Merge remote-tracking branch 'regulator/topic/core' into regulator-next
[sfrench/cifs-2.6.git] / fs / xfs / libxfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_sb.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_trans.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_bmap.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_attr_remote.h"
37 #include "xfs_attr.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_error.h"
40 #include "xfs_trace.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_cksum.h"
43 #include "xfs_dir2.h"
44 #include "xfs_log.h"
45
46
47 /*
48  * xfs_attr_leaf.c
49  *
50  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
51  */
52
53 /*========================================================================
54  * Function prototypes for the kernel.
55  *========================================================================*/
56
57 /*
58  * Routines used for growing the Btree.
59  */
60 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
61                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
62 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
63                                    struct xfs_attr3_icleaf_hdr *ichdr,
64                                    struct xfs_da_args *args, int freemap_index);
65 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
66                                    struct xfs_attr3_icleaf_hdr *ichdr,
67                                    struct xfs_buf *leaf_buffer);
68 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
69                                                    xfs_da_state_blk_t *blk1,
70                                                    xfs_da_state_blk_t *blk2);
71 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
72                         xfs_da_state_blk_t *leaf_blk_1,
73                         struct xfs_attr3_icleaf_hdr *ichdr1,
74                         xfs_da_state_blk_t *leaf_blk_2,
75                         struct xfs_attr3_icleaf_hdr *ichdr2,
76                         int *number_entries_in_blk1,
77                         int *number_usedbytes_in_blk1);
78
79 /*
80  * Utility routines.
81  */
82 STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
83                         struct xfs_attr_leafblock *src_leaf,
84                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
85                         struct xfs_attr_leafblock *dst_leaf,
86                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
87                         int move_count);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
89
90 /*
91  * attr3 block 'firstused' conversion helpers.
92  *
93  * firstused refers to the offset of the first used byte of the nameval region
94  * of an attr leaf block. The region starts at the tail of the block and expands
95  * backwards towards the middle. As such, firstused is initialized to the block
96  * size for an empty leaf block and is reduced from there.
97  *
98  * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
99  * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
100  * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
101  * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
102  * the attr block size. The following helpers manage the conversion between the
103  * in-core and on-disk formats.
104  */
105
106 static void
107 xfs_attr3_leaf_firstused_from_disk(
108         struct xfs_da_geometry          *geo,
109         struct xfs_attr3_icleaf_hdr     *to,
110         struct xfs_attr_leafblock       *from)
111 {
112         struct xfs_attr3_leaf_hdr       *hdr3;
113
114         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
115                 hdr3 = (struct xfs_attr3_leaf_hdr *) from;
116                 to->firstused = be16_to_cpu(hdr3->firstused);
117         } else {
118                 to->firstused = be16_to_cpu(from->hdr.firstused);
119         }
120
121         /*
122          * Convert from the magic fsb size value to actual blocksize. This
123          * should only occur for empty blocks when the block size overflows
124          * 16-bits.
125          */
126         if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
127                 ASSERT(!to->count && !to->usedbytes);
128                 ASSERT(geo->blksize > USHRT_MAX);
129                 to->firstused = geo->blksize;
130         }
131 }
132
133 static void
134 xfs_attr3_leaf_firstused_to_disk(
135         struct xfs_da_geometry          *geo,
136         struct xfs_attr_leafblock       *to,
137         struct xfs_attr3_icleaf_hdr     *from)
138 {
139         struct xfs_attr3_leaf_hdr       *hdr3;
140         uint32_t                        firstused;
141
142         /* magic value should only be seen on disk */
143         ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
144
145         /*
146          * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
147          * value. This only overflows at the max supported value of 64k. Use the
148          * magic on-disk value to represent block size in this case.
149          */
150         firstused = from->firstused;
151         if (firstused > USHRT_MAX) {
152                 ASSERT(from->firstused == geo->blksize);
153                 firstused = XFS_ATTR3_LEAF_NULLOFF;
154         }
155
156         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
157                 hdr3 = (struct xfs_attr3_leaf_hdr *) to;
158                 hdr3->firstused = cpu_to_be16(firstused);
159         } else {
160                 to->hdr.firstused = cpu_to_be16(firstused);
161         }
162 }
163
164 void
165 xfs_attr3_leaf_hdr_from_disk(
166         struct xfs_da_geometry          *geo,
167         struct xfs_attr3_icleaf_hdr     *to,
168         struct xfs_attr_leafblock       *from)
169 {
170         int     i;
171
172         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
173                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
174
175         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
176                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
177
178                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
179                 to->back = be32_to_cpu(hdr3->info.hdr.back);
180                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
181                 to->count = be16_to_cpu(hdr3->count);
182                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
183                 xfs_attr3_leaf_firstused_from_disk(geo, to, from);
184                 to->holes = hdr3->holes;
185
186                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
187                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
188                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
189                 }
190                 return;
191         }
192         to->forw = be32_to_cpu(from->hdr.info.forw);
193         to->back = be32_to_cpu(from->hdr.info.back);
194         to->magic = be16_to_cpu(from->hdr.info.magic);
195         to->count = be16_to_cpu(from->hdr.count);
196         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
197         xfs_attr3_leaf_firstused_from_disk(geo, to, from);
198         to->holes = from->hdr.holes;
199
200         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
201                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
202                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
203         }
204 }
205
206 void
207 xfs_attr3_leaf_hdr_to_disk(
208         struct xfs_da_geometry          *geo,
209         struct xfs_attr_leafblock       *to,
210         struct xfs_attr3_icleaf_hdr     *from)
211 {
212         int                             i;
213
214         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
215                from->magic == XFS_ATTR3_LEAF_MAGIC);
216
217         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
218                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
219
220                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
221                 hdr3->info.hdr.back = cpu_to_be32(from->back);
222                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
223                 hdr3->count = cpu_to_be16(from->count);
224                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
225                 xfs_attr3_leaf_firstused_to_disk(geo, to, from);
226                 hdr3->holes = from->holes;
227                 hdr3->pad1 = 0;
228
229                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
230                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
231                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
232                 }
233                 return;
234         }
235         to->hdr.info.forw = cpu_to_be32(from->forw);
236         to->hdr.info.back = cpu_to_be32(from->back);
237         to->hdr.info.magic = cpu_to_be16(from->magic);
238         to->hdr.count = cpu_to_be16(from->count);
239         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
240         xfs_attr3_leaf_firstused_to_disk(geo, to, from);
241         to->hdr.holes = from->holes;
242         to->hdr.pad1 = 0;
243
244         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
245                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
246                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
247         }
248 }
249
250 static bool
251 xfs_attr3_leaf_verify(
252         struct xfs_buf          *bp)
253 {
254         struct xfs_mount        *mp = bp->b_target->bt_mount;
255         struct xfs_attr_leafblock *leaf = bp->b_addr;
256         struct xfs_perag *pag = bp->b_pag;
257         struct xfs_attr3_icleaf_hdr ichdr;
258
259         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
260
261         if (xfs_sb_version_hascrc(&mp->m_sb)) {
262                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
263
264                 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
265                         return false;
266
267                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
268                         return false;
269                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
270                         return false;
271                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
272                         return false;
273         } else {
274                 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
275                         return false;
276         }
277         /*
278          * In recovery there is a transient state where count == 0 is valid
279          * because we may have transitioned an empty shortform attr to a leaf
280          * if the attr didn't fit in shortform.
281          */
282         if (pag && pag->pagf_init && ichdr.count == 0)
283                 return false;
284
285         /* XXX: need to range check rest of attr header values */
286         /* XXX: hash order check? */
287
288         return true;
289 }
290
291 static void
292 xfs_attr3_leaf_write_verify(
293         struct xfs_buf  *bp)
294 {
295         struct xfs_mount        *mp = bp->b_target->bt_mount;
296         struct xfs_buf_log_item *bip = bp->b_fspriv;
297         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
298
299         if (!xfs_attr3_leaf_verify(bp)) {
300                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
301                 xfs_verifier_error(bp);
302                 return;
303         }
304
305         if (!xfs_sb_version_hascrc(&mp->m_sb))
306                 return;
307
308         if (bip)
309                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
310
311         xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
312 }
313
314 /*
315  * leaf/node format detection on trees is sketchy, so a node read can be done on
316  * leaf level blocks when detection identifies the tree as a node format tree
317  * incorrectly. In this case, we need to swap the verifier to match the correct
318  * format of the block being read.
319  */
320 static void
321 xfs_attr3_leaf_read_verify(
322         struct xfs_buf          *bp)
323 {
324         struct xfs_mount        *mp = bp->b_target->bt_mount;
325
326         if (xfs_sb_version_hascrc(&mp->m_sb) &&
327              !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
328                 xfs_buf_ioerror(bp, -EFSBADCRC);
329         else if (!xfs_attr3_leaf_verify(bp))
330                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
331
332         if (bp->b_error)
333                 xfs_verifier_error(bp);
334 }
335
336 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
337         .name = "xfs_attr3_leaf",
338         .verify_read = xfs_attr3_leaf_read_verify,
339         .verify_write = xfs_attr3_leaf_write_verify,
340 };
341
342 int
343 xfs_attr3_leaf_read(
344         struct xfs_trans        *tp,
345         struct xfs_inode        *dp,
346         xfs_dablk_t             bno,
347         xfs_daddr_t             mappedbno,
348         struct xfs_buf          **bpp)
349 {
350         int                     err;
351
352         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
353                                 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
354         if (!err && tp && *bpp)
355                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
356         return err;
357 }
358
359 /*========================================================================
360  * Namespace helper routines
361  *========================================================================*/
362
363 /*
364  * If namespace bits don't match return 0.
365  * If all match then return 1.
366  */
367 STATIC int
368 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
369 {
370         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
371 }
372
373
374 /*========================================================================
375  * External routines when attribute fork size < XFS_LITINO(mp).
376  *========================================================================*/
377
378 /*
379  * Query whether the requested number of additional bytes of extended
380  * attribute space will be able to fit inline.
381  *
382  * Returns zero if not, else the di_forkoff fork offset to be used in the
383  * literal area for attribute data once the new bytes have been added.
384  *
385  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
386  * special case for dev/uuid inodes, they have fixed size data forks.
