Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
[firefly-linux-kernel-4.4.55.git] / fs / xfs / libxfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_attr_remote.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_cksum.h"
44 #include "xfs_dinode.h"
45 #include "xfs_dir2.h"
46
47
48 /*
49  * xfs_attr_leaf.c
50  *
51  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52  */
53
54 /*========================================================================
55  * Function prototypes for the kernel.
56  *========================================================================*/
57
58 /*
59  * Routines used for growing the Btree.
60  */
61 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
62                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
63 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
64                                    struct xfs_attr3_icleaf_hdr *ichdr,
65                                    struct xfs_da_args *args, int freemap_index);
66 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
67                                    struct xfs_attr3_icleaf_hdr *ichdr,
68                                    struct xfs_buf *leaf_buffer);
69 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
70                                                    xfs_da_state_blk_t *blk1,
71                                                    xfs_da_state_blk_t *blk2);
72 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
73                         xfs_da_state_blk_t *leaf_blk_1,
74                         struct xfs_attr3_icleaf_hdr *ichdr1,
75                         xfs_da_state_blk_t *leaf_blk_2,
76                         struct xfs_attr3_icleaf_hdr *ichdr2,
77                         int *number_entries_in_blk1,
78                         int *number_usedbytes_in_blk1);
79
80 /*
81  * Utility routines.
82  */
83 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
84                         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);
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_buf_ioerror(bp, -EFSCORRUPTED);
218                 xfs_verifier_error(bp);
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_buf_update_cksum(bp, 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_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
245                 xfs_buf_ioerror(bp, -EFSBADCRC);
246         else if (!xfs_attr3_leaf_verify(bp))
247                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
248
249         if (bp->b_error)
250                 xfs_verifier_error(bp);
251 }
252
253 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
254         .verify_read = xfs_attr3_leaf_read_verify,
255         .verify_write = xfs_attr3_leaf_write_verify,
256 };
257
258 int
259 xfs_attr3_leaf_read(
260         struct xfs_trans        *tp,
261         struct xfs_inode        *dp,
262         xfs_dablk_t             bno,
263         xfs_daddr_t             mappedbno,
264         struct xfs_buf          **bpp)
265 {
266         int                     err;
267
268         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
269                                 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
270         if (!err && tp)
271                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
272         return err;
273 }
274
275 /*========================================================================
276  * Namespace helper routines
277  *========================================================================*/
278
279 /*
280  * If namespace bits don't match return 0.
281  * If all match then return 1.
282  */
283 STATIC int
284 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
285 {
286         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
287 }
288
289
290 /*========================================================================
291  * External routines when attribute fork size < XFS_LITINO(mp).
292  *========================================================================*/
293
294 /*
295  * Query whether the requested number of additional bytes of extended
296  * attribute space will be able to fit inline.
297  *
298  * Returns zero if not, else the di_forkoff fork offset to be used in the
299  * literal area for attribute data once the new bytes have been added.
300  *
301  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
302  * special case for dev/uuid inodes, they have fixed size data forks.
303  */
304 int
305 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
306 {
307         int offset;
308         int minforkoff; /* lower limit on valid forkoff locations */
309         int maxforkoff; /* upper limit on valid forkoff locations */
310         int dsize;
311         xfs_mount_t *mp = dp->i_mount;
312
313         /* rounded down */
314         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
315
316         switch (dp->i_d.di_format) {
317         case XFS_DINODE_FMT_DEV:
318                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
319                 return (offset >= minforkoff) ? minforkoff : 0;
320         case XFS_DINODE_FMT_UUID:
321                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
322                 return (offset >= minforkoff) ? minforkoff : 0;
323         }
324
325         /*
326          * If the requested numbers of bytes is smaller or equal to the
327          * current attribute fork size we can always proceed.
328          *
329          * Note that if_bytes in the data fork might actually be larger than
330          * the current data fork size is due to delalloc extents. In that
331          * case either the extent count will go down when they are converted
332          * to real extents, or the delalloc conversion will take care of the
333          * literal area rebalancing.
334          */
335         if (bytes <= XFS_IFORK_ASIZE(dp))
336                 return dp->i_d.di_forkoff;
337
338         /*
339          * For attr2 we can try to move the forkoff if there is space in the
340          * literal area, but for the old format we are done if there is no
341          * space in the fixed attribute fork.
342          */
343         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
344                 return 0;
345
346         dsize = dp->i_df.if_bytes;
347
348         switch (dp->i_d.di_format) {
349         case XFS_DINODE_FMT_EXTENTS:
350                 /*
351                  * If there is no attr fork and the data fork is extents, 
352                  * determine if creating the default attr fork will result
353                  * in the extents form migrating to btree. If so, the
354                  * minimum offset only needs to be the space required for
355                  * the btree root.
356                  */
357                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
358                     xfs_default_attroffset(dp))
359                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
360                 break;
361         case XFS_DINODE_FMT_BTREE:
362                 /*
363                  * If we have a data btree then keep forkoff if we have one,
364                  * otherwise we are adding a new attr, so then we set
365                  * minforkoff to where the btree root can finish so we have
366                  * plenty of room for attrs
367                  */
368                 if (dp->i_d.di_forkoff) {
369                         if (offset < dp->i_d.di_forkoff)
370                                 return 0;
371                         return dp->i_d.di_forkoff;
372                 }
373                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
374                 break;
375         }
376
377         /*
378          * A data fork btree root must have space for at least
379          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
380          */
381         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
382         minforkoff = roundup(minforkoff, 8) >> 3;
383
384         /* attr fork btree root can have at least this many key/ptr pairs */
385         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
386                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
387         maxforkoff = maxforkoff >> 3;   /* rounded down */
388
389         if (offset >= maxforkoff)
390                 return maxforkoff;
391         if (offset >= minforkoff)
392                 return offset;
393         return 0;
394 }
395
396 /*
397  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
398  */
399 STATIC void
400 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
401 {
402         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
403             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
404                 spin_lock(&mp->m_sb_lock);
405                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
406                         xfs_sb_version_addattr2(&mp->m_sb);
407                         spin_unlock(&mp->m_sb_lock);
408                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
409                 } else
410                         spin_unlock(&mp->m_sb_lock);
411         }
412 }
413
414 /*
415  * Create the initial contents of a shortform attribute list.
416  */
417 void
418 xfs_attr_shortform_create(xfs_da_args_t *args)
419 {
420         xfs_attr_sf_hdr_t *hdr;
421         xfs_inode_t *dp;
422         xfs_ifork_t *ifp;
423
424         trace_xfs_attr_sf_create(args);
425
426         dp = args->dp;
427         ASSERT(dp != NULL);
428         ifp = dp->i_afp;
429         ASSERT(ifp != NULL);
430         ASSERT(ifp->if_bytes == 0);
431         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
432                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
433                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
434                 ifp->if_flags |= XFS_IFINLINE;
435         } else {
436                 ASSERT(ifp->if_flags & XFS_IFINLINE);
437         }
438         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
439         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
440         hdr->count = 0;
441         hdr->totsize = cpu_to_be16(sizeof(*hdr));
442         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
443 }
444
445 /*
446  * Add a name/value pair to the shortform attribute list.
447  * Overflow from the inode has already been checked for.
448  */
449 void
450 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
451 {
452         xfs_attr_shortform_t *sf;
453         xfs_attr_sf_entry_t *sfe;
454         int i, offset, size;
455         xfs_mount_t *mp;
456         xfs_inode_t *dp;
457         xfs_ifork_t *ifp;
458
459         trace_xfs_attr_sf_add(args);
460
461         dp = args->dp;
462         mp = dp->i_mount;
463         dp->i_d.di_forkoff = forkoff;
464
465         ifp = dp->i_afp;
466         ASSERT(ifp->if_flags & XFS_IFINLINE);
467         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
468         sfe = &sf->list[0];
469         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
470 #ifdef DEBUG
471                 if (sfe->namelen != args->namelen)
472                         continue;
473                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
474                         continue;
475                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
476                         continue;
477                 ASSERT(0);
478 #endif
479         }
480
481         offset = (char *)sfe - (char *)sf;
482         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
483         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
484         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
485         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
486
487         sfe->namelen = args->namelen;
488         sfe->valuelen = args->valuelen;
489         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
490         memcpy(sfe->nameval, args->name, args->namelen);
491         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
492         sf->hdr.count++;
493         be16_add_cpu(&sf->hdr.totsize, size);
494         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
495
496         xfs_sbversion_add_attr2(mp, args->trans);
497 }
498
499 /*
500  * After the last attribute is removed revert to original inode format,
501  * making all literal area available to the data fork once more.
502  */
503 STATIC void
504 xfs_attr_fork_reset(
505         struct xfs_inode        *ip,
506         struct xfs_trans        *tp)
507 {
508         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
509         ip->i_d.di_forkoff = 0;
510         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
511
512         ASSERT(ip->i_d.di_anextents == 0);
513         ASSERT(ip->i_afp == NULL);
514
515         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
516 }
517
518 /*
519  * Remove an attribute from the shortform attribute list structure.
