]> rtime.felk.cvut.cz Git - linux-imx.git/blob - fs/xfs/xfs_attr_leaf.c
drm/radeon: only save UVD bo when we have open handles
[linux-imx.git] / fs / xfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_types.h"
22 #include "xfs_bit.h"
23 #include "xfs_log.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_alloc.h"
33 #include "xfs_btree.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_attr_remote.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_inode_item.h"
39 #include "xfs_bmap.h"
40 #include "xfs_attr.h"
41 #include "xfs_attr_leaf.h"
42 #include "xfs_error.h"
43 #include "xfs_trace.h"
44 #include "xfs_buf_item.h"
45 #include "xfs_cksum.h"
46
47
48 /*
49  * xfs_attr_leaf.c
50  *
51  * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52  */
53
54 /*========================================================================
55  * Function prototypes for the kernel.
56  *========================================================================*/
57
58 /*
59  * Routines used for growing the Btree.
60  */
61 STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
62                                  xfs_dablk_t which_block, struct xfs_buf **bpp);
63 STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
64                                    struct xfs_attr3_icleaf_hdr *ichdr,
65                                    struct xfs_da_args *args, int freemap_index);
66 STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
67                                    struct xfs_attr3_icleaf_hdr *ichdr,
68                                    struct xfs_buf *leaf_buffer);
69 STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
70                                                    xfs_da_state_blk_t *blk1,
71                                                    xfs_da_state_blk_t *blk2);
72 STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
73                         xfs_da_state_blk_t *leaf_blk_1,
74                         struct xfs_attr3_icleaf_hdr *ichdr1,
75                         xfs_da_state_blk_t *leaf_blk_2,
76                         struct xfs_attr3_icleaf_hdr *ichdr2,
77                         int *number_entries_in_blk1,
78                         int *number_usedbytes_in_blk1);
79
80 /*
81  * Routines used for shrinking the Btree.
82  */
83 STATIC int xfs_attr3_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
84                                   struct xfs_buf *bp, int level);
85 STATIC int xfs_attr3_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
86                                   struct xfs_buf *bp);
87 STATIC int xfs_attr3_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
88                                    xfs_dablk_t blkno, int blkcnt);
89
90 /*
91  * Utility routines.
92  */
93 STATIC void xfs_attr3_leaf_moveents(struct xfs_attr_leafblock *src_leaf,
94                         struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
95                         struct xfs_attr_leafblock *dst_leaf,
96                         struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
97                         int move_count, struct xfs_mount *mp);
98 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
99
100 void
101 xfs_attr3_leaf_hdr_from_disk(
102         struct xfs_attr3_icleaf_hdr     *to,
103         struct xfs_attr_leafblock       *from)
104 {
105         int     i;
106
107         ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
108                from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
109
110         if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
111                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
112
113                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
114                 to->back = be32_to_cpu(hdr3->info.hdr.back);
115                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
116                 to->count = be16_to_cpu(hdr3->count);
117                 to->usedbytes = be16_to_cpu(hdr3->usedbytes);
118                 to->firstused = be16_to_cpu(hdr3->firstused);
119                 to->holes = hdr3->holes;
120
121                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
122                         to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
123                         to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
124                 }
125                 return;
126         }
127         to->forw = be32_to_cpu(from->hdr.info.forw);
128         to->back = be32_to_cpu(from->hdr.info.back);
129         to->magic = be16_to_cpu(from->hdr.info.magic);
130         to->count = be16_to_cpu(from->hdr.count);
131         to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
132         to->firstused = be16_to_cpu(from->hdr.firstused);
133         to->holes = from->hdr.holes;
134
135         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
136                 to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
137                 to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
138         }
139 }
140
141 void
142 xfs_attr3_leaf_hdr_to_disk(
143         struct xfs_attr_leafblock       *to,
144         struct xfs_attr3_icleaf_hdr     *from)
145 {
146         int     i;
147
148         ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
149                from->magic == XFS_ATTR3_LEAF_MAGIC);
150
151         if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
152                 struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
153
154                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
155                 hdr3->info.hdr.back = cpu_to_be32(from->back);
156                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
157                 hdr3->count = cpu_to_be16(from->count);
158                 hdr3->usedbytes = cpu_to_be16(from->usedbytes);
159                 hdr3->firstused = cpu_to_be16(from->firstused);
160                 hdr3->holes = from->holes;
161                 hdr3->pad1 = 0;
162
163                 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
164                         hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
165                         hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
166                 }
167                 return;
168         }
169         to->hdr.info.forw = cpu_to_be32(from->forw);
170         to->hdr.info.back = cpu_to_be32(from->back);
171         to->hdr.info.magic = cpu_to_be16(from->magic);
172         to->hdr.count = cpu_to_be16(from->count);
173         to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
174         to->hdr.firstused = cpu_to_be16(from->firstused);
175         to->hdr.holes = from->holes;
176         to->hdr.pad1 = 0;
177
178         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
179                 to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
180                 to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
181         }
182 }
183
184 static bool
185 xfs_attr3_leaf_verify(
186         struct xfs_buf          *bp)
187 {
188         struct xfs_mount        *mp = bp->b_target->bt_mount;
189         struct xfs_attr_leafblock *leaf = bp->b_addr;
190         struct xfs_attr3_icleaf_hdr ichdr;
191
192         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
193
194         if (xfs_sb_version_hascrc(&mp->m_sb)) {
195                 struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
196
197                 if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
198                         return false;
199
200                 if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_uuid))
201                         return false;
202                 if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
203                         return false;
204         } else {
205                 if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
206                         return false;
207         }
208         if (ichdr.count == 0)
209                 return false;
210
211         /* XXX: need to range check rest of attr header values */
212         /* XXX: hash order check? */
213
214         return true;
215 }
216
217 static void
218 xfs_attr3_leaf_write_verify(
219         struct xfs_buf  *bp)
220 {
221         struct xfs_mount        *mp = bp->b_target->bt_mount;
222         struct xfs_buf_log_item *bip = bp->b_fspriv;
223         struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
224
225         if (!xfs_attr3_leaf_verify(bp)) {
226                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
227                 xfs_buf_ioerror(bp, EFSCORRUPTED);
228                 return;
229         }
230
231         if (!xfs_sb_version_hascrc(&mp->m_sb))
232                 return;
233
234         if (bip)
235                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
236
237         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_ATTR3_LEAF_CRC_OFF);
238 }
239
240 /*
241  * leaf/node format detection on trees is sketchy, so a node read can be done on
242  * leaf level blocks when detection identifies the tree as a node format tree
243  * incorrectly. In this case, we need to swap the verifier to match the correct
244  * format of the block being read.
245  */
246 static void
247 xfs_attr3_leaf_read_verify(
248         struct xfs_buf          *bp)
249 {
250         struct xfs_mount        *mp = bp->b_target->bt_mount;
251
252         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
253              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
254                                           XFS_ATTR3_LEAF_CRC_OFF)) ||
255             !xfs_attr3_leaf_verify(bp)) {
256                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
257                 xfs_buf_ioerror(bp, EFSCORRUPTED);
258         }
259 }
260
261 const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
262         .verify_read = xfs_attr3_leaf_read_verify,
263         .verify_write = xfs_attr3_leaf_write_verify,
264 };
265
266 int
267 xfs_attr3_leaf_read(
268         struct xfs_trans        *tp,
269         struct xfs_inode        *dp,
270         xfs_dablk_t             bno,
271         xfs_daddr_t             mappedbno,
272         struct xfs_buf          **bpp)
273 {
274         int                     err;
275
276         err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
277                                 XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
278         if (!err && tp)
279                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
280         return err;
281 }
282
283 /*========================================================================
284  * Namespace helper routines
285  *========================================================================*/
286
287 /*
288  * If namespace bits don't match return 0.
289  * If all match then return 1.
290  */
291 STATIC int
292 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
293 {
294         return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
295 }
296
297
298 /*========================================================================
299  * External routines when attribute fork size < XFS_LITINO(mp).
300  *========================================================================*/
301
302 /*
303  * Query whether the requested number of additional bytes of extended
304  * attribute space will be able to fit inline.
305  *
306  * Returns zero if not, else the di_forkoff fork offset to be used in the
307  * literal area for attribute data once the new bytes have been added.
308  *
309  * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
310  * special case for dev/uuid inodes, they have fixed size data forks.
311  */
312 int
313 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
314 {
315         int offset;
316         int minforkoff; /* lower limit on valid forkoff locations */
317         int maxforkoff; /* upper limit on valid forkoff locations */
318         int dsize;
319         xfs_mount_t *mp = dp->i_mount;
320
321         /* rounded down */
322         offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
323
324         switch (dp->i_d.di_format) {
325         case XFS_DINODE_FMT_DEV:
326                 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
327                 return (offset >= minforkoff) ? minforkoff : 0;
328         case XFS_DINODE_FMT_UUID:
329                 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
330                 return (offset >= minforkoff) ? minforkoff : 0;
331         }
332
333         /*
334          * If the requested numbers of bytes is smaller or equal to the
335          * current attribute fork size we can always proceed.
336          *
337          * Note that if_bytes in the data fork might actually be larger than
338          * the current data fork size is due to delalloc extents. In that
339          * case either the extent count will go down when they are converted
340          * to real extents, or the delalloc conversion will take care of the
341          * literal area rebalancing.
342          */
343         if (bytes <= XFS_IFORK_ASIZE(dp))
344                 return dp->i_d.di_forkoff;
345
346         /*
347          * For attr2 we can try to move the forkoff if there is space in the
348          * literal area, but for the old format we are done if there is no
349          * space in the fixed attribute fork.
350          */
351         if (!(mp->m_flags & XFS_MOUNT_ATTR2))
352                 return 0;
353
354         dsize = dp->i_df.if_bytes;
355
356         switch (dp->i_d.di_format) {
357         case XFS_DINODE_FMT_EXTENTS:
358                 /*
359                  * If there is no attr fork and the data fork is extents, 
360                  * determine if creating the default attr fork will result
361                  * in the extents form migrating to btree. If so, the
362                  * minimum offset only needs to be the space required for
363                  * the btree root.
364                  */
365                 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
366                     xfs_default_attroffset(dp))
367                         dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
368                 break;
369         case XFS_DINODE_FMT_BTREE:
370                 /*
371                  * If we have a data btree then keep forkoff if we have one,
372                  * otherwise we are adding a new attr, so then we set
373                  * minforkoff to where the btree root can finish so we have
374                  * plenty of room for attrs
375                  */
376                 if (dp->i_d.di_forkoff) {
377                         if (offset < dp->i_d.di_forkoff)
378                                 return 0;
379                         return dp->i_d.di_forkoff;
380                 }
381                 dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
382                 break;
383         }
384
385         /*
386          * A data fork btree root must have space for at least
387          * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
388          */
389         minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
390         minforkoff = roundup(minforkoff, 8) >> 3;
391
392         /* attr fork btree root can have at least this many key/ptr pairs */
393         maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
394                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
395         maxforkoff = maxforkoff >> 3;   /* rounded down */
396
397         if (offset >= maxforkoff)
398                 return maxforkoff;
399         if (offset >= minforkoff)
400                 return offset;
401         return 0;
402 }
403
404 /*
405  * Switch on the ATTR2 superblock bit (implies also FEATURES2)
406  */
407 STATIC void
408 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
409 {
410         if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
411             !(xfs_sb_version_hasattr2(&mp->m_sb))) {
412                 spin_lock(&mp->m_sb_lock);
413                 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
414                         xfs_sb_version_addattr2(&mp->m_sb);
415                         spin_unlock(&mp->m_sb_lock);
416                         xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
417                 } else
418                         spin_unlock(&mp->m_sb_lock);
419         }
420 }
421
422 /*
423  * Create the initial contents of a shortform attribute list.
424  */
425 void
426 xfs_attr_shortform_create(xfs_da_args_t *args)
427 {
428         xfs_attr_sf_hdr_t *hdr;
429         xfs_inode_t *dp;
430         xfs_ifork_t *ifp;
431
432         trace_xfs_attr_sf_create(args);
433
434         dp = args->dp;
435         ASSERT(dp != NULL);
436         ifp = dp->i_afp;
437         ASSERT(ifp != NULL);
438         ASSERT(ifp->if_bytes == 0);
439         if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
440                 ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
441                 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
442                 ifp->if_flags |= XFS_IFINLINE;
443         } else {
444                 ASSERT(ifp->if_flags & XFS_IFINLINE);
445         }
446         xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
447         hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
448         hdr->count = 0;
449         hdr->totsize = cpu_to_be16(sizeof(*hdr));
450         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
451 }
452
453 /*
454  * Add a name/value pair to the shortform attribute list.
