88cd84a31f82378deccfe9074e3a05524355884a
[firefly-linux-kernel-4.4.55.git] / fs / xfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_attr_remote.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_cksum.h"
44 #include "xfs_dinode.h"
45 #include "xfs_dir2.h"
46
47
48 /*
49  * xfs_attr_leaf.c
50  *
51  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52  */
53
54 /*========================================================================
55  * Function prototypes for the kernel.
56  *========================================================================*/
57
58 /*
59  * Routines used for growing the Btree.
60  */
61 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
62                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
63 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
64                                    struct xfs_attr3_icleaf_hdr *ichdr,
65                                    struct xfs_da_args *args, int freemap_index);
66 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
67                                    struct xfs_attr3_icleaf_hdr *ichdr,
68                                    struct xfs_buf *leaf_buffer);
69 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
70                                                    xfs_da_state_blk_t *blk1,
71                                                    xfs_da_state_blk_t *blk2);
72 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
73                         xfs_da_state_blk_t *leaf_blk_1,
74                         struct xfs_attr3_icleaf_hdr *ichdr1,
75                         xfs_da_state_blk_t *leaf_blk_2,
76                         struct xfs_attr3_icleaf_hdr *ichdr2,
77                         int *number_entries_in_blk1,
78                         int *number_usedbytes_in_blk1);
79
80 /*
81  * Utility routines.
82  */
83 STATIC void xfs_attr3_leaf_moveents(struct xfs_attr_leafblock *src_leaf,
84                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
85                         struct xfs_attr_leafblock *dst_leaf,
86                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87                         int move_count, struct xfs_mount *mp);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
89
90 void
91 xfs_attr3_leaf_hdr_from_disk(
92         struct xfs_attr3_icleaf_hdr     *to,
93         struct xfs_attr_leafblock       *from)
94 {
95         int     i;
96
97         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
98                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
99
100         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
101                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
102
103                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
104                 to->back = be32_to_cpu(hdr3->info.hdr.back);
105                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
106                 to->count = be16_to_cpu(hdr3->count);
107                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
108                 to->firstused = be16_to_cpu(hdr3->firstused);
109                 to->holes = hdr3->holes;
110
111                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
112                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
113                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
114                 }
115                 return;
116         }
117         to->forw = be32_to_cpu(from->hdr.info.forw);
118         to->back = be32_to_cpu(from->hdr.info.back);
119         to->magic = be16_to_cpu(from->hdr.info.magic);
120         to->count = be16_to_cpu(from->hdr.count);
121         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
122         to->firstused = be16_to_cpu(from->hdr.firstused);
123         to->holes = from->hdr.holes;
124
125         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
126                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
127                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
128         }
129 }
130
131 void
132 xfs_attr3_leaf_hdr_to_disk(
133         struct xfs_attr_leafblock       *to,
134         struct xfs_attr3_icleaf_hdr     *from)
135 {
136         int     i;
137
138         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
139                from->magic == XFS_ATTR3_LEAF_MAGIC);
140
141         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
142                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
143
144                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
145                 hdr3->info.hdr.back = cpu_to_be32(from->back);
146                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
147                 hdr3->count = cpu_to_be16(from->count);
148                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
149                 hdr3->firstused = cpu_to_be16(from->firstused);
150                 hdr3->holes = from->holes;
151                 hdr3->pad1 = 0;
152
153                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
154                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
155                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
156                 }
157                 return;
158         }
159         to->hdr.info.forw = cpu_to_be32(from->forw);
160         to->hdr.info.back = cpu_to_be32(from->back);
161         to->hdr.info.magic = cpu_to_be16(from->magic);
162         to->hdr.count = cpu_to_be16(from->count);
163         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
164         to->hdr.firstused = cpu_to_be16(from->firstused);
165         to->hdr.holes = from->holes;
166         to->hdr.pad1 = 0;
167
168         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
169                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
170                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
171         }
172 }
173
174 static bool
175 xfs_attr3_leaf_verify(
176         struct xfs_buf          *bp)
177 {
178         struct xfs_mount        *mp = bp->b_target->bt_mount;
179         struct xfs_attr_leafblock *leaf = bp->b_addr;
180         struct xfs_attr3_icleaf_hdr ichdr;
181
182         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
183
184         if (xfs_sb_version_hascrc(&mp->m_sb)) {
185                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
186
187                 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
188                         return false;
189
190                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
191                         return false;
192                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
193                         return false;
194         } else {
195                 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
196                         return false;
197         }
198         if (ichdr.count == 0)
199                 return false;
200
201         /* XXX: need to range check rest of attr header values */
202         /* XXX: hash order check? */
203
204         return true;
205 }
206
207 static void
208 xfs_attr3_leaf_write_verify(
209         struct xfs_buf  *bp)
210 {
211         struct xfs_mount        *mp = bp->b_target->bt_mount;
212         struct xfs_buf_log_item *bip = bp->b_fspriv;
213         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
214
215         if (!xfs_attr3_leaf_verify(bp)) {
216                 xfs_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(XFS_ERROR(ENOATTR));
550
551         /*
552          * Fix up the attribute fork data, covering the hole
553          */
554         end = base + size;
555         totsize = be16_to_cpu(sf->hdr.totsize);
556         if (end != totsize)
557                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
558         sf->hdr.count--;
559         be16_add_cpu(&sf->hdr.totsize, -size);
560
561         /*
562          * Fix up the start offset of the attribute fork
563          */
564         totsize -= size;
565         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
566             (mp->m_flags & XFS_MOUNT_ATTR2) &&
567             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
568             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
569                 xfs_attr_fork_reset(dp, args->trans);
570         } else {
571                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
572                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
573                 ASSERT(dp->i_d.di_forkoff);
574                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
575                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
576                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
577                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
578                 xfs_trans_log_inode(args->trans, dp,
579                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
580         }
581
582         xfs_sbversion_add_attr2(mp, args->trans);
583
584         return(0);
585 }
586
587 /*
588  * Look up a name in a shortform attribute list structure.
589  */
590 /*ARGSUSED*/
591 int
592 xfs_attr_shortform_lookup(xfs_da_args_t *args)
593 {
594         xfs_attr_shortform_t *sf;
595         xfs_attr_sf_entry_t *sfe;
596         int i;
597         xfs_ifork_t *ifp;
598
599         trace_xfs_attr_sf_lookup(args);
600
601         ifp = args->dp->i_afp;
602         ASSERT(ifp->if_flags & XFS_IFINLINE);
603         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
604         sfe = &sf->list[0];
605         for (i = 0; i < sf->hdr.count;
606                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
607                 if (sfe->namelen != args->namelen)
608                         continue;
609                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
610                         continue;
611                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
612                         continue;
613                 return(XFS_ERROR(EEXIST));
614         }
615         return(XFS_ERROR(ENOATTR));
616 }
617
618 /*
619  * Look up a name in a shortform attribute list structure.
620  */
621 /*ARGSUSED*/
622 int
623 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
624 {
625         xfs_attr_shortform_t *sf;
626         xfs_attr_sf_entry_t *sfe;
627         int i;
628
629         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
630         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
631         sfe = &sf->list[0];
632         for (i = 0; i < sf->hdr.count;
633                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
634                 if (sfe->namelen != args->namelen)
635                         continue;
636                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
637                         continue;
638                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
639                         continue;
640                 if (args->flags & ATTR_KERNOVAL) {
641                         args->valuelen = sfe->valuelen;
642                         return(XFS_ERROR(EEXIST));
643                 }
644                 if (args->valuelen < sfe->valuelen) {
645                         args->valuelen = sfe->valuelen;
646                         return(XFS_ERROR(ERANGE));
647                 }
648                 args->valuelen = sfe->valuelen;
649                 memcpy(args->value, &sfe->nameval[args->namelen],
650                                                     args->valuelen);
651                 return(XFS_ERROR(EEXIST));
652         }
653         return(XFS_ERROR(ENOATTR));
654 }
655
656 /*
657  * Convert from using the shortform to the leaf.