520  */
521 int
522 xfs_attr_shortform_remove(xfs_da_args_t *args)
523 {
524         xfs_attr_shortform_t *sf;
525         xfs_attr_sf_entry_t *sfe;
526         int base, size=0, end, totsize, i;
527         xfs_mount_t *mp;
528         xfs_inode_t *dp;
529
530         trace_xfs_attr_sf_remove(args);
531
532         dp = args->dp;
533         mp = dp->i_mount;
534         base = sizeof(xfs_attr_sf_hdr_t);
535         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
536         sfe = &sf->list[0];
537         end = sf->hdr.count;
538         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
539                                         base += size, i++) {
540                 size = XFS_ATTR_SF_ENTSIZE(sfe);
541                 if (sfe->namelen != args->namelen)
542                         continue;
543                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
544                         continue;
545                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
546                         continue;
547                 break;
548         }
549         if (i == end)
550                 return -ENOATTR;
551
552         /*
553          * Fix up the attribute fork data, covering the hole
554          */
555         end = base + size;
556         totsize = be16_to_cpu(sf->hdr.totsize);
557         if (end != totsize)
558                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
559         sf->hdr.count--;
560         be16_add_cpu(&sf->hdr.totsize, -size);
561
562         /*
563          * Fix up the start offset of the attribute fork
564          */
565         totsize -= size;
566         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
567             (mp->m_flags & XFS_MOUNT_ATTR2) &&
568             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
569             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
570                 xfs_attr_fork_reset(dp, args->trans);
571         } else {
572                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
573                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
574                 ASSERT(dp->i_d.di_forkoff);
575                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
576                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
577                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
578                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
579                 xfs_trans_log_inode(args->trans, dp,
580                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
581         }
582
583         xfs_sbversion_add_attr2(mp, args->trans);
584
585         return 0;
586 }
587
588 /*
589  * Look up a name in a shortform attribute list structure.
590  */
591 /*ARGSUSED*/
592 int
593 xfs_attr_shortform_lookup(xfs_da_args_t *args)
594 {
595         xfs_attr_shortform_t *sf;
596         xfs_attr_sf_entry_t *sfe;
597         int i;
598         xfs_ifork_t *ifp;
599
600         trace_xfs_attr_sf_lookup(args);
601
602         ifp = args->dp->i_afp;
603         ASSERT(ifp->if_flags & XFS_IFINLINE);
604         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
605         sfe = &sf->list[0];
606         for (i = 0; i < sf->hdr.count;
607                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
608                 if (sfe->namelen != args->namelen)
609                         continue;
610                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
611                         continue;
612                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
613                         continue;
614                 return -EEXIST;
615         }
616         return -ENOATTR;
617 }
618
619 /*
620  * Look up a name in a shortform attribute list structure.
621  */
622 /*ARGSUSED*/
623 int
624 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
625 {
626         xfs_attr_shortform_t *sf;
627         xfs_attr_sf_entry_t *sfe;
628         int i;
629
630         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
631         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
632         sfe = &sf->list[0];
633         for (i = 0; i < sf->hdr.count;
634                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
635                 if (sfe->namelen != args->namelen)
636                         continue;
637                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
638                         continue;
639                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
640                         continue;
641                 if (args->flags & ATTR_KERNOVAL) {
642                         args->valuelen = sfe->valuelen;
643                         return -EEXIST;
644                 }
645                 if (args->valuelen < sfe->valuelen) {
646                         args->valuelen = sfe->valuelen;
647                         return -ERANGE;
648                 }
649                 args->valuelen = sfe->valuelen;
650                 memcpy(args->value, &sfe->nameval[args->namelen],
651                                                     args->valuelen);
652                 return -EEXIST;
653         }
654         return -ENOATTR;
655 }
656
657 /*
658  * Convert from using the shortform to the leaf.
659  */
660 int
661 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
662 {
663         xfs_inode_t *dp;
664         xfs_attr_shortform_t *sf;
665         xfs_attr_sf_entry_t *sfe;
666         xfs_da_args_t nargs;
667         char *tmpbuffer;
668         int error, i, size;
669         xfs_dablk_t blkno;
670         struct xfs_buf *bp;
671         xfs_ifork_t *ifp;
672
673         trace_xfs_attr_sf_to_leaf(args);
674
675         dp = args->dp;
676         ifp = dp->i_afp;
677         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
678         size = be16_to_cpu(sf->hdr.totsize);
679         tmpbuffer = kmem_alloc(size, KM_SLEEP);
680         ASSERT(tmpbuffer != NULL);
681         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
682         sf = (xfs_attr_shortform_t *)tmpbuffer;
683
684         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
685         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
686
687         bp = NULL;
688         error = xfs_da_grow_inode(args, &blkno);
689         if (error) {
690                 /*
691                  * If we hit an IO error middle of the transaction inside
692                  * grow_inode(), we may have inconsistent data. Bail out.
693                  */
694                 if (error == -EIO)
695                         goto out;
696                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
697                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
698                 goto out;
699         }
700
701         ASSERT(blkno == 0);
702         error = xfs_attr3_leaf_create(args, blkno, &bp);
703         if (error) {
704                 error = xfs_da_shrink_inode(args, 0, bp);
705                 bp = NULL;
706                 if (error)
707                         goto out;
708                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
709                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
710                 goto out;
711         }
712
713         memset((char *)&nargs, 0, sizeof(nargs));
714         nargs.dp = dp;
715         nargs.geo = args->geo;
716         nargs.firstblock = args->firstblock;
717         nargs.flist = args->flist;
718         nargs.total = args->total;
719         nargs.whichfork = XFS_ATTR_FORK;
720         nargs.trans = args->trans;
721         nargs.op_flags = XFS_DA_OP_OKNOENT;
722
723         sfe = &sf->list[0];
724         for (i = 0; i < sf->hdr.count; i++) {
725                 nargs.name = sfe->nameval;
726                 nargs.namelen = sfe->namelen;
727                 nargs.value = &sfe->nameval[nargs.namelen];
728                 nargs.valuelen = sfe->valuelen;
729                 nargs.hashval = xfs_da_hashname(sfe->nameval,
730                                                 sfe->namelen);
731                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
732                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
733                 ASSERT(error == -ENOATTR);
734                 error = xfs_attr3_leaf_add(bp, &nargs);
735                 ASSERT(error != -ENOSPC);
736                 if (error)
737                         goto out;
738                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
739         }
740         error = 0;
741
742 out:
743         kmem_free(tmpbuffer);
744         return error;
745 }
746
747 /*
748  * Check a leaf attribute block to see if all the entries would fit into
749  * a shortform attribute list.
750  */
751 int
752 xfs_attr_shortform_allfit(
753         struct xfs_buf          *bp,
754         struct xfs_inode        *dp)
755 {
756         struct xfs_attr_leafblock *leaf;
757         struct xfs_attr_leaf_entry *entry;
758         xfs_attr_leaf_name_local_t *name_loc;
759         struct xfs_attr3_icleaf_hdr leafhdr;
760         int                     bytes;
761         int                     i;
762
763         leaf = bp->b_addr;
764         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
765         entry = xfs_attr3_leaf_entryp(leaf);
766
767         bytes = sizeof(struct xfs_attr_sf_hdr);
768         for (i = 0; i < leafhdr.count; entry++, i++) {
769                 if (entry->flags & XFS_ATTR_INCOMPLETE)
770                         continue;               /* don't copy partial entries */
771                 if (!(entry->flags & XFS_ATTR_LOCAL))
772                         return 0;
773                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
774                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
775                         return 0;
776                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
777                         return 0;
778                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
779                                 + name_loc->namelen
780                                 + be16_to_cpu(name_loc->valuelen);
781         }
782         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
783             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
784             (bytes == sizeof(struct xfs_attr_sf_hdr)))
785                 return -1;
786         return xfs_attr_shortform_bytesfit(dp, bytes);
787 }
788
789 /*
790  * Convert a leaf attribute list to shortform attribute list
791  */
792 int
793 xfs_attr3_leaf_to_shortform(
794         struct xfs_buf          *bp,
795         struct xfs_da_args      *args,
796         int                     forkoff)
797 {
798         struct xfs_attr_leafblock *leaf;
799         struct xfs_attr3_icleaf_hdr ichdr;
800         struct xfs_attr_leaf_entry *entry;
801         struct xfs_attr_leaf_name_local *name_loc;
802         struct xfs_da_args      nargs;
803         struct xfs_inode        *dp = args->dp;
804         char                    *tmpbuffer;
805         int                     error;
806         int                     i;
807
808         trace_xfs_attr_leaf_to_sf(args);
809
810         tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
811         if (!tmpbuffer)
812                 return -ENOMEM;
813
814         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
815
816         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
817         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
818         entry = xfs_attr3_leaf_entryp(leaf);
819
820         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
821         memset(bp->b_addr, 0, args->geo->blksize);
822
823         /*
824          * Clean out the prior contents of the attribute list.
825          */
826         error = xfs_da_shrink_inode(args, 0, bp);
827         if (error)
828                 goto out;
829
830         if (forkoff == -1) {
831                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
832                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
833                 xfs_attr_fork_reset(dp, args->trans);
834                 goto out;
835         }
836
837         xfs_attr_shortform_create(args);
838
839         /*
840          * Copy the attributes
841          */
842         memset((char *)&nargs, 0, sizeof(nargs));
843         nargs.geo = args->geo;
844         nargs.dp = dp;
845         nargs.firstblock = args->firstblock;
846         nargs.flist = args->flist;
847         nargs.total = args->total;
848         nargs.whichfork = XFS_ATTR_FORK;
849         nargs.trans = args->trans;
850         nargs.op_flags = XFS_DA_OP_OKNOENT;
851
852         for (i = 0; i < ichdr.count; entry++, i++) {
853                 if (entry->flags & XFS_ATTR_INCOMPLETE)
854                         continue;       /* don't copy partial entries */
855                 if (!entry->nameidx)
856                         continue;
857                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
858                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
859                 nargs.name = name_loc->nameval;
860                 nargs.namelen = name_loc->namelen;
861                 nargs.value = &name_loc->nameval[nargs.namelen];
862                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
863                 nargs.hashval = be32_to_cpu(entry->hashval);
864                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
865                 xfs_attr_shortform_add(&nargs, forkoff);
866         }
867         error = 0;
868
869 out:
870         kmem_free(tmpbuffer);
871         return error;
872 }
873
874 /*
875  * Convert from using a single leaf to a root node and a leaf.
