xfs: merge xfs_dinode.h into xfs_format.h
[firefly-linux-kernel-4.4.55.git] / fs / xfs / libxfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_attr_remote.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_cksum.h"
44 #include "xfs_dir2.h"
45
46
47 /*
48  * xfs_attr_leaf.c
49  *
50  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
51  */
52
53 /*========================================================================
54  * Function prototypes for the kernel.
55  *========================================================================*/
56
57 /*
58  * Routines used for growing the Btree.
59  */
60 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
61                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
62 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
63                                    struct xfs_attr3_icleaf_hdr *ichdr,
64                                    struct xfs_da_args *args, int freemap_index);
65 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
66                                    struct xfs_attr3_icleaf_hdr *ichdr,
67                                    struct xfs_buf *leaf_buffer);
68 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
69                                                    xfs_da_state_blk_t *blk1,
70                                                    xfs_da_state_blk_t *blk2);
71 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
72                         xfs_da_state_blk_t *leaf_blk_1,
73                         struct xfs_attr3_icleaf_hdr *ichdr1,
74                         xfs_da_state_blk_t *leaf_blk_2,
75                         struct xfs_attr3_icleaf_hdr *ichdr2,
76                         int *number_entries_in_blk1,
77                         int *number_usedbytes_in_blk1);
78
79 /*
80  * Utility routines.
81  */
82 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
83                         struct xfs_attr_leafblock *src_leaf,
84                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
85                         struct xfs_attr_leafblock *dst_leaf,
86                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87                         int move_count);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
89
90 void
91 xfs_attr3_leaf_hdr_from_disk(
92         struct xfs_attr3_icleaf_hdr     *to,
93         struct xfs_attr_leafblock       *from)
94 {
95         int     i;
96
97         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
98                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
99
100         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
101                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
102
103                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
104                 to->back = be32_to_cpu(hdr3->info.hdr.back);
105                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
106                 to->count = be16_to_cpu(hdr3->count);
107                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
108                 to->firstused = be16_to_cpu(hdr3->firstused);
109                 to->holes = hdr3->holes;
110
111                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
112                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
113                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
114                 }
115                 return;
116         }
117         to->forw = be32_to_cpu(from->hdr.info.forw);
118         to->back = be32_to_cpu(from->hdr.info.back);
119         to->magic = be16_to_cpu(from->hdr.info.magic);
120         to->count = be16_to_cpu(from->hdr.count);
121         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
122         to->firstused = be16_to_cpu(from->hdr.firstused);
123         to->holes = from->hdr.holes;
124
125         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
126                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
127                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
128         }
129 }
130
131 void
132 xfs_attr3_leaf_hdr_to_disk(
133         struct xfs_attr_leafblock       *to,
134         struct xfs_attr3_icleaf_hdr     *from)
135 {
136         int     i;
137
138         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
139                from->magic == XFS_ATTR3_LEAF_MAGIC);
140
141         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
142                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
143
144                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
145                 hdr3->info.hdr.back = cpu_to_be32(from->back);
146                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
147                 hdr3->count = cpu_to_be16(from->count);
148                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
149                 hdr3->firstused = cpu_to_be16(from->firstused);
150                 hdr3->holes = from->holes;
151                 hdr3->pad1 = 0;
152
153                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
154                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
155                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
156                 }
157                 return;
158         }
159         to->hdr.info.forw = cpu_to_be32(from->forw);
160         to->hdr.info.back = cpu_to_be32(from->back);
161         to->hdr.info.magic = cpu_to_be16(from->magic);
162         to->hdr.count = cpu_to_be16(from->count);
163         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
164         to->hdr.firstused = cpu_to_be16(from->firstused);
165         to->hdr.holes = from->holes;
166         to->hdr.pad1 = 0;
167
168         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
169                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
170                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
171         }
172 }
173
174 static bool
175 xfs_attr3_leaf_verify(
176         struct xfs_buf          *bp)
177 {
178         struct xfs_mount        *mp = bp->b_target->bt_mount;
179         struct xfs_attr_leafblock *leaf = bp->b_addr;
180         struct xfs_attr3_icleaf_hdr ichdr;
181
182         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
183
184         if (xfs_sb_version_hascrc(&mp->m_sb)) {
185                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
186
187                 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
188                         return false;
189
190                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
191                         return false;
192                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
193                         return false;
194         } else {
195                 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
196                         return false;
197         }
198         if (ichdr.count == 0)
199                 return false;
200
201         /* XXX: need to range check rest of attr header values */
202         /* XXX: hash order check? */
203
204         return true;
205 }
206
207 static void
208 xfs_attr3_leaf_write_verify(
209         struct xfs_buf  *bp)
210 {
211         struct xfs_mount        *mp = bp->b_target->bt_mount;
212         struct xfs_buf_log_item *bip = bp->b_fspriv;
213         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
214
215         if (!xfs_attr3_leaf_verify(bp)) {
216                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
217                 xfs_verifier_error(bp);
218                 return;
219         }
220
221         if (!xfs_sb_version_hascrc(&mp->m_sb))
222                 return;
223
224         if (bip)
225                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
226
227         xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
228 }
229
230 /*
231  * leaf/node format detection on trees is sketchy, so a node read can be done on
232  * leaf level blocks when detection identifies the tree as a node format tree
233  * incorrectly. In this case, we need to swap the verifier to match the correct
234  * format of the block being read.
235  */
236 static void
237 xfs_attr3_leaf_read_verify(
238         struct xfs_buf          *bp)
239 {
240         struct xfs_mount        *mp = bp->b_target->bt_mount;
241
242         if (xfs_sb_version_hascrc(&mp->m_sb) &&
243              !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
244                 xfs_buf_ioerror(bp, -EFSBADCRC);
245         else if (!xfs_attr3_leaf_verify(bp))
246                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
247
248         if (bp->b_error)
249                 xfs_verifier_error(bp);
250 }
251
252 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
253         .verify_read = xfs_attr3_leaf_read_verify,
254         .verify_write = xfs_attr3_leaf_write_verify,
255 };
256
257 int
258 xfs_attr3_leaf_read(
259         struct xfs_trans        *tp,
260         struct xfs_inode        *dp,
261         xfs_dablk_t             bno,
262         xfs_daddr_t             mappedbno,
263         struct xfs_buf          **bpp)
264 {
265         int                     err;
266
267         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
268                                 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
269         if (!err && tp)
270                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
271         return err;
272 }
273
274 /*========================================================================
275  * Namespace helper routines
276  *========================================================================*/
277
278 /*
279  * If namespace bits don't match return 0.
280  * If all match then return 1.
281  */
282 STATIC int
283 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
284 {
285         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
286 }
287
288
289 /*========================================================================
290  * External routines when attribute fork size < XFS_LITINO(mp).
291  *========================================================================*/
292
293 /*
294  * Query whether the requested number of additional bytes of extended
295  * attribute space will be able to fit inline.
296  *
297  * Returns zero if not, else the di_forkoff fork offset to be used in the
298  * literal area for attribute data once the new bytes have been added.
299  *
300  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
301  * special case for dev/uuid inodes, they have fixed size data forks.
302  */
303 int
304 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
305 {
306         int offset;
307         int minforkoff; /* lower limit on valid forkoff locations */
308         int maxforkoff; /* upper limit on valid forkoff locations */
309         int dsize;
310         xfs_mount_t *mp = dp->i_mount;
311
312         /* rounded down */
313         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
314
315         switch (dp->i_d.di_format) {
316         case XFS_DINODE_FMT_DEV:
317                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
318                 return (offset >= minforkoff) ? minforkoff : 0;
319         case XFS_DINODE_FMT_UUID:
320                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
321                 return (offset >= minforkoff) ? minforkoff : 0;
322         }
323
324         /*
325          * If the requested numbers of bytes is smaller or equal to the
326          * current attribute fork size we can always proceed.
327          *
328          * Note that if_bytes in the data fork might actually be larger than
329          * the current data fork size is due to delalloc extents. In that
330          * case either the extent count will go down when they are converted
331          * to real extents, or the delalloc conversion will take care of the
332          * literal area rebalancing.
333          */
334         if (bytes <= XFS_IFORK_ASIZE(dp))
335                 return dp->i_d.di_forkoff;
336
337         /*
338          * For attr2 we can try to move the forkoff if there is space in the
339          * literal area, but for the old format we are done if there is no
340          * space in the fixed attribute fork.
341          */
342         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
343                 return 0;
344
345         dsize = dp->i_df.if_bytes;
346
347         switch (dp->i_d.di_format) {
348         case XFS_DINODE_FMT_EXTENTS:
349                 /*
350                  * If there is no attr fork and the data fork is extents, 
351                  * determine if creating the default attr fork will result
352                  * in the extents form migrating to btree. If so, the
353                  * minimum offset only needs to be the space required for
354                  * the btree root.
355                  */
356                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
357                     xfs_default_attroffset(dp))
358                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
359                 break;
360         case XFS_DINODE_FMT_BTREE:
361                 /*
362                  * If we have a data btree then keep forkoff if we have one,
363                  * otherwise we are adding a new attr, so then we set
364                  * minforkoff to where the btree root can finish so we have
365                  * plenty of room for attrs
366                  */
367                 if (dp->i_d.di_forkoff) {
368                         if (offset < dp->i_d.di_forkoff)
369                                 return 0;
370                         return dp->i_d.di_forkoff;
371                 }
372                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
373                 break;
374         }
375
376         /*
377          * A data fork btree root must have space for at least
378          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
379          */
380         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
381         minforkoff = roundup(minforkoff, 8) >> 3;
382
383         /* attr fork btree root can have at least this many key/ptr pairs */
384         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
385                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
386         maxforkoff = maxforkoff >> 3;   /* rounded down */
387
388         if (offset >= maxforkoff)
389                 return maxforkoff;
390         if (offset >= minforkoff)
391                 return offset;
392         return 0;
393 }
394
395 /*
396  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
397  */
398 STATIC void
399 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
400 {
401         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
402             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
403                 spin_lock(&mp->m_sb_lock);
404                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
405                         xfs_sb_version_addattr2(&mp->m_sb);
406                         spin_unlock(&mp->m_sb_lock);
407                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
408                 } else
409                         spin_unlock(&mp->m_sb_lock);
410         }
411 }
412
413 /*
414  * Create the initial contents of a shortform attribute list.