455  * Overflow from the inode has already been checked for.
456  */
457 void
458 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
459 {
460         xfs_attr_shortform_t *sf;
461         xfs_attr_sf_entry_t *sfe;
462         int i, offset, size;
463         xfs_mount_t *mp;
464         xfs_inode_t *dp;
465         xfs_ifork_t *ifp;
466
467         trace_xfs_attr_sf_add(args);
468
469         dp = args->dp;
470         mp = dp->i_mount;
471         dp->i_d.di_forkoff = forkoff;
472
473         ifp = dp->i_afp;
474         ASSERT(ifp->if_flags & XFS_IFINLINE);
475         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
476         sfe = &sf->list[0];
477         for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
478 #ifdef DEBUG
479                 if (sfe->namelen != args->namelen)
480                         continue;
481                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
482                         continue;
483                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
484                         continue;
485                 ASSERT(0);
486 #endif
487         }
488
489         offset = (char *)sfe - (char *)sf;
490         size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
491         xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
492         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
493         sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
494
495         sfe->namelen = args->namelen;
496         sfe->valuelen = args->valuelen;
497         sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
498         memcpy(sfe->nameval, args->name, args->namelen);
499         memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
500         sf->hdr.count++;
501         be16_add_cpu(&sf->hdr.totsize, size);
502         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
503
504         xfs_sbversion_add_attr2(mp, args->trans);
505 }
506
507 /*
508  * After the last attribute is removed revert to original inode format,
509  * making all literal area available to the data fork once more.
510  */
511 STATIC void
512 xfs_attr_fork_reset(
513         struct xfs_inode        *ip,
514         struct xfs_trans        *tp)
515 {
516         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
517         ip->i_d.di_forkoff = 0;
518         ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
519
520         ASSERT(ip->i_d.di_anextents == 0);
521         ASSERT(ip->i_afp == NULL);
522
523         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
524 }
525
526 /*
527  * Remove an attribute from the shortform attribute list structure.
528  */
529 int
530 xfs_attr_shortform_remove(xfs_da_args_t *args)
531 {
532         xfs_attr_shortform_t *sf;
533         xfs_attr_sf_entry_t *sfe;
534         int base, size=0, end, totsize, i;
535         xfs_mount_t *mp;
536         xfs_inode_t *dp;
537
538         trace_xfs_attr_sf_remove(args);
539
540         dp = args->dp;
541         mp = dp->i_mount;
542         base = sizeof(xfs_attr_sf_hdr_t);
543         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
544         sfe = &sf->list[0];
545         end = sf->hdr.count;
546         for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
547                                         base += size, i++) {
548                 size = XFS_ATTR_SF_ENTSIZE(sfe);
549                 if (sfe->namelen != args->namelen)
550                         continue;
551                 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
552                         continue;
553                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
554                         continue;
555                 break;
556         }
557         if (i == end)
558                 return(XFS_ERROR(ENOATTR));
559
560         /*
561          * Fix up the attribute fork data, covering the hole
562          */
563         end = base + size;
564         totsize = be16_to_cpu(sf->hdr.totsize);
565         if (end != totsize)
566                 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
567         sf->hdr.count--;
568         be16_add_cpu(&sf->hdr.totsize, -size);
569
570         /*
571          * Fix up the start offset of the attribute fork
572          */
573         totsize -= size;
574         if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
575             (mp->m_flags & XFS_MOUNT_ATTR2) &&
576             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
577             !(args->op_flags & XFS_DA_OP_ADDNAME)) {
578                 xfs_attr_fork_reset(dp, args->trans);
579         } else {
580                 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
581                 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
582                 ASSERT(dp->i_d.di_forkoff);
583                 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
584                                 (args->op_flags & XFS_DA_OP_ADDNAME) ||
585                                 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
586                                 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
587                 xfs_trans_log_inode(args->trans, dp,
588                                         XFS_ILOG_CORE | XFS_ILOG_ADATA);
589         }
590
591         xfs_sbversion_add_attr2(mp, args->trans);
592
593         return(0);
594 }
595
596 /*
597  * Look up a name in a shortform attribute list structure.
598  */
599 /*ARGSUSED*/
600 int
601 xfs_attr_shortform_lookup(xfs_da_args_t *args)
602 {
603         xfs_attr_shortform_t *sf;
604         xfs_attr_sf_entry_t *sfe;
605         int i;
606         xfs_ifork_t *ifp;
607
608         trace_xfs_attr_sf_lookup(args);
609
610         ifp = args->dp->i_afp;
611         ASSERT(ifp->if_flags & XFS_IFINLINE);
612         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
613         sfe = &sf->list[0];
614         for (i = 0; i < sf->hdr.count;
615                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
616                 if (sfe->namelen != args->namelen)
617                         continue;
618                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
619                         continue;
620                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
621                         continue;
622                 return(XFS_ERROR(EEXIST));
623         }
624         return(XFS_ERROR(ENOATTR));
625 }
626
627 /*
628  * Look up a name in a shortform attribute list structure.
629  */
630 /*ARGSUSED*/
631 int
632 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
633 {
634         xfs_attr_shortform_t *sf;
635         xfs_attr_sf_entry_t *sfe;
636         int i;
637
638         ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
639         sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
640         sfe = &sf->list[0];
641         for (i = 0; i < sf->hdr.count;
642                                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
643                 if (sfe->namelen != args->namelen)
644                         continue;
645                 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
646                         continue;
647                 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
648                         continue;
649                 if (args->flags & ATTR_KERNOVAL) {
650                         args->valuelen = sfe->valuelen;
651                         return(XFS_ERROR(EEXIST));
652                 }
653                 if (args->valuelen < sfe->valuelen) {
654                         args->valuelen = sfe->valuelen;
655                         return(XFS_ERROR(ERANGE));
656                 }
657                 args->valuelen = sfe->valuelen;
658                 memcpy(args->value, &sfe->nameval[args->namelen],
659                                                     args->valuelen);
660                 return(XFS_ERROR(EEXIST));
661         }
662         return(XFS_ERROR(ENOATTR));
663 }
664
665 /*
666  * Convert from using the shortform to the leaf.
667  */
668 int
669 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
670 {
671         xfs_inode_t *dp;
672         xfs_attr_shortform_t *sf;
673         xfs_attr_sf_entry_t *sfe;
674         xfs_da_args_t nargs;
675         char *tmpbuffer;
676         int error, i, size;
677         xfs_dablk_t blkno;
678         struct xfs_buf *bp;
679         xfs_ifork_t *ifp;
680
681         trace_xfs_attr_sf_to_leaf(args);
682
683         dp = args->dp;
684         ifp = dp->i_afp;
685         sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
686         size = be16_to_cpu(sf->hdr.totsize);
687         tmpbuffer = kmem_alloc(size, KM_SLEEP);
688         ASSERT(tmpbuffer != NULL);
689         memcpy(tmpbuffer, ifp->if_u1.if_data, size);
690         sf = (xfs_attr_shortform_t *)tmpbuffer;
691
692         xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
693         xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
694
695         bp = NULL;
696         error = xfs_da_grow_inode(args, &blkno);
697         if (error) {
698                 /*
699                  * If we hit an IO error middle of the transaction inside
700                  * grow_inode(), we may have inconsistent data. Bail out.
701                  */
702                 if (error == EIO)
703                         goto out;
704                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
705                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
706                 goto out;
707         }
708
709         ASSERT(blkno == 0);
710         error = xfs_attr3_leaf_create(args, blkno, &bp);
711         if (error) {
712                 error = xfs_da_shrink_inode(args, 0, bp);
713                 bp = NULL;
714                 if (error)
715                         goto out;
716                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
717                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
718                 goto out;
719         }
720
721         memset((char *)&nargs, 0, sizeof(nargs));
722         nargs.dp = dp;
723         nargs.firstblock = args->firstblock;
724         nargs.flist = args->flist;
725         nargs.total = args->total;
726         nargs.whichfork = XFS_ATTR_FORK;
727         nargs.trans = args->trans;
728         nargs.op_flags = XFS_DA_OP_OKNOENT;
729
730         sfe = &sf->list[0];
731         for (i = 0; i < sf->hdr.count; i++) {
732                 nargs.name = sfe->nameval;
733                 nargs.namelen = sfe->namelen;
734                 nargs.value = &sfe->nameval[nargs.namelen];
735                 nargs.valuelen = sfe->valuelen;
736                 nargs.hashval = xfs_da_hashname(sfe->nameval,
737                                                 sfe->namelen);
738                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
739                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
740                 ASSERT(error == ENOATTR);
741                 error = xfs_attr3_leaf_add(bp, &nargs);
742                 ASSERT(error != ENOSPC);
743                 if (error)
744                         goto out;
745                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
746         }
747         error = 0;
748
749 out:
750         kmem_free(tmpbuffer);
751         return(error);
752 }
753
754 STATIC int
755 xfs_attr_shortform_compare(const void *a, const void *b)
756 {
757         xfs_attr_sf_sort_t *sa, *sb;
758
759         sa = (xfs_attr_sf_sort_t *)a;
760         sb = (xfs_attr_sf_sort_t *)b;
761         if (sa->hash < sb->hash) {
762                 return(-1);
763         } else if (sa->hash > sb->hash) {
764                 return(1);
765         } else {
766                 return(sa->entno - sb->entno);
767         }
768 }
769
770
771 #define XFS_ISRESET_CURSOR(cursor) \
772         (!((cursor)->initted) && !((cursor)->hashval) && \
773          !((cursor)->blkno) && !((cursor)->offset))
774 /*
775  * Copy out entries of shortform attribute lists for attr_list().
776  * Shortform attribute lists are not stored in hashval sorted order.
777  * If the output buffer is not large enough to hold them all, then we
778  * we have to calculate each entries' hashvalue and sort them before
779  * we can begin returning them to the user.
780  */
781 /*ARGSUSED*/
782 int
783 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
784 {
785         attrlist_cursor_kern_t *cursor;
786         xfs_attr_sf_sort_t *sbuf, *sbp;
787         xfs_attr_shortform_t *sf;
788         xfs_attr_sf_entry_t *sfe;
789         xfs_inode_t *dp;
790         int sbsize, nsbuf, count, i;
791         int error;
792
793         ASSERT(context != NULL);
794         dp = context->dp;
795         ASSERT(dp != NULL);
796         ASSERT(dp->i_afp != NULL);
797         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
798         ASSERT(sf != NULL);
799         if (!sf->hdr.count)
800                 return(0);
801         cursor = context->cursor;
802         ASSERT(cursor != NULL);
803
804         trace_xfs_attr_list_sf(context);
805
806         /*
807          * If the buffer is large enough and the cursor is at the start,
808          * do not bother with sorting since we will return everything in
809          * one buffer and another call using the cursor won't need to be
810          * made.
811          * Note the generous fudge factor of 16 overhead bytes per entry.
812          * If bufsize is zero then put_listent must be a search function
813          * and can just scan through what we have.
814          */
815         if (context->bufsize == 0 ||
816             (XFS_ISRESET_CURSOR(cursor) &&
817              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
818                 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
819                         error = context->put_listent(context,
820                                            sfe->flags,
821                                            sfe->nameval,
822                                            (int)sfe->namelen,
823                                            (int)sfe->valuelen,
824                                            &sfe->nameval[sfe->namelen]);
825
826                         /*
827                          * Either search callback finished early or
828                          * didn't fit it all in the buffer after all.
829                          */
830                         if (context->seen_enough)
831                                 break;
832
833                         if (error)
834                                 return error;
835                         sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
836                 }
837                 trace_xfs_attr_list_sf_all(context);
838                 return(0);
839         }
840
841         /* do no more for a search callback */
842         if (context->bufsize == 0)
843                 return 0;
844
845         /*
846          * It didn't all fit, so we have to sort everything on hashval.
847          */
848         sbsize = sf->hdr.count * sizeof(*sbuf);
849         sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
850
851         /*
852          * Scan the attribute list for the rest of the entries, storing
853          * the relevant info from only those that match into a buffer.
854          */
855         nsbuf = 0;
856         for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
857                 if (unlikely(
858                     ((char *)sfe < (char *)sf) ||
859                     ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
860                         XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
861                                              XFS_ERRLEVEL_LOW,
862                                              context->dp->i_mount, sfe);
863                         kmem_free(sbuf);
864                         return XFS_ERROR(EFSCORRUPTED);
865                 }
866
867                 sbp->entno = i;
868                 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
869                 sbp->name = sfe->nameval;
870                 sbp->namelen = sfe->namelen;
871                 /* These are bytes, and both on-disk, don't endian-flip */
872                 sbp->valuelen = sfe->valuelen;
873                 sbp->flags = sfe->flags;
874                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
875                 sbp++;
876                 nsbuf++;
877         }
878
879         /*
880          * Sort the entries on hash then entno.