876  */
877 int
878 xfs_attr3_leaf_to_node(
879         struct xfs_da_args      *args)
880 {
881         struct xfs_attr_leafblock *leaf;
882         struct xfs_attr3_icleaf_hdr icleafhdr;
883         struct xfs_attr_leaf_entry *entries;
884         struct xfs_da_node_entry *btree;
885         struct xfs_da3_icnode_hdr icnodehdr;
886         struct xfs_da_intnode   *node;
887         struct xfs_inode        *dp = args->dp;
888         struct xfs_mount        *mp = dp->i_mount;
889         struct xfs_buf          *bp1 = NULL;
890         struct xfs_buf          *bp2 = NULL;
891         xfs_dablk_t             blkno;
892         int                     error;
893
894         trace_xfs_attr_leaf_to_node(args);
895
896         error = xfs_da_grow_inode(args, &blkno);
897         if (error)
898                 goto out;
899         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
900         if (error)
901                 goto out;
902
903         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
904         if (error)
905                 goto out;
906
907         /* copy leaf to new buffer, update identifiers */
908         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
909         bp2->b_ops = bp1->b_ops;
910         memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
911         if (xfs_sb_version_hascrc(&mp->m_sb)) {
912                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
913                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
914         }
915         xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
916
917         /*
918          * Set up the new root node.
919          */
920         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
921         if (error)
922                 goto out;
923         node = bp1->b_addr;
924         dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
925         btree = dp->d_ops->node_tree_p(node);
926
927         leaf = bp2->b_addr;
928         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
929         entries = xfs_attr3_leaf_entryp(leaf);
930
931         /* both on-disk, don't endian-flip twice */
932         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
933         btree[0].before = cpu_to_be32(blkno);
934         icnodehdr.count = 1;
935         dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
936         xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
937         error = 0;
938 out:
939         return error;
940 }
941
942 /*========================================================================
943  * Routines used for growing the Btree.
944  *========================================================================*/
945
946 /*
947  * Create the initial contents of a leaf attribute list
948  * or a leaf in a node attribute list.
949  */
950 STATIC int
951 xfs_attr3_leaf_create(
952         struct xfs_da_args      *args,
953         xfs_dablk_t             blkno,
954         struct xfs_buf          **bpp)
955 {
956         struct xfs_attr_leafblock *leaf;
957         struct xfs_attr3_icleaf_hdr ichdr;
958         struct xfs_inode        *dp = args->dp;
959         struct xfs_mount        *mp = dp->i_mount;
960         struct xfs_buf          *bp;
961         int                     error;
962
963         trace_xfs_attr_leaf_create(args);
964
965         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
966                                             XFS_ATTR_FORK);
967         if (error)
968                 return error;
969         bp->b_ops = &xfs_attr3_leaf_buf_ops;
970         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
971         leaf = bp->b_addr;
972         memset(leaf, 0, args->geo->blksize);
973
974         memset(&ichdr, 0, sizeof(ichdr));
975         ichdr.firstused = args->geo->blksize;
976
977         if (xfs_sb_version_hascrc(&mp->m_sb)) {
978                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
979
980                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
981
982                 hdr3->blkno = cpu_to_be64(bp->b_bn);
983                 hdr3->owner = cpu_to_be64(dp->i_ino);
984                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
985
986                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
987         } else {
988                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
989                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
990         }
991         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
992
993         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
994         xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
995
996         *bpp = bp;
997         return 0;
998 }
999
1000 /*
1001  * Split the leaf node, rebalance, then add the new entry.
1002  */
1003 int
1004 xfs_attr3_leaf_split(
1005         struct xfs_da_state     *state,
1006         struct xfs_da_state_blk *oldblk,
1007         struct xfs_da_state_blk *newblk)
1008 {
1009         xfs_dablk_t blkno;
1010         int error;
1011
1012         trace_xfs_attr_leaf_split(state->args);
1013
1014         /*
1015          * Allocate space for a new leaf node.
1016          */
1017         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1018         error = xfs_da_grow_inode(state->args, &blkno);
1019         if (error)
1020                 return error;
1021         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1022         if (error)
1023                 return error;
1024         newblk->blkno = blkno;
1025         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1026
1027         /*
1028          * Rebalance the entries across the two leaves.
1029          * NOTE: rebalance() currently depends on the 2nd block being empty.
1030          */
1031         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1032         error = xfs_da3_blk_link(state, oldblk, newblk);
1033         if (error)
1034                 return error;
1035
1036         /*
1037          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1038          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1039          * "new" attrs info.  Will need the "old" info to remove it later.
1040          *
1041          * Insert the "new" entry in the correct block.
1042          */
1043         if (state->inleaf) {
1044                 trace_xfs_attr_leaf_add_old(state->args);
1045                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1046         } else {
1047                 trace_xfs_attr_leaf_add_new(state->args);
1048                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1049         }
1050
1051         /*
1052          * Update last hashval in each block since we added the name.
1053          */
1054         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1055         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1056         return error;
1057 }
1058
1059 /*
1060  * Add a name to the leaf attribute list structure.
1061  */
1062 int
1063 xfs_attr3_leaf_add(
1064         struct xfs_buf          *bp,
1065         struct xfs_da_args      *args)
1066 {
1067         struct xfs_attr_leafblock *leaf;
1068         struct xfs_attr3_icleaf_hdr ichdr;
1069         int                     tablesize;
1070         int                     entsize;
1071         int                     sum;
1072         int                     tmp;
1073         int                     i;
1074
1075         trace_xfs_attr_leaf_add(args);
1076
1077         leaf = bp->b_addr;
1078         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1079         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1080         entsize = xfs_attr_leaf_newentsize(args, NULL);
1081
1082         /*
1083          * Search through freemap for first-fit on new name length.
1084          * (may need to figure in size of entry struct too)
1085          */
1086         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1087                                         + xfs_attr3_leaf_hdr_size(leaf);
1088         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1089                 if (tablesize > ichdr.firstused) {
1090                         sum += ichdr.freemap[i].size;
1091                         continue;
1092                 }
1093                 if (!ichdr.freemap[i].size)
1094                         continue;       /* no space in this map */
1095                 tmp = entsize;
1096                 if (ichdr.freemap[i].base < ichdr.firstused)
1097                         tmp += sizeof(xfs_attr_leaf_entry_t);
1098                 if (ichdr.freemap[i].size >= tmp) {
1099                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1100                         goto out_log_hdr;
1101                 }
1102                 sum += ichdr.freemap[i].size;
1103         }
1104
1105         /*
1106          * If there are no holes in the address space of the block,
1107          * and we don't have enough freespace, then compaction will do us
1108          * no good and we should just give up.
1109          */
1110         if (!ichdr.holes && sum < entsize)
1111                 return -ENOSPC;
1112
1113         /*
1114          * Compact the entries to coalesce free space.
1115          * This may change the hdr->count via dropping INCOMPLETE entries.
1116          */
1117         xfs_attr3_leaf_compact(args, &ichdr, bp);
1118
1119         /*
1120          * After compaction, the block is guaranteed to have only one
1121          * free region, in freemap[0].  If it is not big enough, give up.
1122          */
1123         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1124                 tmp = -ENOSPC;
1125                 goto out_log_hdr;
1126         }
1127
1128         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1129
1130 out_log_hdr:
1131         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1132         xfs_trans_log_buf(args->trans, bp,
1133                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1134                                 xfs_attr3_leaf_hdr_size(leaf)));
1135         return tmp;
1136 }
1137
1138 /*
1139  * Add a name to a leaf attribute list structure.
1140  */
1141 STATIC int
1142 xfs_attr3_leaf_add_work(
1143         struct xfs_buf          *bp,
1144         struct xfs_attr3_icleaf_hdr *ichdr,
1145         struct xfs_da_args      *args,
1146         int                     mapindex)
1147 {
1148         struct xfs_attr_leafblock *leaf;
1149         struct xfs_attr_leaf_entry *entry;
1150         struct xfs_attr_leaf_name_local *name_loc;
1151         struct xfs_attr_leaf_name_remote *name_rmt;
1152         struct xfs_mount        *mp;
1153         int                     tmp;
1154         int                     i;
1155
1156         trace_xfs_attr_leaf_add_work(args);
1157
1158         leaf = bp->b_addr;
1159         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1160         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1161
1162         /*
1163          * Force open some space in the entry array and fill it in.
1164          */
1165         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1166         if (args->index < ichdr->count) {
1167                 tmp  = ichdr->count - args->index;
1168                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1169                 memmove(entry + 1, entry, tmp);
1170                 xfs_trans_log_buf(args->trans, bp,
1171                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1172         }
1173         ichdr->count++;
1174
1175         /*
1176          * Allocate space for the new string (at the end of the run).
