelen);
2209                 if (args->flags & ATTR_KERNOVAL) {
2210                         args->valuelen = valuelen;
2211                         return 0;
2212                 }
2213                 if (args->valuelen < valuelen) {
2214                         args->valuelen = valuelen;
2215                         return XFS_ERROR(ERANGE);
2216                 }
2217                 args->valuelen = valuelen;
2218                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2219         } else {
2220                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2221                 ASSERT(name_rmt->namelen == args->namelen);
2222                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2223                 valuelen = be32_to_cpu(name_rmt->valuelen);
2224                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2225                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2226                                                        valuelen);
2227                 if (args->flags & ATTR_KERNOVAL) {
2228                         args->valuelen = valuelen;
2229                         return 0;
2230                 }
2231                 if (args->valuelen < valuelen) {
2232                         args->valuelen = valuelen;
2233                         return XFS_ERROR(ERANGE);
2234                 }
2235                 args->valuelen = valuelen;
2236         }
2237         return 0;
2238 }
2239
2240 /*========================================================================
2241  * Utility routines.
2242  *========================================================================*/
2243
2244 /*
2245  * Move the indicated entries from one leaf to another.
2246  * NOTE: this routine modifies both source and destination leaves.
2247  */
2248 /*ARGSUSED*/
2249 STATIC void
2250 xfs_attr3_leaf_moveents(
2251         struct xfs_attr_leafblock       *leaf_s,
2252         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2253         int                             start_s,
2254         struct xfs_attr_leafblock       *leaf_d,
2255         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2256         int                             start_d,
2257         int                             count,
2258         struct xfs_mount                *mp)
2259 {
2260         struct xfs_attr_leaf_entry      *entry_s;
2261         struct xfs_attr_leaf_entry      *entry_d;
2262         int                             desti;
2263         int                             tmp;
2264         int                             i;
2265
2266         /*
2267          * Check for nothing to do.
2268          */
2269         if (count == 0)
2270                 return;
2271
2272         /*
2273          * Set up environment.
2274          */
2275         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2276                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2277         ASSERT(ichdr_s->magic == ichdr_d->magic);
2278         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2279         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2280                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2281         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2282         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2283                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2284
2285         ASSERT(start_s < ichdr_s->count);
2286         ASSERT(start_d <= ichdr_d->count);
2287         ASSERT(count <= ichdr_s->count);
2288
2289
2290         /*
2291          * Move the entries in the destination leaf up to make a hole?
2292          */
2293         if (start_d < ichdr_d->count) {
2294                 tmp  = ichdr_d->count - start_d;
2295                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2296                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2297                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2298                 memmove(entry_d, entry_s, tmp);
2299         }
2300
2301         /*
2302          * Copy all entry's in the same (sorted) order,
2303          * but allocate attribute info packed and in sequence.
2304          */
2305         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2306         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2307         desti = start_d;
2308         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2309                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2310                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2311 #ifdef GROT
2312                 /*
2313                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2314                  * may also need to change the insertion index.  Code turned
2315                  * off for 6.2, should be revisited later.
2316                  */
2317                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2318                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2319                         ichdr_s->usedbytes -= tmp;
2320                         ichdr_s->count -= 1;
2321                         entry_d--;      /* to compensate for ++ in loop hdr */
2322                         desti--;
2323                         if ((start_s + i) < offset)
2324                                 result++;       /* insertion index adjustment */
2325                 } else {
2326 #endif /* GROT */
2327                         ichdr_d->firstused -= tmp;
2328                         /* both on-disk, don't endian flip twice */
2329                         entry_d->hashval = entry_s->hashval;
2330                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2331                         entry_d->flags = entry_s->flags;
2332                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2333                                                         <= XFS_LBSIZE(mp));
2334                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2335                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2336                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2337                                                         <= XFS_LBSIZE(mp));
2338                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2339                         ichdr_s->usedbytes -= tmp;
2340                         ichdr_d->usedbytes += tmp;
2341                         ichdr_s->count -= 1;
2342                         ichdr_d->count += 1;
2343                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2344                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2345                         ASSERT(ichdr_d->firstused >= tmp);
2346 #ifdef GROT
2347                 }
2348 #endif /* GROT */
2349         }
2350
2351         /*
2352          * Zero out the entries we just copied.
2353          */
2354         if (start_s == ichdr_s->count) {
2355                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2356                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2357                 ASSERT(((char *)entry_s + tmp) <=
2358                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2359                 memset(entry_s, 0, tmp);
2360         } else {
2361                 /*
2362                  * Move the remaining entries down to fill the hole,
2363                  * then zero the entries at the top.