658  */
659 int
660 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
661 {
662         xfs_inode_t *dp;
663         xfs_attr_shortform_t *sf;
664         xfs_attr_sf_entry_t *sfe;
665         xfs_da_args_t nargs;
666         char *tmpbuffer;
667         int error, i, size;
668         xfs_dablk_t blkno;
669         struct xfs_buf *bp;
670         xfs_ifork_t *ifp;
671
672         trace_xfs_attr_sf_to_leaf(args);
673
674         dp = args->dp;
675         ifp = dp->i_afp;
676         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
677         size = be16_to_cpu(sf->hdr.totsize);
678         tmpbuffer = kmem_alloc(size, KM_SLEEP);
679         ASSERT(tmpbuffer != NULL);
680         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
681         sf = (xfs_attr_shortform_t *)tmpbuffer;
682
683         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
684         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
685
686         bp = NULL;
687         error = xfs_da_grow_inode(args, &blkno);
688         if (error) {
689                 /*
690                  * If we hit an IO error middle of the transaction inside
691                  * grow_inode(), we may have inconsistent data. Bail out.
692                  */
693                 if (error == EIO)
694                         goto out;
695                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
696                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
697                 goto out;
698         }
699
700         ASSERT(blkno == 0);
701         error = xfs_attr3_leaf_create(args, blkno, &bp);
702         if (error) {
703                 error = xfs_da_shrink_inode(args, 0, bp);
704                 bp = NULL;
705                 if (error)
706                         goto out;
707                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
708                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
709                 goto out;
710         }
711
712         memset((char *)&nargs, 0, sizeof(nargs));
713         nargs.dp = dp;
714         nargs.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(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
810         if (!tmpbuffer)
811                 return ENOMEM;
812
813         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
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, XFS_LBSIZE(dp->i_mount));
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, XFS_LBSIZE(mp));
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, XFS_LBSIZE(mp) - 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, XFS_LBSIZE(mp) - 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, XFS_LBSIZE(mp));
972
973         memset(&ichdr, 0, sizeof(ichdr));
974         ichdr.firstused = XFS_LBSIZE(mp);
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, XFS_LBSIZE(mp) - 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->namelen, args->valuelen,
1080                            args->geo->blksize, NULL);
1081
1082         /*
1083          * Search through freemap for first-fit on new name length.
1084          * (may need to figure in size of entry struct too)
1085          */
1086         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1087                                         + xfs_attr3_leaf_hdr_size(leaf);
1088         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1089                 if (tablesize > ichdr.firstused) {
1090                         sum += ichdr.freemap[i].size;
1091                         continue;
1092                 }
1093                 if (!ichdr.freemap[i].size)
1094                         continue;       /* no space in this map */
1095                 tmp = entsize;
1096                 if (ichdr.freemap[i].base < ichdr.firstused)
1097                         tmp += sizeof(xfs_attr_leaf_entry_t);
1098                 if (ichdr.freemap[i].size >= tmp) {
1099                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1100                         goto out_log_hdr;
1101                 }
1102                 sum += ichdr.freemap[i].size;
1103         }
1104
1105         /*
1106          * If there are no holes in the address space of the block,
1107          * and we don't have enough freespace, then compaction will do us
1108          * no good and we should just give up.
1109          */
1110         if (!ichdr.holes && sum < entsize)
1111                 return XFS_ERROR(ENOSPC);
1112
1113         /*
1114          * Compact the entries to coalesce free space.
1115          * This may change the hdr->count via dropping INCOMPLETE entries.
1116          */
1117         xfs_attr3_leaf_compact(args, &ichdr, bp);
1118
1119         /*
1120          * After compaction, the block is guaranteed to have only one
1121          * free region, in freemap[0].  If it is not big enough, give up.
1122          */
1123         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1124                 tmp = ENOSPC;
1125                 goto out_log_hdr;
1126         }
1127
1128         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1129
1130 out_log_hdr:
1131         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1132         xfs_trans_log_buf(args->trans, bp,
1133                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1134                                 xfs_attr3_leaf_hdr_size(leaf)));
1135         return tmp;
1136 }
1137
1138 /*
1139  * Add a name to a leaf attribute list structure.
1140  */
1141 STATIC int
1142 xfs_attr3_leaf_add_work(
1143         struct xfs_buf          *bp,
1144         struct xfs_attr3_icleaf_hdr *ichdr,
1145         struct xfs_da_args      *args,
1146         int                     mapindex)
1147 {
1148         struct xfs_attr_leafblock *leaf;
1149         struct xfs_attr_leaf_entry *entry;
1150         struct xfs_attr_leaf_name_local *name_loc;
1151         struct xfs_attr_leaf_name_remote *name_rmt;
1152         struct xfs_mount        *mp;
1153         int                     tmp;
1154         int                     i;
1155
1156         trace_xfs_attr_leaf_add_work(args);
1157
1158         leaf = bp->b_addr;
1159         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1160         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1161
1162         /*
1163          * Force open some space in the entry array and fill it in.
1164          */
1165         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1166         if (args->index < ichdr->count) {
1167                 tmp  = ichdr->count - args->index;
1168                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1169                 memmove(entry + 1, entry, tmp);
1170                 xfs_trans_log_buf(args->trans, bp,
1171                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1172         }
1173         ichdr->count++;
1174
1175         /*
1176          * Allocate space for the new string (at the end of the run).
1177          */
1178         mp = args->trans->t_mountp;
1179         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1180         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1181         ASSERT(ichdr->freemap[mapindex].size >=
1182                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1183                                          args->geo->blksize, NULL));
1184         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1185         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1186
1187         ichdr->freemap[mapindex].size -=
1188                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1189                                                  args->geo->blksize, &tmp);
1190
1191         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1192                                      ichdr->freemap[mapindex].size);
1193         entry->hashval = cpu_to_be32(args->hashval);
1194         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1195         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1196         if (args->op_flags & XFS_DA_OP_RENAME) {
1197                 entry->flags |= XFS_ATTR_INCOMPLETE;
1198                 if ((args->blkno2 == args->blkno) &&
1199                     (args->index2 <= args->index)) {
1200                         args->index2++;
1201                 }
1202         }
1203         xfs_trans_log_buf(args->trans, bp,
1204                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1205         ASSERT((args->index == 0) ||
1206                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1207         ASSERT((args->index == ichdr->count - 1) ||
1208                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1209
1210         /*
1211          * For "remote" attribute values, simply note that we need to
1212          * allocate space for the "remote" value.  We can't actually
1213          * allocate the extents in this transaction, and we can't decide
1214          * which blocks they should be as we might allocate more blocks
1215          * as part of this transaction (a split operation for example).