415  */
416 void
417 xfs_attr_shortform_create(xfs_da_args_t *args)
418 {
419         xfs_attr_sf_hdr_t *hdr;
420         xfs_inode_t *dp;
421         xfs_ifork_t *ifp;
422
423         trace_xfs_attr_sf_create(args);
424
425         dp = args->dp;
426         ASSERT(dp != NULL);
427         ifp = dp->i_afp;
428         ASSERT(ifp != NULL);
429         ASSERT(ifp->if_bytes == 0);
430         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
431                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
432                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
433                 ifp->if_flags |= XFS_IFINLINE;
434         } else {
435                 ASSERT(ifp->if_flags & XFS_IFINLINE);
436         }
437         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
438         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
439         hdr->count = 0;
440         hdr->totsize = cpu_to_be16(sizeof(*hdr));
441         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
442 }
443
444 /*
445  * Add a name/value pair to the shortform attribute list.
446  * Overflow from the inode has already been checked for.
447  */
448 void
449 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
450 {
451         xfs_attr_shortform_t *sf;
452         xfs_attr_sf_entry_t *sfe;
453         int i, offset, size;
454         xfs_mount_t *mp;
455         xfs_inode_t *dp;
456         xfs_ifork_t *ifp;
457
458         trace_xfs_attr_sf_add(args);
459
460         dp = args->dp;
461         mp = dp->i_mount;
462         dp->i_d.di_forkoff = forkoff;
463
464         ifp = dp->i_afp;
465         ASSERT(ifp->if_flags & XFS_IFINLINE);
466         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
467         sfe = &sf->list[0];
468         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
469 #ifdef DEBUG
470                 if (sfe->namelen != args->namelen)
471                         continue;
472                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
473                         continue;
474                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
475                         continue;
476                 ASSERT(0);
477 #endif
478         }
479
480         offset = (char *)sfe - (char *)sf;
481         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
482         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
483         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
484         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
485
486         sfe->namelen = args->namelen;
487         sfe->valuelen = args->valuelen;
488         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
489         memcpy(sfe->nameval, args->name, args->namelen);
490         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
491         sf->hdr.count++;
492         be16_add_cpu(&sf->hdr.totsize, size);
493         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
494
495         xfs_sbversion_add_attr2(mp, args->trans);
496 }
497
498 /*
499  * After the last attribute is removed revert to original inode format,
500  * making all literal area available to the data fork once more.
501  */
502 STATIC void
503 xfs_attr_fork_reset(
504         struct xfs_inode        *ip,
505         struct xfs_trans        *tp)
506 {
507         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
508         ip->i_d.di_forkoff = 0;
509         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
510
511         ASSERT(ip->i_d.di_anextents == 0);
512         ASSERT(ip->i_afp == NULL);
513
514         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
515 }
516
517 /*
518  * Remove an attribute from the shortform attribute list structure.
519  */
520 int
521 xfs_attr_shortform_remove(xfs_da_args_t *args)
522 {
523         xfs_attr_shortform_t *sf;
524         xfs_attr_sf_entry_t *sfe;
525         int base, size=0, end, totsize, i;
526         xfs_mount_t *mp;
527         xfs_inode_t *dp;
528
529         trace_xfs_attr_sf_remove(args);
530
531         dp = args->dp;
532         mp = dp->i_mount;
533         base = sizeof(xfs_attr_sf_hdr_t);
534         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
535         sfe = &sf->list[0];
536         end = sf->hdr.count;
537         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
538                                         base += size, i++) {
539                 size = XFS_ATTR_SF_ENTSIZE(sfe);
540                 if (sfe->namelen != args->namelen)
541                         continue;
542                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
543                         continue;
544                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
545                         continue;
546                 break;
547         }
548         if (i == end)
549                 return -ENOATTR;
550
551         /*
552          * Fix up the attribute fork data, covering the hole
553          */
554         end = base + size;
555         totsize = be16_to_cpu(sf->hdr.totsize);
556         if (end != totsize)
557                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
558         sf->hdr.count--;
559         be16_add_cpu(&sf->hdr.totsize, -size);
560
561         /*
562          * Fix up the start offset of the attribute fork
563          */
564         totsize -= size;
565         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
566             (mp->m_flags & XFS_MOUNT_ATTR2) &&
567             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
568             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
569                 xfs_attr_fork_reset(dp, args->trans);
570         } else {
571                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
572                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
573                 ASSERT(dp->i_d.di_forkoff);
574                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
575                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
576                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
577                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
578                 xfs_trans_log_inode(args->trans, dp,
579                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
580         }
581
582         xfs_sbversion_add_attr2(mp, args->trans);
583
584         return 0;
585 }
586
587 /*
588  * Look up a name in a shortform attribute list structure.
589  */
590 /*ARGSUSED*/
591 int
592 xfs_attr_shortform_lookup(xfs_da_args_t *args)
593 {
594         xfs_attr_shortform_t *sf;
595         xfs_attr_sf_entry_t *sfe;
596         int i;
597         xfs_ifork_t *ifp;
598
599         trace_xfs_attr_sf_lookup(args);
600
601         ifp = args->dp->i_afp;
602         ASSERT(ifp->if_flags & XFS_IFINLINE);
603         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
604         sfe = &sf->list[0];
605         for (i = 0; i < sf->hdr.count;
606                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
607                 if (sfe->namelen != args->namelen)
608                         continue;
609                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
610                         continue;
611                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
612                         continue;
613                 return -EEXIST;
614         }
615         return -ENOATTR;
616 }
617
618 /*
619  * Look up a name in a shortform attribute list structure.
620  */
621 /*ARGSUSED*/
622 int
623 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
624 {
625         xfs_attr_shortform_t *sf;
626         xfs_attr_sf_entry_t *sfe;
627         int i;
628
629         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
630         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
631         sfe = &sf->list[0];
632         for (i = 0; i < sf->hdr.count;
633                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
634                 if (sfe->namelen != args->namelen)
635                         continue;
636                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
637                         continue;
638                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
639                         continue;
640                 if (args->flags & ATTR_KERNOVAL) {
641                         args->valuelen = sfe->valuelen;
642                         return -EEXIST;
643                 }
644                 if (args->valuelen < sfe->valuelen) {
645                         args->valuelen = sfe->valuelen;
646                         return -ERANGE;
647                 }
648                 args->valuelen = sfe->valuelen;
649                 memcpy(args->value, &sfe->nameval[args->namelen],
650                                                     args->valuelen);
651                 return -EEXIST;
652         }
653         return -ENOATTR;
654 }
655
656 /*
657  * Convert from using the shortform to the leaf.
658  */
659 int
660 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
661 {
662         xfs_inode_t *dp;
663         xfs_attr_shortform_t *sf;
664         xfs_attr_sf_entry_t *sfe;
665         xfs_da_args_t nargs;
666         char *tmpbuffer;
667         int error, i, size;
668         xfs_dablk_t blkno;
669         struct xfs_buf *bp;
670         xfs_ifork_t *ifp;
671
672         trace_xfs_attr_sf_to_leaf(args);
673
674         dp = args->dp;
675         ifp = dp->i_afp;
676         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
677         size = be16_to_cpu(sf->hdr.totsize);
678         tmpbuffer = kmem_alloc(size, KM_SLEEP);
679         ASSERT(tmpbuffer != NULL);
680         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
681         sf = (xfs_attr_shortform_t *)tmpbuffer;
682
683         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
684         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
685
686         bp = NULL;
687         error = xfs_da_grow_inode(args, &blkno);
688         if (error) {
689                 /*
690                  * If we hit an IO error middle of the transaction inside
691                  * grow_inode(), we may have inconsistent data. Bail out.
692                  */
693                 if (error == -EIO)
694                         goto out;
695                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
696                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
697                 goto out;
698         }
699
700         ASSERT(blkno == 0);
701         error = xfs_attr3_leaf_create(args, blkno, &bp);
702         if (error) {
703                 error = xfs_da_shrink_inode(args, 0, bp);
704                 bp = NULL;
705                 if (error)
706                         goto out;
707                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
708                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
709                 goto out;
710         }
711
712         memset((char *)&nargs, 0, sizeof(nargs));
713         nargs.dp = dp;
714         nargs.geo = args->geo;
715         nargs.firstblock = args->firstblock;
716         nargs.flist = args->flist;
717         nargs.total = args->total;
718         nargs.whichfork = XFS_ATTR_FORK;
719         nargs.trans = args->trans;
720         nargs.op_flags = XFS_DA_OP_OKNOENT;
721
722         sfe = &sf->list[0];
723         for (i = 0; i < sf->hdr.count; i++) {
724                 nargs.name = sfe->nameval;
725                 nargs.namelen = sfe->namelen;
726                 nargs.value = &sfe->nameval[nargs.namelen];
727                 nargs.valuelen = sfe->valuelen;
728                 nargs.hashval = xfs_da_hashname(sfe->nameval,
729                                                 sfe->namelen);
730                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
731                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
732                 ASSERT(error == -ENOATTR);
733                 error = xfs_attr3_leaf_add(bp, &nargs);
734                 ASSERT(error != -ENOSPC);
735                 if (error)
736                         goto out;
737                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
738         }
739         error = 0;
740
741 out:
742         kmem_free(tmpbuffer);
743         return error;
744 }
745
746 /*
747  * Check a leaf attribute block to see if all the entries would fit into
748  * a shortform attribute list.