881          */
882         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
883
884         /*
885          * Re-find our place IN THE SORTED LIST.
886          */
887         count = 0;
888         cursor->initted = 1;
889         cursor->blkno = 0;
890         for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
891                 if (sbp->hash == cursor->hashval) {
892                         if (cursor->offset == count) {
893                                 break;
894                         }
895                         count++;
896                 } else if (sbp->hash > cursor->hashval) {
897                         break;
898                 }
899         }
900         if (i == nsbuf) {
901                 kmem_free(sbuf);
902                 return(0);
903         }
904
905         /*
906          * Loop putting entries into the user buffer.
907          */
908         for ( ; i < nsbuf; i++, sbp++) {
909                 if (cursor->hashval != sbp->hash) {
910                         cursor->hashval = sbp->hash;
911                         cursor->offset = 0;
912                 }
913                 error = context->put_listent(context,
914                                         sbp->flags,
915                                         sbp->name,
916                                         sbp->namelen,
917                                         sbp->valuelen,
918                                         &sbp->name[sbp->namelen]);
919                 if (error)
920                         return error;
921                 if (context->seen_enough)
922                         break;
923                 cursor->offset++;
924         }
925
926         kmem_free(sbuf);
927         return(0);
928 }
929
930 /*
931  * Check a leaf attribute block to see if all the entries would fit into
932  * a shortform attribute list.
933  */
934 int
935 xfs_attr_shortform_allfit(
936         struct xfs_buf          *bp,
937         struct xfs_inode        *dp)
938 {
939         struct xfs_attr_leafblock *leaf;
940         struct xfs_attr_leaf_entry *entry;
941         xfs_attr_leaf_name_local_t *name_loc;
942         struct xfs_attr3_icleaf_hdr leafhdr;
943         int                     bytes;
944         int                     i;
945
946         leaf = bp->b_addr;
947         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
948         entry = xfs_attr3_leaf_entryp(leaf);
949
950         bytes = sizeof(struct xfs_attr_sf_hdr);
951         for (i = 0; i < leafhdr.count; entry++, i++) {
952                 if (entry->flags & XFS_ATTR_INCOMPLETE)
953                         continue;               /* don't copy partial entries */
954                 if (!(entry->flags & XFS_ATTR_LOCAL))
955                         return(0);
956                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
957                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
958                         return(0);
959                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
960                         return(0);
961                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
962                                 + name_loc->namelen
963                                 + be16_to_cpu(name_loc->valuelen);
964         }
965         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
966             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
967             (bytes == sizeof(struct xfs_attr_sf_hdr)))
968                 return -1;
969         return xfs_attr_shortform_bytesfit(dp, bytes);
970 }
971
972 /*
973  * Convert a leaf attribute list to shortform attribute list
974  */
975 int
976 xfs_attr3_leaf_to_shortform(
977         struct xfs_buf          *bp,
978         struct xfs_da_args      *args,
979         int                     forkoff)
980 {
981         struct xfs_attr_leafblock *leaf;
982         struct xfs_attr3_icleaf_hdr ichdr;
983         struct xfs_attr_leaf_entry *entry;
984         struct xfs_attr_leaf_name_local *name_loc;
985         struct xfs_da_args      nargs;
986         struct xfs_inode        *dp = args->dp;
987         char                    *tmpbuffer;
988         int                     error;
989         int                     i;
990
991         trace_xfs_attr_leaf_to_sf(args);
992
993         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
994         if (!tmpbuffer)
995                 return ENOMEM;
996
997         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
998
999         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1000         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1001         entry = xfs_attr3_leaf_entryp(leaf);
1002
1003         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1004         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
1005
1006         /*
1007          * Clean out the prior contents of the attribute list.
1008          */
1009         error = xfs_da_shrink_inode(args, 0, bp);
1010         if (error)
1011                 goto out;
1012
1013         if (forkoff == -1) {
1014                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1015                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1016                 xfs_attr_fork_reset(dp, args->trans);
1017                 goto out;
1018         }
1019
1020         xfs_attr_shortform_create(args);
1021
1022         /*
1023          * Copy the attributes
1024          */
1025         memset((char *)&nargs, 0, sizeof(nargs));
1026         nargs.dp = dp;
1027         nargs.firstblock = args->firstblock;
1028         nargs.flist = args->flist;
1029         nargs.total = args->total;
1030         nargs.whichfork = XFS_ATTR_FORK;
1031         nargs.trans = args->trans;
1032         nargs.op_flags = XFS_DA_OP_OKNOENT;
1033
1034         for (i = 0; i < ichdr.count; entry++, i++) {
1035                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1036                         continue;       /* don't copy partial entries */
1037                 if (!entry->nameidx)
1038                         continue;
1039                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1040                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1041                 nargs.name = name_loc->nameval;
1042                 nargs.namelen = name_loc->namelen;
1043                 nargs.value = &name_loc->nameval[nargs.namelen];
1044                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1045                 nargs.hashval = be32_to_cpu(entry->hashval);
1046                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1047                 xfs_attr_shortform_add(&nargs, forkoff);
1048         }
1049         error = 0;
1050
1051 out:
1052         kmem_free(tmpbuffer);
1053         return error;
1054 }
1055
1056 /*
1057  * Convert from using a single leaf to a root node and a leaf.
1058  */
1059 int
1060 xfs_attr3_leaf_to_node(
1061         struct xfs_da_args      *args)
1062 {
1063         struct xfs_attr_leafblock *leaf;
1064         struct xfs_attr3_icleaf_hdr icleafhdr;
1065         struct xfs_attr_leaf_entry *entries;
1066         struct xfs_da_node_entry *btree;
1067         struct xfs_da3_icnode_hdr icnodehdr;
1068         struct xfs_da_intnode   *node;
1069         struct xfs_inode        *dp = args->dp;
1070         struct xfs_mount        *mp = dp->i_mount;
1071         struct xfs_buf          *bp1 = NULL;
1072         struct xfs_buf          *bp2 = NULL;
1073         xfs_dablk_t             blkno;
1074         int                     error;
1075
1076         trace_xfs_attr_leaf_to_node(args);
1077
1078         error = xfs_da_grow_inode(args, &blkno);
1079         if (error)
1080                 goto out;
1081         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1082         if (error)
1083                 goto out;
1084
1085         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1086         if (error)
1087                 goto out;
1088
1089         /* copy leaf to new buffer, update identifiers */
1090         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1091         bp2->b_ops = bp1->b_ops;
1092         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
1093         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1094                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1095                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1096         }
1097         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
1098
1099         /*
1100          * Set up the new root node.
1101          */
1102         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1103         if (error)
1104                 goto out;
1105         node = bp1->b_addr;
1106         xfs_da3_node_hdr_from_disk(&icnodehdr, node);
1107         btree = xfs_da3_node_tree_p(node);
1108
1109         leaf = bp2->b_addr;
1110         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
1111         entries = xfs_attr3_leaf_entryp(leaf);
1112
1113         /* both on-disk, don't endian-flip twice */
1114         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1115         btree[0].before = cpu_to_be32(blkno);
1116         icnodehdr.count = 1;
1117         xfs_da3_node_hdr_to_disk(node, &icnodehdr);
1118         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
1119         error = 0;
1120 out:
1121         return error;
1122 }
1123
1124
1125 /*========================================================================
1126  * Routines used for growing the Btree.
1127  *========================================================================*/
1128
1129 /*
1130  * Create the initial contents of a leaf attribute list
1131  * or a leaf in a node attribute list.
1132  */
1133 STATIC int
1134 xfs_attr3_leaf_create(
1135         struct xfs_da_args      *args,
1136         xfs_dablk_t             blkno,
1137         struct xfs_buf          **bpp)
1138 {
1139         struct xfs_attr_leafblock *leaf;
1140         struct xfs_attr3_icleaf_hdr ichdr;
1141         struct xfs_inode        *dp = args->dp;
1142         struct xfs_mount        *mp = dp->i_mount;
1143         struct xfs_buf          *bp;
1144         int                     error;
1145
1146         trace_xfs_attr_leaf_create(args);
1147
1148         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1149                                             XFS_ATTR_FORK);
1150         if (error)
1151                 return error;
1152         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1153         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1154         leaf = bp->b_addr;
1155         memset(leaf, 0, XFS_LBSIZE(mp));
1156
1157         memset(&ichdr, 0, sizeof(ichdr));
1158         ichdr.firstused = XFS_LBSIZE(mp);
1159
1160         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1161                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1162
1163                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1164
1165                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1166                 hdr3->owner = cpu_to_be64(dp->i_ino);
1167                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
1168
1169                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1170         } else {
1171                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1172                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1173         }
1174         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1175
1176         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1177         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
1178
1179         *bpp = bp;
1180         return 0;
1181 }
1182
1183 /*
1184  * Split the leaf node, rebalance, then add the new entry.
1185  */
1186 int
1187 xfs_attr3_leaf_split(
1188         struct xfs_da_state     *state,
1189         struct xfs_da_state_blk *oldblk,
1190         struct xfs_da_state_blk *newblk)
1191 {
1192         xfs_dablk_t blkno;
1193         int error;
1194
1195         trace_xfs_attr_leaf_split(state->args);
1196
1197         /*
1198          * Allocate space for a new leaf node.
1199          */
1200         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1201         error = xfs_da_grow_inode(state->args, &blkno);
1202         if (error)
1203                 return(error);
1204         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1205         if (error)
1206                 return(error);
1207         newblk->blkno = blkno;
1208         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1209
1210         /*
1211          * Rebalance the entries across the two leaves.
1212          * NOTE: rebalance() currently depends on the 2nd block being empty.
1213          */
1214         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1215         error = xfs_da3_blk_link(state, oldblk, newblk);
1216         if (error)
1217                 return(error);
1218
1219         /*
1220          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1221          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1222          * "new" attrs info.  Will need the "old" info to remove it later.
1223          *
1224          * Insert the "new" entry in the correct block.
1225          */
1226         if (state->inleaf) {
1227                 trace_xfs_attr_leaf_add_old(state->args);
1228                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1229         } else {
1230                 trace_xfs_attr_leaf_add_new(state->args);
1231                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1232         }
1233
1234         /*
1235          * Update last hashval in each block since we added the name.
1236          */
1237         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1238         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1239         return(error);
1240 }
1241
1242 /*
1243  * Add a name to the leaf attribute list structure.
1244  */
1245 int
1246 xfs_attr3_leaf_add(
1247         struct xfs_buf          *bp,
1248         struct xfs_da_args      *args)
1249 {
1250         struct xfs_attr_leafblock *leaf;
1251         struct xfs_attr3_icleaf_hdr ichdr;
1252         int                     tablesize;
1253         int                     entsize;
1254         int                     sum;
1255         int                     tmp;
1256         int                     i;
1257
1258         trace_xfs_attr_leaf_add(args);
1259
1260         leaf = bp->b_addr;
1261         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1262         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1263         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1264                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1265
1266         /*
1267          * Search through freemap for first-fit on new name length.
1268          * (may need to figure in size of entry struct too)
1269          */
1270         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1271                                         + xfs_attr3_leaf_hdr_size(leaf);
1272         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1273                 if (tablesize > ichdr.firstused) {
1274                         sum += ichdr.freemap[i].size;
1275                         continue;
1276                 }
1277                 if (!ichdr.freemap[i].size)
1278                         continue;       /* no space in this map */
1279                 tmp = entsize;
1280                 if (ichdr.freemap[i].base < ichdr.firstused)
1281                         tmp += sizeof(xfs_attr_leaf_entry_t);
1282                 if (ichdr.freemap[i].size >= tmp) {
1283                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1284                         goto out_log_hdr;
1285                 }
1286                 sum += ichdr.freemap[i].size;
1287         }
1288
1289         /*
1290          * If there are no holes in the address space of the block,
1291          * and we don't have enough freespace, then compaction will do us
1292          * no good and we should just give up.
1293          */
1294         if (!ichdr.holes && sum < entsize)
1295                 return XFS_ERROR(ENOSPC);
1296
1297         /*
1298          * Compact the entries to coalesce free space.
1299          * This may change the hdr->count via dropping INCOMPLETE entries.
1300          */
1301         xfs_attr3_leaf_compact(args, &ichdr, bp);
1302
1303         /*
1304          * After compaction, the block is guaranteed to have only one
1305          * free region, in freemap[0].  If it is not big enough, give up.