387  */
388 int
389 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
390 {
391         int offset;
392         int minforkoff; /* lower limit on valid forkoff locations */
393         int maxforkoff; /* upper limit on valid forkoff locations */
394         int dsize;
395         xfs_mount_t *mp = dp->i_mount;
396
397         /* rounded down */
398         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
399
400         if (dp->i_d.di_format == XFS_DINODE_FMT_DEV) {
401                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
402                 return (offset >= minforkoff) ? minforkoff : 0;
403         }
404
405         /*
406          * If the requested numbers of bytes is smaller or equal to the
407          * current attribute fork size we can always proceed.
408          *
409          * Note that if_bytes in the data fork might actually be larger than
410          * the current data fork size is due to delalloc extents. In that
411          * case either the extent count will go down when they are converted
412          * to real extents, or the delalloc conversion will take care of the
413          * literal area rebalancing.
414          */
415         if (bytes <= XFS_IFORK_ASIZE(dp))
416                 return dp->i_d.di_forkoff;
417
418         /*
419          * For attr2 we can try to move the forkoff if there is space in the
420          * literal area, but for the old format we are done if there is no
421          * space in the fixed attribute fork.
422          */
423         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
424                 return 0;
425
426         dsize = dp->i_df.if_bytes;
427
428         switch (dp->i_d.di_format) {
429         case XFS_DINODE_FMT_EXTENTS:
430                 /*
431                  * If there is no attr fork and the data fork is extents, 
432                  * determine if creating the default attr fork will result
433                  * in the extents form migrating to btree. If so, the
434                  * minimum offset only needs to be the space required for
435                  * the btree root.
436                  */
437                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
438                     xfs_default_attroffset(dp))
439                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
440                 break;
441         case XFS_DINODE_FMT_BTREE:
442                 /*
443                  * If we have a data btree then keep forkoff if we have one,
444                  * otherwise we are adding a new attr, so then we set
445                  * minforkoff to where the btree root can finish so we have
446                  * plenty of room for attrs
447                  */
448                 if (dp->i_d.di_forkoff) {
449                         if (offset < dp->i_d.di_forkoff)
450                                 return 0;
451                         return dp->i_d.di_forkoff;
452                 }
453                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
454                 break;
455         }
456
457         /*
458          * A data fork btree root must have space for at least
459          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
460          */
461         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
462         minforkoff = roundup(minforkoff, 8) >> 3;
463
464         /* attr fork btree root can have at least this many key/ptr pairs */
465         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
466                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
467         maxforkoff = maxforkoff >> 3;   /* rounded down */
468
469         if (offset >= maxforkoff)
470                 return maxforkoff;
471         if (offset >= minforkoff)
472                 return offset;
473         return 0;
474 }
475
476 /*
477  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
478  */
479 STATIC void
480 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
481 {
482         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
483             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
484                 spin_lock(&mp->m_sb_lock);
485                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
486                         xfs_sb_version_addattr2(&mp->m_sb);
487                         spin_unlock(&mp->m_sb_lock);
488                         xfs_log_sb(tp);
489                 } else
490                         spin_unlock(&mp->m_sb_lock);
491         }
492 }
493
494 /*
495  * Create the initial contents of a shortform attribute list.
496  */
497 void
498 xfs_attr_shortform_create(xfs_da_args_t *args)
499 {
500         xfs_attr_sf_hdr_t *hdr;
501         xfs_inode_t *dp;
502         xfs_ifork_t *ifp;
503
504         trace_xfs_attr_sf_create(args);
505
506         dp = args->dp;
507         ASSERT(dp != NULL);
508         ifp = dp->i_afp;
509         ASSERT(ifp != NULL);
510         ASSERT(ifp->if_bytes == 0);
511         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
512                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
513                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
514                 ifp->if_flags |= XFS_IFINLINE;
515         } else {
516                 ASSERT(ifp->if_flags & XFS_IFINLINE);
517         }
518         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
519         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
520         hdr->count = 0;
521         hdr->totsize = cpu_to_be16(sizeof(*hdr));
522         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
523 }
524
525 /*
526  * Add a name/value pair to the shortform attribute list.
527  * Overflow from the inode has already been checked for.
528  */
529 void
530 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
531 {
532         xfs_attr_shortform_t *sf;
533         xfs_attr_sf_entry_t *sfe;
534         int i, offset, size;
535         xfs_mount_t *mp;
536         xfs_inode_t *dp;
537         xfs_ifork_t *ifp;
538
539         trace_xfs_attr_sf_add(args);
540
541         dp = args->dp;
542         mp = dp->i_mount;
543         dp->i_d.di_forkoff = forkoff;
544
545         ifp = dp->i_afp;
546         ASSERT(ifp->if_flags & XFS_IFINLINE);
547         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
548         sfe = &sf->list[0];
549         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
550 #ifdef DEBUG
551                 if (sfe->namelen != args->namelen)
552                         continue;
553                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
554                         continue;
555                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
556                         continue;
557                 ASSERT(0);
558 #endif
559         }
560
561         offset = (char *)sfe - (char *)sf;
562         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
563         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
564         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
565         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
566
567         sfe->namelen = args->namelen;
568         sfe->valuelen = args->valuelen;
569         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
570         memcpy(sfe->nameval, args->name, args->namelen);
571         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
572         sf->hdr.count++;
573         be16_add_cpu(&sf->hdr.totsize, size);
574         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
575
576         xfs_sbversion_add_attr2(mp, args->trans);
577 }
578
579 /*
580  * After the last attribute is removed revert to original inode format,
581  * making all literal area available to the data fork once more.
582  */
583 void
584 xfs_attr_fork_remove(
585         struct xfs_inode        *ip,
586         struct xfs_trans        *tp)
587 {
588         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
589         ip->i_d.di_forkoff = 0;
590         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
591
592         ASSERT(ip->i_d.di_anextents == 0);
593         ASSERT(ip->i_afp == NULL);
594
595         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
596 }
597
598 /*
599  * Remove an attribute from the shortform attribute list structure.
600  */
601 int
602 xfs_attr_shortform_remove(xfs_da_args_t *args)
603 {
604         xfs_attr_shortform_t *sf;
605         xfs_attr_sf_entry_t *sfe;
606         int base, size=0, end, totsize, i;
607         xfs_mount_t *mp;
608         xfs_inode_t *dp;
609
610         trace_xfs_attr_sf_remove(args);
611
612         dp = args->dp;
613         mp = dp->i_mount;
614         base = sizeof(xfs_attr_sf_hdr_t);
615         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
616         sfe = &sf->list[0];
617         end = sf->hdr.count;
618         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
619                                         base += size, i++) {
620                 size = XFS_ATTR_SF_ENTSIZE(sfe);
621                 if (sfe->namelen != args->namelen)
622                         continue;
623                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
624                         continue;
625                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
626                         continue;
627                 break;
628         }
629         if (i == end)
630                 return -ENOATTR;
631
632         /*
633          * Fix up the attribute fork data, covering the hole
634          */
635         end = base + size;
636         totsize = be16_to_cpu(sf->hdr.totsize);
637         if (end != totsize)
638                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
639         sf->hdr.count--;
640         be16_add_cpu(&sf->hdr.totsize, -size);
641
642         /*
643          * Fix up the start offset of the attribute fork
644          */
645         totsize -= size;
646         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
647             (mp->m_flags & XFS_MOUNT_ATTR2) &&
648             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
649             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
650                 xfs_attr_fork_remove(dp, args->trans);
651         } else {
652                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
653                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
654                 ASSERT(dp->i_d.di_forkoff);
655                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
656                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
657                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
658                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
659                 xfs_trans_log_inode(args->trans, dp,
660                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
661         }
662
663         xfs_sbversion_add_attr2(mp, args->trans);
664
665         return 0;
666 }
667
668 /*
669  * Look up a name in a shortform attribute list structure.
670  */
671 /*ARGSUSED*/
672 int
673 xfs_attr_shortform_lookup(xfs_da_args_t *args)
674 {
675         xfs_attr_shortform_t *sf;
676         xfs_attr_sf_entry_t *sfe;
677         int i;
678         xfs_ifork_t *ifp;
679
680         trace_xfs_attr_sf_lookup(args);
681
682         ifp = args->dp->i_afp;
683         ASSERT(ifp->if_flags & XFS_IFINLINE);
684         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
685         sfe = &sf->list[0];
686         for (i = 0; i < sf->hdr.count;
687                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
688                 if (sfe->namelen != args->namelen)
689                         continue;
690                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
691                         continue;
692                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
693                         continue;
694                 return -EEXIST;
695         }
696         return -ENOATTR;
697 }
698
699 /*
700  * Look up a name in a shortform attribute list structure.
701  */
702 /*ARGSUSED*/
703 int
704 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
705 {
706         xfs_attr_shortform_t *sf;
707         xfs_attr_sf_entry_t *sfe;
708         int i;
709
710         ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
711         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
712         sfe = &sf->list[0];
713         for (i = 0; i < sf->hdr.count;
714                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
715                 if (sfe->namelen != args->namelen)
716                         continue;
717                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
718                         continue;
719                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
720                         continue;
721                 if (args->flags & ATTR_KERNOVAL) {
722                         args->valuelen = sfe->valuelen;
723                         return -EEXIST;
724                 }
725                 if (args->valuelen < sfe->valuelen) {
726                         args->valuelen = sfe->valuelen;
727                         return -ERANGE;
728                 }
729                 args->valuelen = sfe->valuelen;
730                 memcpy(args->value, &sfe->nameval[args->namelen],
731                                                     args->valuelen);
732                 return -EEXIST;
733         }
734         return -ENOATTR;
735 }
736
737 /*
738  * Convert from using the shortform to the leaf.  On success, return the
739  * buffer so that we can keep it locked until we're totally done with it.
740  */
741 int
742 xfs_attr_shortform_to_leaf(
743         struct xfs_da_args      *args,
744         struct xfs_buf          **leaf_bp)
745 {
746         xfs_inode_t *dp;
747         xfs_attr_shortform_t *sf;
748         xfs_attr_sf_entry_t *sfe;
749         xfs_da_args_t nargs;
750         char *tmpbuffer;
751         int error, i, size;
752         xfs_dablk_t blkno;
753         struct xfs_buf *bp;
754         xfs_ifork_t *ifp;
755
756         trace_xfs_attr_sf_to_leaf(args);
757
758         dp = args->dp;
759         ifp = dp->i_afp;
760         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
761         size = be16_to_cpu(sf->hdr.totsize);
762         tmpbuffer = kmem_alloc(size, KM_SLEEP);
763         ASSERT(tmpbuffer != NULL);
764         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
765         sf = (xfs_attr_shortform_t *)tmpbuffer;
766
767         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
768         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
769
770         bp = NULL;
771         error = xfs_da_grow_inode(args, &blkno);
772         if (error) {
773                 /*
774                  * If we hit an IO error middle of the transaction inside
775                  * grow_inode(), we may have inconsistent data. Bail out.