749  */
750 int
751 xfs_attr_shortform_allfit(
752         struct xfs_buf          *bp,
753         struct xfs_inode        *dp)
754 {
755         struct xfs_attr_leafblock *leaf;
756         struct xfs_attr_leaf_entry *entry;
757         xfs_attr_leaf_name_local_t *name_loc;
758         struct xfs_attr3_icleaf_hdr leafhdr;
759         int                     bytes;
760         int                     i;
761
762         leaf = bp->b_addr;
763         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
764         entry = xfs_attr3_leaf_entryp(leaf);
765
766         bytes = sizeof(struct xfs_attr_sf_hdr);
767         for (i = 0; i < leafhdr.count; entry++, i++) {
768                 if (entry->flags & XFS_ATTR_INCOMPLETE)
769                         continue;               /* don't copy partial entries */
770                 if (!(entry->flags & XFS_ATTR_LOCAL))
771                         return 0;
772                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
773                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
774                         return 0;
775                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
776                         return 0;
777                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
778                                 + name_loc->namelen
779                                 + be16_to_cpu(name_loc->valuelen);
780         }
781         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
782             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
783             (bytes == sizeof(struct xfs_attr_sf_hdr)))
784                 return -1;
785         return xfs_attr_shortform_bytesfit(dp, bytes);
786 }
787
788 /*
789  * Convert a leaf attribute list to shortform attribute list
790  */
791 int
792 xfs_attr3_leaf_to_shortform(
793         struct xfs_buf          *bp,
794         struct xfs_da_args      *args,
795         int                     forkoff)
796 {
797         struct xfs_attr_leafblock *leaf;
798         struct xfs_attr3_icleaf_hdr ichdr;
799         struct xfs_attr_leaf_entry *entry;
800         struct xfs_attr_leaf_name_local *name_loc;
801         struct xfs_da_args      nargs;
802         struct xfs_inode        *dp = args->dp;
803         char                    *tmpbuffer;
804         int                     error;
805         int                     i;
806
807         trace_xfs_attr_leaf_to_sf(args);
808
809         tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
810         if (!tmpbuffer)
811                 return -ENOMEM;
812
813         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
814
815         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
816         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
817         entry = xfs_attr3_leaf_entryp(leaf);
818
819         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
820         memset(bp->b_addr, 0, args->geo->blksize);
821
822         /*
823          * Clean out the prior contents of the attribute list.
824          */
825         error = xfs_da_shrink_inode(args, 0, bp);
826         if (error)
827                 goto out;
828
829         if (forkoff == -1) {
830                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
831                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
832                 xfs_attr_fork_reset(dp, args->trans);
833                 goto out;
834         }
835
836         xfs_attr_shortform_create(args);
837
838         /*
839          * Copy the attributes
840          */
841         memset((char *)&nargs, 0, sizeof(nargs));
842         nargs.geo = args->geo;
843         nargs.dp = dp;
844         nargs.firstblock = args->firstblock;
845         nargs.flist = args->flist;
846         nargs.total = args->total;
847         nargs.whichfork = XFS_ATTR_FORK;
848         nargs.trans = args->trans;
849         nargs.op_flags = XFS_DA_OP_OKNOENT;
850
851         for (i = 0; i < ichdr.count; entry++, i++) {
852                 if (entry->flags & XFS_ATTR_INCOMPLETE)
853                         continue;       /* don't copy partial entries */
854                 if (!entry->nameidx)
855                         continue;
856                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
857                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
858                 nargs.name = name_loc->nameval;
859                 nargs.namelen = name_loc->namelen;
860                 nargs.value = &name_loc->nameval[nargs.namelen];
861                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
862                 nargs.hashval = be32_to_cpu(entry->hashval);
863                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
864                 xfs_attr_shortform_add(&nargs, forkoff);
865         }
866         error = 0;
867
868 out:
869         kmem_free(tmpbuffer);
870         return error;
871 }
872
873 /*
874  * Convert from using a single leaf to a root node and a leaf.
875  */
876 int
877 xfs_attr3_leaf_to_node(
878         struct xfs_da_args      *args)
879 {
880         struct xfs_attr_leafblock *leaf;
881         struct xfs_attr3_icleaf_hdr icleafhdr;
882         struct xfs_attr_leaf_entry *entries;
883         struct xfs_da_node_entry *btree;
884         struct xfs_da3_icnode_hdr icnodehdr;
885         struct xfs_da_intnode   *node;
886         struct xfs_inode        *dp = args->dp;
887         struct xfs_mount        *mp = dp->i_mount;
888         struct xfs_buf          *bp1 = NULL;
889         struct xfs_buf          *bp2 = NULL;
890         xfs_dablk_t             blkno;
891         int                     error;
892
893         trace_xfs_attr_leaf_to_node(args);
894
895         error = xfs_da_grow_inode(args, &blkno);
896         if (error)
897                 goto out;
898         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
899         if (error)
900                 goto out;
901
902         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
903         if (error)
904                 goto out;
905
906         /* copy leaf to new buffer, update identifiers */
907         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
908         bp2->b_ops = bp1->b_ops;
909         memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
910         if (xfs_sb_version_hascrc(&mp->m_sb)) {
911                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
912                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
913         }
914         xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
915
916         /*
917          * Set up the new root node.
918          */
919         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
920         if (error)
921                 goto out;
922         node = bp1->b_addr;
923         dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
924         btree = dp->d_ops->node_tree_p(node);
925
926         leaf = bp2->b_addr;
927         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
928         entries = xfs_attr3_leaf_entryp(leaf);
929
930         /* both on-disk, don't endian-flip twice */
931         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
932         btree[0].before = cpu_to_be32(blkno);
933         icnodehdr.count = 1;
934         dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
935         xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
936         error = 0;
937 out:
938         return error;
939 }
940
941 /*========================================================================
942  * Routines used for growing the Btree.
943  *========================================================================*/
944
945 /*
946  * Create the initial contents of a leaf attribute list
947  * or a leaf in a node attribute list.
948  */
949 STATIC int
950 xfs_attr3_leaf_create(
951         struct xfs_da_args      *args,
952         xfs_dablk_t             blkno,
953         struct xfs_buf          **bpp)
954 {
955         struct xfs_attr_leafblock *leaf;
956         struct xfs_attr3_icleaf_hdr ichdr;
957         struct xfs_inode        *dp = args->dp;
958         struct xfs_mount        *mp = dp->i_mount;
959         struct xfs_buf          *bp;
960         int                     error;
961
962         trace_xfs_attr_leaf_create(args);
963
964         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
965                                             XFS_ATTR_FORK);
966         if (error)
967                 return error;
968         bp->b_ops = &xfs_attr3_leaf_buf_ops;
969         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
970         leaf = bp->b_addr;
971         memset(leaf, 0, args->geo->blksize);
972
973         memset(&ichdr, 0, sizeof(ichdr));
974         ichdr.firstused = args->geo->blksize;
975
976         if (xfs_sb_version_hascrc(&mp->m_sb)) {
977                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
978
979                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
980
981                 hdr3->blkno = cpu_to_be64(bp->b_bn);
982                 hdr3->owner = cpu_to_be64(dp->i_ino);
983                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
984
985                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
986         } else {
987                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
988                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
989         }
990         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
991
992         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
993         xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
994
995         *bpp = bp;
996         return 0;
997 }
998
999 /*
1000  * Split the leaf node, rebalance, then add the new entry.
1001  */
1002 int
1003 xfs_attr3_leaf_split(
1004         struct xfs_da_state     *state,
1005         struct xfs_da_state_blk *oldblk,
1006         struct xfs_da_state_blk *newblk)
1007 {
1008         xfs_dablk_t blkno;
1009         int error;
1010
1011         trace_xfs_attr_leaf_split(state->args);
1012
1013         /*
1014          * Allocate space for a new leaf node.
1015          */
1016         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1017         error = xfs_da_grow_inode(state->args, &blkno);
1018         if (error)
1019                 return error;
1020         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1021         if (error)
1022                 return error;
1023         newblk->blkno = blkno;
1024         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1025
1026         /*
1027          * Rebalance the entries across the two leaves.
1028          * NOTE: rebalance() currently depends on the 2nd block being empty.
1029          */
1030         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1031         error = xfs_da3_blk_link(state, oldblk, newblk);
1032         if (error)
1033                 return error;
1034
1035         /*
1036          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1037          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1038          * "new" attrs info.  Will need the "old" info to remove it later.
1039          *
1040          * Insert the "new" entry in the correct block.
1041          */
1042         if (state->inleaf) {
1043                 trace_xfs_attr_leaf_add_old(state->args);
1044                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1045         } else {
1046                 trace_xfs_attr_leaf_add_new(state->args);
1047                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1048         }
1049
1050         /*
1051          * Update last hashval in each block since we added the name.
1052          */
1053         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1054         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1055         return error;
1056 }
1057
1058 /*
1059  * Add a name to the leaf attribute list structure.
1060  */
1061 int
1062 xfs_attr3_leaf_add(
1063         struct xfs_buf          *bp,
1064         struct xfs_da_args      *args)
1065 {
1066         struct xfs_attr_leafblock *leaf;
1067         struct xfs_attr3_icleaf_hdr ichdr;
1068         int                     tablesize;
1069         int                     entsize;
1070         int                     sum;
1071         int                     tmp;
1072         int                     i;
1073
1074         trace_xfs_attr_leaf_add(args);
1075
1076         leaf = bp->b_addr;
1077         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1078         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1079         entsize = xfs_attr_leaf_newentsize(args, NULL);
1080
1081         /*
1082          * Search through freemap for first-fit on new name length.
1083          * (may need to figure in size of entry struct too)
1084          */
1085         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1086                                         + xfs_attr3_leaf_hdr_size(leaf);
1087         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1088                 if (tablesize > ichdr.firstused) {
1089                         sum += ichdr.freemap[i].size;
1090                         continue;
1091                 }
1092                 if (!ichdr.freemap[i].size)
1093                         continue;       /* no space in this map */
1094                 tmp = entsize;
1095                 if (ichdr.freemap[i].base < ichdr.firstused)
1096                         tmp += sizeof(xfs_attr_leaf_entry_t);
1097                 if (ichdr.freemap[i].size >= tmp) {
1098                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1099                         goto out_log_hdr;
1100                 }
1101                 sum += ichdr.freemap[i].size;
1102         }
1103
1104         /*
1105          * If there are no holes in the address space of the block,
1106          * and we don't have enough freespace, then compaction will do us
1107          * no good and we should just give up.