1216          */
1217         if (entry->flags & XFS_ATTR_LOCAL) {
1218                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1219                 name_loc->namelen = args->namelen;
1220                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1221                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1222                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1223                                    be16_to_cpu(name_loc->valuelen));
1224         } else {
1225                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1226                 name_rmt->namelen = args->namelen;
1227                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1228                 entry->flags |= XFS_ATTR_INCOMPLETE;
1229                 /* just in case */
1230                 name_rmt->valuelen = 0;
1231                 name_rmt->valueblk = 0;
1232                 args->rmtblkno = 1;
1233                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1234                 args->rmtvaluelen = args->valuelen;
1235         }
1236         xfs_trans_log_buf(args->trans, bp,
1237              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1238                                    xfs_attr_leaf_entsize(leaf, args->index)));
1239
1240         /*
1241          * Update the control info for this leaf node
1242          */
1243         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1244                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1245
1246         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1247                                         + xfs_attr3_leaf_hdr_size(leaf));
1248         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1249                                         + xfs_attr3_leaf_hdr_size(leaf);
1250
1251         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1252                 if (ichdr->freemap[i].base == tmp) {
1253                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1254                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1255                 }
1256         }
1257         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1258         return 0;
1259 }
1260
1261 /*
1262  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1263  */
1264 STATIC void
1265 xfs_attr3_leaf_compact(
1266         struct xfs_da_args      *args,
1267         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1268         struct xfs_buf          *bp)
1269 {
1270         struct xfs_attr_leafblock *leaf_src;
1271         struct xfs_attr_leafblock *leaf_dst;
1272         struct xfs_attr3_icleaf_hdr ichdr_src;
1273         struct xfs_trans        *trans = args->trans;
1274         struct xfs_mount        *mp = trans->t_mountp;
1275         char                    *tmpbuffer;
1276
1277         trace_xfs_attr_leaf_compact(args);
1278
1279         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1280         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1281         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1282         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1283         leaf_dst = bp->b_addr;
1284
1285         /*
1286          * Copy the on-disk header back into the destination buffer to ensure
1287          * all the information in the header that is not part of the incore
1288          * header structure is preserved.
1289          */
1290         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1291
1292         /* Initialise the incore headers */
1293         ichdr_src = *ichdr_dst; /* struct copy */
1294         ichdr_dst->firstused = XFS_LBSIZE(mp);
1295         ichdr_dst->usedbytes = 0;
1296         ichdr_dst->count = 0;
1297         ichdr_dst->holes = 0;
1298         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1299         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1300                                                 ichdr_dst->freemap[0].base;
1301
1302         /* write the header back to initialise the underlying buffer */
1303         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1304
1305         /*
1306          * Copy all entry's in the same (sorted) order,
1307          * but allocate name/value pairs packed and in sequence.
1308          */
1309         xfs_attr3_leaf_moveents(leaf_src, &ichdr_src, 0, leaf_dst, ichdr_dst, 0,
1310                                 ichdr_src.count, mp);
1311         /*
1312          * this logs the entire buffer, but the caller must write the header
1313          * back to the buffer when it is finished modifying it.
1314          */
1315         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1316
1317         kmem_free(tmpbuffer);
1318 }
1319
1320 /*
1321  * Compare two leaf blocks "order".
1322  * Return 0 unless leaf2 should go before leaf1.
1323  */
1324 static int
1325 xfs_attr3_leaf_order(
1326         struct xfs_buf  *leaf1_bp,
1327         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1328         struct xfs_buf  *leaf2_bp,
1329         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1330 {
1331         struct xfs_attr_leaf_entry *entries1;
1332         struct xfs_attr_leaf_entry *entries2;
1333
1334         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1335         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1336         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1337             ((be32_to_cpu(entries2[0].hashval) <
1338               be32_to_cpu(entries1[0].hashval)) ||
1339              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1340               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1341                 return 1;
1342         }
1343         return 0;
1344 }
1345
1346 int
1347 xfs_attr_leaf_order(
1348         struct xfs_buf  *leaf1_bp,
1349         struct xfs_buf  *leaf2_bp)
1350 {
1351         struct xfs_attr3_icleaf_hdr ichdr1;
1352         struct xfs_attr3_icleaf_hdr ichdr2;
1353
1354         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1355         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1356         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1357 }
1358
1359 /*
1360  * Redistribute the attribute list entries between two leaf nodes,
1361  * taking into account the size of the new entry.
1362  *
1363  * NOTE: if new block is empty, then it will get the upper half of the
1364  * old block.  At present, all (one) callers pass in an empty second block.
1365  *
1366  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1367  * to match what it is doing in splitting the attribute leaf block.  Those
1368  * values are used in "atomic rename" operations on attributes.  Note that
1369  * the "new" and "old" values can end up in different blocks.
1370  */
1371 STATIC void
1372 xfs_attr3_leaf_rebalance(
1373         struct xfs_da_state     *state,
1374         struct xfs_da_state_blk *blk1,
1375         struct xfs_da_state_blk *blk2)
1376 {
1377         struct xfs_da_args      *args;
1378         struct xfs_attr_leafblock *leaf1;
1379         struct xfs_attr_leafblock *leaf2;
1380         struct xfs_attr3_icleaf_hdr ichdr1;
1381         struct xfs_attr3_icleaf_hdr ichdr2;
1382         struct xfs_attr_leaf_entry *entries1;
1383         struct xfs_attr_leaf_entry *entries2;
1384         int                     count;
1385         int                     totallen;
1386         int                     max;
1387         int                     space;
1388         int                     swap;
1389
1390         /*
1391          * Set up environment.
1392          */
1393         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1394         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1395         leaf1 = blk1->bp->b_addr;
1396         leaf2 = blk2->bp->b_addr;
1397         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1398         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1399         ASSERT(ichdr2.count == 0);
1400         args = state->args;
1401
1402         trace_xfs_attr_leaf_rebalance(args);
1403
1404         /*
1405          * Check ordering of blocks, reverse if it makes things simpler.
1406          *
1407          * NOTE: Given that all (current) callers pass in an empty
1408          * second block, this code should never set "swap".
1409          */
1410         swap = 0;
1411         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1412                 struct xfs_da_state_blk *tmp_blk;
1413                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1414
1415                 tmp_blk = blk1;
1416                 blk1 = blk2;
1417                 blk2 = tmp_blk;
1418
1419                 /* struct copies to swap them rather than reconverting */
1420                 tmp_ichdr = ichdr1;
1421                 ichdr1 = ichdr2;
1422                 ichdr2 = tmp_ichdr;
1423
1424                 leaf1 = blk1->bp->b_addr;
1425                 leaf2 = blk2->bp->b_addr;
1426                 swap = 1;
1427         }
1428
1429         /*
1430          * Examine entries until we reduce the absolute difference in
1431          * byte usage between the two blocks to a minimum.  Then get
1432          * the direction to copy and the number of elements to move.
1433          *
1434          * "inleaf" is true if the new entry should be inserted into blk1.
1435          * If "swap" is also true, then reverse the sense of "inleaf".
1436          */
1437         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1438                                                       blk2, &ichdr2,
1439                                                       &count, &totallen);
1440         if (swap)
1441                 state->inleaf = !state->inleaf;
1442
1443         /*
1444          * Move any entries required from leaf to leaf:
1445          */
1446         if (count < ichdr1.count) {
1447                 /*
1448                  * Figure the total bytes to be added to the destination leaf.
1449                  */
1450                 /* number entries being moved */
1451                 count = ichdr1.count - count;
1452                 space  = ichdr1.usedbytes - totallen;
1453                 space += count * sizeof(xfs_attr_leaf_entry_t);
1454
1455                 /*
1456                  * leaf2 is the destination, compact it if it looks tight.
1457                  */
1458                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1459                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1460                 if (space > max)
1461                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1462
1463                 /*
1464                  * Move high entries from leaf1 to low end of leaf2.
1465                  */
1466                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1467                                 leaf2, &ichdr2, 0, count, state->mp);
1468
1469         } else if (count > ichdr1.count) {
1470                 /*
1471                  * I assert that since all callers pass in an empty
1472                  * second buffer, this code should never execute.
1473                  */
1474                 ASSERT(0);
1475
1476                 /*
1477                  * Figure the total bytes to be added to the destination leaf.