776                  */
777                 if (error == -EIO)
778                         goto out;
779                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
780                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
781                 goto out;
782         }
783
784         ASSERT(blkno == 0);
785         error = xfs_attr3_leaf_create(args, blkno, &bp);
786         if (error) {
787                 error = xfs_da_shrink_inode(args, 0, bp);
788                 bp = NULL;
789                 if (error)
790                         goto out;
791                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
792                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
793                 goto out;
794         }
795
796         memset((char *)&nargs, 0, sizeof(nargs));
797         nargs.dp = dp;
798         nargs.geo = args->geo;
799         nargs.firstblock = args->firstblock;
800         nargs.dfops = args->dfops;
801         nargs.total = args->total;
802         nargs.whichfork = XFS_ATTR_FORK;
803         nargs.trans = args->trans;
804         nargs.op_flags = XFS_DA_OP_OKNOENT;
805
806         sfe = &sf->list[0];
807         for (i = 0; i < sf->hdr.count; i++) {
808                 nargs.name = sfe->nameval;
809                 nargs.namelen = sfe->namelen;
810                 nargs.value = &sfe->nameval[nargs.namelen];
811                 nargs.valuelen = sfe->valuelen;
812                 nargs.hashval = xfs_da_hashname(sfe->nameval,
813                                                 sfe->namelen);
814                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
815                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
816                 ASSERT(error == -ENOATTR);
817                 error = xfs_attr3_leaf_add(bp, &nargs);
818                 ASSERT(error != -ENOSPC);
819                 if (error)
820                         goto out;
821                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
822         }
823         error = 0;
824         *leaf_bp = bp;
825 out:
826         kmem_free(tmpbuffer);
827         return error;
828 }
829
830 /*
831  * Check a leaf attribute block to see if all the entries would fit into
832  * a shortform attribute list.
833  */
834 int
835 xfs_attr_shortform_allfit(
836         struct xfs_buf          *bp,
837         struct xfs_inode        *dp)
838 {
839         struct xfs_attr_leafblock *leaf;
840         struct xfs_attr_leaf_entry *entry;
841         xfs_attr_leaf_name_local_t *name_loc;
842         struct xfs_attr3_icleaf_hdr leafhdr;
843         int                     bytes;
844         int                     i;
845         struct xfs_mount        *mp = bp->b_target->bt_mount;
846
847         leaf = bp->b_addr;
848         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
849         entry = xfs_attr3_leaf_entryp(leaf);
850
851         bytes = sizeof(struct xfs_attr_sf_hdr);
852         for (i = 0; i < leafhdr.count; entry++, i++) {
853                 if (entry->flags & XFS_ATTR_INCOMPLETE)
854                         continue;               /* don't copy partial entries */
855                 if (!(entry->flags & XFS_ATTR_LOCAL))
856                         return 0;
857                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
858                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
859                         return 0;
860                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
861                         return 0;
862                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
863                                 + name_loc->namelen
864                                 + be16_to_cpu(name_loc->valuelen);
865         }
866         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
867             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
868             (bytes == sizeof(struct xfs_attr_sf_hdr)))
869                 return -1;
870         return xfs_attr_shortform_bytesfit(dp, bytes);
871 }
872
873 /*
874  * Convert a leaf attribute list to shortform attribute list
875  */
876 int
877 xfs_attr3_leaf_to_shortform(
878         struct xfs_buf          *bp,
879         struct xfs_da_args      *args,
880         int                     forkoff)
881 {
882         struct xfs_attr_leafblock *leaf;
883         struct xfs_attr3_icleaf_hdr ichdr;
884         struct xfs_attr_leaf_entry *entry;
885         struct xfs_attr_leaf_name_local *name_loc;
886         struct xfs_da_args      nargs;
887         struct xfs_inode        *dp = args->dp;
888         char                    *tmpbuffer;
889         int                     error;
890         int                     i;
891
892         trace_xfs_attr_leaf_to_sf(args);
893
894         tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
895         if (!tmpbuffer)
896                 return -ENOMEM;
897
898         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
899
900         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
901         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
902         entry = xfs_attr3_leaf_entryp(leaf);
903
904         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
905         memset(bp->b_addr, 0, args->geo->blksize);
906
907         /*
908          * Clean out the prior contents of the attribute list.
909          */
910         error = xfs_da_shrink_inode(args, 0, bp);
911         if (error)
912                 goto out;
913
914         if (forkoff == -1) {
915                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
916                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
917                 xfs_attr_fork_remove(dp, args->trans);
918                 goto out;
919         }
920
921         xfs_attr_shortform_create(args);
922
923         /*
924          * Copy the attributes
925          */
926         memset((char *)&nargs, 0, sizeof(nargs));
927         nargs.geo = args->geo;
928         nargs.dp = dp;
929         nargs.firstblock = args->firstblock;
930         nargs.dfops = args->dfops;
931         nargs.total = args->total;
932         nargs.whichfork = XFS_ATTR_FORK;
933         nargs.trans = args->trans;
934         nargs.op_flags = XFS_DA_OP_OKNOENT;
935
936         for (i = 0; i < ichdr.count; entry++, i++) {
937                 if (entry->flags & XFS_ATTR_INCOMPLETE)
938                         continue;       /* don't copy partial entries */
939                 if (!entry->nameidx)
940                         continue;
941                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
942                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
943                 nargs.name = name_loc->nameval;
944                 nargs.namelen = name_loc->namelen;
945                 nargs.value = &name_loc->nameval[nargs.namelen];
946                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
947                 nargs.hashval = be32_to_cpu(entry->hashval);
948                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
949                 xfs_attr_shortform_add(&nargs, forkoff);
950         }
951         error = 0;
952
953 out:
954         kmem_free(tmpbuffer);
955         return error;
956 }
957
958 /*
959  * Convert from using a single leaf to a root node and a leaf.
960  */
961 int
962 xfs_attr3_leaf_to_node(
963         struct xfs_da_args      *args)
964 {
965         struct xfs_attr_leafblock *leaf;
966         struct xfs_attr3_icleaf_hdr icleafhdr;
967         struct xfs_attr_leaf_entry *entries;
968         struct xfs_da_node_entry *btree;
969         struct xfs_da3_icnode_hdr icnodehdr;
970         struct xfs_da_intnode   *node;
971         struct xfs_inode        *dp = args->dp;
972         struct xfs_mount        *mp = dp->i_mount;
973         struct xfs_buf          *bp1 = NULL;
974         struct xfs_buf          *bp2 = NULL;
975         xfs_dablk_t             blkno;
976         int                     error;
977
978         trace_xfs_attr_leaf_to_node(args);
979
980         error = xfs_da_grow_inode(args, &blkno);
981         if (error)
982                 goto out;
983         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
984         if (error)
985                 goto out;
986
987         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
988         if (error)
989                 goto out;
990
991         /* copy leaf to new buffer, update identifiers */
992         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
993         bp2->b_ops = bp1->b_ops;
994         memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
995         if (xfs_sb_version_hascrc(&mp->m_sb)) {
996                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
997                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
998         }
999         xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
1000
1001         /*
1002          * Set up the new root node.
1003          */
1004         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1005         if (error)
1006                 goto out;
1007         node = bp1->b_addr;
1008         dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1009         btree = dp->d_ops->node_tree_p(node);
1010
1011         leaf = bp2->b_addr;
1012         xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1013         entries = xfs_attr3_leaf_entryp(leaf);
1014
1015         /* both on-disk, don't endian-flip twice */
1016         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1017         btree[0].before = cpu_to_be32(blkno);
1018         icnodehdr.count = 1;
1019         dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1020         xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1021         error = 0;
1022 out:
1023         return error;
1024 }
1025
1026 /*========================================================================
1027  * Routines used for growing the Btree.
1028  *========================================================================*/
1029
1030 /*
1031  * Create the initial contents of a leaf attribute list
1032  * or a leaf in a node attribute list.
1033  */
1034 STATIC int
1035 xfs_attr3_leaf_create(
1036         struct xfs_da_args      *args,
1037         xfs_dablk_t             blkno,
1038         struct xfs_buf          **bpp)
1039 {
1040         struct xfs_attr_leafblock *leaf;
1041         struct xfs_attr3_icleaf_hdr ichdr;
1042         struct xfs_inode        *dp = args->dp;
1043         struct xfs_mount        *mp = dp->i_mount;
1044         struct xfs_buf          *bp;
1045         int                     error;
1046
1047         trace_xfs_attr_leaf_create(args);
1048
1049         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1050                                             XFS_ATTR_FORK);
1051         if (error)
1052                 return error;
1053         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1054         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1055         leaf = bp->b_addr;
1056         memset(leaf, 0, args->geo->blksize);
1057
1058         memset(&ichdr, 0, sizeof(ichdr));
1059         ichdr.firstused = args->geo->blksize;
1060
1061         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1062                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1063
1064                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1065
1066                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1067                 hdr3->owner = cpu_to_be64(dp->i_ino);
1068                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1069
1070                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1071         } else {
1072                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1073                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1074         }
1075         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1076
1077         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1078         xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1079
1080         *bpp = bp;
1081         return 0;
1082 }
1083
1084 /*
1085  * Split the leaf node, rebalance, then add the new entry.
1086  */
1087 int
1088 xfs_attr3_leaf_split(
1089         struct xfs_da_state     *state,
1090         struct xfs_da_state_blk *oldblk,
1091         struct xfs_da_state_blk *newblk)
1092 {
1093         xfs_dablk_t blkno;
1094         int error;
1095
1096         trace_xfs_attr_leaf_split(state->args);
1097
1098         /*
1099          * Allocate space for a new leaf node.
1100          */
1101         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1102         error = xfs_da_grow_inode(state->args, &blkno);
1103         if (error)
1104                 return error;
1105         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1106         if (error)
1107                 return error;
1108         newblk->blkno = blkno;
1109         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1110
1111         /*
1112          * Rebalance the entries across the two leaves.
1113          * NOTE: rebalance() currently depends on the 2nd block being empty.
1114          */
1115         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1116         error = xfs_da3_blk_link(state, oldblk, newblk);
1117         if (error)
1118                 return error;
1119
1120         /*
1121          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1122          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1123          * "new" attrs info.  Will need the "old" info to remove it later.
1124          *
1125          * Insert the "new" entry in the correct block.