1108          */
1109         if (!ichdr.holes && sum < entsize)
1110                 return -ENOSPC;
1111
1112         /*
1113          * Compact the entries to coalesce free space.
1114          * This may change the hdr->count via dropping INCOMPLETE entries.
1115          */
1116         xfs_attr3_leaf_compact(args, &ichdr, bp);
1117
1118         /*
1119          * After compaction, the block is guaranteed to have only one
1120          * free region, in freemap[0].  If it is not big enough, give up.
1121          */
1122         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1123                 tmp = -ENOSPC;
1124                 goto out_log_hdr;
1125         }
1126
1127         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1128
1129 out_log_hdr:
1130         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1131         xfs_trans_log_buf(args->trans, bp,
1132                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1133                                 xfs_attr3_leaf_hdr_size(leaf)));
1134         return tmp;
1135 }
1136
1137 /*
1138  * Add a name to a leaf attribute list structure.
1139  */
1140 STATIC int
1141 xfs_attr3_leaf_add_work(
1142         struct xfs_buf          *bp,
1143         struct xfs_attr3_icleaf_hdr *ichdr,
1144         struct xfs_da_args      *args,
1145         int                     mapindex)
1146 {
1147         struct xfs_attr_leafblock *leaf;
1148         struct xfs_attr_leaf_entry *entry;
1149         struct xfs_attr_leaf_name_local *name_loc;
1150         struct xfs_attr_leaf_name_remote *name_rmt;
1151         struct xfs_mount        *mp;
1152         int                     tmp;
1153         int                     i;
1154
1155         trace_xfs_attr_leaf_add_work(args);
1156
1157         leaf = bp->b_addr;
1158         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1159         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1160
1161         /*
1162          * Force open some space in the entry array and fill it in.
1163          */
1164         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1165         if (args->index < ichdr->count) {
1166                 tmp  = ichdr->count - args->index;
1167                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1168                 memmove(entry + 1, entry, tmp);
1169                 xfs_trans_log_buf(args->trans, bp,
1170                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1171         }
1172         ichdr->count++;
1173
1174         /*
1175          * Allocate space for the new string (at the end of the run).
1176          */
1177         mp = args->trans->t_mountp;
1178         ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1179         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1180         ASSERT(ichdr->freemap[mapindex].size >=
1181                 xfs_attr_leaf_newentsize(args, NULL));
1182         ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1183         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1184
1185         ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1186
1187         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1188                                      ichdr->freemap[mapindex].size);
1189         entry->hashval = cpu_to_be32(args->hashval);
1190         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1191         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1192         if (args->op_flags & XFS_DA_OP_RENAME) {
1193                 entry->flags |= XFS_ATTR_INCOMPLETE;
1194                 if ((args->blkno2 == args->blkno) &&
1195                     (args->index2 <= args->index)) {
1196                         args->index2++;
1197                 }
1198         }
1199         xfs_trans_log_buf(args->trans, bp,
1200                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1201         ASSERT((args->index == 0) ||
1202                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1203         ASSERT((args->index == ichdr->count - 1) ||
1204                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1205
1206         /*
1207          * For "remote" attribute values, simply note that we need to
1208          * allocate space for the "remote" value.  We can't actually
1209          * allocate the extents in this transaction, and we can't decide
1210          * which blocks they should be as we might allocate more blocks
1211          * as part of this transaction (a split operation for example).
1212          */
1213         if (entry->flags & XFS_ATTR_LOCAL) {
1214                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1215                 name_loc->namelen = args->namelen;
1216                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1217                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1218                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1219                                    be16_to_cpu(name_loc->valuelen));
1220         } else {
1221                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1222                 name_rmt->namelen = args->namelen;
1223                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1224                 entry->flags |= XFS_ATTR_INCOMPLETE;
1225                 /* just in case */
1226                 name_rmt->valuelen = 0;
1227                 name_rmt->valueblk = 0;
1228                 args->rmtblkno = 1;
1229                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1230                 args->rmtvaluelen = args->valuelen;
1231         }
1232         xfs_trans_log_buf(args->trans, bp,
1233              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1234                                    xfs_attr_leaf_entsize(leaf, args->index)));
1235
1236         /*
1237          * Update the control info for this leaf node
1238          */
1239         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1240                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1241
1242         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1243                                         + xfs_attr3_leaf_hdr_size(leaf));
1244         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1245                                         + xfs_attr3_leaf_hdr_size(leaf);
1246
1247         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1248                 if (ichdr->freemap[i].base == tmp) {
1249                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1250                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1251                 }
1252         }
1253         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1254         return 0;
1255 }
1256
1257 /*
1258  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1259  */
1260 STATIC void
1261 xfs_attr3_leaf_compact(
1262         struct xfs_da_args      *args,
1263         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1264         struct xfs_buf          *bp)
1265 {
1266         struct xfs_attr_leafblock *leaf_src;
1267         struct xfs_attr_leafblock *leaf_dst;
1268         struct xfs_attr3_icleaf_hdr ichdr_src;
1269         struct xfs_trans        *trans = args->trans;
1270         char                    *tmpbuffer;
1271
1272         trace_xfs_attr_leaf_compact(args);
1273
1274         tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1275         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1276         memset(bp->b_addr, 0, args->geo->blksize);
1277         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1278         leaf_dst = bp->b_addr;
1279
1280         /*
1281          * Copy the on-disk header back into the destination buffer to ensure
1282          * all the information in the header that is not part of the incore
1283          * header structure is preserved.
1284          */
1285         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1286
1287         /* Initialise the incore headers */
1288         ichdr_src = *ichdr_dst; /* struct copy */
1289         ichdr_dst->firstused = args->geo->blksize;
1290         ichdr_dst->usedbytes = 0;
1291         ichdr_dst->count = 0;
1292         ichdr_dst->holes = 0;
1293         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1294         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1295                                                 ichdr_dst->freemap[0].base;
1296
1297         /* write the header back to initialise the underlying buffer */
1298         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1299
1300         /*
1301          * Copy all entry's in the same (sorted) order,
1302          * but allocate name/value pairs packed and in sequence.
1303          */
1304         xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1305                                 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1306         /*
1307          * this logs the entire buffer, but the caller must write the header
1308          * back to the buffer when it is finished modifying it.
1309          */
1310         xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1311
1312         kmem_free(tmpbuffer);
1313 }
1314
1315 /*
1316  * Compare two leaf blocks "order".
1317  * Return 0 unless leaf2 should go before leaf1.
1318  */
1319 static int
1320 xfs_attr3_leaf_order(
1321         struct xfs_buf  *leaf1_bp,
1322         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1323         struct xfs_buf  *leaf2_bp,
1324         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1325 {
1326         struct xfs_attr_leaf_entry *entries1;
1327         struct xfs_attr_leaf_entry *entries2;
1328
1329         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1330         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1331         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1332             ((be32_to_cpu(entries2[0].hashval) <
1333               be32_to_cpu(entries1[0].hashval)) ||
1334              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1335               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1336                 return 1;
1337         }
1338         return 0;
1339 }
1340
1341 int
1342 xfs_attr_leaf_order(
1343         struct xfs_buf  *leaf1_bp,
1344         struct xfs_buf  *leaf2_bp)
1345 {
1346         struct xfs_attr3_icleaf_hdr ichdr1;
1347         struct xfs_attr3_icleaf_hdr ichdr2;
1348
1349         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1350         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1351         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1352 }
1353
1354 /*
1355  * Redistribute the attribute list entries between two leaf nodes,
1356  * taking into account the size of the new entry.
1357  *
1358  * NOTE: if new block is empty, then it will get the upper half of the
1359  * old block.  At present, all (one) callers pass in an empty second block.
1360  *
1361  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1362  * to match what it is doing in splitting the attribute leaf block.  Those
1363  * values are used in "atomic rename" operations on attributes.  Note that
1364  * the "new" and "old" values can end up in different blocks.
1365  */
1366 STATIC void
1367 xfs_attr3_leaf_rebalance(
1368         struct xfs_da_state     *state,
1369         struct xfs_da_state_blk *blk1,
1370         struct xfs_da_state_blk *blk2)
1371 {
1372         struct xfs_da_args      *args;
1373         struct xfs_attr_leafblock *leaf1;
1374         struct xfs_attr_leafblock *leaf2;
1375         struct xfs_attr3_icleaf_hdr ichdr1;
1376         struct xfs_attr3_icleaf_hdr ichdr2;
1377         struct xfs_attr_leaf_entry *entries1;
1378         struct xfs_attr_leaf_entry *entries2;
1379         int                     count;
1380         int                     totallen;
1381         int                     max;
1382         int                     space;
1383         int                     swap;
1384
1385         /*
1386          * Set up environment.
1387          */
1388         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1389         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1390         leaf1 = blk1->bp->b_addr;
1391         leaf2 = blk2->bp->b_addr;
1392         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1393         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1394         ASSERT(ichdr2.count == 0);
1395         args = state->args;
1396
1397         trace_xfs_attr_leaf_rebalance(args);
1398
1399         /*
1400          * Check ordering of blocks, reverse if it makes things simpler.
1401          *
1402          * NOTE: Given that all (current) callers pass in an empty
1403          * second block, this code should never set "swap".
1404          */
1405         swap = 0;
1406         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1407                 struct xfs_da_state_blk *tmp_blk;
1408                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1409
1410                 tmp_blk = blk1;
1411                 blk1 = blk2;
1412                 blk2 = tmp_blk;
1413
1414                 /* struct copies to swap them rather than reconverting */
1415                 tmp_ichdr = ichdr1;
1416                 ichdr1 = ichdr2;
1417                 ichdr2 = tmp_ichdr;
1418
1419                 leaf1 = blk1->bp->b_addr;
1420                 leaf2 = blk2->bp->b_addr;
1421                 swap = 1;
1422         }
1423
1424         /*
1425          * Examine entries until we reduce the absolute difference in
1426          * byte usage between the two blocks to a minimum.  Then get
1427          * the direction to copy and the number of elements to move.