1177          */
1178         mp = args->trans->t_mountp;
1179         ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1180         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1181         ASSERT(ichdr->freemap[mapindex].size >=
1182                 xfs_attr_leaf_newentsize(args, NULL));
1183         ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1184         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1185
1186         ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1187
1188         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1189                                      ichdr->freemap[mapindex].size);
1190         entry->hashval = cpu_to_be32(args->hashval);
1191         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1192         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1193         if (args->op_flags & XFS_DA_OP_RENAME) {
1194                 entry->flags |= XFS_ATTR_INCOMPLETE;
1195                 if ((args->blkno2 == args->blkno) &&
1196                     (args->index2 <= args->index)) {
1197                         args->index2++;
1198                 }
1199         }
1200         xfs_trans_log_buf(args->trans, bp,
1201                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1202         ASSERT((args->index == 0) ||
1203                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1204         ASSERT((args->index == ichdr->count - 1) ||
1205                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1206
1207         /*
1208          * For "remote" attribute values, simply note that we need to
1209          * allocate space for the "remote" value.  We can't actually
1210          * allocate the extents in this transaction, and we can't decide
1211          * which blocks they should be as we might allocate more blocks
1212          * as part of this transaction (a split operation for example).
1213          */
1214         if (entry->flags & XFS_ATTR_LOCAL) {
1215                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1216                 name_loc->namelen = args->namelen;
1217                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1218                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1219                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1220                                    be16_to_cpu(name_loc->valuelen));
1221         } else {
1222                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1223                 name_rmt->namelen = args->namelen;
1224                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1225                 entry->flags |= XFS_ATTR_INCOMPLETE;
1226                 /* just in case */
1227                 name_rmt->valuelen = 0;
1228                 name_rmt->valueblk = 0;
1229                 args->rmtblkno = 1;
1230                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1231                 args->rmtvaluelen = 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         char                    *tmpbuffer;
1272
1273         trace_xfs_attr_leaf_compact(args);
1274
1275         tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1276         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1277         memset(bp->b_addr, 0, args->geo->blksize);
1278         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1279         leaf_dst = bp->b_addr;
1280
1281         /*
1282          * Copy the on-disk header back into the destination buffer to ensure
1283          * all the information in the header that is not part of the incore
1284          * header structure is preserved.
1285          */
1286         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1287
1288         /* Initialise the incore headers */
1289         ichdr_src = *ichdr_dst; /* struct copy */
1290         ichdr_dst->firstused = args->geo->blksize;
1291         ichdr_dst->usedbytes = 0;
1292         ichdr_dst->count = 0;
1293         ichdr_dst->holes = 0;
1294         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1295         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1296                                                 ichdr_dst->freemap[0].base;
1297
1298         /* write the header back to initialise the underlying buffer */
1299         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1300
1301         /*
1302          * Copy all entry's in the same (sorted) order,
1303          * but allocate name/value pairs packed and in sequence.
1304          */
1305         xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1306                                 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1307         /*
1308          * this logs the entire buffer, but the caller must write the header
1309          * back to the buffer when it is finished modifying it.
1310          */
1311         xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1312
1313         kmem_free(tmpbuffer);
1314 }
1315
1316 /*
1317  * Compare two leaf blocks "order".
1318  * Return 0 unless leaf2 should go before leaf1.
1319  */
1320 static int
1321 xfs_attr3_leaf_order(
1322         struct xfs_buf  *leaf1_bp,
1323         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1324         struct xfs_buf  *leaf2_bp,
1325         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1326 {
1327         struct xfs_attr_leaf_entry *entries1;
1328         struct xfs_attr_leaf_entry *entries2;
1329
1330         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1331         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1332         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1333             ((be32_to_cpu(entries2[0].hashval) <
1334               be32_to_cpu(entries1[0].hashval)) ||
1335              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1336               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1337                 return 1;
1338         }
1339         return 0;
1340 }
1341
1342 int
1343 xfs_attr_leaf_order(
1344         struct xfs_buf  *leaf1_bp,
1345         struct xfs_buf  *leaf2_bp)
1346 {
1347         struct xfs_attr3_icleaf_hdr ichdr1;
1348         struct xfs_attr3_icleaf_hdr ichdr2;
1349
1350         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1351         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1352         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1353 }
1354
1355 /*
1356  * Redistribute the attribute list entries between two leaf nodes,
1357  * taking into account the size of the new entry.
1358  *
1359  * NOTE: if new block is empty, then it will get the upper half of the
1360  * old block.  At present, all (one) callers pass in an empty second block.
1361  *
1362  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1363  * to match what it is doing in splitting the attribute leaf block.  Those
1364  * values are used in "atomic rename" operations on attributes.  Note that
1365  * the "new" and "old" values can end up in different blocks.
1366  */
1367 STATIC void
1368 xfs_attr3_leaf_rebalance(
1369         struct xfs_da_state     *state,
1370         struct xfs_da_state_blk *blk1,
1371         struct xfs_da_state_blk *blk2)
1372 {
1373         struct xfs_da_args      *args;
1374         struct xfs_attr_leafblock *leaf1;
1375         struct xfs_attr_leafblock *leaf2;
1376         struct xfs_attr3_icleaf_hdr ichdr1;
1377         struct xfs_attr3_icleaf_hdr ichdr2;
1378         struct xfs_attr_leaf_entry *entries1;
1379         struct xfs_attr_leaf_entry *entries2;
1380         int                     count;
1381         int                     totallen;
1382         int                     max;
1383         int                     space;
1384         int                     swap;
1385
1386         /*
1387          * Set up environment.
1388          */
1389         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1390         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1391         leaf1 = blk1->bp->b_addr;
1392         leaf2 = blk2->bp->b_addr;
1393         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1394         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1395         ASSERT(ichdr2.count == 0);
1396         args = state->args;
1397
1398         trace_xfs_attr_leaf_rebalance(args);
1399
1400         /*
1401          * Check ordering of blocks, reverse if it makes things simpler.
1402          *
1403          * NOTE: Given that all (current) callers pass in an empty
1404          * second block, this code should never set "swap".
1405          */
1406         swap = 0;
1407         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1408                 struct xfs_da_state_blk *tmp_blk;
1409                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1410
1411                 tmp_blk = blk1;
1412                 blk1 = blk2;
1413                 blk2 = tmp_blk;
1414
1415                 /* struct copies to swap them rather than reconverting */
1416                 tmp_ichdr = ichdr1;
1417                 ichdr1 = ichdr2;
1418                 ichdr2 = tmp_ichdr;
1419
1420                 leaf1 = blk1->bp->b_addr;
1421                 leaf2 = blk2->bp->b_addr;
1422                 swap = 1;
1423         }
1424
1425         /*
1426          * Examine entries until we reduce the absolute difference in
1427          * byte usage between the two blocks to a minimum.  Then get
1428          * the direction to copy and the number of elements to move.
1429          *
1430          * "inleaf" is true if the new entry should be inserted into blk1.
1431          * If "swap" is also true, then reverse the sense of "inleaf".
1432          */
1433         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1434                                                       blk2, &ichdr2,
1435                                                       &count, &totallen);
1436         if (swap)
1437                 state->inleaf = !state->inleaf;
1438
1439         /*
1440          * Move any entries required from leaf to leaf:
1441          */
1442         if (count < ichdr1.count) {
1443                 /*
1444                  * Figure the total bytes to be added to the destination leaf.
1445                  */
1446                 /* number entries being moved */
1447                 count = ichdr1.count - count;
1448                 space  = ichdr1.usedbytes - totallen;
1449                 space += count * sizeof(xfs_attr_leaf_entry_t);
1450
1451                 /*
1452                  * leaf2 is the destination, compact it if it looks tight.
1453                  */
1454                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1455                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1456                 if (space > max)
1457                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1458
1459                 /*
1460                  * Move high entries from leaf1 to low end of leaf2.
1461                  */
1462                 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1463                                 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1464
1465         } else if (count > ichdr1.count) {
1466                 /*
1467                  * I assert that since all callers pass in an empty
1468                  * second buffer, this code should never execute.
1469                  */
1470                 ASSERT(0);
1471
1472                 /*
1473                  * Figure the total bytes to be added to the destination leaf.
1474                  */
1475                 /* number entries being moved */
1476                 count -= ichdr1.count;
1477                 space  = totallen - ichdr1.usedbytes;
1478                 space += count * sizeof(xfs_attr_leaf_entry_t);
1479
1480                 /*
1481                  * leaf1 is the destination, compact it if it looks tight.
1482                  */
1483                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1484                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1485                 if (space > max)
1486                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1487
1488                 /*
1489                  * Move low entries from leaf2 to high end of leaf1.
1490                  */
1491                 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1492                                         ichdr1.count, count);
1493         }
1494
1495         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1496         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1497         xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1498         xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1499
1500         /*
1501          * Copy out last hashval in each block for B-tree code.
1502          */
1503         entries1 = xfs_attr3_leaf_entryp(leaf1);
1504         entries2 = xfs_attr3_leaf_entryp(leaf2);
1505         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1506         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1507
1508         /*
1509          * Adjust the expected index for insertion.
1510          * NOTE: this code depends on the (current) situation that the
1511          * second block was originally empty.