1126          */
1127         if (state->inleaf) {
1128                 trace_xfs_attr_leaf_add_old(state->args);
1129                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1130         } else {
1131                 trace_xfs_attr_leaf_add_new(state->args);
1132                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1133         }
1134
1135         /*
1136          * Update last hashval in each block since we added the name.
1137          */
1138         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1139         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1140         return error;
1141 }
1142
1143 /*
1144  * Add a name to the leaf attribute list structure.
1145  */
1146 int
1147 xfs_attr3_leaf_add(
1148         struct xfs_buf          *bp,
1149         struct xfs_da_args      *args)
1150 {
1151         struct xfs_attr_leafblock *leaf;
1152         struct xfs_attr3_icleaf_hdr ichdr;
1153         int                     tablesize;
1154         int                     entsize;
1155         int                     sum;
1156         int                     tmp;
1157         int                     i;
1158
1159         trace_xfs_attr_leaf_add(args);
1160
1161         leaf = bp->b_addr;
1162         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1163         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1164         entsize = xfs_attr_leaf_newentsize(args, NULL);
1165
1166         /*
1167          * Search through freemap for first-fit on new name length.
1168          * (may need to figure in size of entry struct too)
1169          */
1170         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1171                                         + xfs_attr3_leaf_hdr_size(leaf);
1172         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1173                 if (tablesize > ichdr.firstused) {
1174                         sum += ichdr.freemap[i].size;
1175                         continue;
1176                 }
1177                 if (!ichdr.freemap[i].size)
1178                         continue;       /* no space in this map */
1179                 tmp = entsize;
1180                 if (ichdr.freemap[i].base < ichdr.firstused)
1181                         tmp += sizeof(xfs_attr_leaf_entry_t);
1182                 if (ichdr.freemap[i].size >= tmp) {
1183                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1184                         goto out_log_hdr;
1185                 }
1186                 sum += ichdr.freemap[i].size;
1187         }
1188
1189         /*
1190          * If there are no holes in the address space of the block,
1191          * and we don't have enough freespace, then compaction will do us
1192          * no good and we should just give up.
1193          */
1194         if (!ichdr.holes && sum < entsize)
1195                 return -ENOSPC;
1196
1197         /*
1198          * Compact the entries to coalesce free space.
1199          * This may change the hdr->count via dropping INCOMPLETE entries.
1200          */
1201         xfs_attr3_leaf_compact(args, &ichdr, bp);
1202
1203         /*
1204          * After compaction, the block is guaranteed to have only one
1205          * free region, in freemap[0].  If it is not big enough, give up.
1206          */
1207         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1208                 tmp = -ENOSPC;
1209                 goto out_log_hdr;
1210         }
1211
1212         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1213
1214 out_log_hdr:
1215         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1216         xfs_trans_log_buf(args->trans, bp,
1217                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1218                                 xfs_attr3_leaf_hdr_size(leaf)));
1219         return tmp;
1220 }
1221
1222 /*
1223  * Add a name to a leaf attribute list structure.
1224  */
1225 STATIC int
1226 xfs_attr3_leaf_add_work(
1227         struct xfs_buf          *bp,
1228         struct xfs_attr3_icleaf_hdr *ichdr,
1229         struct xfs_da_args      *args,
1230         int                     mapindex)
1231 {
1232         struct xfs_attr_leafblock *leaf;
1233         struct xfs_attr_leaf_entry *entry;
1234         struct xfs_attr_leaf_name_local *name_loc;
1235         struct xfs_attr_leaf_name_remote *name_rmt;
1236         struct xfs_mount        *mp;
1237         int                     tmp;
1238         int                     i;
1239
1240         trace_xfs_attr_leaf_add_work(args);
1241
1242         leaf = bp->b_addr;
1243         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1244         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1245
1246         /*
1247          * Force open some space in the entry array and fill it in.
1248          */
1249         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1250         if (args->index < ichdr->count) {
1251                 tmp  = ichdr->count - args->index;
1252                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1253                 memmove(entry + 1, entry, tmp);
1254                 xfs_trans_log_buf(args->trans, bp,
1255                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1256         }
1257         ichdr->count++;
1258
1259         /*
1260          * Allocate space for the new string (at the end of the run).
1261          */
1262         mp = args->trans->t_mountp;
1263         ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1264         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1265         ASSERT(ichdr->freemap[mapindex].size >=
1266                 xfs_attr_leaf_newentsize(args, NULL));
1267         ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1268         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1269
1270         ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1271
1272         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1273                                      ichdr->freemap[mapindex].size);
1274         entry->hashval = cpu_to_be32(args->hashval);
1275         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1276         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1277         if (args->op_flags & XFS_DA_OP_RENAME) {
1278                 entry->flags |= XFS_ATTR_INCOMPLETE;
1279                 if ((args->blkno2 == args->blkno) &&
1280                     (args->index2 <= args->index)) {
1281                         args->index2++;
1282                 }
1283         }
1284         xfs_trans_log_buf(args->trans, bp,
1285                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1286         ASSERT((args->index == 0) ||
1287                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1288         ASSERT((args->index == ichdr->count - 1) ||
1289                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1290
1291         /*
1292          * For "remote" attribute values, simply note that we need to
1293          * allocate space for the "remote" value.  We can't actually
1294          * allocate the extents in this transaction, and we can't decide
1295          * which blocks they should be as we might allocate more blocks
1296          * as part of this transaction (a split operation for example).
1297          */
1298         if (entry->flags & XFS_ATTR_LOCAL) {
1299                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1300                 name_loc->namelen = args->namelen;
1301                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1302                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1303                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1304                                    be16_to_cpu(name_loc->valuelen));
1305         } else {
1306                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1307                 name_rmt->namelen = args->namelen;
1308                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1309                 entry->flags |= XFS_ATTR_INCOMPLETE;
1310                 /* just in case */
1311                 name_rmt->valuelen = 0;
1312                 name_rmt->valueblk = 0;
1313                 args->rmtblkno = 1;
1314                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1315                 args->rmtvaluelen = args->valuelen;
1316         }
1317         xfs_trans_log_buf(args->trans, bp,
1318              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1319                                    xfs_attr_leaf_entsize(leaf, args->index)));
1320
1321         /*
1322          * Update the control info for this leaf node
1323          */
1324         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1325                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1326
1327         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1328                                         + xfs_attr3_leaf_hdr_size(leaf));
1329         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1330                                         + xfs_attr3_leaf_hdr_size(leaf);
1331
1332         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1333                 if (ichdr->freemap[i].base == tmp) {
1334                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1335                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1336                 }
1337         }
1338         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1339         return 0;
1340 }
1341
1342 /*
1343  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1344  */
1345 STATIC void
1346 xfs_attr3_leaf_compact(
1347         struct xfs_da_args      *args,
1348         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1349         struct xfs_buf          *bp)
1350 {
1351         struct xfs_attr_leafblock *leaf_src;
1352         struct xfs_attr_leafblock *leaf_dst;
1353         struct xfs_attr3_icleaf_hdr ichdr_src;
1354         struct xfs_trans        *trans = args->trans;
1355         char                    *tmpbuffer;
1356
1357         trace_xfs_attr_leaf_compact(args);
1358
1359         tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1360         memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1361         memset(bp->b_addr, 0, args->geo->blksize);
1362         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1363         leaf_dst = bp->b_addr;
1364
1365         /*
1366          * Copy the on-disk header back into the destination buffer to ensure
1367          * all the information in the header that is not part of the incore
1368          * header structure is preserved.
1369          */
1370         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1371
1372         /* Initialise the incore headers */
1373         ichdr_src = *ichdr_dst; /* struct copy */
1374         ichdr_dst->firstused = args->geo->blksize;
1375         ichdr_dst->usedbytes = 0;
1376         ichdr_dst->count = 0;
1377         ichdr_dst->holes = 0;
1378         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1379         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1380                                                 ichdr_dst->freemap[0].base;
1381
1382         /* write the header back to initialise the underlying buffer */
1383         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1384
1385         /*
1386          * Copy all entry's in the same (sorted) order,
1387          * but allocate name/value pairs packed and in sequence.
1388          */
1389         xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1390                                 leaf_dst, ichdr_dst, 0, ichdr_src.count);
1391         /*
1392          * this logs the entire buffer, but the caller must write the header
1393          * back to the buffer when it is finished modifying it.
1394          */
1395         xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1396
1397         kmem_free(tmpbuffer);
1398 }
1399
1400 /*
1401  * Compare two leaf blocks "order".
1402  * Return 0 unless leaf2 should go before leaf1.
1403  */
1404 static int
1405 xfs_attr3_leaf_order(
1406         struct xfs_buf  *leaf1_bp,
1407         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1408         struct xfs_buf  *leaf2_bp,
1409         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1410 {
1411         struct xfs_attr_leaf_entry *entries1;
1412         struct xfs_attr_leaf_entry *entries2;
1413
1414         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1415         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1416         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1417             ((be32_to_cpu(entries2[0].hashval) <
1418               be32_to_cpu(entries1[0].hashval)) ||
1419              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1420               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1421                 return 1;
1422         }
1423         return 0;
1424 }
1425
1426 int
1427 xfs_attr_leaf_order(
1428         struct xfs_buf  *leaf1_bp,
1429         struct xfs_buf  *leaf2_bp)
1430 {
1431         struct xfs_attr3_icleaf_hdr ichdr1;
1432         struct xfs_attr3_icleaf_hdr ichdr2;
1433         struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1434
1435         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1436         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1437         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1438 }
1439
1440 /*
1441  * Redistribute the attribute list entries between two leaf nodes,
1442  * taking into account the size of the new entry.
1443  *
1444  * NOTE: if new block is empty, then it will get the upper half of the
1445  * old block.  At present, all (one) callers pass in an empty second block.
1446  *
1447  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1448  * to match what it is doing in splitting the attribute leaf block.  Those
1449  * values are used in "atomic rename" operations on attributes.  Note that
1450  * the "new" and "old" values can end up in different blocks.
1451  */
1452 STATIC void
1453 xfs_attr3_leaf_rebalance(
1454         struct xfs_da_state     *state,
1455         struct xfs_da_state_blk *blk1,
1456         struct xfs_da_state_blk *blk2)
1457 {
1458         struct xfs_da_args      *args;
1459         struct xfs_attr_leafblock *leaf1;
1460         struct xfs_attr_leafblock *leaf2;
1461         struct xfs_attr3_icleaf_hdr ichdr1;
1462         struct xfs_attr3_icleaf_hdr ichdr2;
1463         struct xfs_attr_leaf_entry *entries1;
1464         struct xfs_attr_leaf_entry *entries2;
1465         int                     count;
1466         int                     totallen;
1467         int                     max;
1468         int                     space;
1469         int                     swap;
1470
1471         /*
1472          * Set up environment.