1478                  */
1479                 /* number entries being moved */
1480                 count -= ichdr1.count;
1481                 space  = totallen - ichdr1.usedbytes;
1482                 space += count * sizeof(xfs_attr_leaf_entry_t);
1483
1484                 /*
1485                  * leaf1 is the destination, compact it if it looks tight.
1486                  */
1487                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1488                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1489                 if (space > max)
1490                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1491
1492                 /*
1493                  * Move low entries from leaf2 to high end of leaf1.
1494                  */
1495                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1496                                         ichdr1.count, count, state->mp);
1497         }
1498
1499         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1500         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1501         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1502         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1503
1504         /*
1505          * Copy out last hashval in each block for B-tree code.
1506          */
1507         entries1 = xfs_attr3_leaf_entryp(leaf1);
1508         entries2 = xfs_attr3_leaf_entryp(leaf2);
1509         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1510         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1511
1512         /*
1513          * Adjust the expected index for insertion.
1514          * NOTE: this code depends on the (current) situation that the
1515          * second block was originally empty.
1516          *
1517          * If the insertion point moved to the 2nd block, we must adjust
1518          * the index.  We must also track the entry just following the
1519          * new entry for use in an "atomic rename" operation, that entry
1520          * is always the "old" entry and the "new" entry is what we are
1521          * inserting.  The index/blkno fields refer to the "old" entry,
1522          * while the index2/blkno2 fields refer to the "new" entry.
1523          */
1524         if (blk1->index > ichdr1.count) {
1525                 ASSERT(state->inleaf == 0);
1526                 blk2->index = blk1->index - ichdr1.count;
1527                 args->index = args->index2 = blk2->index;
1528                 args->blkno = args->blkno2 = blk2->blkno;
1529         } else if (blk1->index == ichdr1.count) {
1530                 if (state->inleaf) {
1531                         args->index = blk1->index;
1532                         args->blkno = blk1->blkno;
1533                         args->index2 = 0;
1534                         args->blkno2 = blk2->blkno;
1535                 } else {
1536                         /*
1537                          * On a double leaf split, the original attr location
1538                          * is already stored in blkno2/index2, so don't
1539                          * overwrite it overwise we corrupt the tree.
1540                          */
1541                         blk2->index = blk1->index - ichdr1.count;
1542                         args->index = blk2->index;
1543                         args->blkno = blk2->blkno;
1544                         if (!state->extravalid) {
1545                                 /*
1546                                  * set the new attr location to match the old
1547                                  * one and let the higher level split code
1548                                  * decide where in the leaf to place it.
1549                                  */
1550                                 args->index2 = blk2->index;
1551                                 args->blkno2 = blk2->blkno;
1552                         }
1553                 }
1554         } else {
1555                 ASSERT(state->inleaf == 1);
1556                 args->index = args->index2 = blk1->index;
1557                 args->blkno = args->blkno2 = blk1->blkno;
1558         }
1559 }
1560
1561 /*
1562  * Examine entries until we reduce the absolute difference in
1563  * byte usage between the two blocks to a minimum.
1564  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1565  * GROT: there will always be enough room in either block for a new entry.
1566  * GROT: Do a double-split for this case?
1567  */
1568 STATIC int
1569 xfs_attr3_leaf_figure_balance(
1570         struct xfs_da_state             *state,
1571         struct xfs_da_state_blk         *blk1,
1572         struct xfs_attr3_icleaf_hdr     *ichdr1,
1573         struct xfs_da_state_blk         *blk2,
1574         struct xfs_attr3_icleaf_hdr     *ichdr2,
1575         int                             *countarg,
1576         int                             *usedbytesarg)
1577 {
1578         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1579         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1580         struct xfs_attr_leaf_entry      *entry;
1581         int                             count;
1582         int                             max;
1583         int                             index;
1584         int                             totallen = 0;
1585         int                             half;
1586         int                             lastdelta;
1587         int                             foundit = 0;
1588         int                             tmp;
1589
1590         /*
1591          * Examine entries until we reduce the absolute difference in
1592          * byte usage between the two blocks to a minimum.
1593          */
1594         max = ichdr1->count + ichdr2->count;
1595         half = (max + 1) * sizeof(*entry);
1596         half += ichdr1->usedbytes + ichdr2->usedbytes +
1597                         xfs_attr_leaf_newentsize(state->args->namelen,
1598                                                  state->args->valuelen,
1599                                                  state->blocksize, NULL);
1600         half /= 2;
1601         lastdelta = state->blocksize;
1602         entry = xfs_attr3_leaf_entryp(leaf1);
1603         for (count = index = 0; count < max; entry++, index++, count++) {
1604
1605 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1606                 /*
1607                  * The new entry is in the first block, account for it.
1608                  */
1609                 if (count == blk1->index) {
1610                         tmp = totallen + sizeof(*entry) +
1611                                 xfs_attr_leaf_newentsize(
1612                                                 state->args->namelen,
1613                                                 state->args->valuelen,
1614                                                 state->blocksize, NULL);
1615                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1616                                 break;
1617                         lastdelta = XFS_ATTR_ABS(half - tmp);
1618                         totallen = tmp;
1619                         foundit = 1;
1620                 }
1621
1622                 /*
1623                  * Wrap around into the second block if necessary.
1624                  */
1625                 if (count == ichdr1->count) {
1626                         leaf1 = leaf2;
1627                         entry = xfs_attr3_leaf_entryp(leaf1);
1628                         index = 0;
1629                 }
1630
1631                 /*
1632                  * Figure out if next leaf entry would be too much.
1633                  */
1634                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1635                                                                         index);
1636                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1637                         break;
1638                 lastdelta = XFS_ATTR_ABS(half - tmp);
1639                 totallen = tmp;
1640 #undef XFS_ATTR_ABS
1641         }
1642
1643         /*
1644          * Calculate the number of usedbytes that will end up in lower block.
1645          * If new entry not in lower block, fix up the count.
1646          */
1647         totallen -= count * sizeof(*entry);
1648         if (foundit) {
1649                 totallen -= sizeof(*entry) +
1650                                 xfs_attr_leaf_newentsize(
1651                                                 state->args->namelen,
1652                                                 state->args->valuelen,
1653                                                 state->blocksize, NULL);
1654         }
1655
1656         *countarg = count;
1657         *usedbytesarg = totallen;
1658         return foundit;
1659 }
1660
1661 /*========================================================================
1662  * Routines used for shrinking the Btree.
1663  *========================================================================*/
1664
1665 /*
1666  * Check a leaf block and its neighbors to see if the block should be
1667  * collapsed into one or the other neighbor.  Always keep the block
1668  * with the smaller block number.
1669  * If the current block is over 50% full, don't try to join it, return 0.
1670  * If the block is empty, fill in the state structure and return 2.
1671  * If it can be collapsed, fill in the state structure and return 1.
1672  * If nothing can be done, return 0.
1673  *
1674  * GROT: allow for INCOMPLETE entries in calculation.
1675  */
1676 int
1677 xfs_attr3_leaf_toosmall(
1678         struct xfs_da_state     *state,
1679         int                     *action)
1680 {
1681         struct xfs_attr_leafblock *leaf;
1682         struct xfs_da_state_blk *blk;
1683         struct xfs_attr3_icleaf_hdr ichdr;
1684         struct xfs_buf          *bp;
1685         xfs_dablk_t             blkno;
1686         int                     bytes;
1687         int                     forward;
1688         int                     error;
1689         int                     retval;
1690         int                     i;
1691
1692         trace_xfs_attr_leaf_toosmall(state->args);
1693
1694         /*
1695          * Check for the degenerate case of the block being over 50% full.
1696          * If so, it's not worth even looking to see if we might be able
1697          * to coalesce with a sibling.