1306          */
1307         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1308                 tmp = ENOSPC;
1309                 goto out_log_hdr;
1310         }
1311
1312         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1313
1314 out_log_hdr:
1315         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1316         xfs_trans_log_buf(args->trans, bp,
1317                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1318                                 xfs_attr3_leaf_hdr_size(leaf)));
1319         return tmp;
1320 }
1321
1322 /*
1323  * Add a name to a leaf attribute list structure.
1324  */
1325 STATIC int
1326 xfs_attr3_leaf_add_work(
1327         struct xfs_buf          *bp,
1328         struct xfs_attr3_icleaf_hdr *ichdr,
1329         struct xfs_da_args      *args,
1330         int                     mapindex)
1331 {
1332         struct xfs_attr_leafblock *leaf;
1333         struct xfs_attr_leaf_entry *entry;
1334         struct xfs_attr_leaf_name_local *name_loc;
1335         struct xfs_attr_leaf_name_remote *name_rmt;
1336         struct xfs_mount        *mp;
1337         int                     tmp;
1338         int                     i;
1339
1340         trace_xfs_attr_leaf_add_work(args);
1341
1342         leaf = bp->b_addr;
1343         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1344         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1345
1346         /*
1347          * Force open some space in the entry array and fill it in.
1348          */
1349         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1350         if (args->index < ichdr->count) {
1351                 tmp  = ichdr->count - args->index;
1352                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1353                 memmove(entry + 1, entry, tmp);
1354                 xfs_trans_log_buf(args->trans, bp,
1355                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1356         }
1357         ichdr->count++;
1358
1359         /*
1360          * Allocate space for the new string (at the end of the run).
1361          */
1362         mp = args->trans->t_mountp;
1363         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1364         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1365         ASSERT(ichdr->freemap[mapindex].size >=
1366                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1367                                          mp->m_sb.sb_blocksize, NULL));
1368         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1369         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1370
1371         ichdr->freemap[mapindex].size -=
1372                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1373                                                  mp->m_sb.sb_blocksize, &tmp);
1374
1375         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1376                                      ichdr->freemap[mapindex].size);
1377         entry->hashval = cpu_to_be32(args->hashval);
1378         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1379         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1380         if (args->op_flags & XFS_DA_OP_RENAME) {
1381                 entry->flags |= XFS_ATTR_INCOMPLETE;
1382                 if ((args->blkno2 == args->blkno) &&
1383                     (args->index2 <= args->index)) {
1384                         args->index2++;
1385                 }
1386         }
1387         xfs_trans_log_buf(args->trans, bp,
1388                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1389         ASSERT((args->index == 0) ||
1390                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1391         ASSERT((args->index == ichdr->count - 1) ||
1392                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1393
1394         /*
1395          * For "remote" attribute values, simply note that we need to
1396          * allocate space for the "remote" value.  We can't actually
1397          * allocate the extents in this transaction, and we can't decide
1398          * which blocks they should be as we might allocate more blocks
1399          * as part of this transaction (a split operation for example).
1400          */
1401         if (entry->flags & XFS_ATTR_LOCAL) {
1402                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1403                 name_loc->namelen = args->namelen;
1404                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1405                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1406                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1407                                    be16_to_cpu(name_loc->valuelen));
1408         } else {
1409                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1410                 name_rmt->namelen = args->namelen;
1411                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1412                 entry->flags |= XFS_ATTR_INCOMPLETE;
1413                 /* just in case */
1414                 name_rmt->valuelen = 0;
1415                 name_rmt->valueblk = 0;
1416                 args->rmtblkno = 1;
1417                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1418         }
1419         xfs_trans_log_buf(args->trans, bp,
1420              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1421                                    xfs_attr_leaf_entsize(leaf, args->index)));
1422
1423         /*
1424          * Update the control info for this leaf node
1425          */
1426         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1427                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1428
1429         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1430                                         + xfs_attr3_leaf_hdr_size(leaf));
1431         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1432                                         + xfs_attr3_leaf_hdr_size(leaf);
1433
1434         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1435                 if (ichdr->freemap[i].base == tmp) {
1436                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1437                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1438                 }
1439         }
1440         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1441         return 0;
1442 }
1443
1444 /*
1445  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1446  */
1447 STATIC void
1448 xfs_attr3_leaf_compact(
1449         struct xfs_da_args      *args,
1450         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1451         struct xfs_buf          *bp)
1452 {
1453         struct xfs_attr_leafblock *leaf_src;
1454         struct xfs_attr_leafblock *leaf_dst;
1455         struct xfs_attr3_icleaf_hdr ichdr_src;
1456         struct xfs_trans        *trans = args->trans;
1457         struct xfs_mount        *mp = trans->t_mountp;
1458         char                    *tmpbuffer;
1459
1460         trace_xfs_attr_leaf_compact(args);
1461
1462         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1463         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1464         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1465         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1466         leaf_dst = bp->b_addr;
1467
1468         /*
1469          * Copy the on-disk header back into the destination buffer to ensure
1470          * all the information in the header that is not part of the incore
1471          * header structure is preserved.
1472          */
1473         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1474
1475         /* Initialise the incore headers */
1476         ichdr_src = *ichdr_dst; /* struct copy */
1477         ichdr_dst->firstused = XFS_LBSIZE(mp);
1478         ichdr_dst->usedbytes = 0;
1479         ichdr_dst->count = 0;
1480         ichdr_dst->holes = 0;
1481         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1482         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1483                                                 ichdr_dst->freemap[0].base;
1484
1485
1486         /* write the header back to initialise the underlying buffer */
1487         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1488
1489         /*
1490          * Copy all entry's in the same (sorted) order,
1491          * but allocate name/value pairs packed and in sequence.
1492          */
1493         xfs_attr3_leaf_moveents(leaf_src, &ichdr_src, 0, leaf_dst, ichdr_dst, 0,
1494                                 ichdr_src.count, mp);
1495         /*
1496          * this logs the entire buffer, but the caller must write the header
1497          * back to the buffer when it is finished modifying it.
1498          */
1499         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1500
1501         kmem_free(tmpbuffer);
1502 }
1503
1504 /*
1505  * Compare two leaf blocks "order".
1506  * Return 0 unless leaf2 should go before leaf1.
1507  */
1508 static int
1509 xfs_attr3_leaf_order(
1510         struct xfs_buf  *leaf1_bp,
1511         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1512         struct xfs_buf  *leaf2_bp,
1513         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1514 {
1515         struct xfs_attr_leaf_entry *entries1;
1516         struct xfs_attr_leaf_entry *entries2;
1517
1518         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1519         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1520         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1521             ((be32_to_cpu(entries2[0].hashval) <
1522               be32_to_cpu(entries1[0].hashval)) ||
1523              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1524               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1525                 return 1;
1526         }
1527         return 0;
1528 }
1529
1530 int
1531 xfs_attr_leaf_order(
1532         struct xfs_buf  *leaf1_bp,
1533         struct xfs_buf  *leaf2_bp)
1534 {
1535         struct xfs_attr3_icleaf_hdr ichdr1;
1536         struct xfs_attr3_icleaf_hdr ichdr2;
1537
1538         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1539         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1540         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1541 }
1542
1543 /*
1544  * Redistribute the attribute list entries between two leaf nodes,
1545  * taking into account the size of the new entry.
1546  *
1547  * NOTE: if new block is empty, then it will get the upper half of the
1548  * old block.  At present, all (one) callers pass in an empty second block.
1549  *
1550  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1551  * to match what it is doing in splitting the attribute leaf block.  Those
1552  * values are used in "atomic rename" operations on attributes.  Note that
1553  * the "new" and "old" values can end up in different blocks.
1554  */
1555 STATIC void
1556 xfs_attr3_leaf_rebalance(
1557         struct xfs_da_state     *state,
1558         struct xfs_da_state_blk *blk1,
1559         struct xfs_da_state_blk *blk2)
1560 {
1561         struct xfs_da_args      *args;
1562         struct xfs_attr_leafblock *leaf1;
1563         struct xfs_attr_leafblock *leaf2;
1564         struct xfs_attr3_icleaf_hdr ichdr1;
1565         struct xfs_attr3_icleaf_hdr ichdr2;
1566         struct xfs_attr_leaf_entry *entries1;
1567         struct xfs_attr_leaf_entry *entries2;
1568         int                     count;
1569         int                     totallen;
1570         int                     max;
1571         int                     space;
1572         int                     swap;
1573
1574         /*
1575          * Set up environment.
1576          */
1577         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1578         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1579         leaf1 = blk1->bp->b_addr;
1580         leaf2 = blk2->bp->b_addr;
1581         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1582         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1583         ASSERT(ichdr2.count == 0);
1584         args = state->args;
1585
1586         trace_xfs_attr_leaf_rebalance(args);
1587
1588         /*
1589          * Check ordering of blocks, reverse if it makes things simpler.
1590          *
1591          * NOTE: Given that all (current) callers pass in an empty
1592          * second block, this code should never set "swap".
1593          */
1594         swap = 0;
1595         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1596                 struct xfs_da_state_blk *tmp_blk;
1597                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1598
1599                 tmp_blk = blk1;
1600                 blk1 = blk2;
1601                 blk2 = tmp_blk;
1602
1603                 /* struct copies to swap them rather than reconverting */
1604                 tmp_ichdr = ichdr1;
1605                 ichdr1 = ichdr2;
1606                 ichdr2 = tmp_ichdr;
1607
1608                 leaf1 = blk1->bp->b_addr;
1609                 leaf2 = blk2->bp->b_addr;
1610                 swap = 1;
1611         }
1612
1613         /*
1614          * Examine entries until we reduce the absolute difference in
1615          * byte usage between the two blocks to a minimum.  Then get
1616          * the direction to copy and the number of elements to move.
1617          *
1618          * "inleaf" is true if the new entry should be inserted into blk1.
1619          * If "swap" is also true, then reverse the sense of "inleaf".
1620          */
1621         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1622                                                       blk2, &ichdr2,
1623                                                       &count, &totallen);
1624         if (swap)
1625                 state->inleaf = !state->inleaf;
1626
1627         /*
1628          * Move any entries required from leaf to leaf:
1629          */
1630         if (count < ichdr1.count) {
1631                 /*
1632                  * Figure the total bytes to be added to the destination leaf.
1633                  */
1634                 /* number entries being moved */
1635                 count = ichdr1.count - count;
1636                 space  = ichdr1.usedbytes - totallen;
1637                 space += count * sizeof(xfs_attr_leaf_entry_t);
1638
1639                 /*
1640                  * leaf2 is the destination, compact it if it looks tight.
1641                  */
1642                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1643                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1644                 if (space > max)
1645                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1646
1647                 /*
1648                  * Move high entries from leaf1 to low end of leaf2.
1649                  */
1650                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1651                                 leaf2, &ichdr2, 0, count, state->mp);
1652
1653         } else if (count > ichdr1.count) {
1654                 /*
1655                  * I assert that since all callers pass in an empty
1656                  * second buffer, this code should never execute.
1657                  */
1658                 ASSERT(0);
1659
1660                 /*
1661                  * Figure the total bytes to be added to the destination leaf.
1662                  */
1663                 /* number entries being moved */
1664                 count -= ichdr1.count;
1665                 space  = totallen - ichdr1.usedbytes;
1666                 space += count * sizeof(xfs_attr_leaf_entry_t);
1667
1668                 /*
1669                  * leaf1 is the destination, compact it if it looks tight.
1670                  */
1671                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1672                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1673                 if (space > max)
1674                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1675
1676                 /*
1677                  * Move low entries from leaf2 to high end of leaf1.
1678                  */
1679                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1680                                         ichdr1.count, count, state->mp);
1681         }
1682
1683         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1684         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1685         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1686         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1687
1688         /*
1689          * Copy out last hashval in each block for B-tree code.
1690          */
1691         entries1 = xfs_attr3_leaf_entryp(leaf1);
1692         entries2 = xfs_attr3_leaf_entryp(leaf2);
1693         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1694         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1695
1696         /*
1697          * Adjust the expected index for insertion.
1698          * NOTE: this code depends on the (current) situation that the
1699          * second block was originally empty.
1700          *
1701          * If the insertion point moved to the 2nd block, we must adjust
1702          * the index.  We must also track the entry just following the
1703          * new entry for use in an "atomic rename" operation, that entry
1704          * is always the "old" entry and the "new" entry is what we are
1705          * inserting.  The index/blkno fields refer to the "old" entry,
1706          * while the index2/blkno2 fields refer to the "new" entry.