1473          */
1474         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1475         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1476         leaf1 = blk1->bp->b_addr;
1477         leaf2 = blk2->bp->b_addr;
1478         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1479         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1480         ASSERT(ichdr2.count == 0);
1481         args = state->args;
1482
1483         trace_xfs_attr_leaf_rebalance(args);
1484
1485         /*
1486          * Check ordering of blocks, reverse if it makes things simpler.
1487          *
1488          * NOTE: Given that all (current) callers pass in an empty
1489          * second block, this code should never set "swap".
1490          */
1491         swap = 0;
1492         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1493                 struct xfs_da_state_blk *tmp_blk;
1494                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1495
1496                 tmp_blk = blk1;
1497                 blk1 = blk2;
1498                 blk2 = tmp_blk;
1499
1500                 /* struct copies to swap them rather than reconverting */
1501                 tmp_ichdr = ichdr1;
1502                 ichdr1 = ichdr2;
1503                 ichdr2 = tmp_ichdr;
1504
1505                 leaf1 = blk1->bp->b_addr;
1506                 leaf2 = blk2->bp->b_addr;
1507                 swap = 1;
1508         }
1509
1510         /*
1511          * Examine entries until we reduce the absolute difference in
1512          * byte usage between the two blocks to a minimum.  Then get
1513          * the direction to copy and the number of elements to move.
1514          *
1515          * "inleaf" is true if the new entry should be inserted into blk1.
1516          * If "swap" is also true, then reverse the sense of "inleaf".
1517          */
1518         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1519                                                       blk2, &ichdr2,
1520                                                       &count, &totallen);
1521         if (swap)
1522                 state->inleaf = !state->inleaf;
1523
1524         /*
1525          * Move any entries required from leaf to leaf:
1526          */
1527         if (count < ichdr1.count) {
1528                 /*
1529                  * Figure the total bytes to be added to the destination leaf.
1530                  */
1531                 /* number entries being moved */
1532                 count = ichdr1.count - count;
1533                 space  = ichdr1.usedbytes - totallen;
1534                 space += count * sizeof(xfs_attr_leaf_entry_t);
1535
1536                 /*
1537                  * leaf2 is the destination, compact it if it looks tight.
1538                  */
1539                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1540                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1541                 if (space > max)
1542                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1543
1544                 /*
1545                  * Move high entries from leaf1 to low end of leaf2.
1546                  */
1547                 xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1548                                 ichdr1.count - count, leaf2, &ichdr2, 0, count);
1549
1550         } else if (count > ichdr1.count) {
1551                 /*
1552                  * I assert that since all callers pass in an empty
1553                  * second buffer, this code should never execute.
1554                  */
1555                 ASSERT(0);
1556
1557                 /*
1558                  * Figure the total bytes to be added to the destination leaf.
1559                  */
1560                 /* number entries being moved */
1561                 count -= ichdr1.count;
1562                 space  = totallen - ichdr1.usedbytes;
1563                 space += count * sizeof(xfs_attr_leaf_entry_t);
1564
1565                 /*
1566                  * leaf1 is the destination, compact it if it looks tight.
1567                  */
1568                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1569                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1570                 if (space > max)
1571                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1572
1573                 /*
1574                  * Move low entries from leaf2 to high end of leaf1.
1575                  */
1576                 xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1577                                         ichdr1.count, count);
1578         }
1579
1580         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1581         xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1582         xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1583         xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1584
1585         /*
1586          * Copy out last hashval in each block for B-tree code.
1587          */
1588         entries1 = xfs_attr3_leaf_entryp(leaf1);
1589         entries2 = xfs_attr3_leaf_entryp(leaf2);
1590         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1591         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1592
1593         /*
1594          * Adjust the expected index for insertion.
1595          * NOTE: this code depends on the (current) situation that the
1596          * second block was originally empty.
1597          *
1598          * If the insertion point moved to the 2nd block, we must adjust
1599          * the index.  We must also track the entry just following the
1600          * new entry for use in an "atomic rename" operation, that entry
1601          * is always the "old" entry and the "new" entry is what we are
1602          * inserting.  The index/blkno fields refer to the "old" entry,
1603          * while the index2/blkno2 fields refer to the "new" entry.
1604          */
1605         if (blk1->index > ichdr1.count) {
1606                 ASSERT(state->inleaf == 0);
1607                 blk2->index = blk1->index - ichdr1.count;
1608                 args->index = args->index2 = blk2->index;
1609                 args->blkno = args->blkno2 = blk2->blkno;
1610         } else if (blk1->index == ichdr1.count) {
1611                 if (state->inleaf) {
1612                         args->index = blk1->index;
1613                         args->blkno = blk1->blkno;
1614                         args->index2 = 0;
1615                         args->blkno2 = blk2->blkno;
1616                 } else {
1617                         /*
1618                          * On a double leaf split, the original attr location
1619                          * is already stored in blkno2/index2, so don't
1620                          * overwrite it overwise we corrupt the tree.
1621                          */
1622                         blk2->index = blk1->index - ichdr1.count;
1623                         args->index = blk2->index;
1624                         args->blkno = blk2->blkno;
1625                         if (!state->extravalid) {
1626                                 /*
1627                                  * set the new attr location to match the old
1628                                  * one and let the higher level split code
1629                                  * decide where in the leaf to place it.
1630                                  */
1631                                 args->index2 = blk2->index;
1632                                 args->blkno2 = blk2->blkno;
1633                         }
1634                 }
1635         } else {
1636                 ASSERT(state->inleaf == 1);
1637                 args->index = args->index2 = blk1->index;
1638                 args->blkno = args->blkno2 = blk1->blkno;
1639         }
1640 }
1641
1642 /*
1643  * Examine entries until we reduce the absolute difference in
1644  * byte usage between the two blocks to a minimum.
1645  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1646  * GROT: there will always be enough room in either block for a new entry.
1647  * GROT: Do a double-split for this case?
1648  */
1649 STATIC int
1650 xfs_attr3_leaf_figure_balance(
1651         struct xfs_da_state             *state,
1652         struct xfs_da_state_blk         *blk1,
1653         struct xfs_attr3_icleaf_hdr     *ichdr1,
1654         struct xfs_da_state_blk         *blk2,
1655         struct xfs_attr3_icleaf_hdr     *ichdr2,
1656         int                             *countarg,
1657         int                             *usedbytesarg)
1658 {
1659         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1660         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1661         struct xfs_attr_leaf_entry      *entry;
1662         int                             count;
1663         int                             max;
1664         int                             index;
1665         int                             totallen = 0;
1666         int                             half;
1667         int                             lastdelta;
1668         int                             foundit = 0;
1669         int                             tmp;
1670
1671         /*
1672          * Examine entries until we reduce the absolute difference in
1673          * byte usage between the two blocks to a minimum.
1674          */
1675         max = ichdr1->count + ichdr2->count;
1676         half = (max + 1) * sizeof(*entry);
1677         half += ichdr1->usedbytes + ichdr2->usedbytes +
1678                         xfs_attr_leaf_newentsize(state->args, NULL);
1679         half /= 2;
1680         lastdelta = state->args->geo->blksize;
1681         entry = xfs_attr3_leaf_entryp(leaf1);
1682         for (count = index = 0; count < max; entry++, index++, count++) {
1683
1684 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1685                 /*
1686                  * The new entry is in the first block, account for it.
1687                  */
1688                 if (count == blk1->index) {
1689                         tmp = totallen + sizeof(*entry) +
1690                                 xfs_attr_leaf_newentsize(state->args, NULL);
1691                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1692                                 break;
1693                         lastdelta = XFS_ATTR_ABS(half - tmp);
1694                         totallen = tmp;
1695                         foundit = 1;
1696                 }
1697
1698                 /*
1699                  * Wrap around into the second block if necessary.
1700                  */
1701                 if (count == ichdr1->count) {
1702                         leaf1 = leaf2;
1703                         entry = xfs_attr3_leaf_entryp(leaf1);
1704                         index = 0;
1705                 }
1706
1707                 /*
1708                  * Figure out if next leaf entry would be too much.
1709                  */
1710                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1711                                                                         index);
1712                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1713                         break;
1714                 lastdelta = XFS_ATTR_ABS(half - tmp);
1715                 totallen = tmp;
1716 #undef XFS_ATTR_ABS
1717         }
1718
1719         /*
1720          * Calculate the number of usedbytes that will end up in lower block.
1721          * If new entry not in lower block, fix up the count.
1722          */
1723         totallen -= count * sizeof(*entry);
1724         if (foundit) {
1725                 totallen -= sizeof(*entry) +
1726                                 xfs_attr_leaf_newentsize(state->args, NULL);
1727         }
1728
1729         *countarg = count;
1730         *usedbytesarg = totallen;
1731         return foundit;
1732 }
1733
1734 /*========================================================================
1735  * Routines used for shrinking the Btree.
1736  *========================================================================*/
1737
1738 /*
1739  * Check a leaf block and its neighbors to see if the block should be
1740  * collapsed into one or the other neighbor.  Always keep the block
1741  * with the smaller block number.
1742  * If the current block is over 50% full, don't try to join it, return 0.
1743  * If the block is empty, fill in the state structure and return 2.
1744  * If it can be collapsed, fill in the state structure and return 1.
1745  * If nothing can be done, return 0.
1746  *
1747  * GROT: allow for INCOMPLETE entries in calculation.
1748  */
1749 int
1750 xfs_attr3_leaf_toosmall(
1751         struct xfs_da_state     *state,
1752         int                     *action)
1753 {
1754         struct xfs_attr_leafblock *leaf;
1755         struct xfs_da_state_blk *blk;
1756         struct xfs_attr3_icleaf_hdr ichdr;
1757         struct xfs_buf          *bp;
1758         xfs_dablk_t             blkno;
1759         int                     bytes;
1760         int                     forward;
1761         int                     error;
1762         int                     retval;
1763         int                     i;
1764
1765         trace_xfs_attr_leaf_toosmall(state->args);
1766
1767         /*
1768          * Check for the degenerate case of the block being over 50% full.
1769          * If so, it's not worth even looking to see if we might be able
1770          * to coalesce with a sibling.