1512          *
1513          * If the insertion point moved to the 2nd block, we must adjust
1514          * the index.  We must also track the entry just following the
1515          * new entry for use in an "atomic rename" operation, that entry
1516          * is always the "old" entry and the "new" entry is what we are
1517          * inserting.  The index/blkno fields refer to the "old" entry,
1518          * while the index2/blkno2 fields refer to the "new" entry.
1519          */
1520         if (blk1->index > ichdr1.count) {
1521                 ASSERT(state->inleaf == 0);
1522                 blk2->index = blk1->index - ichdr1.count;
1523                 args->index = args->index2 = blk2->index;
1524                 args->blkno = args->blkno2 = blk2->blkno;
1525         } else if (blk1->index == ichdr1.count) {
1526                 if (state->inleaf) {
1527                         args->index = blk1->index;
1528                         args->blkno = blk1->blkno;
1529                         args->index2 = 0;
1530                         args->blkno2 = blk2->blkno;
1531                 } else {
1532                         /*
1533                          * On a double leaf split, the original attr location
1534                          * is already stored in blkno2/index2, so don't
1535                          * overwrite it overwise we corrupt the tree.
1536                          */
1537                         blk2->index = blk1->index - ichdr1.count;
1538                         args->index = blk2->index;
1539                         args->blkno = blk2->blkno;
1540                         if (!state->extravalid) {
1541                                 /*
1542                                  * set the new attr location to match the old
1543                                  * one and let the higher level split code
1544                                  * decide where in the leaf to place it.
1545                                  */
1546                                 args->index2 = blk2->index;
1547                                 args->blkno2 = blk2->blkno;
1548                         }
1549                 }
1550         } else {
1551                 ASSERT(state->inleaf == 1);
1552                 args->index = args->index2 = blk1->index;
1553                 args->blkno = args->blkno2 = blk1->blkno;
1554         }
1555 }
1556
1557 /*
1558  * Examine entries until we reduce the absolute difference in
1559  * byte usage between the two blocks to a minimum.
1560  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1561  * GROT: there will always be enough room in either block for a new entry.
1562  * GROT: Do a double-split for this case?
1563  */
1564 STATIC int
1565 xfs_attr3_leaf_figure_balance(
1566         struct xfs_da_state             *state,
1567         struct xfs_da_state_blk         *blk1,
1568         struct xfs_attr3_icleaf_hdr     *ichdr1,
1569         struct xfs_da_state_blk         *blk2,
1570         struct xfs_attr3_icleaf_hdr     *ichdr2,
1571         int                             *countarg,
1572         int                             *usedbytesarg)
1573 {
1574         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1575         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1576         struct xfs_attr_leaf_entry      *entry;
1577         int                             count;
1578         int                             max;
1579         int                             index;
1580         int                             totallen = 0;
1581         int                             half;
1582         int                             lastdelta;
1583         int                             foundit = 0;
1584         int                             tmp;
1585
1586         /*
1587          * Examine entries until we reduce the absolute difference in
1588          * byte usage between the two blocks to a minimum.
1589          */
1590         max = ichdr1->count + ichdr2->count;
1591         half = (max + 1) * sizeof(*entry);
1592         half += ichdr1->usedbytes + ichdr2->usedbytes +
1593                         xfs_attr_leaf_newentsize(state->args, NULL);
1594         half /= 2;
1595         lastdelta = state->args->geo->blksize;
1596         entry = xfs_attr3_leaf_entryp(leaf1);
1597         for (count = index = 0; count < max; entry++, index++, count++) {
1598
1599 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1600                 /*
1601                  * The new entry is in the first block, account for it.
1602                  */
1603                 if (count == blk1->index) {
1604                         tmp = totallen + sizeof(*entry) +
1605                                 xfs_attr_leaf_newentsize(state->args, NULL);
1606                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1607                                 break;
1608                         lastdelta = XFS_ATTR_ABS(half - tmp);
1609                         totallen = tmp;
1610                         foundit = 1;
1611                 }
1612
1613                 /*
1614                  * Wrap around into the second block if necessary.
1615                  */
1616                 if (count == ichdr1->count) {
1617                         leaf1 = leaf2;
1618                         entry = xfs_attr3_leaf_entryp(leaf1);
1619                         index = 0;
1620                 }
1621
1622                 /*
1623                  * Figure out if next leaf entry would be too much.
1624                  */
1625                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1626                                                                         index);
1627                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1628                         break;
1629                 lastdelta = XFS_ATTR_ABS(half - tmp);
1630                 totallen = tmp;
1631 #undef XFS_ATTR_ABS
1632         }
1633
1634         /*
1635          * Calculate the number of usedbytes that will end up in lower block.
1636          * If new entry not in lower block, fix up the count.
1637          */
1638         totallen -= count * sizeof(*entry);
1639         if (foundit) {
1640                 totallen -= sizeof(*entry) +
1641                                 xfs_attr_leaf_newentsize(state->args, NULL);
1642         }
1643
1644         *countarg = count;
1645         *usedbytesarg = totallen;
1646         return foundit;
1647 }
1648
1649 /*========================================================================
1650  * Routines used for shrinking the Btree.
1651  *========================================================================*/
1652
1653 /*
1654  * Check a leaf block and its neighbors to see if the block should be
1655  * collapsed into one or the other neighbor.  Always keep the block
1656  * with the smaller block number.
1657  * If the current block is over 50% full, don't try to join it, return 0.
1658  * If the block is empty, fill in the state structure and return 2.
1659  * If it can be collapsed, fill in the state structure and return 1.
1660  * If nothing can be done, return 0.
1661  *
1662  * GROT: allow for INCOMPLETE entries in calculation.
1663  */
1664 int
1665 xfs_attr3_leaf_toosmall(
1666         struct xfs_da_state     *state,
1667         int                     *action)
1668 {
1669         struct xfs_attr_leafblock *leaf;
1670         struct xfs_da_state_blk *blk;
1671         struct xfs_attr3_icleaf_hdr ichdr;
1672         struct xfs_buf          *bp;
1673         xfs_dablk_t             blkno;
1674         int                     bytes;
1675         int                     forward;
1676         int                     error;
1677         int                     retval;
1678         int                     i;
1679
1680         trace_xfs_attr_leaf_toosmall(state->args);
1681
1682         /*
1683          * Check for the degenerate case of the block being over 50% full.
1684          * If so, it's not worth even looking to see if we might be able
1685          * to coalesce with a sibling.
1686          */
1687         blk = &state->path.blk[ state->path.active-1 ];
1688         leaf = blk->bp->b_addr;
1689         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1690         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1691                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1692                 ichdr.usedbytes;
1693         if (bytes > (state->args->geo->blksize >> 1)) {
1694                 *action = 0;    /* blk over 50%, don't try to join */
1695                 return 0;
1696         }
1697
1698         /*
1699          * Check for the degenerate case of the block being empty.
1700          * If the block is empty, we'll simply delete it, no need to
1701          * coalesce it with a sibling block.  We choose (arbitrarily)
1702          * to merge with the forward block unless it is NULL.
1703          */
1704         if (ichdr.count == 0) {
1705                 /*
1706                  * Make altpath point to the block we want to keep and
1707                  * path point to the block we want to drop (this one).
1708                  */
1709                 forward = (ichdr.forw != 0);
1710                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1711                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1712                                                  0, &retval);
1713                 if (error)
1714                         return error;
1715                 if (retval) {
1716                         *action = 0;
1717                 } else {
1718                         *action = 2;
1719                 }
1720                 return 0;
1721         }
1722
1723         /*
1724          * Examine each sibling block to see if we can coalesce with
1725          * at least 25% free space to spare.  We need to figure out
1726          * whether to merge with the forward or the backward block.
1727          * We prefer coalescing with the lower numbered sibling so as
1728          * to shrink an attribute list over time.
1729          */
1730         /* start with smaller blk num */
1731         forward = ichdr.forw < ichdr.back;
1732         for (i = 0; i < 2; forward = !forward, i++) {
1733                 struct xfs_attr3_icleaf_hdr ichdr2;
1734                 if (forward)
1735                         blkno = ichdr.forw;
1736                 else
1737                         blkno = ichdr.back;
1738                 if (blkno == 0)
1739                         continue;
1740                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1741                                         blkno, -1, &bp);
1742                 if (error)
1743                         return error;
1744
1745                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1746
1747                 bytes = state->args->geo->blksize -
1748                         (state->args->geo->blksize >> 2) -
1749                         ichdr.usedbytes - ichdr2.usedbytes -
1750                         ((ichdr.count + ichdr2.count) *
1751                                         sizeof(xfs_attr_leaf_entry_t)) -
1752                         xfs_attr3_leaf_hdr_size(leaf);
1753
1754                 xfs_trans_brelse(state->args->trans, bp);
1755                 if (bytes >= 0)
1756                         break;  /* fits with at least 25% to spare */
1757         }
1758         if (i >= 2) {
1759                 *action = 0;
1760                 return 0;
1761         }
1762
1763         /*
1764          * Make altpath point to the block we want to keep (the lower
1765          * numbered block) and path point to the block we want to drop.
1766          */
1767         memcpy(&state->altpath, &state->path, sizeof(state->path));
1768         if (blkno < blk->blkno) {
1769                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1770                                                  0, &retval);
1771         } else {
1772                 error = xfs_da3_path_shift(state, &state->path, forward,
1773                                                  0, &retval);
1774         }
1775         if (error)
1776                 return error;
1777         if (retval) {
1778                 *action = 0;
1779         } else {
1780                 *action = 1;
1781         }
1782         return 0;
1783 }
1784
1785 /*
1786  * Remove a name from the leaf attribute list structure.