1428          *
1429          * "inleaf" is true if the new entry should be inserted into blk1.
1430          * If "swap" is also true, then reverse the sense of "inleaf".
1431          */
1432         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1433                                                       blk2, &ichdr2,
1434                                                       &count, &totallen);
1435         if (swap)
1436                 state->inleaf = !state->inleaf;
1437
1438         /*
1439          * Move any entries required from leaf to leaf:
1440          */
1441         if (count < ichdr1.count) {
1442                 /*
1443                  * Figure the total bytes to be added to the destination leaf.
1444                  */
1445                 /* number entries being moved */
1446                 count = ichdr1.count - count;
1447                 space  = ichdr1.usedbytes - totallen;
1448                 space += count * sizeof(xfs_attr_leaf_entry_t);
1449
1450                 /*
1451                  * leaf2 is the destination, compact it if it looks tight.
1452                  */
1453                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1454                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1455                 if (space > max)
1456                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1457
1458                 /*
1459                  * Move high entries from leaf1 to low end of leaf2.
1460                  */
1461                 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1462                                 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1463
1464         } else if (count > ichdr1.count) {
1465                 /*
1466                  * I assert that since all callers pass in an empty
1467                  * second buffer, this code should never execute.
1468                  */
1469                 ASSERT(0);
1470
1471                 /*
1472                  * Figure the total bytes to be added to the destination leaf.
1473                  */
1474                 /* number entries being moved */
1475                 count -= ichdr1.count;
1476                 space  = totallen - ichdr1.usedbytes;
1477                 space += count * sizeof(xfs_attr_leaf_entry_t);
1478
1479                 /*
1480                  * leaf1 is the destination, compact it if it looks tight.
1481                  */
1482                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1483                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1484                 if (space > max)
1485                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1486
1487                 /*
1488                  * Move low entries from leaf2 to high end of leaf1.
1489                  */
1490                 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1491                                         ichdr1.count, count);
1492         }
1493
1494         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1495         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1496         xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1497         xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1498
1499         /*
1500          * Copy out last hashval in each block for B-tree code.
1501          */
1502         entries1 = xfs_attr3_leaf_entryp(leaf1);
1503         entries2 = xfs_attr3_leaf_entryp(leaf2);
1504         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1505         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1506
1507         /*
1508          * Adjust the expected index for insertion.
1509          * NOTE: this code depends on the (current) situation that the
1510          * second block was originally empty.
1511          *
1512          * If the insertion point moved to the 2nd block, we must adjust
1513          * the index.  We must also track the entry just following the
1514          * new entry for use in an "atomic rename" operation, that entry
1515          * is always the "old" entry and the "new" entry is what we are
1516          * inserting.  The index/blkno fields refer to the "old" entry,
1517          * while the index2/blkno2 fields refer to the "new" entry.
1518          */
1519         if (blk1->index > ichdr1.count) {
1520                 ASSERT(state->inleaf == 0);
1521                 blk2->index = blk1->index - ichdr1.count;
1522                 args->index = args->index2 = blk2->index;
1523                 args->blkno = args->blkno2 = blk2->blkno;
1524         } else if (blk1->index == ichdr1.count) {
1525                 if (state->inleaf) {
1526                         args->index = blk1->index;
1527                         args->blkno = blk1->blkno;
1528                         args->index2 = 0;
1529                         args->blkno2 = blk2->blkno;
1530                 } else {
1531                         /*
1532                          * On a double leaf split, the original attr location
1533                          * is already stored in blkno2/index2, so don't
1534                          * overwrite it overwise we corrupt the tree.
1535                          */
1536                         blk2->index = blk1->index - ichdr1.count;
1537                         args->index = blk2->index;
1538                         args->blkno = blk2->blkno;
1539                         if (!state->extravalid) {
1540                                 /*
1541                                  * set the new attr location to match the old
1542                                  * one and let the higher level split code
1543                                  * decide where in the leaf to place it.
1544                                  */
1545                                 args->index2 = blk2->index;
1546                                 args->blkno2 = blk2->blkno;
1547                         }
1548                 }
1549         } else {
1550                 ASSERT(state->inleaf == 1);
1551                 args->index = args->index2 = blk1->index;
1552                 args->blkno = args->blkno2 = blk1->blkno;
1553         }
1554 }
1555
1556 /*
1557  * Examine entries until we reduce the absolute difference in
1558  * byte usage between the two blocks to a minimum.
1559  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1560  * GROT: there will always be enough room in either block for a new entry.
1561  * GROT: Do a double-split for this case?
1562  */
1563 STATIC int
1564 xfs_attr3_leaf_figure_balance(
1565         struct xfs_da_state             *state,
1566         struct xfs_da_state_blk         *blk1,
1567         struct xfs_attr3_icleaf_hdr     *ichdr1,
1568         struct xfs_da_state_blk         *blk2,
1569         struct xfs_attr3_icleaf_hdr     *ichdr2,
1570         int                             *countarg,
1571         int                             *usedbytesarg)
1572 {
1573         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1574         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1575         struct xfs_attr_leaf_entry      *entry;
1576         int                             count;
1577         int                             max;
1578         int                             index;
1579         int                             totallen = 0;
1580         int                             half;
1581         int                             lastdelta;
1582         int                             foundit = 0;
1583         int                             tmp;
1584
1585         /*
1586          * Examine entries until we reduce the absolute difference in
1587          * byte usage between the two blocks to a minimum.
1588          */
1589         max = ichdr1->count + ichdr2->count;
1590         half = (max + 1) * sizeof(*entry);
1591         half += ichdr1->usedbytes + ichdr2->usedbytes +
1592                         xfs_attr_leaf_newentsize(state->args, NULL);
1593         half /= 2;
1594         lastdelta = state->args->geo->blksize;
1595         entry = xfs_attr3_leaf_entryp(leaf1);
1596         for (count = index = 0; count < max; entry++, index++, count++) {
1597
1598 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1599                 /*
1600                  * The new entry is in the first block, account for it.
1601                  */
1602                 if (count == blk1->index) {
1603                         tmp = totallen + sizeof(*entry) +
1604                                 xfs_attr_leaf_newentsize(state->args, NULL);
1605                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1606                                 break;
1607                         lastdelta = XFS_ATTR_ABS(half - tmp);
1608                         totallen = tmp;
1609                         foundit = 1;
1610                 }
1611
1612                 /*
1613                  * Wrap around into the second block if necessary.
1614                  */
1615                 if (count == ichdr1->count) {
1616                         leaf1 = leaf2;
1617                         entry = xfs_attr3_leaf_entryp(leaf1);
1618                         index = 0;
1619                 }
1620
1621                 /*
1622                  * Figure out if next leaf entry would be too much.
1623                  */
1624                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1625                                                                         index);
1626                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1627                         break;
1628                 lastdelta = XFS_ATTR_ABS(half - tmp);
1629                 totallen = tmp;
1630 #undef XFS_ATTR_ABS
1631         }
1632
1633         /*
1634          * Calculate the number of usedbytes that will end up in lower block.
1635          * If new entry not in lower block, fix up the count.
1636          */
1637         totallen -= count * sizeof(*entry);
1638         if (foundit) {
1639                 totallen -= sizeof(*entry) +
1640                                 xfs_attr_leaf_newentsize(state->args, NULL);
1641         }
1642
1643         *countarg = count;
1644         *usedbytesarg = totallen;
1645         return foundit;
1646 }
1647
1648 /*========================================================================
1649  * Routines used for shrinking the Btree.
1650  *========================================================================*/
1651
1652 /*
1653  * Check a leaf block and its neighbors to see if the block should be
1654  * collapsed into one or the other neighbor.  Always keep the block
1655  * with the smaller block number.
1656  * If the current block is over 50% full, don't try to join it, return 0.
1657  * If the block is empty, fill in the state structure and return 2.
1658  * If it can be collapsed, fill in the state structure and return 1.
1659  * If nothing can be done, return 0.
1660  *
1661  * GROT: allow for INCOMPLETE entries in calculation.
1662  */
1663 int
1664 xfs_attr3_leaf_toosmall(
1665         struct xfs_da_state     *state,
1666         int                     *action)
1667 {
1668         struct xfs_attr_leafblock *leaf;
1669         struct xfs_da_state_blk *blk;
1670         struct xfs_attr3_icleaf_hdr ichdr;
1671         struct xfs_buf          *bp;
1672         xfs_dablk_t             blkno;
1673         int                     bytes;
1674         int                     forward;
1675         int                     error;
1676         int                     retval;
1677         int                     i;
1678
1679         trace_xfs_attr_leaf_toosmall(state->args);
1680
1681         /*
1682          * Check for the degenerate case of the block being over 50% full.
1683          * If so, it's not worth even looking to see if we might be able
1684          * to coalesce with a sibling.
1685          */
1686         blk = &state->path.blk[ state->path.active-1 ];
1687         leaf = blk->bp->b_addr;
1688         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1689         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1690                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1691                 ichdr.usedbytes;
1692         if (bytes > (state->args->geo->blksize >> 1)) {
1693                 *action = 0;    /* blk over 50%, don't try to join */
1694                 return 0;
1695         }
1696
1697         /*
1698          * Check for the degenerate case of the block being empty.
1699          * If the block is empty, we'll simply delete it, no need to
1700          * coalesce it with a sibling block.  We choose (arbitrarily)
1701          * to merge with the forward block unless it is NULL.
1702          */
1703         if (ichdr.count == 0) {
1704                 /*
1705                  * Make altpath point to the block we want to keep and
1706                  * path point to the block we want to drop (this one).
1707                  */
1708                 forward = (ichdr.forw != 0);
1709                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1710                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1711                                                  0, &retval);
1712                 if (error)
1713                         return error;
1714                 if (retval) {
1715                         *action = 0;
1716                 } else {
1717                         *action = 2;
1718                 }
1719                 return 0;
1720         }
1721
1722         /*
1723          * Examine each sibling block to see if we can coalesce with
1724          * at least 25% free space to spare.  We need to figure out
1725          * whether to merge with the forward or the backward block.