1771          */
1772         blk = &state->path.blk[ state->path.active-1 ];
1773         leaf = blk->bp->b_addr;
1774         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1775         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1776                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1777                 ichdr.usedbytes;
1778         if (bytes > (state->args->geo->blksize >> 1)) {
1779                 *action = 0;    /* blk over 50%, don't try to join */
1780                 return 0;
1781         }
1782
1783         /*
1784          * Check for the degenerate case of the block being empty.
1785          * If the block is empty, we'll simply delete it, no need to
1786          * coalesce it with a sibling block.  We choose (arbitrarily)
1787          * to merge with the forward block unless it is NULL.
1788          */
1789         if (ichdr.count == 0) {
1790                 /*
1791                  * Make altpath point to the block we want to keep and
1792                  * path point to the block we want to drop (this one).
1793                  */
1794                 forward = (ichdr.forw != 0);
1795                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1796                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1797                                                  0, &retval);
1798                 if (error)
1799                         return error;
1800                 if (retval) {
1801                         *action = 0;
1802                 } else {
1803                         *action = 2;
1804                 }
1805                 return 0;
1806         }
1807
1808         /*
1809          * Examine each sibling block to see if we can coalesce with
1810          * at least 25% free space to spare.  We need to figure out
1811          * whether to merge with the forward or the backward block.
1812          * We prefer coalescing with the lower numbered sibling so as
1813          * to shrink an attribute list over time.
1814          */
1815         /* start with smaller blk num */
1816         forward = ichdr.forw < ichdr.back;
1817         for (i = 0; i < 2; forward = !forward, i++) {
1818                 struct xfs_attr3_icleaf_hdr ichdr2;
1819                 if (forward)
1820                         blkno = ichdr.forw;
1821                 else
1822                         blkno = ichdr.back;
1823                 if (blkno == 0)
1824                         continue;
1825                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1826                                         blkno, -1, &bp);
1827                 if (error)
1828                         return error;
1829
1830                 xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1831
1832                 bytes = state->args->geo->blksize -
1833                         (state->args->geo->blksize >> 2) -
1834                         ichdr.usedbytes - ichdr2.usedbytes -
1835                         ((ichdr.count + ichdr2.count) *
1836                                         sizeof(xfs_attr_leaf_entry_t)) -
1837                         xfs_attr3_leaf_hdr_size(leaf);
1838
1839                 xfs_trans_brelse(state->args->trans, bp);
1840                 if (bytes >= 0)
1841                         break;  /* fits with at least 25% to spare */
1842         }
1843         if (i >= 2) {
1844                 *action = 0;
1845                 return 0;
1846         }
1847
1848         /*
1849          * Make altpath point to the block we want to keep (the lower
1850          * numbered block) and path point to the block we want to drop.
1851          */
1852         memcpy(&state->altpath, &state->path, sizeof(state->path));
1853         if (blkno < blk->blkno) {
1854                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1855                                                  0, &retval);
1856         } else {
1857                 error = xfs_da3_path_shift(state, &state->path, forward,
1858                                                  0, &retval);
1859         }
1860         if (error)
1861                 return error;
1862         if (retval) {
1863                 *action = 0;
1864         } else {
1865                 *action = 1;
1866         }
1867         return 0;
1868 }
1869
1870 /*
1871  * Remove a name from the leaf attribute list structure.
1872  *
1873  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1874  * If two leaves are 37% full, when combined they will leave 25% free.
1875  */
1876 int
1877 xfs_attr3_leaf_remove(
1878         struct xfs_buf          *bp,
1879         struct xfs_da_args      *args)
1880 {
1881         struct xfs_attr_leafblock *leaf;
1882         struct xfs_attr3_icleaf_hdr ichdr;
1883         struct xfs_attr_leaf_entry *entry;
1884         int                     before;
1885         int                     after;
1886         int                     smallest;
1887         int                     entsize;
1888         int                     tablesize;
1889         int                     tmp;
1890         int                     i;
1891
1892         trace_xfs_attr_leaf_remove(args);
1893
1894         leaf = bp->b_addr;
1895         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1896
1897         ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1898         ASSERT(args->index >= 0 && args->index < ichdr.count);
1899         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1900                                         xfs_attr3_leaf_hdr_size(leaf));
1901
1902         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1903
1904         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1905         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1906
1907         /*
1908          * Scan through free region table:
1909          *    check for adjacency of free'd entry with an existing one,
1910          *    find smallest free region in case we need to replace it,
1911          *    adjust any map that borders the entry table,
1912          */
1913         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1914                                         + xfs_attr3_leaf_hdr_size(leaf);
1915         tmp = ichdr.freemap[0].size;
1916         before = after = -1;
1917         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1918         entsize = xfs_attr_leaf_entsize(leaf, args->index);
1919         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1920                 ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1921                 ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1922                 if (ichdr.freemap[i].base == tablesize) {
1923                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1924                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1925                 }
1926
1927                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1928                                 be16_to_cpu(entry->nameidx)) {
1929                         before = i;
1930                 } else if (ichdr.freemap[i].base ==
1931                                 (be16_to_cpu(entry->nameidx) + entsize)) {
1932                         after = i;
1933                 } else if (ichdr.freemap[i].size < tmp) {
1934                         tmp = ichdr.freemap[i].size;
1935                         smallest = i;
1936                 }
1937         }
1938
1939         /*
1940          * Coalesce adjacent freemap regions,
1941          * or replace the smallest region.
1942          */
1943         if ((before >= 0) || (after >= 0)) {
1944                 if ((before >= 0) && (after >= 0)) {
1945                         ichdr.freemap[before].size += entsize;
1946                         ichdr.freemap[before].size += ichdr.freemap[after].size;
1947                         ichdr.freemap[after].base = 0;
1948                         ichdr.freemap[after].size = 0;
1949                 } else if (before >= 0) {
1950                         ichdr.freemap[before].size += entsize;
1951                 } else {
1952                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1953                         ichdr.freemap[after].size += entsize;
1954                 }
1955         } else {
1956                 /*
1957                  * Replace smallest region (if it is smaller than free'd entry)
1958                  */
1959                 if (ichdr.freemap[smallest].size < entsize) {
1960                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1961                         ichdr.freemap[smallest].size = entsize;
1962                 }
1963         }
1964
1965         /*
1966          * Did we remove the first entry?
1967          */
1968         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1969                 smallest = 1;
1970         else
1971                 smallest = 0;
1972
1973         /*
1974          * Compress the remaining entries and zero out the removed stuff.
1975          */
1976         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1977         ichdr.usedbytes -= entsize;
1978         xfs_trans_log_buf(args->trans, bp,
1979              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1980                                    entsize));
1981
1982         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1983         memmove(entry, entry + 1, tmp);
1984         ichdr.count--;
1985         xfs_trans_log_buf(args->trans, bp,
1986             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1987
1988         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1989         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1990
1991         /*
1992          * If we removed the first entry, re-find the first used byte
1993          * in the name area.  Note that if the entry was the "firstused",
1994          * then we don't have a "hole" in our block resulting from
1995          * removing the name.
1996          */
1997         if (smallest) {
1998                 tmp = args->geo->blksize;
1999                 entry = xfs_attr3_leaf_entryp(leaf);
2000                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2001                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2002                         ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
2003
2004                         if (be16_to_cpu(entry->nameidx) < tmp)
2005                                 tmp = be16_to_cpu(entry->nameidx);
2006                 }
2007                 ichdr.firstused = tmp;
2008                 ASSERT(ichdr.firstused != 0);
2009         } else {
2010                 ichdr.holes = 1;        /* mark as needing compaction */
2011         }
2012         xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2013         xfs_trans_log_buf(args->trans, bp,
2014                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2015                                           xfs_attr3_leaf_hdr_size(leaf)));
2016
2017         /*
2018          * Check if leaf is less than 50% full, caller may want to
2019          * "join" the leaf with a sibling if so.
2020          */
2021         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2022               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2023
2024         return tmp < args->geo->magicpct; /* leaf is < 37% full */
2025 }
2026
2027 /*
2028  * Move all the attribute list entries from drop_leaf into save_leaf.
2029  */
2030 void
2031 xfs_attr3_leaf_unbalance(
2032         struct xfs_da_state     *state,
2033         struct xfs_da_state_blk *drop_blk,
2034         struct xfs_da_state_blk *save_blk)
2035 {
2036         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2037         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2038         struct xfs_attr3_icleaf_hdr drophdr;
2039         struct xfs_attr3_icleaf_hdr savehdr;
2040         struct xfs_attr_leaf_entry *entry;
2041
2042         trace_xfs_attr_leaf_unbalance(state->args);
2043
2044         drop_leaf = drop_blk->bp->b_addr;
2045         save_leaf = save_blk->bp->b_addr;
2046         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2047         xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2048         entry = xfs_attr3_leaf_entryp(drop_leaf);
2049
2050         /*
2051          * Save last hashval from dying block for later Btree fixup.
2052          */
2053         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2054
2055         /*
2056          * Check if we need a temp buffer, or can we do it in place.
2057          * Note that we don't check "leaf" for holes because we will
2058          * always be dropping it, toosmall() decided that for us already.
2059          */
2060         if (savehdr.holes == 0) {
2061                 /*
2062                  * dest leaf has no holes, so we add there.  May need
2063                  * to make some room in the entry array.
2064                  */
2065                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2066                                          drop_blk->bp, &drophdr)) {
2067                         xfs_attr3_leaf_moveents(state->args,
2068                                                 drop_leaf, &drophdr, 0,
2069                                                 save_leaf, &savehdr, 0,
2070                                                 drophdr.count);
2071                 } else {
2072                         xfs_attr3_leaf_moveents(state->args,
2073                                                 drop_leaf, &drophdr, 0,
2074                                                 save_leaf, &savehdr,
2075                                                 savehdr.count, drophdr.count);
2076                 }
2077         } else {
2078                 /*
2079                  * Destination has holes, so we make a temporary copy
2080                  * of the leaf and add them both to that.
2081                  */
2082                 struct xfs_attr_leafblock *tmp_leaf;
2083                 struct xfs_attr3_icleaf_hdr tmphdr;
2084
2085                 tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2086
2087                 /*
2088                  * Copy the header into the temp leaf so that all the stuff
2089                  * not in the incore header is present and gets copied back in
2090                  * once we've moved all the entries.