1787  *
1788  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1789  * If two leaves are 37% full, when combined they will leave 25% free.
1790  */
1791 int
1792 xfs_attr3_leaf_remove(
1793         struct xfs_buf          *bp,
1794         struct xfs_da_args      *args)
1795 {
1796         struct xfs_attr_leafblock *leaf;
1797         struct xfs_attr3_icleaf_hdr ichdr;
1798         struct xfs_attr_leaf_entry *entry;
1799         int                     before;
1800         int                     after;
1801         int                     smallest;
1802         int                     entsize;
1803         int                     tablesize;
1804         int                     tmp;
1805         int                     i;
1806
1807         trace_xfs_attr_leaf_remove(args);
1808
1809         leaf = bp->b_addr;
1810         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1811
1812         ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1813         ASSERT(args->index >= 0 && args->index < ichdr.count);
1814         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1815                                         xfs_attr3_leaf_hdr_size(leaf));
1816
1817         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1818
1819         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1820         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1821
1822         /*
1823          * Scan through free region table:
1824          *    check for adjacency of free'd entry with an existing one,
1825          *    find smallest free region in case we need to replace it,
1826          *    adjust any map that borders the entry table,
1827          */
1828         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1829                                         + xfs_attr3_leaf_hdr_size(leaf);
1830         tmp = ichdr.freemap[0].size;
1831         before = after = -1;
1832         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1833         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1834         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1835                 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1836                 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1837                 if (ichdr.freemap[i].base == tablesize) {
1838                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1839                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1840                 }
1841
1842                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1843                                 be16_to_cpu(entry->nameidx)) {
1844                         before = i;
1845                 } else if (ichdr.freemap[i].base ==
1846                                 (be16_to_cpu(entry->nameidx) + entsize)) {
1847                         after = i;
1848                 } else if (ichdr.freemap[i].size < tmp) {
1849                         tmp = ichdr.freemap[i].size;
1850                         smallest = i;
1851                 }
1852         }
1853
1854         /*
1855          * Coalesce adjacent freemap regions,
1856          * or replace the smallest region.
1857          */
1858         if ((before >= 0) || (after >= 0)) {
1859                 if ((before >= 0) && (after >= 0)) {
1860                         ichdr.freemap[before].size += entsize;
1861                         ichdr.freemap[before].size += ichdr.freemap[after].size;
1862                         ichdr.freemap[after].base = 0;
1863                         ichdr.freemap[after].size = 0;
1864                 } else if (before >= 0) {
1865                         ichdr.freemap[before].size += entsize;
1866                 } else {
1867                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1868                         ichdr.freemap[after].size += entsize;
1869                 }
1870         } else {
1871                 /*
1872                  * Replace smallest region (if it is smaller than free'd entry)
1873                  */
1874                 if (ichdr.freemap[smallest].size < entsize) {
1875                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1876                         ichdr.freemap[smallest].size = entsize;
1877                 }
1878         }
1879
1880         /*
1881          * Did we remove the first entry?
1882          */
1883         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1884                 smallest = 1;
1885         else
1886                 smallest = 0;
1887
1888         /*
1889          * Compress the remaining entries and zero out the removed stuff.
1890          */
1891         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1892         ichdr.usedbytes -= entsize;
1893         xfs_trans_log_buf(args->trans, bp,
1894              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1895                                    entsize));
1896
1897         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1898         memmove(entry, entry + 1, tmp);
1899         ichdr.count--;
1900         xfs_trans_log_buf(args->trans, bp,
1901             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1902
1903         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1904         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1905
1906         /*
1907          * If we removed the first entry, re-find the first used byte
1908          * in the name area.  Note that if the entry was the "firstused",
1909          * then we don't have a "hole" in our block resulting from
1910          * removing the name.
1911          */
1912         if (smallest) {
1913                 tmp = args->geo->blksize;
1914                 entry = xfs_attr3_leaf_entryp(leaf);
1915                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1916                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1917                         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1918
1919                         if (be16_to_cpu(entry->nameidx) < tmp)
1920                                 tmp = be16_to_cpu(entry->nameidx);
1921                 }
1922                 ichdr.firstused = tmp;
1923                 if (!ichdr.firstused)
1924                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
1925         } else {
1926                 ichdr.holes = 1;        /* mark as needing compaction */
1927         }
1928         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1929         xfs_trans_log_buf(args->trans, bp,
1930                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1931                                           xfs_attr3_leaf_hdr_size(leaf)));
1932
1933         /*
1934          * Check if leaf is less than 50% full, caller may want to
1935          * "join" the leaf with a sibling if so.
1936          */
1937         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
1938               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
1939
1940         return tmp < args->geo->magicpct; /* leaf is < 37% full */
1941 }
1942
1943 /*
1944  * Move all the attribute list entries from drop_leaf into save_leaf.
1945  */
1946 void
1947 xfs_attr3_leaf_unbalance(
1948         struct xfs_da_state     *state,
1949         struct xfs_da_state_blk *drop_blk,
1950         struct xfs_da_state_blk *save_blk)
1951 {
1952         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
1953         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
1954         struct xfs_attr3_icleaf_hdr drophdr;
1955         struct xfs_attr3_icleaf_hdr savehdr;
1956         struct xfs_attr_leaf_entry *entry;
1957
1958         trace_xfs_attr_leaf_unbalance(state->args);
1959
1960         drop_leaf = drop_blk->bp->b_addr;
1961         save_leaf = save_blk->bp->b_addr;
1962         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1963         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
1964         entry = xfs_attr3_leaf_entryp(drop_leaf);
1965
1966         /*
1967          * Save last hashval from dying block for later Btree fixup.
1968          */
1969         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
1970
1971         /*
1972          * Check if we need a temp buffer, or can we do it in place.
1973          * Note that we don't check "leaf" for holes because we will
1974          * always be dropping it, toosmall() decided that for us already.
1975          */
1976         if (savehdr.holes == 0) {
1977                 /*
1978                  * dest leaf has no holes, so we add there.  May need
1979                  * to make some room in the entry array.
1980                  */
1981                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
1982                                          drop_blk->bp, &drophdr)) {
1983                         xfs_attr3_leaf_moveents(state->args,
1984                                                 drop_leaf, &drophdr, 0,
1985                                                 save_leaf, &savehdr, 0,
1986                                                 drophdr.count);
1987                 } else {
1988                         xfs_attr3_leaf_moveents(state->args,
1989                                                 drop_leaf, &drophdr, 0,
1990                                                 save_leaf, &savehdr,
1991                                                 savehdr.count, drophdr.count);
1992                 }
1993         } else {
1994                 /*
1995                  * Destination has holes, so we make a temporary copy
1996                  * of the leaf and add them both to that.
1997                  */
1998                 struct xfs_attr_leafblock *tmp_leaf;
1999                 struct xfs_attr3_icleaf_hdr tmphdr;
2000
2001                 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2002
2003                 /*
2004                  * Copy the header into the temp leaf so that all the stuff
2005                  * not in the incore header is present and gets copied back in
2006                  * once we've moved all the entries.
2007                  */
2008                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2009
2010                 memset(&tmphdr, 0, sizeof(tmphdr));
2011                 tmphdr.magic = savehdr.magic;
2012                 tmphdr.forw = savehdr.forw;
2013                 tmphdr.back = savehdr.back;
2014                 tmphdr.firstused = state->args->geo->blksize;
2015
2016                 /* write the header to the temp buffer to initialise it */
2017                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2018
2019                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2020                                          drop_blk->bp, &drophdr)) {
2021                         xfs_attr3_leaf_moveents(state->args,
2022                                                 drop_leaf, &drophdr, 0,
2023                                                 tmp_leaf, &tmphdr, 0,
2024                                                 drophdr.count);
2025                         xfs_attr3_leaf_moveents(state->args,
2026                                                 save_leaf, &savehdr, 0,
2027                                                 tmp_leaf, &tmphdr, tmphdr.count,
2028                                                 savehdr.count);
2029                 } else {
2030                         xfs_attr3_leaf_moveents(state->args,
2031                                                 save_leaf, &savehdr, 0,
2032                                                 tmp_leaf, &tmphdr, 0,
2033                                                 savehdr.count);
2034                         xfs_attr3_leaf_moveents(state->args,
2035                                                 drop_leaf, &drophdr, 0,
2036                                                 tmp_leaf, &tmphdr, tmphdr.count,
2037                                                 drophdr.count);
2038                 }
2039                 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2040                 savehdr = tmphdr; /* struct copy */
2041                 kmem_free(tmp_leaf);
2042         }
2043
2044         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2045         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2046                                            state->args->geo->blksize - 1);
2047
2048         /*
2049          * Copy out last hashval in each block for B-tree code.
2050          */
2051         entry = xfs_attr3_leaf_entryp(save_leaf);
2052         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2053 }
2054
2055 /*========================================================================
2056  * Routines used for finding things in the Btree.
2057  *========================================================================*/
2058
2059 /*
2060  * Look up a name in a leaf attribute list structure.
2061  * This is the internal routine, it uses the caller's buffer.
2062  *
2063  * Note that duplicate keys are allowed, but only check within the
2064  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2065  *
2066  * Return in args->index the index into the entry[] array of either
2067  * the found entry, or where the entry should have been (insert before
2068  * that entry).