1726          * We prefer coalescing with the lower numbered sibling so as
1727          * to shrink an attribute list over time.
1728          */
1729         /* start with smaller blk num */
1730         forward = ichdr.forw < ichdr.back;
1731         for (i = 0; i < 2; forward = !forward, i++) {
1732                 struct xfs_attr3_icleaf_hdr ichdr2;
1733                 if (forward)
1734                         blkno = ichdr.forw;
1735                 else
1736                         blkno = ichdr.back;
1737                 if (blkno == 0)
1738                         continue;
1739                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1740                                         blkno, -1, &bp);
1741                 if (error)
1742                         return error;
1743
1744                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1745
1746                 bytes = state->args->geo->blksize -
1747                         (state->args->geo->blksize >> 2) -
1748                         ichdr.usedbytes - ichdr2.usedbytes -
1749                         ((ichdr.count + ichdr2.count) *
1750                                         sizeof(xfs_attr_leaf_entry_t)) -
1751                         xfs_attr3_leaf_hdr_size(leaf);
1752
1753                 xfs_trans_brelse(state->args->trans, bp);
1754                 if (bytes >= 0)
1755                         break;  /* fits with at least 25% to spare */
1756         }
1757         if (i >= 2) {
1758                 *action = 0;
1759                 return 0;
1760         }
1761
1762         /*
1763          * Make altpath point to the block we want to keep (the lower
1764          * numbered block) and path point to the block we want to drop.
1765          */
1766         memcpy(&state->altpath, &state->path, sizeof(state->path));
1767         if (blkno < blk->blkno) {
1768                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1769                                                  0, &retval);
1770         } else {
1771                 error = xfs_da3_path_shift(state, &state->path, forward,
1772                                                  0, &retval);
1773         }
1774         if (error)
1775                 return error;
1776         if (retval) {
1777                 *action = 0;
1778         } else {
1779                 *action = 1;
1780         }
1781         return 0;
1782 }
1783
1784 /*
1785  * Remove a name from the leaf attribute list structure.
1786  *
1787  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1788  * If two leaves are 37% full, when combined they will leave 25% free.
1789  */
1790 int
1791 xfs_attr3_leaf_remove(
1792         struct xfs_buf          *bp,
1793         struct xfs_da_args      *args)
1794 {
1795         struct xfs_attr_leafblock *leaf;
1796         struct xfs_attr3_icleaf_hdr ichdr;
1797         struct xfs_attr_leaf_entry *entry;
1798         int                     before;
1799         int                     after;
1800         int                     smallest;
1801         int                     entsize;
1802         int                     tablesize;
1803         int                     tmp;
1804         int                     i;
1805
1806         trace_xfs_attr_leaf_remove(args);
1807
1808         leaf = bp->b_addr;
1809         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1810
1811         ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1812         ASSERT(args->index >= 0 && args->index < ichdr.count);
1813         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1814                                         xfs_attr3_leaf_hdr_size(leaf));
1815
1816         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1817
1818         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1819         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1820
1821         /*
1822          * Scan through free region table:
1823          *    check for adjacency of free'd entry with an existing one,
1824          *    find smallest free region in case we need to replace it,
1825          *    adjust any map that borders the entry table,
1826          */
1827         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1828                                         + xfs_attr3_leaf_hdr_size(leaf);
1829         tmp = ichdr.freemap[0].size;
1830         before = after = -1;
1831         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1832         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1833         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1834                 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1835                 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1836                 if (ichdr.freemap[i].base == tablesize) {
1837                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1838                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1839                 }
1840
1841                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1842                                 be16_to_cpu(entry->nameidx)) {
1843                         before = i;
1844                 } else if (ichdr.freemap[i].base ==
1845                                 (be16_to_cpu(entry->nameidx) + entsize)) {
1846                         after = i;
1847                 } else if (ichdr.freemap[i].size < tmp) {
1848                         tmp = ichdr.freemap[i].size;
1849                         smallest = i;
1850                 }
1851         }
1852
1853         /*
1854          * Coalesce adjacent freemap regions,
1855          * or replace the smallest region.
1856          */
1857         if ((before >= 0) || (after >= 0)) {
1858                 if ((before >= 0) && (after >= 0)) {
1859                         ichdr.freemap[before].size += entsize;
1860                         ichdr.freemap[before].size += ichdr.freemap[after].size;
1861                         ichdr.freemap[after].base = 0;
1862                         ichdr.freemap[after].size = 0;
1863                 } else if (before >= 0) {
1864                         ichdr.freemap[before].size += entsize;
1865                 } else {
1866                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1867                         ichdr.freemap[after].size += entsize;
1868                 }
1869         } else {
1870                 /*
1871                  * Replace smallest region (if it is smaller than free'd entry)
1872                  */
1873                 if (ichdr.freemap[smallest].size < entsize) {
1874                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1875                         ichdr.freemap[smallest].size = entsize;
1876                 }
1877         }
1878
1879         /*
1880          * Did we remove the first entry?
1881          */
1882         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1883                 smallest = 1;
1884         else
1885                 smallest = 0;
1886
1887         /*
1888          * Compress the remaining entries and zero out the removed stuff.
1889          */
1890         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1891         ichdr.usedbytes -= entsize;
1892         xfs_trans_log_buf(args->trans, bp,
1893              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1894                                    entsize));
1895
1896         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1897         memmove(entry, entry + 1, tmp);
1898         ichdr.count--;
1899         xfs_trans_log_buf(args->trans, bp,
1900             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1901
1902         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1903         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1904
1905         /*
1906          * If we removed the first entry, re-find the first used byte
1907          * in the name area.  Note that if the entry was the "firstused",
1908          * then we don't have a "hole" in our block resulting from
1909          * removing the name.
1910          */
1911         if (smallest) {
1912                 tmp = args->geo->blksize;
1913                 entry = xfs_attr3_leaf_entryp(leaf);
1914                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1915                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1916                         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1917
1918                         if (be16_to_cpu(entry->nameidx) < tmp)
1919                                 tmp = be16_to_cpu(entry->nameidx);
1920                 }
1921                 ichdr.firstused = tmp;
1922                 if (!ichdr.firstused)
1923                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
1924         } else {
1925                 ichdr.holes = 1;        /* mark as needing compaction */
1926         }
1927         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1928         xfs_trans_log_buf(args->trans, bp,
1929                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1930                                           xfs_attr3_leaf_hdr_size(leaf)));
1931
1932         /*
1933          * Check if leaf is less than 50% full, caller may want to
1934          * "join" the leaf with a sibling if so.
1935          */
1936         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
1937               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
1938
1939         return tmp < args->geo->magicpct; /* leaf is < 37% full */
1940 }
1941
1942 /*
1943  * Move all the attribute list entries from drop_leaf into save_leaf.
1944  */
1945 void
1946 xfs_attr3_leaf_unbalance(
1947         struct xfs_da_state     *state,
1948         struct xfs_da_state_blk *drop_blk,
1949         struct xfs_da_state_blk *save_blk)
1950 {
1951         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
1952         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
1953         struct xfs_attr3_icleaf_hdr drophdr;
1954         struct xfs_attr3_icleaf_hdr savehdr;
1955         struct xfs_attr_leaf_entry *entry;
1956
1957         trace_xfs_attr_leaf_unbalance(state->args);
1958
1959         drop_leaf = drop_blk->bp->b_addr;
1960         save_leaf = save_blk->bp->b_addr;
1961         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1962         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
1963         entry = xfs_attr3_leaf_entryp(drop_leaf);
1964
1965         /*
1966          * Save last hashval from dying block for later Btree fixup.
1967          */
1968         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
1969
1970         /*
1971          * Check if we need a temp buffer, or can we do it in place.
1972          * Note that we don't check "leaf" for holes because we will
1973          * always be dropping it, toosmall() decided that for us already.
1974          */
1975         if (savehdr.holes == 0) {
1976                 /*
1977                  * dest leaf has no holes, so we add there.  May need
1978                  * to make some room in the entry array.
1979                  */
1980                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
1981                                          drop_blk->bp, &drophdr)) {
1982                         xfs_attr3_leaf_moveents(state->args,
1983                                                 drop_leaf, &drophdr, 0,
1984                                                 save_leaf, &savehdr, 0,
1985                                                 drophdr.count);
1986                 } else {
1987                         xfs_attr3_leaf_moveents(state->args,
1988                                                 drop_leaf, &drophdr, 0,
1989                                                 save_leaf, &savehdr,
1990                                                 savehdr.count, drophdr.count);
1991                 }
1992         } else {
1993                 /*
1994                  * Destination has holes, so we make a temporary copy
1995                  * of the leaf and add them both to that.
1996                  */
1997                 struct xfs_attr_leafblock *tmp_leaf;
1998                 struct xfs_attr3_icleaf_hdr tmphdr;
1999
2000                 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2001
2002                 /*
2003                  * Copy the header into the temp leaf so that all the stuff
2004                  * not in the incore header is present and gets copied back in
2005                  * once we've moved all the entries.
2006                  */
2007                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2008
2009                 memset(&tmphdr, 0, sizeof(tmphdr));
2010                 tmphdr.magic = savehdr.magic;
2011                 tmphdr.forw = savehdr.forw;
2012                 tmphdr.back = savehdr.back;
2013                 tmphdr.firstused = state->args->geo->blksize;
2014
2015                 /* write the header to the temp buffer to initialise it */
2016                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2017
2018                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2019                                          drop_blk->bp, &drophdr)) {
2020                         xfs_attr3_leaf_moveents(state->args,
2021                                                 drop_leaf, &drophdr, 0,
2022                                                 tmp_leaf, &tmphdr, 0,
2023                                                 drophdr.count);
2024                         xfs_attr3_leaf_moveents(state->args,
2025                                                 save_leaf, &savehdr, 0,
2026                                                 tmp_leaf, &tmphdr, tmphdr.count,
2027                                                 savehdr.count);
2028                 } else {
2029                         xfs_attr3_leaf_moveents(state->args,
2030                                                 save_leaf, &savehdr, 0,
2031                                                 tmp_leaf, &tmphdr, 0,
2032                                                 savehdr.count);
2033                         xfs_attr3_leaf_moveents(state->args,
2034                                                 drop_leaf, &drophdr, 0,
2035                                                 tmp_leaf, &tmphdr, tmphdr.count,
2036                                                 drophdr.count);
2037                 }
2038                 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2039                 savehdr = tmphdr; /* struct copy */
2040                 kmem_free(tmp_leaf);
2041         }
2042
2043         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2044         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2045                                            state->args->geo->blksize - 1);
2046
2047         /*
2048          * Copy out last hashval in each block for B-tree code.