1698          */
1699         blk = &state->path.blk[ state->path.active-1 ];
1700         leaf = blk->bp->b_addr;
1701         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1702         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1703                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1704                 ichdr.usedbytes;
1705         if (bytes > (state->blocksize >> 1)) {
1706                 *action = 0;    /* blk over 50%, don't try to join */
1707                 return(0);
1708         }
1709
1710         /*
1711          * Check for the degenerate case of the block being empty.
1712          * If the block is empty, we'll simply delete it, no need to
1713          * coalesce it with a sibling block.  We choose (arbitrarily)
1714          * to merge with the forward block unless it is NULL.
1715          */
1716         if (ichdr.count == 0) {
1717                 /*
1718                  * Make altpath point to the block we want to keep and
1719                  * path point to the block we want to drop (this one).
1720                  */
1721                 forward = (ichdr.forw != 0);
1722                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1723                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1724                                                  0, &retval);
1725                 if (error)
1726                         return(error);
1727                 if (retval) {
1728                         *action = 0;
1729                 } else {
1730                         *action = 2;
1731                 }
1732                 return 0;
1733         }
1734
1735         /*
1736          * Examine each sibling block to see if we can coalesce with
1737          * at least 25% free space to spare.  We need to figure out
1738          * whether to merge with the forward or the backward block.
1739          * We prefer coalescing with the lower numbered sibling so as
1740          * to shrink an attribute list over time.
1741          */
1742         /* start with smaller blk num */
1743         forward = ichdr.forw < ichdr.back;
1744         for (i = 0; i < 2; forward = !forward, i++) {
1745                 struct xfs_attr3_icleaf_hdr ichdr2;
1746                 if (forward)
1747                         blkno = ichdr.forw;
1748                 else
1749                         blkno = ichdr.back;
1750                 if (blkno == 0)
1751                         continue;
1752                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1753                                         blkno, -1, &bp);
1754                 if (error)
1755                         return(error);
1756
1757                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1758
1759                 bytes = state->blocksize - (state->blocksize >> 2) -
1760                         ichdr.usedbytes - ichdr2.usedbytes -
1761                         ((ichdr.count + ichdr2.count) *
1762                                         sizeof(xfs_attr_leaf_entry_t)) -
1763                         xfs_attr3_leaf_hdr_size(leaf);
1764
1765                 xfs_trans_brelse(state->args->trans, bp);
1766                 if (bytes >= 0)
1767                         break;  /* fits with at least 25% to spare */
1768         }
1769         if (i >= 2) {
1770                 *action = 0;
1771                 return(0);
1772         }
1773
1774         /*
1775          * Make altpath point to the block we want to keep (the lower
1776          * numbered block) and path point to the block we want to drop.
1777          */
1778         memcpy(&state->altpath, &state->path, sizeof(state->path));
1779         if (blkno < blk->blkno) {
1780                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1781                                                  0, &retval);
1782         } else {
1783                 error = xfs_da3_path_shift(state, &state->path, forward,
1784                                                  0, &retval);
1785         }
1786         if (error)
1787                 return(error);
1788         if (retval) {
1789                 *action = 0;
1790         } else {
1791                 *action = 1;
1792         }
1793         return(0);
1794 }
1795
1796 /*
1797  * Remove a name from the leaf attribute list structure.
1798  *
1799  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1800  * If two leaves are 37% full, when combined they will leave 25% free.
1801  */
1802 int
1803 xfs_attr3_leaf_remove(
1804         struct xfs_buf          *bp,
1805         struct xfs_da_args      *args)
1806 {
1807         struct xfs_attr_leafblock *leaf;
1808         struct xfs_attr3_icleaf_hdr ichdr;
1809         struct xfs_attr_leaf_entry *entry;
1810         struct xfs_mount        *mp = args->trans->t_mountp;
1811         int                     before;
1812         int                     after;
1813         int                     smallest;
1814         int                     entsize;
1815         int                     tablesize;
1816         int                     tmp;
1817         int                     i;
1818
1819         trace_xfs_attr_leaf_remove(args);
1820
1821         leaf = bp->b_addr;
1822         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1823
1824         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
1825         ASSERT(args->index >= 0 && args->index < ichdr.count);
1826         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1827                                         xfs_attr3_leaf_hdr_size(leaf));
1828
1829         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1830
1831         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1832         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1833
1834         /*
1835          * Scan through free region table:
1836          *    check for adjacency of free'd entry with an existing one,
1837          *    find smallest free region in case we need to replace it,
1838          *    adjust any map that borders the entry table,
1839          */
1840         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1841                                         + xfs_attr3_leaf_hdr_size(leaf);
1842         tmp = ichdr.freemap[0].size;
1843         before = after = -1;
1844         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1845         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1846         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1847                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
1848                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
1849                 if (ichdr.freemap[i].base == tablesize) {
1850                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1851                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1852                 }
1853
1854                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1855                                 be16_to_cpu(entry->nameidx)) {
1856                         before = i;
1857                 } else if (ichdr.freemap[i].base ==
1858                                 (be16_to_cpu(entry->nameidx) + entsize)) {
1859                         after = i;
1860                 } else if (ichdr.freemap[i].size < tmp) {
1861                         tmp = ichdr.freemap[i].size;
1862                         smallest = i;
1863                 }
1864         }
1865
1866         /*
1867          * Coalesce adjacent freemap regions,
1868          * or replace the smallest region.
1869          */
1870         if ((before >= 0) || (after >= 0)) {
1871                 if ((before >= 0) && (after >= 0)) {
1872                         ichdr.freemap[before].size += entsize;
1873                         ichdr.freemap[before].size += ichdr.freemap[after].size;
1874                         ichdr.freemap[after].base = 0;
1875                         ichdr.freemap[after].size = 0;
1876                 } else if (before >= 0) {
1877                         ichdr.freemap[before].size += entsize;
1878                 } else {
1879                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1880                         ichdr.freemap[after].size += entsize;
1881                 }
1882         } else {
1883                 /*
1884                  * Replace smallest region (if it is smaller than free'd entry)
1885                  */
1886                 if (ichdr.freemap[smallest].size < entsize) {
1887                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1888                         ichdr.freemap[smallest].size = entsize;
1889                 }
1890         }
1891
1892         /*
1893          * Did we remove the first entry?
1894          */
1895         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1896                 smallest = 1;
1897         else
1898                 smallest = 0;
1899
1900         /*
1901          * Compress the remaining entries and zero out the removed stuff.
1902          */
1903         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1904         ichdr.usedbytes -= entsize;
1905         xfs_trans_log_buf(args->trans, bp,
1906              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1907                                    entsize));
1908
1909         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1910         memmove(entry, entry + 1, tmp);
1911         ichdr.count--;
1912         xfs_trans_log_buf(args->trans, bp,
1913             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1914
1915         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1916         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1917
1918         /*
1919          * If we removed the first entry, re-find the first used byte
1920          * in the name area.  Note that if the entry was the "firstused",
1921          * then we don't have a "hole" in our block resulting from
1922          * removing the name.
1923          */
1924         if (smallest) {
1925                 tmp = XFS_LBSIZE(mp);
1926                 entry = xfs_attr3_leaf_entryp(leaf);
1927                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1928                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1929                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1930
1931                         if (be16_to_cpu(entry->nameidx) < tmp)
1932                                 tmp = be16_to_cpu(entry->nameidx);
1933                 }
1934                 ichdr.firstused = tmp;
1935                 if (!ichdr.firstused)
1936                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
1937         } else {
1938                 ichdr.holes = 1;        /* mark as needing compaction */
1939         }
1940         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1941         xfs_trans_log_buf(args->trans, bp,
1942                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1943                                           xfs_attr3_leaf_hdr_size(leaf)));
1944
1945         /*
1946          * Check if leaf is less than 50% full, caller may want to
1947          * "join" the leaf with a sibling if so.