2364                  */
2365                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2366                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2367                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2368                 memmove(entry_d, entry_s, tmp);
2369
2370                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2371                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2372                 ASSERT(((char *)entry_s + tmp) <=
2373                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2374                 memset(entry_s, 0, tmp);
2375         }
2376
2377         /*
2378          * Fill in the freemap information
2379          */
2380         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2381         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2382         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2383         ichdr_d->freemap[1].base = 0;
2384         ichdr_d->freemap[2].base = 0;
2385         ichdr_d->freemap[1].size = 0;
2386         ichdr_d->freemap[2].size = 0;
2387         ichdr_s->holes = 1;     /* leaf may not be compact */
2388 }
2389
2390 /*
2391  * Pick up the last hashvalue from a leaf block.
2392  */
2393 xfs_dahash_t
2394 xfs_attr_leaf_lasthash(
2395         struct xfs_buf  *bp,
2396         int             *count)
2397 {
2398         struct xfs_attr3_icleaf_hdr ichdr;
2399         struct xfs_attr_leaf_entry *entries;
2400
2401         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2402         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2403         if (count)
2404                 *count = ichdr.count;
2405         if (!ichdr.count)
2406                 return 0;
2407         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2408 }
2409
2410 /*
2411  * Calculate the number of bytes used to store the indicated attribute
2412  * (whether local or remote only calculate bytes in this block).
2413  */
2414 STATIC int
2415 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2416 {
2417         struct xfs_attr_leaf_entry *entries;
2418         xfs_attr_leaf_name_local_t *name_loc;
2419         xfs_attr_leaf_name_remote_t *name_rmt;
2420         int size;
2421
2422         entries = xfs_attr3_leaf_entryp(leaf);
2423         if (entries[index].flags & XFS_ATTR_LOCAL) {
2424                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2425                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2426                                                    be16_to_cpu(name_loc->valuelen));
2427         } else {
2428                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2429                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2430         }
2431         return size;
2432 }
2433
2434 /*
2435  * Calculate the number of bytes that would be required to store the new
2436  * attribute (whether local or remote only calculate bytes in this block).
2437  * This routine decides as a side effect whether the attribute will be
2438  * a "local" or a "remote" attribute.
2439  */
2440 int
2441 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2442 {
2443         int size;
2444
2445         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2446         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2447                 if (local) {
2448                         *local = 1;
2449                 }
2450         } else {
2451                 size = xfs_attr_leaf_entsize_remote(namelen);
2452                 if (local) {
2453                         *local = 0;
2454                 }
2455         }
2456         return size;
2457 }
2458
2459
2460 /*========================================================================
2461  * Manage the INCOMPLETE flag in a leaf entry
2462  *========================================================================*/
2463
2464 /*
2465  * Clear the INCOMPLETE flag on an entry in a leaf block.
2466  */
2467 int
2468 xfs_attr3_leaf_clearflag(
2469         struct xfs_da_args      *args)
2470 {
2471         struct xfs_attr_leafblock *leaf;
2472         struct xfs_attr_leaf_entry *entry;
2473         struct xfs_attr_leaf_name_remote *name_rmt;
2474         struct xfs_buf          *bp;
2475         int                     error;
2476 #ifdef DEBUG
2477         struct xfs_attr3_icleaf_hdr ichdr;
2478         xfs_attr_leaf_name_local_t *name_loc;
2479         int namelen;
2480         char *name;
2481 #endif /* DEBUG */
2482
2483         trace_xfs_attr_leaf_clearflag(args);
2484         /*
2485          * Set up the operation.
2486          */
2487         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2488         if (error)
2489                 return(error);
2490
2491         leaf = bp->b_addr;
2492         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2493         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2494
2495 #ifdef DEBUG
2496         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2497         ASSERT(args->index < ichdr.count);
2498         ASSERT(args->index >= 0);
2499
2500         if (entry->flags & XFS_ATTR_LOCAL) {
2501                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2502                 namelen = name_loc->namelen;
2503                 name = (char *)name_loc->nameval;
2504         } else {
2505                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2506                 namelen = name_rmt->namelen;
2507                 name = (char *)name_rmt->name;
2508         }
2509         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2510         ASSERT(namelen == args->namelen);
2511         ASSERT(memcmp(name, args->name, namelen) == 0);
2512 #endif /* DEBUG */
2513
2514         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2515         xfs_trans_log_buf(args->trans, bp,
2516                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2517
2518         if (args->rmtblkno) {
2519                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2520                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2521                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2522                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2523                 xfs_trans_log_buf(args->trans, bp,
2524                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2525         }
2526
2527         /*
2528          * Commit the flag value change and start the next trans in series.
2529          */
2530         return xfs_trans_roll(&args->trans, args->dp);
2531 }
2532
2533 /*
2534  * Set the INCOMPLETE flag on an entry in a leaf block.