2049          */
2050         entry = xfs_attr3_leaf_entryp(save_leaf);
2051         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2052 }
2053
2054 /*========================================================================
2055  * Routines used for finding things in the Btree.
2056  *========================================================================*/
2057
2058 /*
2059  * Look up a name in a leaf attribute list structure.
2060  * This is the internal routine, it uses the caller's buffer.
2061  *
2062  * Note that duplicate keys are allowed, but only check within the
2063  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2064  *
2065  * Return in args->index the index into the entry[] array of either
2066  * the found entry, or where the entry should have been (insert before
2067  * that entry).
2068  *
2069  * Don't change the args->value unless we find the attribute.
2070  */
2071 int
2072 xfs_attr3_leaf_lookup_int(
2073         struct xfs_buf          *bp,
2074         struct xfs_da_args      *args)
2075 {
2076         struct xfs_attr_leafblock *leaf;
2077         struct xfs_attr3_icleaf_hdr ichdr;
2078         struct xfs_attr_leaf_entry *entry;
2079         struct xfs_attr_leaf_entry *entries;
2080         struct xfs_attr_leaf_name_local *name_loc;
2081         struct xfs_attr_leaf_name_remote *name_rmt;
2082         xfs_dahash_t            hashval;
2083         int                     probe;
2084         int                     span;
2085
2086         trace_xfs_attr_leaf_lookup(args);
2087
2088         leaf = bp->b_addr;
2089         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2090         entries = xfs_attr3_leaf_entryp(leaf);
2091         ASSERT(ichdr.count < args->geo->blksize / 8);
2092
2093         /*
2094          * Binary search.  (note: small blocks will skip this loop)
2095          */
2096         hashval = args->hashval;
2097         probe = span = ichdr.count / 2;
2098         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2099                 span /= 2;
2100                 if (be32_to_cpu(entry->hashval) < hashval)
2101                         probe += span;
2102                 else if (be32_to_cpu(entry->hashval) > hashval)
2103                         probe -= span;
2104                 else
2105                         break;
2106         }
2107         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2108         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2109
2110         /*
2111          * Since we may have duplicate hashval's, find the first matching
2112          * hashval in the leaf.
2113          */
2114         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2115                 entry--;
2116                 probe--;
2117         }
2118         while (probe < ichdr.count &&
2119                be32_to_cpu(entry->hashval) < hashval) {
2120                 entry++;
2121                 probe++;
2122         }
2123         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2124                 args->index = probe;
2125                 return -ENOATTR;
2126         }
2127
2128         /*
2129          * Duplicate keys may be present, so search all of them for a match.
2130          */
2131         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2132                         entry++, probe++) {
2133 /*
2134  * GROT: Add code to remove incomplete entries.
2135  */
2136                 /*
2137                  * If we are looking for INCOMPLETE entries, show only those.
2138                  * If we are looking for complete entries, show only those.
2139                  */
2140                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2141                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2142                         continue;
2143                 }
2144                 if (entry->flags & XFS_ATTR_LOCAL) {
2145                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2146                         if (name_loc->namelen != args->namelen)
2147                                 continue;
2148                         if (memcmp(args->name, name_loc->nameval,
2149                                                         args->namelen) != 0)
2150                                 continue;
2151                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2152                                 continue;
2153                         args->index = probe;
2154                         return -EEXIST;
2155                 } else {
2156                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2157                         if (name_rmt->namelen != args->namelen)
2158                                 continue;
2159                         if (memcmp(args->name, name_rmt->name,
2160                                                         args->namelen) != 0)
2161                                 continue;
2162                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2163                                 continue;
2164                         args->index = probe;
2165                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2166                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2167                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2168                                                         args->dp->i_mount,
2169                                                         args->rmtvaluelen);
2170                         return -EEXIST;
2171                 }
2172         }
2173         args->index = probe;
2174         return -ENOATTR;
2175 }
2176
2177 /*
2178  * Get the value associated with an attribute name from a leaf attribute
2179  * list structure.
2180  */
2181 int
2182 xfs_attr3_leaf_getvalue(
2183         struct xfs_buf          *bp,
2184         struct xfs_da_args      *args)
2185 {
2186         struct xfs_attr_leafblock *leaf;
2187         struct xfs_attr3_icleaf_hdr ichdr;
2188         struct xfs_attr_leaf_entry *entry;
2189         struct xfs_attr_leaf_name_local *name_loc;
2190         struct xfs_attr_leaf_name_remote *name_rmt;
2191         int                     valuelen;
2192
2193         leaf = bp->b_addr;
2194         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2195         ASSERT(ichdr.count < args->geo->blksize / 8);
2196         ASSERT(args->index < ichdr.count);
2197
2198         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2199         if (entry->flags & XFS_ATTR_LOCAL) {
2200                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2201                 ASSERT(name_loc->namelen == args->namelen);
2202                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2203                 valuelen = be16_to_cpu(name_loc->valuelen);
2204                 if (args->flags & ATTR_KERNOVAL) {
2205                         args->valuelen = valuelen;
2206                         return 0;
2207                 }
2208                 if (args->valuelen < valuelen) {
2209                         args->valuelen = valuelen;
2210                         return -ERANGE;
2211                 }
2212                 args->valuelen = valuelen;
2213                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2214         } else {
2215                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2216                 ASSERT(name_rmt->namelen == args->namelen);
2217                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2218                 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2219                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2220                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2221                                                        args->rmtvaluelen);
2222                 if (args->flags & ATTR_KERNOVAL) {
2223                         args->valuelen = args->rmtvaluelen;
2224                         return 0;
2225                 }
2226                 if (args->valuelen < args->rmtvaluelen) {
2227                         args->valuelen = args->rmtvaluelen;
2228                         return -ERANGE;
2229                 }
2230                 args->valuelen = args->rmtvaluelen;
2231         }
2232         return 0;
2233 }
2234
2235 /*========================================================================
2236  * Utility routines.
2237  *========================================================================*/
2238
2239 /*
2240  * Move the indicated entries from one leaf to another.
2241  * NOTE: this routine modifies both source and destination leaves.
2242  */
2243 /*ARGSUSED*/
2244 STATIC void
2245 xfs_attr3_leaf_moveents(
2246         struct xfs_da_args              *args,
2247         struct xfs_attr_leafblock       *leaf_s,
2248         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2249         int                             start_s,
2250         struct xfs_attr_leafblock       *leaf_d,
2251         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2252         int                             start_d,
2253         int                             count)
2254 {
2255         struct xfs_attr_leaf_entry      *entry_s;
2256         struct xfs_attr_leaf_entry      *entry_d;
2257         int                             desti;
2258         int                             tmp;
2259         int                             i;
2260
2261         /*
2262          * Check for nothing to do.
2263          */
2264         if (count == 0)
2265                 return;
2266
2267         /*
2268          * Set up environment.
2269          */
2270         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2271                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2272         ASSERT(ichdr_s->magic == ichdr_d->magic);
2273         ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2274         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2275                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2276         ASSERT(ichdr_d->count < args->geo->blksize / 8);
2277         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2278                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2279
2280         ASSERT(start_s < ichdr_s->count);
2281         ASSERT(start_d <= ichdr_d->count);
2282         ASSERT(count <= ichdr_s->count);
2283
2284
2285         /*
2286          * Move the entries in the destination leaf up to make a hole?
2287          */
2288         if (start_d < ichdr_d->count) {
2289                 tmp  = ichdr_d->count - start_d;
2290                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2291                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2292                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2293                 memmove(entry_d, entry_s, tmp);
2294         }
2295
2296         /*
2297          * Copy all entry's in the same (sorted) order,
2298          * but allocate attribute info packed and in sequence.
2299          */
2300         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2301         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2302         desti = start_d;
2303         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2304                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2305                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2306 #ifdef GROT
2307                 /*
2308                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2309                  * may also need to change the insertion index.  Code turned
2310                  * off for 6.2, should be revisited later.
2311                  */
2312                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2313                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2314                         ichdr_s->usedbytes -= tmp;
2315                         ichdr_s->count -= 1;
2316                         entry_d--;      /* to compensate for ++ in loop hdr */
2317                         desti--;
2318                         if ((start_s + i) < offset)
2319                                 result++;       /* insertion index adjustment */
2320                 } else {
2321 #endif /* GROT */
2322                         ichdr_d->firstused -= tmp;
2323                         /* both on-disk, don't endian flip twice */
2324                         entry_d->hashval = entry_s->hashval;
2325                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2326                         entry_d->flags = entry_s->flags;
2327                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2328                                                         <= args->geo->blksize);
2329                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2330                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2331                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2332                                                         <= args->geo->blksize);
2333                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2334                         ichdr_s->usedbytes -= tmp;
2335                         ichdr_d->usedbytes += tmp;
2336                         ichdr_s->count -= 1;
2337                         ichdr_d->count += 1;
2338                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2339                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2340                         ASSERT(ichdr_d->firstused >= tmp);
2341 #ifdef GROT
2342                 }
2343 #endif /* GROT */
2344         }
2345
2346         /*
2347          * Zero out the entries we just copied.
2348          */
2349         if (start_s == ichdr_s->count) {
2350                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2351                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2352                 ASSERT(((char *)entry_s + tmp) <=
2353                        ((char *)leaf_s + args->geo->blksize));
2354                 memset(entry_s, 0, tmp);
2355         } else {
2356                 /*
2357                  * Move the remaining entries down to fill the hole,
2358                  * then zero the entries at the top.