2069  *
2070  * Don't change the args->value unless we find the attribute.
2071  */
2072 int
2073 xfs_attr3_leaf_lookup_int(
2074         struct xfs_buf          *bp,
2075         struct xfs_da_args      *args)
2076 {
2077         struct xfs_attr_leafblock *leaf;
2078         struct xfs_attr3_icleaf_hdr ichdr;
2079         struct xfs_attr_leaf_entry *entry;
2080         struct xfs_attr_leaf_entry *entries;
2081         struct xfs_attr_leaf_name_local *name_loc;
2082         struct xfs_attr_leaf_name_remote *name_rmt;
2083         xfs_dahash_t            hashval;
2084         int                     probe;
2085         int                     span;
2086
2087         trace_xfs_attr_leaf_lookup(args);
2088
2089         leaf = bp->b_addr;
2090         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2091         entries = xfs_attr3_leaf_entryp(leaf);
2092         ASSERT(ichdr.count < args->geo->blksize / 8);
2093
2094         /*
2095          * Binary search.  (note: small blocks will skip this loop)
2096          */
2097         hashval = args->hashval;
2098         probe = span = ichdr.count / 2;
2099         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2100                 span /= 2;
2101                 if (be32_to_cpu(entry->hashval) < hashval)
2102                         probe += span;
2103                 else if (be32_to_cpu(entry->hashval) > hashval)
2104                         probe -= span;
2105                 else
2106                         break;
2107         }
2108         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2109         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2110
2111         /*
2112          * Since we may have duplicate hashval's, find the first matching
2113          * hashval in the leaf.
2114          */
2115         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2116                 entry--;
2117                 probe--;
2118         }
2119         while (probe < ichdr.count &&
2120                be32_to_cpu(entry->hashval) < hashval) {
2121                 entry++;
2122                 probe++;
2123         }
2124         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2125                 args->index = probe;
2126                 return -ENOATTR;
2127         }
2128
2129         /*
2130          * Duplicate keys may be present, so search all of them for a match.
2131          */
2132         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2133                         entry++, probe++) {
2134 /*
2135  * GROT: Add code to remove incomplete entries.
2136  */
2137                 /*
2138                  * If we are looking for INCOMPLETE entries, show only those.
2139                  * If we are looking for complete entries, show only those.
2140                  */
2141                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2142                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2143                         continue;
2144                 }
2145                 if (entry->flags & XFS_ATTR_LOCAL) {
2146                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2147                         if (name_loc->namelen != args->namelen)
2148                                 continue;
2149                         if (memcmp(args->name, name_loc->nameval,
2150                                                         args->namelen) != 0)
2151                                 continue;
2152                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2153                                 continue;
2154                         args->index = probe;
2155                         return -EEXIST;
2156                 } else {
2157                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2158                         if (name_rmt->namelen != args->namelen)
2159                                 continue;
2160                         if (memcmp(args->name, name_rmt->name,
2161                                                         args->namelen) != 0)
2162                                 continue;
2163                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2164                                 continue;
2165                         args->index = probe;
2166                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2167                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2168                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2169                                                         args->dp->i_mount,
2170                                                         args->rmtvaluelen);
2171                         return -EEXIST;
2172                 }
2173         }
2174         args->index = probe;
2175         return -ENOATTR;
2176 }
2177
2178 /*
2179  * Get the value associated with an attribute name from a leaf attribute
2180  * list structure.
2181  */
2182 int
2183 xfs_attr3_leaf_getvalue(
2184         struct xfs_buf          *bp,
2185         struct xfs_da_args      *args)
2186 {
2187         struct xfs_attr_leafblock *leaf;
2188         struct xfs_attr3_icleaf_hdr ichdr;
2189         struct xfs_attr_leaf_entry *entry;
2190         struct xfs_attr_leaf_name_local *name_loc;
2191         struct xfs_attr_leaf_name_remote *name_rmt;
2192         int                     valuelen;
2193
2194         leaf = bp->b_addr;
2195         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2196         ASSERT(ichdr.count < args->geo->blksize / 8);
2197         ASSERT(args->index < ichdr.count);
2198
2199         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2200         if (entry->flags & XFS_ATTR_LOCAL) {
2201                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2202                 ASSERT(name_loc->namelen == args->namelen);
2203                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2204                 valuelen = be16_to_cpu(name_loc->valuelen);
2205                 if (args->flags & ATTR_KERNOVAL) {
2206                         args->valuelen = valuelen;
2207                         return 0;
2208                 }
2209                 if (args->valuelen < valuelen) {
2210                         args->valuelen = valuelen;
2211                         return -ERANGE;
2212                 }
2213                 args->valuelen = valuelen;
2214                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2215         } else {
2216                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2217                 ASSERT(name_rmt->namelen == args->namelen);
2218                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2219                 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2220                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2221                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2222                                                        args->rmtvaluelen);
2223                 if (args->flags & ATTR_KERNOVAL) {
2224                         args->valuelen = args->rmtvaluelen;
2225                         return 0;
2226                 }
2227                 if (args->valuelen < args->rmtvaluelen) {
2228                         args->valuelen = args->rmtvaluelen;
2229                         return -ERANGE;
2230                 }
2231                 args->valuelen = args->rmtvaluelen;
2232         }
2233         return 0;
2234 }
2235
2236 /*========================================================================
2237  * Utility routines.
2238  *========================================================================*/
2239
2240 /*
2241  * Move the indicated entries from one leaf to another.
2242  * NOTE: this routine modifies both source and destination leaves.
2243  */
2244 /*ARGSUSED*/
2245 STATIC void
2246 xfs_attr3_leaf_moveents(
2247         struct xfs_da_args              *args,
2248         struct xfs_attr_leafblock       *leaf_s,
2249         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2250         int                             start_s,
2251         struct xfs_attr_leafblock       *leaf_d,
2252         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2253         int                             start_d,
2254         int                             count)
2255 {
2256         struct xfs_attr_leaf_entry      *entry_s;
2257         struct xfs_attr_leaf_entry      *entry_d;
2258         int                             desti;
2259         int                             tmp;
2260         int                             i;
2261
2262         /*
2263          * Check for nothing to do.
2264          */
2265         if (count == 0)
2266                 return;
2267
2268         /*
2269          * Set up environment.
2270          */
2271         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2272                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2273         ASSERT(ichdr_s->magic == ichdr_d->magic);
2274         ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2275         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2276                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2277         ASSERT(ichdr_d->count < args->geo->blksize / 8);
2278         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2279                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2280
2281         ASSERT(start_s < ichdr_s->count);
2282         ASSERT(start_d <= ichdr_d->count);
2283         ASSERT(count <= ichdr_s->count);
2284
2285
2286         /*
2287          * Move the entries in the destination leaf up to make a hole?
2288          */
2289         if (start_d < ichdr_d->count) {
2290                 tmp  = ichdr_d->count - start_d;
2291                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2292                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2293                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2294                 memmove(entry_d, entry_s, tmp);
2295         }
2296
2297         /*
2298          * Copy all entry's in the same (sorted) order,
2299          * but allocate attribute info packed and in sequence.
2300          */
2301         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2302         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2303         desti = start_d;
2304         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2305                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2306                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2307 #ifdef GROT
2308                 /*
2309                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2310                  * may also need to change the insertion index.  Code turned
2311                  * off for 6.2, should be revisited later.
2312                  */
2313                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2314                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2315                         ichdr_s->usedbytes -= tmp;
2316                         ichdr_s->count -= 1;
2317                         entry_d--;      /* to compensate for ++ in loop hdr */
2318                         desti--;
2319                         if ((start_s + i) < offset)
2320                                 result++;       /* insertion index adjustment */
2321                 } else {
2322 #endif /* GROT */
2323                         ichdr_d->firstused -= tmp;
2324                         /* both on-disk, don't endian flip twice */
2325                         entry_d->hashval = entry_s->hashval;
2326                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2327                         entry_d->flags = entry_s->flags;
2328                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2329                                                         <= args->geo->blksize);
2330                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2331                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2332                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2333                                                         <= args->geo->blksize);
2334                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2335                         ichdr_s->usedbytes -= tmp;
2336                         ichdr_d->usedbytes += tmp;
2337                         ichdr_s->count -= 1;
2338                         ichdr_d->count += 1;
2339                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2340                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2341                         ASSERT(ichdr_d->firstused >= tmp);
2342 #ifdef GROT
2343                 }
2344 #endif /* GROT */
2345         }
2346
2347         /*
2348          * Zero out the entries we just copied.
2349          */
2350         if (start_s == ichdr_s->count) {
2351                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2352                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2353                 ASSERT(((char *)entry_s + tmp) <=
2354                        ((char *)leaf_s + args->geo->blksize));
2355                 memset(entry_s, 0, tmp);
2356         } else {
2357                 /*
2358                  * Move the remaining entries down to fill the hole,
2359                  * then zero the entries at the top.
2360                  */
2361                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2362                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2363                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2364                 memmove(entry_d, entry_s, tmp);
2365
2366                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2367                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2368                 ASSERT(((char *)entry_s + tmp) <=
2369                        ((char *)leaf_s + args->geo->blksize));
2370                 memset(entry_s, 0, tmp);
2371         }
2372
2373         /*
2374          * Fill in the freemap information
2375          */
2376         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2377         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2378         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2379         ichdr_d->freemap[1].base = 0;
2380         ichdr_d->freemap[2].base = 0;
2381         ichdr_d->freemap[1].size = 0;
2382         ichdr_d->freemap[2].size = 0;
2383         ichdr_s->holes = 1;     /* leaf may not be compact */
2384 }
2385
2386 /*
2387  * Pick up the last hashvalue from a leaf block.