1948          */
1949         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
1950               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
1951
1952         return tmp < args->geo->magicpct; /* leaf is < 37% full */
1953 }
1954
1955 /*
1956  * Move all the attribute list entries from drop_leaf into save_leaf.
1957  */
1958 void
1959 xfs_attr3_leaf_unbalance(
1960         struct xfs_da_state     *state,
1961         struct xfs_da_state_blk *drop_blk,
1962         struct xfs_da_state_blk *save_blk)
1963 {
1964         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
1965         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
1966         struct xfs_attr3_icleaf_hdr drophdr;
1967         struct xfs_attr3_icleaf_hdr savehdr;
1968         struct xfs_attr_leaf_entry *entry;
1969         struct xfs_mount        *mp = state->mp;
1970
1971         trace_xfs_attr_leaf_unbalance(state->args);
1972
1973         drop_leaf = drop_blk->bp->b_addr;
1974         save_leaf = save_blk->bp->b_addr;
1975         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1976         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
1977         entry = xfs_attr3_leaf_entryp(drop_leaf);
1978
1979         /*
1980          * Save last hashval from dying block for later Btree fixup.
1981          */
1982         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
1983
1984         /*
1985          * Check if we need a temp buffer, or can we do it in place.
1986          * Note that we don't check "leaf" for holes because we will
1987          * always be dropping it, toosmall() decided that for us already.
1988          */
1989         if (savehdr.holes == 0) {
1990                 /*
1991                  * dest leaf has no holes, so we add there.  May need
1992                  * to make some room in the entry array.
1993                  */
1994                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
1995                                          drop_blk->bp, &drophdr)) {
1996                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
1997                                                 save_leaf, &savehdr, 0,
1998                                                 drophdr.count, mp);
1999                 } else {
2000                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2001                                                 save_leaf, &savehdr,
2002                                                 savehdr.count, drophdr.count, mp);
2003                 }
2004         } else {
2005                 /*
2006                  * Destination has holes, so we make a temporary copy
2007                  * of the leaf and add them both to that.
2008                  */
2009                 struct xfs_attr_leafblock *tmp_leaf;
2010                 struct xfs_attr3_icleaf_hdr tmphdr;
2011
2012                 tmp_leaf = kmem_zalloc(state->blocksize, KM_SLEEP);
2013
2014                 /*
2015                  * Copy the header into the temp leaf so that all the stuff
2016                  * not in the incore header is present and gets copied back in
2017                  * once we've moved all the entries.
2018                  */
2019                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2020
2021                 memset(&tmphdr, 0, sizeof(tmphdr));
2022                 tmphdr.magic = savehdr.magic;
2023                 tmphdr.forw = savehdr.forw;
2024                 tmphdr.back = savehdr.back;
2025                 tmphdr.firstused = state->blocksize;
2026
2027                 /* write the header to the temp buffer to initialise it */
2028                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2029
2030                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2031                                          drop_blk->bp, &drophdr)) {
2032                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2033                                                 tmp_leaf, &tmphdr, 0,
2034                                                 drophdr.count, mp);
2035                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2036                                                 tmp_leaf, &tmphdr, tmphdr.count,
2037                                                 savehdr.count, mp);
2038                 } else {
2039                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2040                                                 tmp_leaf, &tmphdr, 0,
2041                                                 savehdr.count, mp);
2042                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2043                                                 tmp_leaf, &tmphdr, tmphdr.count,
2044                                                 drophdr.count, mp);
2045                 }
2046                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2047                 savehdr = tmphdr; /* struct copy */
2048                 kmem_free(tmp_leaf);
2049         }
2050
2051         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2052         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2053                                            state->blocksize - 1);
2054
2055         /*
2056          * Copy out last hashval in each block for B-tree code.
2057          */
2058         entry = xfs_attr3_leaf_entryp(save_leaf);
2059         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2060 }
2061
2062 /*========================================================================
2063  * Routines used for finding things in the Btree.
2064  *========================================================================*/
2065
2066 /*
2067  * Look up a name in a leaf attribute list structure.
2068  * This is the internal routine, it uses the caller's buffer.
2069  *
2070  * Note that duplicate keys are allowed, but only check within the
2071  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2072  *
2073  * Return in args->index the index into the entry[] array of either
2074  * the found entry, or where the entry should have been (insert before
2075  * that entry).
2076  *
2077  * Don't change the args->value unless we find the attribute.
2078  */
2079 int
2080 xfs_attr3_leaf_lookup_int(
2081         struct xfs_buf          *bp,
2082         struct xfs_da_args      *args)
2083 {
2084         struct xfs_attr_leafblock *leaf;
2085         struct xfs_attr3_icleaf_hdr ichdr;
2086         struct xfs_attr_leaf_entry *entry;
2087         struct xfs_attr_leaf_entry *entries;
2088         struct xfs_attr_leaf_name_local *name_loc;
2089         struct xfs_attr_leaf_name_remote *name_rmt;
2090         xfs_dahash_t            hashval;
2091         int                     probe;
2092         int                     span;
2093
2094         trace_xfs_attr_leaf_lookup(args);
2095
2096         leaf = bp->b_addr;
2097         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2098         entries = xfs_attr3_leaf_entryp(leaf);
2099         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2100
2101         /*
2102          * Binary search.  (note: small blocks will skip this loop)
2103          */
2104         hashval = args->hashval;
2105         probe = span = ichdr.count / 2;
2106         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2107                 span /= 2;
2108                 if (be32_to_cpu(entry->hashval) < hashval)
2109                         probe += span;
2110                 else if (be32_to_cpu(entry->hashval) > hashval)
2111                         probe -= span;
2112                 else
2113                         break;
2114         }
2115         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2116         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2117
2118         /*
2119          * Since we may have duplicate hashval's, find the first matching
2120          * hashval in the leaf.
2121          */
2122         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2123                 entry--;
2124                 probe--;
2125         }
2126         while (probe < ichdr.count &&
2127                be32_to_cpu(entry->hashval) < hashval) {
2128                 entry++;
2129                 probe++;
2130         }
2131         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2132                 args->index = probe;
2133                 return XFS_ERROR(ENOATTR);
2134         }
2135
2136         /*
2137          * Duplicate keys may be present, so search all of them for a match.
2138          */
2139         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2140                         entry++, probe++) {
2141 /*
2142  * GROT: Add code to remove incomplete entries.
2143  */
2144                 /*
2145                  * If we are looking for INCOMPLETE entries, show only those.
2146                  * If we are looking for complete entries, show only those.
2147                  */
2148                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2149                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2150                         continue;
2151                 }
2152                 if (entry->flags & XFS_ATTR_LOCAL) {
2153                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2154                         if (name_loc->namelen != args->namelen)
2155                                 continue;
2156                         if (memcmp(args->name, name_loc->nameval,
2157                                                         args->namelen) != 0)
2158                                 continue;
2159                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2160                                 continue;
2161                         args->index = probe;
2162                         return XFS_ERROR(EEXIST);
2163                 } else {
2164                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2165                         if (name_rmt->namelen != args->namelen)
2166                                 continue;
2167                         if (memcmp(args->name, name_rmt->name,
2168                                                         args->namelen) != 0)
2169                                 continue;
2170                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2171                                 continue;
2172                         args->index = probe;
2173                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2174                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2175                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2176                                                         args->dp->i_mount,
2177                                                         args->rmtvaluelen);
2178                         return XFS_ERROR(EEXIST);
2179                 }
2180         }
2181         args->index = probe;
2182         return XFS_ERROR(ENOATTR);
2183 }
2184
2185 /*
2186  * Get the value associated with an attribute name from a leaf attribute
2187  * list structure.