2359                  */
2360                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2361                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2362                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2363                 memmove(entry_d, entry_s, tmp);
2364
2365                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2366                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2367                 ASSERT(((char *)entry_s + tmp) <=
2368                        ((char *)leaf_s + args->geo->blksize));
2369                 memset(entry_s, 0, tmp);
2370         }
2371
2372         /*
2373          * Fill in the freemap information
2374          */
2375         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2376         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2377         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2378         ichdr_d->freemap[1].base = 0;
2379         ichdr_d->freemap[2].base = 0;
2380         ichdr_d->freemap[1].size = 0;
2381         ichdr_d->freemap[2].size = 0;
2382         ichdr_s->holes = 1;     /* leaf may not be compact */
2383 }
2384
2385 /*
2386  * Pick up the last hashvalue from a leaf block.
2387  */
2388 xfs_dahash_t
2389 xfs_attr_leaf_lasthash(
2390         struct xfs_buf  *bp,
2391         int             *count)
2392 {
2393         struct xfs_attr3_icleaf_hdr ichdr;
2394         struct xfs_attr_leaf_entry *entries;
2395
2396         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2397         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2398         if (count)
2399                 *count = ichdr.count;
2400         if (!ichdr.count)
2401                 return 0;
2402         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2403 }
2404
2405 /*
2406  * Calculate the number of bytes used to store the indicated attribute
2407  * (whether local or remote only calculate bytes in this block).
2408  */
2409 STATIC int
2410 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2411 {
2412         struct xfs_attr_leaf_entry *entries;
2413         xfs_attr_leaf_name_local_t *name_loc;
2414         xfs_attr_leaf_name_remote_t *name_rmt;
2415         int size;
2416
2417         entries = xfs_attr3_leaf_entryp(leaf);
2418         if (entries[index].flags & XFS_ATTR_LOCAL) {
2419                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2420                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2421                                                    be16_to_cpu(name_loc->valuelen));
2422         } else {
2423                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2424                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2425         }
2426         return size;
2427 }
2428
2429 /*
2430  * Calculate the number of bytes that would be required to store the new
2431  * attribute (whether local or remote only calculate bytes in this block).
2432  * This routine decides as a side effect whether the attribute will be
2433  * a "local" or a "remote" attribute.
2434  */
2435 int
2436 xfs_attr_leaf_newentsize(
2437         struct xfs_da_args      *args,
2438         int                     *local)
2439 {
2440         int                     size;
2441
2442         size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2443         if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2444                 if (local)
2445                         *local = 1;
2446                 return size;
2447         }
2448         if (local)
2449                 *local = 0;
2450         return xfs_attr_leaf_entsize_remote(args->namelen);
2451 }
2452
2453
2454 /*========================================================================
2455  * Manage the INCOMPLETE flag in a leaf entry
2456  *========================================================================*/
2457
2458 /*
2459  * Clear the INCOMPLETE flag on an entry in a leaf block.
2460  */
2461 int
2462 xfs_attr3_leaf_clearflag(
2463         struct xfs_da_args      *args)
2464 {
2465         struct xfs_attr_leafblock *leaf;
2466         struct xfs_attr_leaf_entry *entry;
2467         struct xfs_attr_leaf_name_remote *name_rmt;
2468         struct xfs_buf          *bp;
2469         int                     error;
2470 #ifdef DEBUG
2471         struct xfs_attr3_icleaf_hdr ichdr;
2472         xfs_attr_leaf_name_local_t *name_loc;
2473         int namelen;
2474         char *name;
2475 #endif /* DEBUG */
2476
2477         trace_xfs_attr_leaf_clearflag(args);
2478         /*
2479          * Set up the operation.
2480          */
2481         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2482         if (error)
2483                 return error;
2484
2485         leaf = bp->b_addr;
2486         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2487         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2488
2489 #ifdef DEBUG
2490         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2491         ASSERT(args->index < ichdr.count);
2492         ASSERT(args->index >= 0);
2493
2494         if (entry->flags & XFS_ATTR_LOCAL) {
2495                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2496                 namelen = name_loc->namelen;
2497                 name = (char *)name_loc->nameval;
2498         } else {
2499                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2500                 namelen = name_rmt->namelen;
2501                 name = (char *)name_rmt->name;
2502         }
2503         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2504         ASSERT(namelen == args->namelen);
2505         ASSERT(memcmp(name, args->name, namelen) == 0);
2506 #endif /* DEBUG */
2507
2508         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2509         xfs_trans_log_buf(args->trans, bp,
2510                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2511
2512         if (args->rmtblkno) {
2513                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2514                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2515                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2516                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2517                 xfs_trans_log_buf(args->trans, bp,
2518                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2519         }
2520
2521         /*
2522          * Commit the flag value change and start the next trans in series.
2523          */
2524         return xfs_trans_roll(&args->trans, args->dp);
2525 }
2526
2527 /*
2528  * Set the INCOMPLETE flag on an entry in a leaf block.
2529  */
2530 int
2531 xfs_attr3_leaf_setflag(
2532         struct xfs_da_args      *args)
2533 {
2534         struct xfs_attr_leafblock *leaf;
2535         struct xfs_attr_leaf_entry *entry;
2536         struct xfs_attr_leaf_name_remote *name_rmt;
2537         struct xfs_buf          *bp;
2538         int error;
2539 #ifdef DEBUG
2540         struct xfs_attr3_icleaf_hdr ichdr;
2541 #endif
2542
2543         trace_xfs_attr_leaf_setflag(args);
2544
2545         /*
2546          * Set up the operation.
2547          */
2548         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2549         if (error)
2550                 return error;
2551
2552         leaf = bp->b_addr;
2553 #ifdef DEBUG
2554         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2555         ASSERT(args->index < ichdr.count);
2556         ASSERT(args->index >= 0);
2557 #endif
2558         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2559
2560         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2561         entry->flags |= XFS_ATTR_INCOMPLETE;
2562         xfs_trans_log_buf(args->trans, bp,
2563                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2564         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2565                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2566                 name_rmt->valueblk = 0;
2567                 name_rmt->valuelen = 0;
2568                 xfs_trans_log_buf(args->trans, bp,
2569                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2570         }
2571
2572         /*
2573          * Commit the flag value change and start the next trans in series.
2574          */
2575         return xfs_trans_roll(&args->trans, args->dp);
2576 }
2577
2578 /*
2579  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2580  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2581  * entry given by args->blkno2/index2.
2582  *
2583  * Note that they could be in different blocks, or in the same block.
2584  */
2585 int
2586 xfs_attr3_leaf_flipflags(
2587         struct xfs_da_args      *args)
2588 {
2589         struct xfs_attr_leafblock *leaf1;
2590         struct xfs_attr_leafblock *leaf2;
2591         struct xfs_attr_leaf_entry *entry1;
2592         struct xfs_attr_leaf_entry *entry2;
2593         struct xfs_attr_leaf_name_remote *name_rmt;
2594         struct xfs_buf          *bp1;
2595         struct xfs_buf          *bp2;
2596         int error;
2597 #ifdef DEBUG
2598         struct xfs_attr3_icleaf_hdr ichdr1;
2599         struct xfs_attr3_icleaf_hdr ichdr2;
2600         xfs_attr_leaf_name_local_t *name_loc;
2601         int namelen1, namelen2;
2602         char *name1, *name2;
2603 #endif /* DEBUG */
2604
2605         trace_xfs_attr_leaf_flipflags(args);
2606
2607         /*
2608          * Read the block containing the "old" attr
2609          */
2610         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2611         if (error)
2612                 return error;
2613
2614         /*
2615          * Read the block containing the "new" attr, if it is different
2616          */
2617         if (args->blkno2 != args->blkno) {
2618                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2619                                            -1, &bp2);
2620                 if (error)
2621                         return error;
2622         } else {
2623                 bp2 = bp1;
2624         }
2625
2626         leaf1 = bp1->b_addr;
2627         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2628
2629         leaf2 = bp2->b_addr;
2630         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2631
2632 #ifdef DEBUG
2633         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2634         ASSERT(args->index < ichdr1.count);
2635         ASSERT(args->index >= 0);
2636
2637         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2638         ASSERT(args->index2 < ichdr2.count);
2639         ASSERT(args->index2 >= 0);
2640
2641         if (entry1->flags & XFS_ATTR_LOCAL) {
2642                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2643                 namelen1 = name_loc->namelen;
2644                 name1 = (char *)name_loc->nameval;
2645         } else {
2646                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2647                 namelen1 = name_rmt->namelen;
2648                 name1 = (char *)name_rmt->name;
2649         }
2650         if (entry2->flags & XFS_ATTR_LOCAL) {
2651                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2652                 namelen2 = name_loc->namelen;
2653                 name2 = (char *)name_loc->nameval;
2654         } else {
2655                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2656                 namelen2 = name_rmt->namelen;
2657                 name2 = (char *)name_rmt->name;
2658         }
2659         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2660         ASSERT(namelen1 == namelen2);
2661         ASSERT(memcmp(name1, name2, namelen1) == 0);
2662 #endif /* DEBUG */
2663
2664         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2665         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2666
2667         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2668         xfs_trans_log_buf(args->trans, bp1,
2669                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2670         if (args->rmtblkno) {
2671                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2672                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2673                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2674                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2675                 xfs_trans_log_buf(args->trans, bp1,
2676                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2677         }
2678
2679         entry2->flags |= XFS_ATTR_INCOMPLETE;
2680         xfs_trans_log_buf(args->trans, bp2,
2681                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2682         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2683                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2684                 name_rmt->valueblk = 0;
2685                 name_rmt->valuelen = 0;
2686                 xfs_trans_log_buf(args->trans, bp2,
2687                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2688         }
2689
2690         /*
2691          * Commit the flag value change and start the next trans in series.
2692          */
2693         error = xfs_trans_roll(&args->trans, args->dp);
2694
2695         return error;
2696 }