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