2188  */
2189 int
2190 xfs_attr3_leaf_getvalue(
2191         struct xfs_buf          *bp,
2192         struct xfs_da_args      *args)
2193 {
2194         struct xfs_attr_leafblock *leaf;
2195         struct xfs_attr3_icleaf_hdr ichdr;
2196         struct xfs_attr_leaf_entry *entry;
2197         struct xfs_attr_leaf_name_local *name_loc;
2198         struct xfs_attr_leaf_name_remote *name_rmt;
2199         int                     valuelen;
2200
2201         leaf = bp->b_addr;
2202         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2203         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2204         ASSERT(args->index < ichdr.count);
2205
2206         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2207         if (entry->flags & XFS_ATTR_LOCAL) {
2208                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2209                 ASSERT(name_loc->namelen == args->namelen);
2210                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2211                 valuelen = be16_to_cpu(name_loc->valuelen);
2212                 if (args->flags & ATTR_KERNOVAL) {
2213                         args->valuelen = valuelen;
2214                         return 0;
2215                 }
2216                 if (args->valuelen < valuelen) {
2217                         args->valuelen = valuelen;
2218                         return XFS_ERROR(ERANGE);
2219                 }
2220                 args->valuelen = valuelen;
2221                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2222         } else {
2223                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2224                 ASSERT(name_rmt->namelen == args->namelen);
2225                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2226                 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2227                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2228                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2229                                                        args->rmtvaluelen);
2230                 if (args->flags & ATTR_KERNOVAL) {
2231                         args->valuelen = args->rmtvaluelen;
2232                         return 0;
2233                 }
2234                 if (args->valuelen < args->rmtvaluelen) {
2235                         args->valuelen = args->rmtvaluelen;
2236                         return XFS_ERROR(ERANGE);
2237                 }
2238                 args->valuelen = args->rmtvaluelen;
2239         }
2240         return 0;
2241 }
2242
2243 /*========================================================================
2244  * Utility routines.
2245  *========================================================================*/
2246
2247 /*
2248  * Move the indicated entries from one leaf to another.
2249  * NOTE: this routine modifies both source and destination leaves.
2250  */
2251 /*ARGSUSED*/
2252 STATIC void
2253 xfs_attr3_leaf_moveents(
2254         struct xfs_attr_leafblock       *leaf_s,
2255         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2256         int                             start_s,
2257         struct xfs_attr_leafblock       *leaf_d,
2258         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2259         int                             start_d,
2260         int                             count,
2261         struct xfs_mount                *mp)
2262 {
2263         struct xfs_attr_leaf_entry      *entry_s;
2264         struct xfs_attr_leaf_entry      *entry_d;
2265         int                             desti;
2266         int                             tmp;
2267         int                             i;
2268
2269         /*
2270          * Check for nothing to do.
2271          */
2272         if (count == 0)
2273                 return;
2274
2275         /*
2276          * Set up environment.
2277          */
2278         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2279                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2280         ASSERT(ichdr_s->magic == ichdr_d->magic);
2281         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2282         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2283                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2284         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2285         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2286                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2287
2288         ASSERT(start_s < ichdr_s->count);
2289         ASSERT(start_d <= ichdr_d->count);
2290         ASSERT(count <= ichdr_s->count);
2291
2292
2293         /*
2294          * Move the entries in the destination leaf up to make a hole?
2295          */
2296         if (start_d < ichdr_d->count) {
2297                 tmp  = ichdr_d->count - start_d;
2298                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2299                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2300                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2301                 memmove(entry_d, entry_s, tmp);
2302         }
2303
2304         /*
2305          * Copy all entry's in the same (sorted) order,
2306          * but allocate attribute info packed and in sequence.
2307          */
2308         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2309         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2310         desti = start_d;
2311         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2312                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2313                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2314 #ifdef GROT
2315                 /*
2316                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2317                  * may also need to change the insertion index.  Code turned
2318                  * off for 6.2, should be revisited later.
2319                  */
2320                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2321                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2322                         ichdr_s->usedbytes -= tmp;
2323                         ichdr_s->count -= 1;
2324                         entry_d--;      /* to compensate for ++ in loop hdr */
2325                         desti--;
2326                         if ((start_s + i) < offset)
2327                                 result++;       /* insertion index adjustment */
2328                 } else {
2329 #endif /* GROT */
2330                         ichdr_d->firstused -= tmp;
2331                         /* both on-disk, don't endian flip twice */
2332                         entry_d->hashval = entry_s->hashval;
2333                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2334                         entry_d->flags = entry_s->flags;
2335                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2336                                                         <= XFS_LBSIZE(mp));
2337                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2338                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2339                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2340                                                         <= XFS_LBSIZE(mp));
2341                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2342                         ichdr_s->usedbytes -= tmp;
2343                         ichdr_d->usedbytes += tmp;
2344                         ichdr_s->count -= 1;
2345                         ichdr_d->count += 1;
2346                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2347                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2348                         ASSERT(ichdr_d->firstused >= tmp);
2349 #ifdef GROT
2350                 }
2351 #endif /* GROT */
2352         }
2353
2354         /*
2355          * Zero out the entries we just copied.
2356          */
2357         if (start_s == ichdr_s->count) {
2358                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2359                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2360                 ASSERT(((char *)entry_s + tmp) <=
2361                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2362                 memset(entry_s, 0, tmp);
2363         } else {
2364                 /*
2365                  * Move the remaining entries down to fill the hole,
2366                  * then zero the entries at the top.
2367                  */
2368                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2369                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2370                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2371                 memmove(entry_d, entry_s, tmp);
2372
2373                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2374                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2375                 ASSERT(((char *)entry_s + tmp) <=
2376                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2377                 memset(entry_s, 0, tmp);
2378         }
2379
2380         /*
2381          * Fill in the freemap information
2382          */
2383         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2384         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2385         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2386         ichdr_d->freemap[1].base = 0;
2387         ichdr_d->freemap[2].base = 0;
2388         ichdr_d->freemap[1].size = 0;
2389         ichdr_d->freemap[2].size = 0;
2390         ichdr_s->holes = 1;     /* leaf may not be compact */
2391 }
2392
2393 /*
2394  * Pick up the last hashvalue from a leaf block.
2395  */
2396 xfs_dahash_t
2397 xfs_attr_leaf_lasthash(
2398         struct xfs_buf  *bp,
2399         int             *count)
2400 {
2401         struct xfs_attr3_icleaf_hdr ichdr;
2402         struct xfs_attr_leaf_entry *entries;
2403
2404         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2405         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2406         if (count)
2407                 *count = ichdr.count;
2408         if (!ichdr.count)
2409                 return 0;
2410         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2411 }
2412
2413 /*
2414  * Calculate the number of bytes used to store the indicated attribute
2415  * (whether local or remote only calculate bytes in this block).
2416  */
2417 STATIC int
2418 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2419 {
2420         struct xfs_attr_leaf_entry *entries;
2421         xfs_attr_leaf_name_local_t *name_loc;
2422         xfs_attr_leaf_name_remote_t *name_rmt;
2423         int size;
2424
2425         entries = xfs_attr3_leaf_entryp(leaf);
2426         if (entries[index].flags & XFS_ATTR_LOCAL) {
2427                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2428                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2429                                                    be16_to_cpu(name_loc->valuelen));
2430         } else {
2431                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2432                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2433         }
2434         return size;
2435 }
2436
2437 /*
2438  * Calculate the number of bytes that would be required to store the new
2439  * attribute (whether local or remote only calculate bytes in this block).
2440  * This routine decides as a side effect whether the attribute will be
2441  * a "local" or a "remote" attribute.