2388  */
2389 xfs_dahash_t
2390 xfs_attr_leaf_lasthash(
2391         struct xfs_buf  *bp,
2392         int             *count)
2393 {
2394         struct xfs_attr3_icleaf_hdr ichdr;
2395         struct xfs_attr_leaf_entry *entries;
2396
2397         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2398         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2399         if (count)
2400                 *count = ichdr.count;
2401         if (!ichdr.count)
2402                 return 0;
2403         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2404 }
2405
2406 /*
2407  * Calculate the number of bytes used to store the indicated attribute
2408  * (whether local or remote only calculate bytes in this block).
2409  */
2410 STATIC int
2411 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2412 {
2413         struct xfs_attr_leaf_entry *entries;
2414         xfs_attr_leaf_name_local_t *name_loc;
2415         xfs_attr_leaf_name_remote_t *name_rmt;
2416         int size;
2417
2418         entries = xfs_attr3_leaf_entryp(leaf);
2419         if (entries[index].flags & XFS_ATTR_LOCAL) {
2420                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2421                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2422                                                    be16_to_cpu(name_loc->valuelen));
2423         } else {
2424                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2425                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2426         }
2427         return size;
2428 }
2429
2430 /*
2431  * Calculate the number of bytes that would be required to store the new
2432  * attribute (whether local or remote only calculate bytes in this block).
2433  * This routine decides as a side effect whether the attribute will be
2434  * a "local" or a "remote" attribute.
2435  */
2436 int
2437 xfs_attr_leaf_newentsize(
2438         struct xfs_da_args      *args,
2439         int                     *local)
2440 {
2441         int                     size;
2442
2443         size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2444         if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2445                 if (local)
2446                         *local = 1;
2447                 return size;
2448         }
2449         if (local)
2450                 *local = 0;
2451         return xfs_attr_leaf_entsize_remote(args->namelen);
2452 }
2453
2454
2455 /*========================================================================
2456  * Manage the INCOMPLETE flag in a leaf entry
2457  *========================================================================*/
2458
2459 /*
2460  * Clear the INCOMPLETE flag on an entry in a leaf block.
2461  */
2462 int
2463 xfs_attr3_leaf_clearflag(
2464         struct xfs_da_args      *args)
2465 {
2466         struct xfs_attr_leafblock *leaf;
2467         struct xfs_attr_leaf_entry *entry;
2468         struct xfs_attr_leaf_name_remote *name_rmt;
2469         struct xfs_buf          *bp;
2470         int                     error;
2471 #ifdef DEBUG
2472         struct xfs_attr3_icleaf_hdr ichdr;
2473         xfs_attr_leaf_name_local_t *name_loc;
2474         int namelen;
2475         char *name;
2476 #endif /* DEBUG */
2477
2478         trace_xfs_attr_leaf_clearflag(args);
2479         /*
2480          * Set up the operation.
2481          */
2482         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2483         if (error)
2484                 return error;
2485
2486         leaf = bp->b_addr;
2487         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2488         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2489
2490 #ifdef DEBUG
2491         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2492         ASSERT(args->index < ichdr.count);
2493         ASSERT(args->index >= 0);
2494
2495         if (entry->flags & XFS_ATTR_LOCAL) {
2496                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2497                 namelen = name_loc->namelen;
2498                 name = (char *)name_loc->nameval;
2499         } else {
2500                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2501                 namelen = name_rmt->namelen;
2502                 name = (char *)name_rmt->name;
2503         }
2504         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2505         ASSERT(namelen == args->namelen);
2506         ASSERT(memcmp(name, args->name, namelen) == 0);
2507 #endif /* DEBUG */
2508
2509         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2510         xfs_trans_log_buf(args->trans, bp,
2511                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2512
2513         if (args->rmtblkno) {
2514                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2515                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2516                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2517                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2518                 xfs_trans_log_buf(args->trans, bp,
2519                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2520         }
2521
2522         /*
2523          * Commit the flag value change and start the next trans in series.
2524          */
2525         return xfs_trans_roll(&args->trans, args->dp);
2526 }
2527
2528 /*
2529  * Set the INCOMPLETE flag on an entry in a leaf block.
2530  */
2531 int
2532 xfs_attr3_leaf_setflag(
2533         struct xfs_da_args      *args)
2534 {
2535         struct xfs_attr_leafblock *leaf;
2536         struct xfs_attr_leaf_entry *entry;
2537         struct xfs_attr_leaf_name_remote *name_rmt;
2538         struct xfs_buf          *bp;
2539         int error;
2540 #ifdef DEBUG
2541         struct xfs_attr3_icleaf_hdr ichdr;
2542 #endif
2543
2544         trace_xfs_attr_leaf_setflag(args);
2545
2546         /*
2547          * Set up the operation.
2548          */
2549         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2550         if (error)
2551                 return error;
2552
2553         leaf = bp->b_addr;
2554 #ifdef DEBUG
2555         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2556         ASSERT(args->index < ichdr.count);
2557         ASSERT(args->index >= 0);
2558 #endif
2559         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2560
2561         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2562         entry->flags |= XFS_ATTR_INCOMPLETE;
2563         xfs_trans_log_buf(args->trans, bp,
2564                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2565         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2566                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2567                 name_rmt->valueblk = 0;
2568                 name_rmt->valuelen = 0;
2569                 xfs_trans_log_buf(args->trans, bp,
2570                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2571         }
2572
2573         /*
2574          * Commit the flag value change and start the next trans in series.
2575          */
2576         return xfs_trans_roll(&args->trans, args->dp);
2577 }
2578
2579 /*
2580  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2581  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2582  * entry given by args->blkno2/index2.
2583  *
2584  * Note that they could be in different blocks, or in the same block.
2585  */
2586 int
2587 xfs_attr3_leaf_flipflags(
2588         struct xfs_da_args      *args)
2589 {
2590         struct xfs_attr_leafblock *leaf1;
2591         struct xfs_attr_leafblock *leaf2;
2592         struct xfs_attr_leaf_entry *entry1;
2593         struct xfs_attr_leaf_entry *entry2;
2594         struct xfs_attr_leaf_name_remote *name_rmt;
2595         struct xfs_buf          *bp1;
2596         struct xfs_buf          *bp2;
2597         int error;
2598 #ifdef DEBUG
2599         struct xfs_attr3_icleaf_hdr ichdr1;
2600         struct xfs_attr3_icleaf_hdr ichdr2;
2601         xfs_attr_leaf_name_local_t *name_loc;
2602         int namelen1, namelen2;
2603         char *name1, *name2;
2604 #endif /* DEBUG */
2605
2606         trace_xfs_attr_leaf_flipflags(args);
2607
2608         /*
2609          * Read the block containing the "old" attr
2610          */
2611         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2612         if (error)
2613                 return error;
2614
2615         /*
2616          * Read the block containing the "new" attr, if it is different
2617          */
2618         if (args->blkno2 != args->blkno) {
2619                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2620                                            -1, &bp2);
2621                 if (error)
2622                         return error;
2623         } else {
2624                 bp2 = bp1;
2625         }
2626
2627         leaf1 = bp1->b_addr;
2628         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2629
2630         leaf2 = bp2->b_addr;
2631         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2632
2633 #ifdef DEBUG
2634         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2635         ASSERT(args->index < ichdr1.count);
2636         ASSERT(args->index >= 0);
2637
2638         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2639         ASSERT(args->index2 < ichdr2.count);
2640         ASSERT(args->index2 >= 0);
2641
2642         if (entry1->flags & XFS_ATTR_LOCAL) {
2643                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2644                 namelen1 = name_loc->namelen;
2645                 name1 = (char *)name_loc->nameval;
2646         } else {
2647                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2648                 namelen1 = name_rmt->namelen;
2649                 name1 = (char *)name_rmt->name;
2650         }
2651         if (entry2->flags & XFS_ATTR_LOCAL) {
2652                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2653                 namelen2 = name_loc->namelen;
2654                 name2 = (char *)name_loc->nameval;
2655         } else {
2656                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2657                 namelen2 = name_rmt->namelen;
2658                 name2 = (char *)name_rmt->name;
2659         }
2660         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2661         ASSERT(namelen1 == namelen2);
2662         ASSERT(memcmp(name1, name2, namelen1) == 0);
2663 #endif /* DEBUG */
2664
2665         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2666         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2667
2668         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2669         xfs_trans_log_buf(args->trans, bp1,
2670                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2671         if (args->rmtblkno) {
2672                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2673                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2674                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2675                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2676                 xfs_trans_log_buf(args->trans, bp1,
2677                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2678         }
2679
2680         entry2->flags |= XFS_ATTR_INCOMPLETE;
2681         xfs_trans_log_buf(args->trans, bp2,
2682                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2683         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2684                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2685                 name_rmt->valueblk = 0;
2686                 name_rmt->valuelen = 0;
2687                 xfs_trans_log_buf(args->trans, bp2,
2688                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2689         }
2690
2691         /*
2692          * Commit the flag value change and start the next trans in series.
2693          */
2694         error = xfs_trans_roll(&args->trans, args->dp);
2695
2696         return error;
2697 }