1707          */
1708         if (blk1->index > ichdr1.count) {
1709                 ASSERT(state->inleaf == 0);
1710                 blk2->index = blk1->index - ichdr1.count;
1711                 args->index = args->index2 = blk2->index;
1712                 args->blkno = args->blkno2 = blk2->blkno;
1713         } else if (blk1->index == ichdr1.count) {
1714                 if (state->inleaf) {
1715                         args->index = blk1->index;
1716                         args->blkno = blk1->blkno;
1717                         args->index2 = 0;
1718                         args->blkno2 = blk2->blkno;
1719                 } else {
1720                         /*
1721                          * On a double leaf split, the original attr location
1722                          * is already stored in blkno2/index2, so don't
1723                          * overwrite it overwise we corrupt the tree.
1724                          */
1725                         blk2->index = blk1->index - ichdr1.count;
1726                         args->index = blk2->index;
1727                         args->blkno = blk2->blkno;
1728                         if (!state->extravalid) {
1729                                 /*
1730                                  * set the new attr location to match the old
1731                                  * one and let the higher level split code
1732                                  * decide where in the leaf to place it.
1733                                  */
1734                                 args->index2 = blk2->index;
1735                                 args->blkno2 = blk2->blkno;
1736                         }
1737                 }
1738         } else {
1739                 ASSERT(state->inleaf == 1);
1740                 args->index = args->index2 = blk1->index;
1741                 args->blkno = args->blkno2 = blk1->blkno;
1742         }
1743 }
1744
1745 /*
1746  * Examine entries until we reduce the absolute difference in
1747  * byte usage between the two blocks to a minimum.
1748  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1749  * GROT: there will always be enough room in either block for a new entry.
1750  * GROT: Do a double-split for this case?
1751  */
1752 STATIC int
1753 xfs_attr3_leaf_figure_balance(
1754         struct xfs_da_state             *state,
1755         struct xfs_da_state_blk         *blk1,
1756         struct xfs_attr3_icleaf_hdr     *ichdr1,
1757         struct xfs_da_state_blk         *blk2,
1758         struct xfs_attr3_icleaf_hdr     *ichdr2,
1759         int                             *countarg,
1760         int                             *usedbytesarg)
1761 {
1762         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1763         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1764         struct xfs_attr_leaf_entry      *entry;
1765         int                             count;
1766         int                             max;
1767         int                             index;
1768         int                             totallen = 0;
1769         int                             half;
1770         int                             lastdelta;
1771         int                             foundit = 0;
1772         int                             tmp;
1773
1774         /*
1775          * Examine entries until we reduce the absolute difference in
1776          * byte usage between the two blocks to a minimum.
1777          */
1778         max = ichdr1->count + ichdr2->count;
1779         half = (max + 1) * sizeof(*entry);
1780         half += ichdr1->usedbytes + ichdr2->usedbytes +
1781                         xfs_attr_leaf_newentsize(state->args->namelen,
1782                                                  state->args->valuelen,
1783                                                  state->blocksize, NULL);
1784         half /= 2;
1785         lastdelta = state->blocksize;
1786         entry = xfs_attr3_leaf_entryp(leaf1);
1787         for (count = index = 0; count < max; entry++, index++, count++) {
1788
1789 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1790                 /*
1791                  * The new entry is in the first block, account for it.
1792                  */
1793                 if (count == blk1->index) {
1794                         tmp = totallen + sizeof(*entry) +
1795                                 xfs_attr_leaf_newentsize(
1796                                                 state->args->namelen,
1797                                                 state->args->valuelen,
1798                                                 state->blocksize, NULL);
1799                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1800                                 break;
1801                         lastdelta = XFS_ATTR_ABS(half - tmp);
1802                         totallen = tmp;
1803                         foundit = 1;
1804                 }
1805
1806                 /*
1807                  * Wrap around into the second block if necessary.
1808                  */
1809                 if (count == ichdr1->count) {
1810                         leaf1 = leaf2;
1811                         entry = xfs_attr3_leaf_entryp(leaf1);
1812                         index = 0;
1813                 }
1814
1815                 /*
1816                  * Figure out if next leaf entry would be too much.
1817                  */
1818                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1819                                                                         index);
1820                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1821                         break;
1822                 lastdelta = XFS_ATTR_ABS(half - tmp);
1823                 totallen = tmp;
1824 #undef XFS_ATTR_ABS
1825         }
1826
1827         /*
1828          * Calculate the number of usedbytes that will end up in lower block.
1829          * If new entry not in lower block, fix up the count.
1830          */
1831         totallen -= count * sizeof(*entry);
1832         if (foundit) {
1833                 totallen -= sizeof(*entry) +
1834                                 xfs_attr_leaf_newentsize(
1835                                                 state->args->namelen,
1836                                                 state->args->valuelen,
1837                                                 state->blocksize, NULL);
1838         }
1839
1840         *countarg = count;
1841         *usedbytesarg = totallen;
1842         return foundit;
1843 }
1844
1845 /*========================================================================
1846  * Routines used for shrinking the Btree.
1847  *========================================================================*/
1848
1849 /*
1850  * Check a leaf block and its neighbors to see if the block should be
1851  * collapsed into one or the other neighbor.  Always keep the block
1852  * with the smaller block number.
1853  * If the current block is over 50% full, don't try to join it, return 0.
1854  * If the block is empty, fill in the state structure and return 2.
1855  * If it can be collapsed, fill in the state structure and return 1.
1856  * If nothing can be done, return 0.
1857  *
1858  * GROT: allow for INCOMPLETE entries in calculation.
1859  */
1860 int
1861 xfs_attr3_leaf_toosmall(
1862         struct xfs_da_state     *state,
1863         int                     *action)
1864 {
1865         struct xfs_attr_leafblock *leaf;
1866         struct xfs_da_state_blk *blk;
1867         struct xfs_attr3_icleaf_hdr ichdr;
1868         struct xfs_buf          *bp;
1869         xfs_dablk_t             blkno;
1870         int                     bytes;
1871         int                     forward;
1872         int                     error;
1873         int                     retval;
1874         int                     i;
1875
1876         trace_xfs_attr_leaf_toosmall(state->args);
1877
1878         /*
1879          * Check for the degenerate case of the block being over 50% full.
1880          * If so, it's not worth even looking to see if we might be able
1881          * to coalesce with a sibling.
1882          */
1883         blk = &state->path.blk[ state->path.active-1 ];
1884         leaf = blk->bp->b_addr;
1885         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1886         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1887                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1888                 ichdr.usedbytes;
1889         if (bytes > (state->blocksize >> 1)) {
1890                 *action = 0;    /* blk over 50%, don't try to join */
1891                 return(0);
1892         }
1893
1894         /*
1895          * Check for the degenerate case of the block being empty.
1896          * If the block is empty, we'll simply delete it, no need to
1897          * coalesce it with a sibling block.  We choose (arbitrarily)
1898          * to merge with the forward block unless it is NULL.
1899          */
1900         if (ichdr.count == 0) {
1901                 /*
1902                  * Make altpath point to the block we want to keep and
1903                  * path point to the block we want to drop (this one).
1904                  */
1905                 forward = (ichdr.forw != 0);
1906                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1907                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1908                                                  0, &retval);
1909                 if (error)
1910                         return(error);
1911                 if (retval) {
1912                         *action = 0;
1913                 } else {
1914                         *action = 2;
1915                 }
1916                 return 0;
1917         }
1918
1919         /*
1920          * Examine each sibling block to see if we can coalesce with
1921          * at least 25% free space to spare.  We need to figure out
1922          * whether to merge with the forward or the backward block.
1923          * We prefer coalescing with the lower numbered sibling so as
1924          * to shrink an attribute list over time.
1925          */
1926         /* start with smaller blk num */
1927         forward = ichdr.forw < ichdr.back;
1928         for (i = 0; i < 2; forward = !forward, i++) {
1929                 struct xfs_attr3_icleaf_hdr ichdr2;
1930                 if (forward)
1931                         blkno = ichdr.forw;
1932                 else
1933                         blkno = ichdr.back;
1934                 if (blkno == 0)
1935                         continue;
1936                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1937                                         blkno, -1, &bp);
1938                 if (error)
1939                         return(error);
1940
1941                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1942
1943                 bytes = state->blocksize - (state->blocksize >> 2) -
1944                         ichdr.usedbytes - ichdr2.usedbytes -
1945                         ((ichdr.count + ichdr2.count) *
1946                                         sizeof(xfs_attr_leaf_entry_t)) -
1947                         xfs_attr3_leaf_hdr_size(leaf);
1948
1949                 xfs_trans_brelse(state->args->trans, bp);
1950                 if (bytes >= 0)
1951                         break;  /* fits with at least 25% to spare */
1952         }
1953         if (i >= 2) {
1954                 *action = 0;
1955                 return(0);
1956         }
1957
1958         /*
1959          * Make altpath point to the block we want to keep (the lower
1960          * numbered block) and path point to the block we want to drop.
1961          */
1962         memcpy(&state->altpath, &state->path, sizeof(state->path));
1963         if (blkno < blk->blkno) {
1964                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1965                                                  0, &retval);
1966         } else {
1967                 error = xfs_da3_path_shift(state, &state->path, forward,
1968                                                  0, &retval);
1969         }
1970         if (error)
1971                 return(error);
1972         if (retval) {
1973                 *action = 0;
1974         } else {
1975                 *action = 1;
1976         }
1977         return(0);
1978 }
1979
1980 /*
1981  * Remove a name from the leaf attribute list structure.
1982  *
1983  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1984  * If two leaves are 37% full, when combined they will leave 25% free.
1985  */
1986 int
1987 xfs_attr3_leaf_remove(
1988         struct xfs_buf          *bp,
1989         struct xfs_da_args      *args)
1990 {
1991         struct xfs_attr_leafblock *leaf;
1992         struct xfs_attr3_icleaf_hdr ichdr;
1993         struct xfs_attr_leaf_entry *entry;
1994         struct xfs_mount        *mp = args->trans->t_mountp;
1995         int                     before;
1996         int                     after;
1997         int                     smallest;
1998         int                     entsize;
1999         int                     tablesize;
2000         int                     tmp;
2001         int                     i;
2002
2003         trace_xfs_attr_leaf_remove(args);
2004
2005         leaf = bp->b_addr;
2006         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2007
2008         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
2009         ASSERT(args->index >= 0 && args->index < ichdr.count);
2010         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2011                                         xfs_attr3_leaf_hdr_size(leaf));
2012
2013         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2014
2015         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2016         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2017
2018         /*
2019          * Scan through free region table:
2020          *    check for adjacency of free'd entry with an existing one,
2021          *    find smallest free region in case we need to replace it,
2022          *    adjust any map that borders the entry table,
2023          */
2024         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2025                                         + xfs_attr3_leaf_hdr_size(leaf);
2026         tmp = ichdr.freemap[0].size;
2027         before = after = -1;
2028         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2029         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2030         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2031                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
2032                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
2033                 if (ichdr.freemap[i].base == tablesize) {
2034                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2035                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2036                 }
2037
2038                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2039                                 be16_to_cpu(entry->nameidx)) {
2040                         before = i;
2041                 } else if (ichdr.freemap[i].base ==
2042                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2043                         after = i;
2044                 } else if (ichdr.freemap[i].size < tmp) {
2045                         tmp = ichdr.freemap[i].size;
2046                         smallest = i;
2047                 }
2048         }
2049
2050         /*
2051          * Coalesce adjacent freemap regions,
2052          * or replace the smallest region.
2053          */
2054         if ((before >= 0) || (after >= 0)) {
2055                 if ((before >= 0) && (after >= 0)) {
2056                         ichdr.freemap[before].size += entsize;
2057                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2058                         ichdr.freemap[after].base = 0;
2059                         ichdr.freemap[after].size = 0;
2060                 } else if (before >= 0) {
2061                         ichdr.freemap[before].size += entsize;
2062                 } else {
2063                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2064                         ichdr.freemap[after].size += entsize;
2065                 }
2066         } else {
2067                 /*
2068                  * Replace smallest region (if it is smaller than free'd entry)
2069                  */
2070                 if (ichdr.freemap[smallest].size < entsize) {
2071                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2072                         ichdr.freemap[smallest].size = entsize;
2073                 }
2074         }
2075
2076         /*
2077          * Did we remove the first entry?
2078          */
2079         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2080                 smallest = 1;
2081         else
2082                 smallest = 0;
2083
2084         /*
2085          * Compress the remaining entries and zero out the removed stuff.
2086          */
2087         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2088         ichdr.usedbytes -= entsize;
2089         xfs_trans_log_buf(args->trans, bp,
2090              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2091                                    entsize));
2092
2093         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2094         memmove(entry, entry + 1, tmp);
2095         ichdr.count--;
2096         xfs_trans_log_buf(args->trans, bp,
2097             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2098
2099         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2100         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2101
2102         /*
2103          * If we removed the first entry, re-find the first used byte
2104          * in the name area.  Note that if the entry was the "firstused",
2105          * then we don't have a "hole" in our block resulting from
2106          * removing the name.