2442  */
2443 int
2444 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2445 {
2446         int size;
2447
2448         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2449         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2450                 if (local) {
2451                         *local = 1;
2452                 }
2453         } else {
2454                 size = xfs_attr_leaf_entsize_remote(namelen);
2455                 if (local) {
2456                         *local = 0;
2457                 }
2458         }
2459         return size;
2460 }
2461
2462
2463 /*========================================================================
2464  * Manage the INCOMPLETE flag in a leaf entry
2465  *========================================================================*/
2466
2467 /*
2468  * Clear the INCOMPLETE flag on an entry in a leaf block.
2469  */
2470 int
2471 xfs_attr3_leaf_clearflag(
2472         struct xfs_da_args      *args)
2473 {
2474         struct xfs_attr_leafblock *leaf;
2475         struct xfs_attr_leaf_entry *entry;
2476         struct xfs_attr_leaf_name_remote *name_rmt;
2477         struct xfs_buf          *bp;
2478         int                     error;
2479 #ifdef DEBUG
2480         struct xfs_attr3_icleaf_hdr ichdr;
2481         xfs_attr_leaf_name_local_t *name_loc;
2482         int namelen;
2483         char *name;
2484 #endif /* DEBUG */
2485
2486         trace_xfs_attr_leaf_clearflag(args);
2487         /*
2488          * Set up the operation.
2489          */
2490         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2491         if (error)
2492                 return(error);
2493
2494         leaf = bp->b_addr;
2495         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2496         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2497
2498 #ifdef DEBUG
2499         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2500         ASSERT(args->index < ichdr.count);
2501         ASSERT(args->index >= 0);
2502
2503         if (entry->flags & XFS_ATTR_LOCAL) {
2504                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2505                 namelen = name_loc->namelen;
2506                 name = (char *)name_loc->nameval;
2507         } else {
2508                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2509                 namelen = name_rmt->namelen;
2510                 name = (char *)name_rmt->name;
2511         }
2512         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2513         ASSERT(namelen == args->namelen);
2514         ASSERT(memcmp(name, args->name, namelen) == 0);
2515 #endif /* DEBUG */
2516
2517         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2518         xfs_trans_log_buf(args->trans, bp,
2519                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2520
2521         if (args->rmtblkno) {
2522                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2523                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2524                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2525                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2526                 xfs_trans_log_buf(args->trans, bp,
2527                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2528         }
2529
2530         /*
2531          * Commit the flag value change and start the next trans in series.
2532          */
2533         return xfs_trans_roll(&args->trans, args->dp);
2534 }
2535
2536 /*
2537  * Set the INCOMPLETE flag on an entry in a leaf block.
2538  */
2539 int
2540 xfs_attr3_leaf_setflag(
2541         struct xfs_da_args      *args)
2542 {
2543         struct xfs_attr_leafblock *leaf;
2544         struct xfs_attr_leaf_entry *entry;
2545         struct xfs_attr_leaf_name_remote *name_rmt;
2546         struct xfs_buf          *bp;
2547         int error;
2548 #ifdef DEBUG
2549         struct xfs_attr3_icleaf_hdr ichdr;
2550 #endif
2551
2552         trace_xfs_attr_leaf_setflag(args);
2553
2554         /*
2555          * Set up the operation.
2556          */
2557         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2558         if (error)
2559                 return(error);
2560
2561         leaf = bp->b_addr;
2562 #ifdef DEBUG
2563         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2564         ASSERT(args->index < ichdr.count);
2565         ASSERT(args->index >= 0);
2566 #endif
2567         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2568
2569         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2570         entry->flags |= XFS_ATTR_INCOMPLETE;
2571         xfs_trans_log_buf(args->trans, bp,
2572                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2573         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2574                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2575                 name_rmt->valueblk = 0;
2576                 name_rmt->valuelen = 0;
2577                 xfs_trans_log_buf(args->trans, bp,
2578                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2579         }
2580
2581         /*
2582          * Commit the flag value change and start the next trans in series.
2583          */
2584         return xfs_trans_roll(&args->trans, args->dp);
2585 }
2586
2587 /*
2588  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2589  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2590  * entry given by args->blkno2/index2.
2591  *
2592  * Note that they could be in different blocks, or in the same block.
2593  */
2594 int
2595 xfs_attr3_leaf_flipflags(
2596         struct xfs_da_args      *args)
2597 {
2598         struct xfs_attr_leafblock *leaf1;
2599         struct xfs_attr_leafblock *leaf2;
2600         struct xfs_attr_leaf_entry *entry1;
2601         struct xfs_attr_leaf_entry *entry2;
2602         struct xfs_attr_leaf_name_remote *name_rmt;
2603         struct xfs_buf          *bp1;
2604         struct xfs_buf          *bp2;
2605         int error;
2606 #ifdef DEBUG
2607         struct xfs_attr3_icleaf_hdr ichdr1;
2608         struct xfs_attr3_icleaf_hdr ichdr2;
2609         xfs_attr_leaf_name_local_t *name_loc;
2610         int namelen1, namelen2;
2611         char *name1, *name2;
2612 #endif /* DEBUG */
2613
2614         trace_xfs_attr_leaf_flipflags(args);
2615
2616         /*
2617          * Read the block containing the "old" attr
2618          */
2619         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2620         if (error)
2621                 return error;
2622
2623         /*
2624          * Read the block containing the "new" attr, if it is different
2625          */
2626         if (args->blkno2 != args->blkno) {
2627                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2628                                            -1, &bp2);
2629                 if (error)
2630                         return error;
2631         } else {
2632                 bp2 = bp1;
2633         }
2634
2635         leaf1 = bp1->b_addr;
2636         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2637
2638         leaf2 = bp2->b_addr;
2639         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2640
2641 #ifdef DEBUG
2642         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2643         ASSERT(args->index < ichdr1.count);
2644         ASSERT(args->index >= 0);
2645
2646         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2647         ASSERT(args->index2 < ichdr2.count);
2648         ASSERT(args->index2 >= 0);
2649
2650         if (entry1->flags & XFS_ATTR_LOCAL) {
2651                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2652                 namelen1 = name_loc->namelen;
2653                 name1 = (char *)name_loc->nameval;
2654         } else {
2655                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2656                 namelen1 = name_rmt->namelen;
2657                 name1 = (char *)name_rmt->name;
2658         }
2659         if (entry2->flags & XFS_ATTR_LOCAL) {
2660                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2661                 namelen2 = name_loc->namelen;
2662                 name2 = (char *)name_loc->nameval;
2663         } else {
2664                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2665                 namelen2 = name_rmt->namelen;
2666                 name2 = (char *)name_rmt->name;
2667         }
2668         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2669         ASSERT(namelen1 == namelen2);
2670         ASSERT(memcmp(name1, name2, namelen1) == 0);
2671 #endif /* DEBUG */
2672
2673         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2674         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2675
2676         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2677         xfs_trans_log_buf(args->trans, bp1,
2678                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2679         if (args->rmtblkno) {
2680                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2681                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2682                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2683                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2684                 xfs_trans_log_buf(args->trans, bp1,
2685                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2686         }
2687
2688         entry2->flags |= XFS_ATTR_INCOMPLETE;
2689         xfs_trans_log_buf(args->trans, bp2,
2690                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2691         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2692                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2693                 name_rmt->valueblk = 0;
2694                 name_rmt->valuelen = 0;
2695                 xfs_trans_log_buf(args->trans, bp2,
2696                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2697         }
2698
2699         /*
2700          * Commit the flag value change and start the next trans in series.
2701          */
2702         error = xfs_trans_roll(&args->trans, args->dp);
2703
2704         return error;
2705 }