2535  */
2536 int
2537 xfs_attr3_leaf_setflag(
2538         struct xfs_da_args      *args)
2539 {
2540         struct xfs_attr_leafblock *leaf;
2541         struct xfs_attr_leaf_entry *entry;
2542         struct xfs_attr_leaf_name_remote *name_rmt;
2543         struct xfs_buf          *bp;
2544         int error;
2545 #ifdef DEBUG
2546         struct xfs_attr3_icleaf_hdr ichdr;
2547 #endif
2548
2549         trace_xfs_attr_leaf_setflag(args);
2550
2551         /*
2552          * Set up the operation.
2553          */
2554         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2555         if (error)
2556                 return(error);
2557
2558         leaf = bp->b_addr;
2559 #ifdef DEBUG
2560         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2561         ASSERT(args->index < ichdr.count);
2562         ASSERT(args->index >= 0);
2563 #endif
2564         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2565
2566         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2567         entry->flags |= XFS_ATTR_INCOMPLETE;
2568         xfs_trans_log_buf(args->trans, bp,
2569                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2570         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2571                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2572                 name_rmt->valueblk = 0;
2573                 name_rmt->valuelen = 0;
2574                 xfs_trans_log_buf(args->trans, bp,
2575                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2576         }
2577
2578         /*
2579          * Commit the flag value change and start the next trans in series.
2580          */
2581         return xfs_trans_roll(&args->trans, args->dp);
2582 }
2583
2584 /*
2585  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2586  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2587  * entry given by args->blkno2/index2.
2588  *
2589  * Note that they could be in different blocks, or in the same block.
2590  */
2591 int
2592 xfs_attr3_leaf_flipflags(
2593         struct xfs_da_args      *args)
2594 {
2595         struct xfs_attr_leafblock *leaf1;
2596         struct xfs_attr_leafblock *leaf2;
2597         struct xfs_attr_leaf_entry *entry1;
2598         struct xfs_attr_leaf_entry *entry2;
2599         struct xfs_attr_leaf_name_remote *name_rmt;
2600         struct xfs_buf          *bp1;
2601         struct xfs_buf          *bp2;
2602         int error;
2603 #ifdef DEBUG
2604         struct xfs_attr3_icleaf_hdr ichdr1;
2605         struct xfs_attr3_icleaf_hdr ichdr2;
2606         xfs_attr_leaf_name_local_t *name_loc;
2607         int namelen1, namelen2;
2608         char *name1, *name2;
2609 #endif /* DEBUG */
2610
2611         trace_xfs_attr_leaf_flipflags(args);
2612
2613         /*
2614          * Read the block containing the "old" attr
2615          */
2616         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2617         if (error)
2618                 return error;
2619
2620         /*
2621          * Read the block containing the "new" attr, if it is different
2622          */
2623         if (args->blkno2 != args->blkno) {
2624                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2625                                            -1, &bp2);
2626                 if (error)
2627                         return error;
2628         } else {
2629                 bp2 = bp1;
2630         }
2631
2632         leaf1 = bp1->b_addr;
2633         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2634
2635         leaf2 = bp2->b_addr;
2636         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2637
2638 #ifdef DEBUG
2639         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2640         ASSERT(args->index < ichdr1.count);
2641         ASSERT(args->index >= 0);
2642
2643         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2644         ASSERT(args->index2 < ichdr2.count);
2645         ASSERT(args->index2 >= 0);
2646
2647         if (entry1->flags & XFS_ATTR_LOCAL) {
2648                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2649                 namelen1 = name_loc->namelen;
2650                 name1 = (char *)name_loc->nameval;
2651         } else {
2652                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2653                 namelen1 = name_rmt->namelen;
2654                 name1 = (char *)name_rmt->name;
2655         }
2656         if (entry2->flags & XFS_ATTR_LOCAL) {
2657                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2658                 namelen2 = name_loc->namelen;
2659                 name2 = (char *)name_loc->nameval;
2660         } else {
2661                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2662                 namelen2 = name_rmt->namelen;
2663                 name2 = (char *)name_rmt->name;
2664         }
2665         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2666         ASSERT(namelen1 == namelen2);
2667         ASSERT(memcmp(name1, name2, namelen1) == 0);
2668 #endif /* DEBUG */
2669
2670         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2671         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2672
2673         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2674         xfs_trans_log_buf(args->trans, bp1,
2675                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2676         if (args->rmtblkno) {
2677                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2678                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2679                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2680                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2681                 xfs_trans_log_buf(args->trans, bp1,
2682                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2683         }
2684
2685         entry2->flags |= XFS_ATTR_INCOMPLETE;
2686         xfs_trans_log_buf(args->trans, bp2,
2687                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2688         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2689                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2690                 name_rmt->valueblk = 0;
2691                 name_rmt->valuelen = 0;
2692                 xfs_trans_log_buf(args->trans, bp2,
2693                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2694         }
2695
2696         /*
2697          * Commit the flag value change and start the next trans in series.
2698          */
2699         error = xfs_trans_roll(&args->trans, args->dp);
2700
2701         return error;
2702 }