2107          */
2108         if (smallest) {
2109                 tmp = XFS_LBSIZE(mp);
2110                 entry = xfs_attr3_leaf_entryp(leaf);
2111                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2112                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2113                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2114
2115                         if (be16_to_cpu(entry->nameidx) < tmp)
2116                                 tmp = be16_to_cpu(entry->nameidx);
2117                 }
2118                 ichdr.firstused = tmp;
2119                 if (!ichdr.firstused)
2120                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
2121         } else {
2122                 ichdr.holes = 1;        /* mark as needing compaction */
2123         }
2124         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
2125         xfs_trans_log_buf(args->trans, bp,
2126                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2127                                           xfs_attr3_leaf_hdr_size(leaf)));
2128
2129         /*
2130          * Check if leaf is less than 50% full, caller may want to
2131          * "join" the leaf with a sibling if so.
2132          */
2133         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2134               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2135
2136         return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
2137 }
2138
2139 /*
2140  * Move all the attribute list entries from drop_leaf into save_leaf.
2141  */
2142 void
2143 xfs_attr3_leaf_unbalance(
2144         struct xfs_da_state     *state,
2145         struct xfs_da_state_blk *drop_blk,
2146         struct xfs_da_state_blk *save_blk)
2147 {
2148         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2149         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2150         struct xfs_attr3_icleaf_hdr drophdr;
2151         struct xfs_attr3_icleaf_hdr savehdr;
2152         struct xfs_attr_leaf_entry *entry;
2153         struct xfs_mount        *mp = state->mp;
2154
2155         trace_xfs_attr_leaf_unbalance(state->args);
2156
2157         drop_leaf = drop_blk->bp->b_addr;
2158         save_leaf = save_blk->bp->b_addr;
2159         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
2160         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
2161         entry = xfs_attr3_leaf_entryp(drop_leaf);
2162
2163         /*
2164          * Save last hashval from dying block for later Btree fixup.
2165          */
2166         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2167
2168         /*
2169          * Check if we need a temp buffer, or can we do it in place.
2170          * Note that we don't check "leaf" for holes because we will
2171          * always be dropping it, toosmall() decided that for us already.
2172          */
2173         if (savehdr.holes == 0) {
2174                 /*
2175                  * dest leaf has no holes, so we add there.  May need
2176                  * to make some room in the entry array.
2177                  */
2178                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2179                                          drop_blk->bp, &drophdr)) {
2180                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2181                                                 save_leaf, &savehdr, 0,
2182                                                 drophdr.count, mp);
2183                 } else {
2184                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2185                                                 save_leaf, &savehdr,
2186                                                 savehdr.count, drophdr.count, mp);
2187                 }
2188         } else {
2189                 /*
2190                  * Destination has holes, so we make a temporary copy
2191                  * of the leaf and add them both to that.
2192                  */
2193                 struct xfs_attr_leafblock *tmp_leaf;
2194                 struct xfs_attr3_icleaf_hdr tmphdr;
2195
2196                 tmp_leaf = kmem_zalloc(state->blocksize, KM_SLEEP);
2197
2198                 /*
2199                  * Copy the header into the temp leaf so that all the stuff
2200                  * not in the incore header is present and gets copied back in
2201                  * once we've moved all the entries.
2202                  */
2203                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2204
2205                 memset(&tmphdr, 0, sizeof(tmphdr));
2206                 tmphdr.magic = savehdr.magic;
2207                 tmphdr.forw = savehdr.forw;
2208                 tmphdr.back = savehdr.back;
2209                 tmphdr.firstused = state->blocksize;
2210
2211                 /* write the header to the temp buffer to initialise it */
2212                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2213
2214                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2215                                          drop_blk->bp, &drophdr)) {
2216                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2217                                                 tmp_leaf, &tmphdr, 0,
2218                                                 drophdr.count, mp);
2219                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2220                                                 tmp_leaf, &tmphdr, tmphdr.count,
2221                                                 savehdr.count, mp);
2222                 } else {
2223                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2224                                                 tmp_leaf, &tmphdr, 0,
2225                                                 savehdr.count, mp);
2226                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2227                                                 tmp_leaf, &tmphdr, tmphdr.count,
2228                                                 drophdr.count, mp);
2229                 }
2230                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2231                 savehdr = tmphdr; /* struct copy */
2232                 kmem_free(tmp_leaf);
2233         }
2234
2235         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2236         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2237                                            state->blocksize - 1);
2238
2239         /*
2240          * Copy out last hashval in each block for B-tree code.
2241          */
2242         entry = xfs_attr3_leaf_entryp(save_leaf);
2243         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2244 }
2245
2246 /*========================================================================
2247  * Routines used for finding things in the Btree.
2248  *========================================================================*/
2249
2250 /*
2251  * Look up a name in a leaf attribute list structure.
2252  * This is the internal routine, it uses the caller's buffer.
2253  *
2254  * Note that duplicate keys are allowed, but only check within the
2255  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2256  *
2257  * Return in args->index the index into the entry[] array of either
2258  * the found entry, or where the entry should have been (insert before
2259  * that entry).
2260  *
2261  * Don't change the args->value unless we find the attribute.
2262  */
2263 int
2264 xfs_attr3_leaf_lookup_int(
2265         struct xfs_buf          *bp,
2266         struct xfs_da_args      *args)
2267 {
2268         struct xfs_attr_leafblock *leaf;
2269         struct xfs_attr3_icleaf_hdr ichdr;
2270         struct xfs_attr_leaf_entry *entry;
2271         struct xfs_attr_leaf_entry *entries;
2272         struct xfs_attr_leaf_name_local *name_loc;
2273         struct xfs_attr_leaf_name_remote *name_rmt;
2274         xfs_dahash_t            hashval;
2275         int                     probe;
2276         int                     span;
2277
2278         trace_xfs_attr_leaf_lookup(args);
2279
2280         leaf = bp->b_addr;
2281         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2282         entries = xfs_attr3_leaf_entryp(leaf);
2283         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2284
2285         /*
2286          * Binary search.  (note: small blocks will skip this loop)
2287          */
2288         hashval = args->hashval;
2289         probe = span = ichdr.count / 2;
2290         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2291                 span /= 2;
2292                 if (be32_to_cpu(entry->hashval) < hashval)
2293                         probe += span;
2294                 else if (be32_to_cpu(entry->hashval) > hashval)
2295                         probe -= span;
2296                 else
2297                         break;
2298         }
2299         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2300         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2301
2302         /*
2303          * Since we may have duplicate hashval's, find the first matching
2304          * hashval in the leaf.
2305          */
2306         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2307                 entry--;
2308                 probe--;
2309         }
2310         while (probe < ichdr.count &&
2311                be32_to_cpu(entry->hashval) < hashval) {
2312                 entry++;
2313                 probe++;
2314         }
2315         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2316                 args->index = probe;
2317                 return XFS_ERROR(ENOATTR);
2318         }
2319
2320         /*
2321          * Duplicate keys may be present, so search all of them for a match.
2322          */
2323         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2324                         entry++, probe++) {
2325 /*
2326  * GROT: Add code to remove incomplete entries.
2327  */
2328                 /*
2329                  * If we are looking for INCOMPLETE entries, show only those.
2330                  * If we are looking for complete entries, show only those.
2331                  */
2332                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2333                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2334                         continue;
2335                 }
2336                 if (entry->flags & XFS_ATTR_LOCAL) {
2337                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2338                         if (name_loc->namelen != args->namelen)
2339                                 continue;
2340                         if (memcmp(args->name, name_loc->nameval,
2341                                                         args->namelen) != 0)
2342                                 continue;
2343                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2344                                 continue;
2345                         args->index = probe;
2346                         return XFS_ERROR(EEXIST);
2347                 } else {
2348                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2349                         if (name_rmt->namelen != args->namelen)
2350                                 continue;
2351                         if (memcmp(args->name, name_rmt->name,
2352                                                         args->namelen) != 0)
2353                                 continue;
2354                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2355                                 continue;
2356                         args->index = probe;
2357                         args->valuelen = be32_to_cpu(name_rmt->valuelen);
2358                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2359                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2360                                                         args->dp->i_mount,
2361                                                         args->valuelen);
2362                         return XFS_ERROR(EEXIST);
2363                 }
2364         }
2365         args->index = probe;
2366         return XFS_ERROR(ENOATTR);
2367 }
2368
2369 /*
2370  * Get the value associated with an attribute name from a leaf attribute
2371  * list structure.
2372  */
2373 int
2374 xfs_attr3_leaf_getvalue(
2375         struct xfs_buf          *bp,
2376         struct xfs_da_args      *args)
2377 {
2378         struct xfs_attr_leafblock *leaf;
2379         struct xfs_attr3_icleaf_hdr ichdr;
2380         struct xfs_attr_leaf_entry *entry;
2381         struct xfs_attr_leaf_name_local *name_loc;
2382         struct xfs_attr_leaf_name_remote *name_rmt;
2383         int                     valuelen;
2384
2385         leaf = bp->b_addr;
2386         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2387         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2388         ASSERT(args->index < ichdr.count);
2389
2390         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2391         if (entry->flags & XFS_ATTR_LOCAL) {
2392                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2393                 ASSERT(name_loc->namelen == args->namelen);
2394                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2395                 valuelen = be16_to_cpu(name_loc->valuelen);
2396                 if (args->flags & ATTR_KERNOVAL) {
2397                         args->valuelen = valuelen;
2398                         return 0;
2399                 }
2400                 if (args->valuelen < valuelen) {
2401                         args->valuelen = valuelen;
2402                         return XFS_ERROR(ERANGE);
2403                 }
2404                 args->valuelen = valuelen;
2405                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2406         } else {
2407                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2408                 ASSERT(name_rmt->namelen == args->namelen);
2409                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2410                 valuelen = be32_to_cpu(name_rmt->valuelen);
2411                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2412                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2413                                                        valuelen);
2414                 if (args->flags & ATTR_KERNOVAL) {
2415                         args->valuelen = valuelen;
2416                         return 0;
2417                 }
2418                 if (args->valuelen < valuelen) {
2419                         args->valuelen = valuelen;
2420                         return XFS_ERROR(ERANGE);
2421                 }
2422                 args->valuelen = valuelen;
2423         }
2424         return 0;
2425 }
2426
2427 /*========================================================================
2428  * Utility routines.
2429  *========================================================================*/
2430
2431 /*
2432  * Move the indicated entries from one leaf to another.
2433  * NOTE: this routine modifies both source and destination leaves.
2434  */
2435 /*ARGSUSED*/
2436 STATIC void
2437 xfs_attr3_leaf_moveents(
2438         struct xfs_attr_leafblock       *leaf_s,
2439         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2440         int                             start_s,
2441         struct xfs_attr_leafblock       *leaf_d,
2442         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2443         int                             start_d,
2444         int                             count,
2445         struct xfs_mount                *mp)
2446 {
2447         struct xfs_attr_leaf_entry      *entry_s;
2448         struct xfs_attr_leaf_entry      *entry_d;
2449         int                             desti;
2450         int                             tmp;
2451         int                             i;
2452
2453         /*
2454          * Check for nothing to do.
2455          */
2456         if (count == 0)
2457                 return;
2458
2459         /*
2460          * Set up environment.
2461          */
2462         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2463                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2464         ASSERT(ichdr_s->magic == ichdr_d->magic);
2465         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2466         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2467                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2468         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2469         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2470                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2471
2472         ASSERT(start_s < ichdr_s->count);
2473         ASSERT(start_d <= ichdr_d->count);
2474         ASSERT(count <= ichdr_s->count);
2475
2476
2477         /*
2478          * Move the entries in the destination leaf up to make a hole?
2479          */
2480         if (start_d < ichdr_d->count) {
2481                 tmp  = ichdr_d->count - start_d;
2482                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2483                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2484                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2485                 memmove(entry_d, entry_s, tmp);
2486         }
2487
2488         /*
2489          * Copy all entry's in the same (sorted) order,
2490          * but allocate attribute info packed and in sequence.
2491          */
2492         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2493         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2494         desti = start_d;
2495         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2496                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2497                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2498 #ifdef GROT
2499                 /*
2500                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2501                  * may also need to change the insertion index.  Code turned
2502                  * off for 6.2, should be revisited later.
