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