2091                  */
2092                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2093
2094                 memset(&tmphdr, 0, sizeof(tmphdr));
2095                 tmphdr.magic = savehdr.magic;
2096                 tmphdr.forw = savehdr.forw;
2097                 tmphdr.back = savehdr.back;
2098                 tmphdr.firstused = state->args->geo->blksize;
2099
2100                 /* write the header to the temp buffer to initialise it */
2101                 xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2102
2103                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2104                                          drop_blk->bp, &drophdr)) {
2105                         xfs_attr3_leaf_moveents(state->args,
2106                                                 drop_leaf, &drophdr, 0,
2107                                                 tmp_leaf, &tmphdr, 0,
2108                                                 drophdr.count);
2109                         xfs_attr3_leaf_moveents(state->args,
2110                                                 save_leaf, &savehdr, 0,
2111                                                 tmp_leaf, &tmphdr, tmphdr.count,
2112                                                 savehdr.count);
2113                 } else {
2114                         xfs_attr3_leaf_moveents(state->args,
2115                                                 save_leaf, &savehdr, 0,
2116                                                 tmp_leaf, &tmphdr, 0,
2117                                                 savehdr.count);
2118                         xfs_attr3_leaf_moveents(state->args,
2119                                                 drop_leaf, &drophdr, 0,
2120                                                 tmp_leaf, &tmphdr, tmphdr.count,
2121                                                 drophdr.count);
2122                 }
2123                 memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2124                 savehdr = tmphdr; /* struct copy */
2125                 kmem_free(tmp_leaf);
2126         }
2127
2128         xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2129         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2130                                            state->args->geo->blksize - 1);
2131
2132         /*
2133          * Copy out last hashval in each block for B-tree code.
2134          */
2135         entry = xfs_attr3_leaf_entryp(save_leaf);
2136         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2137 }
2138
2139 /*========================================================================
2140  * Routines used for finding things in the Btree.
2141  *========================================================================*/
2142
2143 /*
2144  * Look up a name in a leaf attribute list structure.
2145  * This is the internal routine, it uses the caller's buffer.
2146  *
2147  * Note that duplicate keys are allowed, but only check within the
2148  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2149  *
2150  * Return in args->index the index into the entry[] array of either
2151  * the found entry, or where the entry should have been (insert before
2152  * that entry).
2153  *
2154  * Don't change the args->value unless we find the attribute.
2155  */
2156 int
2157 xfs_attr3_leaf_lookup_int(
2158         struct xfs_buf          *bp,
2159         struct xfs_da_args      *args)
2160 {
2161         struct xfs_attr_leafblock *leaf;
2162         struct xfs_attr3_icleaf_hdr ichdr;
2163         struct xfs_attr_leaf_entry *entry;
2164         struct xfs_attr_leaf_entry *entries;
2165         struct xfs_attr_leaf_name_local *name_loc;
2166         struct xfs_attr_leaf_name_remote *name_rmt;
2167         xfs_dahash_t            hashval;
2168         int                     probe;
2169         int                     span;
2170
2171         trace_xfs_attr_leaf_lookup(args);
2172
2173         leaf = bp->b_addr;
2174         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2175         entries = xfs_attr3_leaf_entryp(leaf);
2176         ASSERT(ichdr.count < args->geo->blksize / 8);
2177
2178         /*
2179          * Binary search.  (note: small blocks will skip this loop)
2180          */
2181         hashval = args->hashval;
2182         probe = span = ichdr.count / 2;
2183         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2184                 span /= 2;
2185                 if (be32_to_cpu(entry->hashval) < hashval)
2186                         probe += span;
2187                 else if (be32_to_cpu(entry->hashval) > hashval)
2188                         probe -= span;
2189                 else
2190                         break;
2191         }
2192         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2193         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2194
2195         /*
2196          * Since we may have duplicate hashval's, find the first matching
2197          * hashval in the leaf.
2198          */
2199         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2200                 entry--;
2201                 probe--;
2202         }
2203         while (probe < ichdr.count &&
2204                be32_to_cpu(entry->hashval) < hashval) {
2205                 entry++;
2206                 probe++;
2207         }
2208         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2209                 args->index = probe;
2210                 return -ENOATTR;
2211         }
2212
2213         /*
2214          * Duplicate keys may be present, so search all of them for a match.
2215          */
2216         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2217                         entry++, probe++) {
2218 /*
2219  * GROT: Add code to remove incomplete entries.
2220  */
2221                 /*
2222                  * If we are looking for INCOMPLETE entries, show only those.
2223                  * If we are looking for complete entries, show only those.
2224                  */
2225                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2226                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2227                         continue;
2228                 }
2229                 if (entry->flags & XFS_ATTR_LOCAL) {
2230                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2231                         if (name_loc->namelen != args->namelen)
2232                                 continue;
2233                         if (memcmp(args->name, name_loc->nameval,
2234                                                         args->namelen) != 0)
2235                                 continue;
2236                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2237                                 continue;
2238                         args->index = probe;
2239                         return -EEXIST;
2240                 } else {
2241                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2242                         if (name_rmt->namelen != args->namelen)
2243                                 continue;
2244                         if (memcmp(args->name, name_rmt->name,
2245                                                         args->namelen) != 0)
2246                                 continue;
2247                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2248                                 continue;
2249                         args->index = probe;
2250                         args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2251                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2252                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2253                                                         args->dp->i_mount,
2254                                                         args->rmtvaluelen);
2255                         return -EEXIST;
2256                 }
2257         }
2258         args->index = probe;
2259         return -ENOATTR;
2260 }
2261
2262 /*
2263  * Get the value associated with an attribute name from a leaf attribute
2264  * list structure.
2265  */
2266 int
2267 xfs_attr3_leaf_getvalue(
2268         struct xfs_buf          *bp,
2269         struct xfs_da_args      *args)
2270 {
2271         struct xfs_attr_leafblock *leaf;
2272         struct xfs_attr3_icleaf_hdr ichdr;
2273         struct xfs_attr_leaf_entry *entry;
2274         struct xfs_attr_leaf_name_local *name_loc;
2275         struct xfs_attr_leaf_name_remote *name_rmt;
2276         int                     valuelen;
2277
2278         leaf = bp->b_addr;
2279         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2280         ASSERT(ichdr.count < args->geo->blksize / 8);
2281         ASSERT(args->index < ichdr.count);
2282
2283         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2284         if (entry->flags & XFS_ATTR_LOCAL) {
2285                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2286                 ASSERT(name_loc->namelen == args->namelen);
2287                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2288                 valuelen = be16_to_cpu(name_loc->valuelen);
2289                 if (args->flags & ATTR_KERNOVAL) {
2290                         args->valuelen = valuelen;
2291                         return 0;
2292                 }
2293                 if (args->valuelen < valuelen) {
2294                         args->valuelen = valuelen;
2295                         return -ERANGE;
2296                 }
2297                 args->valuelen = valuelen;
2298                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2299         } else {
2300                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2301                 ASSERT(name_rmt->namelen == args->namelen);
2302                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2303                 args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2304                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2305                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2306                                                        args->rmtvaluelen);
2307                 if (args->flags & ATTR_KERNOVAL) {
2308                         args->valuelen = args->rmtvaluelen;
2309                         return 0;
2310                 }
2311                 if (args->valuelen < args->rmtvaluelen) {
2312                         args->valuelen = args->rmtvaluelen;
2313                         return -ERANGE;
2314                 }
2315                 args->valuelen = args->rmtvaluelen;
2316         }
2317         return 0;
2318 }
2319
2320 /*========================================================================
2321  * Utility routines.
2322  *========================================================================*/
2323
2324 /*
2325  * Move the indicated entries from one leaf to another.
2326  * NOTE: this routine modifies both source and destination leaves.
2327  */
2328 /*ARGSUSED*/
2329 STATIC void
2330 xfs_attr3_leaf_moveents(
2331         struct xfs_da_args              *args,
2332         struct xfs_attr_leafblock       *leaf_s,
2333         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2334         int                             start_s,
2335         struct xfs_attr_leafblock       *leaf_d,
2336         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2337         int                             start_d,
2338         int                             count)
2339 {
2340         struct xfs_attr_leaf_entry      *entry_s;
2341         struct xfs_attr_leaf_entry      *entry_d;
2342         int                             desti;
2343         int                             tmp;
2344         int                             i;
2345
2346         /*
2347          * Check for nothing to do.
2348          */
2349         if (count == 0)
2350                 return;
2351
2352         /*
2353          * Set up environment.
2354          */
2355         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2356                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2357         ASSERT(ichdr_s->magic == ichdr_d->magic);
2358         ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2359         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2360                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2361         ASSERT(ichdr_d->count < args->geo->blksize / 8);
2362         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2363                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2364
2365         ASSERT(start_s < ichdr_s->count);
2366         ASSERT(start_d <= ichdr_d->count);
2367         ASSERT(count <= ichdr_s->count);
2368
2369
2370         /*
2371          * Move the entries in the destination leaf up to make a hole?
2372          */
2373         if (start_d < ichdr_d->count) {
2374                 tmp  = ichdr_d->count - start_d;
2375                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2376                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2377                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2378                 memmove(entry_d, entry_s, tmp);
2379         }
2380
2381         /*
2382          * Copy all entry's in the same (sorted) order,
2383          * but allocate attribute info packed and in sequence.
2384          */
2385         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2386         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2387         desti = start_d;
2388         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2389                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2390                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2391 #ifdef GROT
2392                 /*
2393                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2394                  * may also need to change the insertion index.  Code turned
2395                  * off for 6.2, should be revisited later.
2396                  */
2397                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2398                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2399                         ichdr_s->usedbytes -= tmp;
2400                         ichdr_s->count -= 1;
2401                         entry_d--;      /* to compensate for ++ in loop hdr */
2402                         desti--;
2403                         if ((start_s + i) < offset)
2404                                 result++;       /* insertion index adjustment */
2405                 } else {
2406 #endif /* GROT */
2407                         ichdr_d->firstused -= tmp;
2408                         /* both on-disk, don't endian flip twice */
2409                         entry_d->hashval = entry_s->hashval;
2410                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2411                         entry_d->flags = entry_s->flags;
2412                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2413                                                         <= args->geo->blksize);
2414                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2415                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2416                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2417                                                         <= args->geo->blksize);
2418                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2419                         ichdr_s->usedbytes -= tmp;
2420                         ichdr_d->usedbytes += tmp;
2421                         ichdr_s->count -= 1;
2422                         ichdr_d->count += 1;
2423                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2424                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2425                         ASSERT(ichdr_d->firstused >= tmp);
2426 #ifdef GROT
2427                 }
2428 #endif /* GROT */
2429         }
2430
2431         /*
2432          * Zero out the entries we just copied.