2503                  */
2504                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2505                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2506                         ichdr_s->usedbytes -= tmp;
2507                         ichdr_s->count -= 1;
2508                         entry_d--;      /* to compensate for ++ in loop hdr */
2509                         desti--;
2510                         if ((start_s + i) < offset)
2511                                 result++;       /* insertion index adjustment */
2512                 } else {
2513 #endif /* GROT */
2514                         ichdr_d->firstused -= tmp;
2515                         /* both on-disk, don't endian flip twice */
2516                         entry_d->hashval = entry_s->hashval;
2517                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2518                         entry_d->flags = entry_s->flags;
2519                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2520                                                         <= XFS_LBSIZE(mp));
2521                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2522                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2523                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2524                                                         <= XFS_LBSIZE(mp));
2525                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2526                         ichdr_s->usedbytes -= tmp;
2527                         ichdr_d->usedbytes += tmp;
2528                         ichdr_s->count -= 1;
2529                         ichdr_d->count += 1;
2530                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2531                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2532                         ASSERT(ichdr_d->firstused >= tmp);
2533 #ifdef GROT
2534                 }
2535 #endif /* GROT */
2536         }
2537
2538         /*
2539          * Zero out the entries we just copied.
2540          */
2541         if (start_s == ichdr_s->count) {
2542                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2543                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2544                 ASSERT(((char *)entry_s + tmp) <=
2545                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2546                 memset(entry_s, 0, tmp);
2547         } else {
2548                 /*
2549                  * Move the remaining entries down to fill the hole,
2550                  * then zero the entries at the top.
2551                  */
2552                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2553                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2554                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2555                 memmove(entry_d, entry_s, tmp);
2556
2557                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2558                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2559                 ASSERT(((char *)entry_s + tmp) <=
2560                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2561                 memset(entry_s, 0, tmp);
2562         }
2563
2564         /*
2565          * Fill in the freemap information
2566          */
2567         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2568         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2569         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2570         ichdr_d->freemap[1].base = 0;
2571         ichdr_d->freemap[2].base = 0;
2572         ichdr_d->freemap[1].size = 0;
2573         ichdr_d->freemap[2].size = 0;
2574         ichdr_s->holes = 1;     /* leaf may not be compact */
2575 }
2576
2577 /*
2578  * Pick up the last hashvalue from a leaf block.
2579  */
2580 xfs_dahash_t
2581 xfs_attr_leaf_lasthash(
2582         struct xfs_buf  *bp,
2583         int             *count)
2584 {
2585         struct xfs_attr3_icleaf_hdr ichdr;
2586         struct xfs_attr_leaf_entry *entries;
2587
2588         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2589         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2590         if (count)
2591                 *count = ichdr.count;
2592         if (!ichdr.count)
2593                 return 0;
2594         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2595 }
2596
2597 /*
2598  * Calculate the number of bytes used to store the indicated attribute
2599  * (whether local or remote only calculate bytes in this block).
2600  */
2601 STATIC int
2602 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2603 {
2604         struct xfs_attr_leaf_entry *entries;
2605         xfs_attr_leaf_name_local_t *name_loc;
2606         xfs_attr_leaf_name_remote_t *name_rmt;
2607         int size;
2608
2609         entries = xfs_attr3_leaf_entryp(leaf);
2610         if (entries[index].flags & XFS_ATTR_LOCAL) {
2611                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2612                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2613                                                    be16_to_cpu(name_loc->valuelen));
2614         } else {
2615                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2616                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2617         }
2618         return size;
2619 }
2620
2621 /*
2622  * Calculate the number of bytes that would be required to store the new
2623  * attribute (whether local or remote only calculate bytes in this block).
2624  * This routine decides as a side effect whether the attribute will be
2625  * a "local" or a "remote" attribute.
2626  */
2627 int
2628 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2629 {
2630         int size;
2631
2632         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2633         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2634                 if (local) {
2635                         *local = 1;
2636                 }
2637         } else {
2638                 size = xfs_attr_leaf_entsize_remote(namelen);
2639                 if (local) {
2640                         *local = 0;
2641                 }
2642         }
2643         return size;
2644 }
2645
2646 /*
2647  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2648  */
2649 int
2650 xfs_attr3_leaf_list_int(
2651         struct xfs_buf                  *bp,
2652         struct xfs_attr_list_context    *context)
2653 {
2654         struct attrlist_cursor_kern     *cursor;
2655         struct xfs_attr_leafblock       *leaf;
2656         struct xfs_attr3_icleaf_hdr     ichdr;
2657         struct xfs_attr_leaf_entry      *entries;
2658         struct xfs_attr_leaf_entry      *entry;
2659         int                             retval;
2660         int                             i;
2661
2662         trace_xfs_attr_list_leaf(context);
2663
2664         leaf = bp->b_addr;
2665         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2666         entries = xfs_attr3_leaf_entryp(leaf);
2667
2668         cursor = context->cursor;
2669         cursor->initted = 1;
2670
2671         /*
2672          * Re-find our place in the leaf block if this is a new syscall.
2673          */
2674         if (context->resynch) {
2675                 entry = &entries[0];
2676                 for (i = 0; i < ichdr.count; entry++, i++) {
2677                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2678                                 if (cursor->offset == context->dupcnt) {
2679                                         context->dupcnt = 0;
2680                                         break;
2681                                 }
2682                                 context->dupcnt++;
2683                         } else if (be32_to_cpu(entry->hashval) >
2684                                         cursor->hashval) {
2685                                 context->dupcnt = 0;
2686                                 break;
2687                         }
2688                 }
2689                 if (i == ichdr.count) {
2690                         trace_xfs_attr_list_notfound(context);
2691                         return 0;
2692                 }
2693         } else {
2694                 entry = &entries[0];
2695                 i = 0;
2696         }
2697         context->resynch = 0;
2698
2699         /*
2700          * We have found our place, start copying out the new attributes.
2701          */
2702         retval = 0;
2703         for (; i < ichdr.count; entry++, i++) {
2704                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2705                         cursor->hashval = be32_to_cpu(entry->hashval);
2706                         cursor->offset = 0;
2707                 }
2708
2709                 if (entry->flags & XFS_ATTR_INCOMPLETE)
2710                         continue;               /* skip incomplete entries */
2711
2712                 if (entry->flags & XFS_ATTR_LOCAL) {
2713                         xfs_attr_leaf_name_local_t *name_loc =
2714                                 xfs_attr3_leaf_name_local(leaf, i);
2715
2716                         retval = context->put_listent(context,
2717                                                 entry->flags,
2718                                                 name_loc->nameval,
2719                                                 (int)name_loc->namelen,
2720                                                 be16_to_cpu(name_loc->valuelen),
2721                                                 &name_loc->nameval[name_loc->namelen]);
2722                         if (retval)
2723                                 return retval;
2724                 } else {
2725                         xfs_attr_leaf_name_remote_t *name_rmt =
2726                                 xfs_attr3_leaf_name_remote(leaf, i);
2727
2728                         int valuelen = be32_to_cpu(name_rmt->valuelen);
2729
2730                         if (context->put_value) {
2731                                 xfs_da_args_t args;
2732
2733                                 memset((char *)&args, 0, sizeof(args));
2734                                 args.dp = context->dp;
2735                                 args.whichfork = XFS_ATTR_FORK;
2736                                 args.valuelen = valuelen;
2737                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2738                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2739                                 args.rmtblkcnt = xfs_attr3_rmt_blocks(
2740                                                         args.dp->i_mount, valuelen);
2741                                 retval = xfs_attr_rmtval_get(&args);
2742                                 if (retval)
2743                                         return retval;
2744                                 retval = context->put_listent(context,
2745                                                 entry->flags,
2746                                                 name_rmt->name,
2747                                                 (int)name_rmt->namelen,
2748                                                 valuelen,
2749                                                 args.value);
2750                                 kmem_free(args.value);
2751                         } else {
2752                                 retval = context->put_listent(context,
2753                                                 entry->flags,
2754                                                 name_rmt->name,
2755                                                 (int)name_rmt->namelen,
2756                                                 valuelen,
2757                                                 NULL);
2758                         }
2759                         if (retval)
2760                                 return retval;
2761                 }
2762                 if (context->seen_enough)
2763                         break;
2764                 cursor->offset++;
2765         }
2766         trace_xfs_attr_list_leaf_end(context);
2767         return retval;
2768 }
2769
2770
2771 /*========================================================================
2772  * Manage the INCOMPLETE flag in a leaf entry
2773  *========================================================================*/
2774
2775 /*
2776  * Clear the INCOMPLETE flag on an entry in a leaf block.
2777  */
2778 int
2779 xfs_attr3_leaf_clearflag(
2780         struct xfs_da_args      *args)
2781 {
2782         struct xfs_attr_leafblock *leaf;
2783         struct xfs_attr_leaf_entry *entry;
2784         struct xfs_attr_leaf_name_remote *name_rmt;
2785         struct xfs_buf          *bp;
2786         int                     error;
2787 #ifdef DEBUG
2788         struct xfs_attr3_icleaf_hdr ichdr;
2789         xfs_attr_leaf_name_local_t *name_loc;
2790         int namelen;
2791         char *name;
2792 #endif /* DEBUG */
2793
2794         trace_xfs_attr_leaf_clearflag(args);
2795         /*
2796          * Set up the operation.
2797          */
2798         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2799         if (error)
2800                 return(error);
2801
2802         leaf = bp->b_addr;
2803         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2804         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2805
2806 #ifdef DEBUG
2807         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2808         ASSERT(args->index < ichdr.count);
2809         ASSERT(args->index >= 0);
2810
2811         if (entry->flags & XFS_ATTR_LOCAL) {
2812                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2813                 namelen = name_loc->namelen;
2814                 name = (char *)name_loc->nameval;
2815         } else {
2816                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2817                 namelen = name_rmt->namelen;
2818                 name = (char *)name_rmt->name;
2819         }
2820         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2821         ASSERT(namelen == args->namelen);
2822         ASSERT(memcmp(name, args->name, namelen) == 0);
2823 #endif /* DEBUG */
2824
2825         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2826         xfs_trans_log_buf(args->trans, bp,
2827                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2828
2829         if (args->rmtblkno) {
2830                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2831                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2832                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2833                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2834                 xfs_trans_log_buf(args->trans, bp,
2835                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2836         }
2837
2838         /*
2839          * Commit the flag value change and start the next trans in series.
2840          */
2841         return xfs_trans_roll(&args->trans, args->dp);
2842 }
2843
2844 /*
2845  * Set the INCOMPLETE flag on an entry in a leaf block.
2846  */
2847 int
2848 xfs_attr3_leaf_setflag(
2849         struct xfs_da_args      *args)
2850 {
2851         struct xfs_attr_leafblock *leaf;
2852         struct xfs_attr_leaf_entry *entry;
2853         struct xfs_attr_leaf_name_remote *name_rmt;
2854         struct xfs_buf          *bp;
2855         int error;
2856 #ifdef DEBUG
2857         struct xfs_attr3_icleaf_hdr ichdr;
2858 #endif
2859
2860         trace_xfs_attr_leaf_setflag(args);
2861
2862         /*
2863          * Set up the operation.
2864          */
2865         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2866         if (error)
2867                 return(error);
2868
2869         leaf = bp->b_addr;
2870 #ifdef DEBUG
2871         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2872         ASSERT(args->index < ichdr.count);
2873         ASSERT(args->index >= 0);
2874 #endif
2875         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2876
2877         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2878         entry->flags |= XFS_ATTR_INCOMPLETE;
2879         xfs_trans_log_buf(args->trans, bp,
2880                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2881         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2882                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2883                 name_rmt->valueblk = 0;
2884                 name_rmt->valuelen = 0;
2885                 xfs_trans_log_buf(args->trans, bp,
2886                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2887         }
2888
2889         /*
2890          * Commit the flag value change and start the next trans in series.
2891          */
2892         return xfs_trans_roll(&args->trans, args->dp);
2893 }
2894
2895 /*
2896  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2897  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2898  * entry given by args->blkno2/index2.
2899  *
2900  * Note that they could be in different blocks, or in the same block.