2433          */
2434         if (start_s == ichdr_s->count) {
2435                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2436                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2437                 ASSERT(((char *)entry_s + tmp) <=
2438                        ((char *)leaf_s + args->geo->blksize));
2439                 memset(entry_s, 0, tmp);
2440         } else {
2441                 /*
2442                  * Move the remaining entries down to fill the hole,
2443                  * then zero the entries at the top.
2444                  */
2445                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2446                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2447                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2448                 memmove(entry_d, entry_s, tmp);
2449
2450                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2451                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2452                 ASSERT(((char *)entry_s + tmp) <=
2453                        ((char *)leaf_s + args->geo->blksize));
2454                 memset(entry_s, 0, tmp);
2455         }
2456
2457         /*
2458          * Fill in the freemap information
2459          */
2460         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2461         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2462         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2463         ichdr_d->freemap[1].base = 0;
2464         ichdr_d->freemap[2].base = 0;
2465         ichdr_d->freemap[1].size = 0;
2466         ichdr_d->freemap[2].size = 0;
2467         ichdr_s->holes = 1;     /* leaf may not be compact */
2468 }
2469
2470 /*
2471  * Pick up the last hashvalue from a leaf block.
2472  */
2473 xfs_dahash_t
2474 xfs_attr_leaf_lasthash(
2475         struct xfs_buf  *bp,
2476         int             *count)
2477 {
2478         struct xfs_attr3_icleaf_hdr ichdr;
2479         struct xfs_attr_leaf_entry *entries;
2480         struct xfs_mount *mp = bp->b_target->bt_mount;
2481
2482         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2483         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2484         if (count)
2485                 *count = ichdr.count;
2486         if (!ichdr.count)
2487                 return 0;
2488         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2489 }
2490
2491 /*
2492  * Calculate the number of bytes used to store the indicated attribute
2493  * (whether local or remote only calculate bytes in this block).
2494  */
2495 STATIC int
2496 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2497 {
2498         struct xfs_attr_leaf_entry *entries;
2499         xfs_attr_leaf_name_local_t *name_loc;
2500         xfs_attr_leaf_name_remote_t *name_rmt;
2501         int size;
2502
2503         entries = xfs_attr3_leaf_entryp(leaf);
2504         if (entries[index].flags & XFS_ATTR_LOCAL) {
2505                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2506                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2507                                                    be16_to_cpu(name_loc->valuelen));
2508         } else {
2509                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2510                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2511         }
2512         return size;
2513 }
2514
2515 /*
2516  * Calculate the number of bytes that would be required to store the new
2517  * attribute (whether local or remote only calculate bytes in this block).
2518  * This routine decides as a side effect whether the attribute will be
2519  * a "local" or a "remote" attribute.
2520  */
2521 int
2522 xfs_attr_leaf_newentsize(
2523         struct xfs_da_args      *args,
2524         int                     *local)
2525 {
2526         int                     size;
2527
2528         size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2529         if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2530                 if (local)
2531                         *local = 1;
2532                 return size;
2533         }
2534         if (local)
2535                 *local = 0;
2536         return xfs_attr_leaf_entsize_remote(args->namelen);
2537 }
2538
2539
2540 /*========================================================================
2541  * Manage the INCOMPLETE flag in a leaf entry
2542  *========================================================================*/
2543
2544 /*
2545  * Clear the INCOMPLETE flag on an entry in a leaf block.
2546  */
2547 int
2548 xfs_attr3_leaf_clearflag(
2549         struct xfs_da_args      *args)
2550 {
2551         struct xfs_attr_leafblock *leaf;
2552         struct xfs_attr_leaf_entry *entry;
2553         struct xfs_attr_leaf_name_remote *name_rmt;
2554         struct xfs_buf          *bp;
2555         int                     error;
2556 #ifdef DEBUG
2557         struct xfs_attr3_icleaf_hdr ichdr;
2558         xfs_attr_leaf_name_local_t *name_loc;
2559         int namelen;
2560         char *name;
2561 #endif /* DEBUG */
2562
2563         trace_xfs_attr_leaf_clearflag(args);
2564         /*
2565          * Set up the operation.
2566          */
2567         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2568         if (error)
2569                 return error;
2570
2571         leaf = bp->b_addr;
2572         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2573         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2574
2575 #ifdef DEBUG
2576         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2577         ASSERT(args->index < ichdr.count);
2578         ASSERT(args->index >= 0);
2579
2580         if (entry->flags & XFS_ATTR_LOCAL) {
2581                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2582                 namelen = name_loc->namelen;
2583                 name = (char *)name_loc->nameval;
2584         } else {
2585                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2586                 namelen = name_rmt->namelen;
2587                 name = (char *)name_rmt->name;
2588         }
2589         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2590         ASSERT(namelen == args->namelen);
2591         ASSERT(memcmp(name, args->name, namelen) == 0);
2592 #endif /* DEBUG */
2593
2594         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2595         xfs_trans_log_buf(args->trans, bp,
2596                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2597
2598         if (args->rmtblkno) {
2599                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2600                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2601                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2602                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2603                 xfs_trans_log_buf(args->trans, bp,
2604                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2605         }
2606
2607         /*
2608          * Commit the flag value change and start the next trans in series.
2609          */
2610         return xfs_trans_roll_inode(&args->trans, args->dp);
2611 }
2612
2613 /*
2614  * Set the INCOMPLETE flag on an entry in a leaf block.
2615  */
2616 int
2617 xfs_attr3_leaf_setflag(
2618         struct xfs_da_args      *args)
2619 {
2620         struct xfs_attr_leafblock *leaf;
2621         struct xfs_attr_leaf_entry *entry;
2622         struct xfs_attr_leaf_name_remote *name_rmt;
2623         struct xfs_buf          *bp;
2624         int error;
2625 #ifdef DEBUG
2626         struct xfs_attr3_icleaf_hdr ichdr;
2627 #endif
2628
2629         trace_xfs_attr_leaf_setflag(args);
2630
2631         /*
2632          * Set up the operation.
2633          */
2634         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2635         if (error)
2636                 return error;
2637
2638         leaf = bp->b_addr;
2639 #ifdef DEBUG
2640         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2641         ASSERT(args->index < ichdr.count);
2642         ASSERT(args->index >= 0);
2643 #endif
2644         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2645
2646         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2647         entry->flags |= XFS_ATTR_INCOMPLETE;
2648         xfs_trans_log_buf(args->trans, bp,
2649                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2650         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2651                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2652                 name_rmt->valueblk = 0;
2653                 name_rmt->valuelen = 0;
2654                 xfs_trans_log_buf(args->trans, bp,
2655                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2656         }
2657
2658         /*
2659          * Commit the flag value change and start the next trans in series.
2660          */
2661         return xfs_trans_roll_inode(&args->trans, args->dp);
2662 }
2663
2664 /*
2665  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2666  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2667  * entry given by args->blkno2/index2.
2668  *
2669  * Note that they could be in different blocks, or in the same block.
2670  */
2671 int
2672 xfs_attr3_leaf_flipflags(
2673         struct xfs_da_args      *args)
2674 {
2675         struct xfs_attr_leafblock *leaf1;
2676         struct xfs_attr_leafblock *leaf2;
2677         struct xfs_attr_leaf_entry *entry1;
2678         struct xfs_attr_leaf_entry *entry2;
2679         struct xfs_attr_leaf_name_remote *name_rmt;
2680         struct xfs_buf          *bp1;
2681         struct xfs_buf          *bp2;
2682         int error;
2683 #ifdef DEBUG
2684         struct xfs_attr3_icleaf_hdr ichdr1;
2685         struct xfs_attr3_icleaf_hdr ichdr2;
2686         xfs_attr_leaf_name_local_t *name_loc;
2687         int namelen1, namelen2;
2688         char *name1, *name2;
2689 #endif /* DEBUG */
2690
2691         trace_xfs_attr_leaf_flipflags(args);
2692
2693         /*
2694          * Read the block containing the "old" attr
2695          */
2696         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2697         if (error)
2698                 return error;
2699
2700         /*
2701          * Read the block containing the "new" attr, if it is different
2702          */
2703         if (args->blkno2 != args->blkno) {
2704                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2705                                            -1, &bp2);
2706                 if (error)
2707                         return error;
2708         } else {
2709                 bp2 = bp1;
2710         }
2711
2712         leaf1 = bp1->b_addr;
2713         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2714
2715         leaf2 = bp2->b_addr;
2716         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2717
2718 #ifdef DEBUG
2719         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2720         ASSERT(args->index < ichdr1.count);
2721         ASSERT(args->index >= 0);
2722
2723         xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2724         ASSERT(args->index2 < ichdr2.count);
2725         ASSERT(args->index2 >= 0);
2726
2727         if (entry1->flags & XFS_ATTR_LOCAL) {
2728                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2729                 namelen1 = name_loc->namelen;
2730                 name1 = (char *)name_loc->nameval;
2731         } else {
2732                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2733                 namelen1 = name_rmt->namelen;
2734                 name1 = (char *)name_rmt->name;
2735         }
2736         if (entry2->flags & XFS_ATTR_LOCAL) {
2737                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2738                 namelen2 = name_loc->namelen;
2739                 name2 = (char *)name_loc->nameval;
2740         } else {
2741                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2742                 namelen2 = name_rmt->namelen;
2743                 name2 = (char *)name_rmt->name;
2744         }
2745         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2746         ASSERT(namelen1 == namelen2);
2747         ASSERT(memcmp(name1, name2, namelen1) == 0);
2748 #endif /* DEBUG */
2749
2750         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2751         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2752
2753         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2754         xfs_trans_log_buf(args->trans, bp1,
2755                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2756         if (args->rmtblkno) {
2757                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2758                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2759                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2760                 name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2761                 xfs_trans_log_buf(args->trans, bp1,
2762                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2763         }
2764
2765         entry2->flags |= XFS_ATTR_INCOMPLETE;
2766         xfs_trans_log_buf(args->trans, bp2,
2767                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2768         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2769                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2770                 name_rmt->valueblk = 0;
2771                 name_rmt->valuelen = 0;
2772                 xfs_trans_log_buf(args->trans, bp2,
2773                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2774         }
2775
2776         /*
2777          * Commit the flag value change and start the next trans in series.
2778          */
2779         error = xfs_trans_roll_inode(&args->trans, args->dp);
2780
2781         return error;
2782 }