2901  */
2902 int
2903 xfs_attr3_leaf_flipflags(
2904         struct xfs_da_args      *args)
2905 {
2906         struct xfs_attr_leafblock *leaf1;
2907         struct xfs_attr_leafblock *leaf2;
2908         struct xfs_attr_leaf_entry *entry1;
2909         struct xfs_attr_leaf_entry *entry2;
2910         struct xfs_attr_leaf_name_remote *name_rmt;
2911         struct xfs_buf          *bp1;
2912         struct xfs_buf          *bp2;
2913         int error;
2914 #ifdef DEBUG
2915         struct xfs_attr3_icleaf_hdr ichdr1;
2916         struct xfs_attr3_icleaf_hdr ichdr2;
2917         xfs_attr_leaf_name_local_t *name_loc;
2918         int namelen1, namelen2;
2919         char *name1, *name2;
2920 #endif /* DEBUG */
2921
2922         trace_xfs_attr_leaf_flipflags(args);
2923
2924         /*
2925          * Read the block containing the "old" attr
2926          */
2927         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2928         if (error)
2929                 return error;
2930
2931         /*
2932          * Read the block containing the "new" attr, if it is different
2933          */
2934         if (args->blkno2 != args->blkno) {
2935                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2936                                            -1, &bp2);
2937                 if (error)
2938                         return error;
2939         } else {
2940                 bp2 = bp1;
2941         }
2942
2943         leaf1 = bp1->b_addr;
2944         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2945
2946         leaf2 = bp2->b_addr;
2947         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2948
2949 #ifdef DEBUG
2950         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2951         ASSERT(args->index < ichdr1.count);
2952         ASSERT(args->index >= 0);
2953
2954         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2955         ASSERT(args->index2 < ichdr2.count);
2956         ASSERT(args->index2 >= 0);
2957
2958         if (entry1->flags & XFS_ATTR_LOCAL) {
2959                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2960                 namelen1 = name_loc->namelen;
2961                 name1 = (char *)name_loc->nameval;
2962         } else {
2963                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2964                 namelen1 = name_rmt->namelen;
2965                 name1 = (char *)name_rmt->name;
2966         }
2967         if (entry2->flags & XFS_ATTR_LOCAL) {
2968                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2969                 namelen2 = name_loc->namelen;
2970                 name2 = (char *)name_loc->nameval;
2971         } else {
2972                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2973                 namelen2 = name_rmt->namelen;
2974                 name2 = (char *)name_rmt->name;
2975         }
2976         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2977         ASSERT(namelen1 == namelen2);
2978         ASSERT(memcmp(name1, name2, namelen1) == 0);
2979 #endif /* DEBUG */
2980
2981         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2982         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2983
2984         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2985         xfs_trans_log_buf(args->trans, bp1,
2986                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2987         if (args->rmtblkno) {
2988                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2989                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2990                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2991                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2992                 xfs_trans_log_buf(args->trans, bp1,
2993                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2994         }
2995
2996         entry2->flags |= XFS_ATTR_INCOMPLETE;
2997         xfs_trans_log_buf(args->trans, bp2,
2998                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2999         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
3000                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
3001                 name_rmt->valueblk = 0;
3002                 name_rmt->valuelen = 0;
3003                 xfs_trans_log_buf(args->trans, bp2,
3004                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3005         }
3006
3007         /*
3008          * Commit the flag value change and start the next trans in series.
3009          */
3010         error = xfs_trans_roll(&args->trans, args->dp);
3011
3012         return error;
3013 }
3014
3015 /*========================================================================
3016  * Indiscriminately delete the entire attribute fork
3017  *========================================================================*/
3018
3019 /*
3020  * Recurse (gasp!) through the attribute nodes until we find leaves.
3021  * We're doing a depth-first traversal in order to invalidate everything.
3022  */
3023 int
3024 xfs_attr3_root_inactive(
3025         struct xfs_trans        **trans,
3026         struct xfs_inode        *dp)
3027 {
3028         struct xfs_da_blkinfo   *info;
3029         struct xfs_buf          *bp;
3030         xfs_daddr_t             blkno;
3031         int                     error;
3032
3033         /*
3034          * Read block 0 to see what we have to work with.
3035          * We only get here if we have extents, since we remove
3036          * the extents in reverse order the extent containing
3037          * block 0 must still be there.
3038          */
3039         error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
3040         if (error)
3041                 return error;
3042         blkno = bp->b_bn;
3043
3044         /*
3045          * Invalidate the tree, even if the "tree" is only a single leaf block.
3046          * This is a depth-first traversal!
3047          */
3048         info = bp->b_addr;
3049         switch (info->magic) {
3050         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3051         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3052                 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
3053                 break;
3054         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3055         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3056                 error = xfs_attr3_leaf_inactive(trans, dp, bp);
3057                 break;
3058         default:
3059                 error = XFS_ERROR(EIO);
3060                 xfs_trans_brelse(*trans, bp);
3061                 break;
3062         }
3063         if (error)
3064                 return error;
3065
3066         /*
3067          * Invalidate the incore copy of the root block.
3068          */
3069         error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
3070         if (error)
3071                 return error;
3072         xfs_trans_binval(*trans, bp);   /* remove from cache */
3073         /*
3074          * Commit the invalidate and start the next transaction.
3075          */
3076         error = xfs_trans_roll(trans, dp);
3077
3078         return error;
3079 }
3080
3081 /*
3082  * Recurse (gasp!) through the attribute nodes until we find leaves.
3083  * We're doing a depth-first traversal in order to invalidate everything.
3084  */
3085 STATIC int
3086 xfs_attr3_node_inactive(
3087         struct xfs_trans **trans,
3088         struct xfs_inode *dp,
3089         struct xfs_buf  *bp,
3090         int             level)
3091 {
3092         xfs_da_blkinfo_t *info;
3093         xfs_da_intnode_t *node;
3094         xfs_dablk_t child_fsb;
3095         xfs_daddr_t parent_blkno, child_blkno;
3096         int error, i;
3097         struct xfs_buf *child_bp;
3098         struct xfs_da_node_entry *btree;
3099         struct xfs_da3_icnode_hdr ichdr;
3100
3101         /*
3102          * Since this code is recursive (gasp!) we must protect ourselves.
3103          */
3104         if (level > XFS_DA_NODE_MAXDEPTH) {
3105                 xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3106                 return XFS_ERROR(EIO);
3107         }
3108
3109         node = bp->b_addr;
3110         xfs_da3_node_hdr_from_disk(&ichdr, node);
3111         parent_blkno = bp->b_bn;
3112         if (!ichdr.count) {
3113                 xfs_trans_brelse(*trans, bp);
3114                 return 0;
3115         }
3116         btree = xfs_da3_node_tree_p(node);
3117         child_fsb = be32_to_cpu(btree[0].before);
3118         xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3119
3120         /*
3121          * If this is the node level just above the leaves, simply loop
3122          * over the leaves removing all of them.  If this is higher up
3123          * in the tree, recurse downward.
3124          */
3125         for (i = 0; i < ichdr.count; i++) {
3126                 /*
3127                  * Read the subsidiary block to see what we have to work with.
3128                  * Don't do this in a transaction.  This is a depth-first
3129                  * traversal of the tree so we may deal with many blocks
3130                  * before we come back to this one.
3131                  */
3132                 error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp,
3133                                                 XFS_ATTR_FORK);
3134                 if (error)
3135                         return(error);
3136                 if (child_bp) {
3137                                                 /* save for re-read later */
3138                         child_blkno = XFS_BUF_ADDR(child_bp);
3139
3140                         /*
3141                          * Invalidate the subtree, however we have to.
3142                          */
3143                         info = child_bp->b_addr;
3144                         switch (info->magic) {
3145                         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3146                         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3147                                 error = xfs_attr3_node_inactive(trans, dp,
3148                                                         child_bp, level + 1);
3149                                 break;
3150                         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3151                         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3152                                 error = xfs_attr3_leaf_inactive(trans, dp,
3153                                                         child_bp);
3154                                 break;
3155                         default:
3156                                 error = XFS_ERROR(EIO);
3157                                 xfs_trans_brelse(*trans, child_bp);
3158                                 break;
3159                         }
3160                         if (error)
3161                                 return error;
3162
3163                         /*
3164                          * Remove the subsidiary block from the cache
3165                          * and from the log.
3166                          */
3167                         error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
3168                                 &child_bp, XFS_ATTR_FORK);
3169                         if (error)
3170                                 return error;
3171                         xfs_trans_binval(*trans, child_bp);
3172                 }
3173
3174                 /*
3175                  * If we're not done, re-read the parent to get the next
3176                  * child block number.
3177                  */
3178                 if (i + 1 < ichdr.count) {
3179                         error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
3180                                                  &bp, XFS_ATTR_FORK);
3181                         if (error)
3182                                 return error;
3183                         child_fsb = be32_to_cpu(btree[i + 1].before);
3184                         xfs_trans_brelse(*trans, bp);
3185                 }
3186                 /*
3187                  * Atomically commit the whole invalidate stuff.
3188                  */
3189                 error = xfs_trans_roll(trans, dp);
3190                 if (error)
3191                         return  error;
3192         }
3193
3194         return 0;
3195 }
3196
3197 /*
3198  * Invalidate all of the "remote" value regions pointed to by a particular
3199  * leaf block.
3200  * Note that we must release the lock on the buffer so that we are not
3201  * caught holding something that the logging code wants to flush to disk.
3202  */
3203 STATIC int
3204 xfs_attr3_leaf_inactive(
3205         struct xfs_trans        **trans,
3206         struct xfs_inode        *dp,
3207         struct xfs_buf          *bp)
3208 {
3209         struct xfs_attr_leafblock *leaf;
3210         struct xfs_attr3_icleaf_hdr ichdr;
3211         struct xfs_attr_leaf_entry *entry;
3212         struct xfs_attr_leaf_name_remote *name_rmt;
3213         struct xfs_attr_inactive_list *list;
3214         struct xfs_attr_inactive_list *lp;
3215         int                     error;
3216         int                     count;
3217         int                     size;
3218         int                     tmp;
3219         int                     i;
3220
3221         leaf = bp->b_addr;
3222         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
3223
3224         /*
3225          * Count the number of "remote" value extents.
3226          */
3227         count = 0;
3228         entry = xfs_attr3_leaf_entryp(leaf);
3229         for (i = 0; i < ichdr.count; entry++, i++) {
3230                 if (be16_to_cpu(entry->nameidx) &&
3231                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3232                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3233                         if (name_rmt->valueblk)
3234                                 count++;
3235                 }
3236         }
3237
3238         /*
3239          * If there are no "remote" values, we're done.
3240          */
3241         if (count == 0) {
3242                 xfs_trans_brelse(*trans, bp);
3243                 return 0;
3244         }
3245
3246         /*
3247          * Allocate storage for a list of all the "remote" value extents.
3248          */
3249         size = count * sizeof(xfs_attr_inactive_list_t);
3250         list = kmem_alloc(size, KM_SLEEP);
3251
3252         /*
3253          * Identify each of the "remote" value extents.
3254          */
3255         lp = list;
3256         entry = xfs_attr3_leaf_entryp(leaf);
3257         for (i = 0; i < ichdr.count; entry++, i++) {
3258                 if (be16_to_cpu(entry->nameidx) &&
3259                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3260                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3261                         if (name_rmt->valueblk) {
3262                                 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
3263                                 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount,
3264                                                     be32_to_cpu(name_rmt->valuelen));
3265                                 lp++;
3266                         }
3267                 }
3268         }
3269         xfs_trans_brelse(*trans, bp);   /* unlock for trans. in freextent() */
3270
3271         /*
3272          * Invalidate each of the "remote" value extents.
3273          */
3274         error = 0;
3275         for (lp = list, i = 0; i < count; i++, lp++) {
3276                 tmp = xfs_attr3_leaf_freextent(trans, dp,
3277                                 lp->valueblk, lp->valuelen);
3278
3279                 if (error == 0)
3280                         error = tmp;    /* save only the 1st errno */
3281         }
3282
3283         kmem_free(list);
3284         return error;
3285 }
3286
3287 /*
3288  * Look at all the extents for this logical region,
3289  * invalidate any buffers that are incore/in transactions.
3290  */
3291 STATIC int
3292 xfs_attr3_leaf_freextent(
3293         struct xfs_trans        **trans,
3294         struct xfs_inode        *dp,
3295         xfs_dablk_t             blkno,
3296         int                     blkcnt)
3297 {
3298         struct xfs_bmbt_irec    map;
3299         struct xfs_buf          *bp;
3300         xfs_dablk_t             tblkno;
3301         xfs_daddr_t             dblkno;
3302         int                     tblkcnt;
3303         int                     dblkcnt;
3304         int                     nmap;
3305         int                     error;
3306
3307         /*
3308          * Roll through the "value", invalidating the attribute value's
3309          * blocks.
3310          */
3311         tblkno = blkno;
3312         tblkcnt = blkcnt;
3313         while (tblkcnt > 0) {
3314                 /*
3315                  * Try to remember where we decided to put the value.
3316                  */
3317                 nmap = 1;
3318                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
3319                                        &map, &nmap, XFS_BMAPI_ATTRFORK);
3320                 if (error) {
3321                         return(error);
3322                 }
3323                 ASSERT(nmap == 1);
3324                 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3325
3326                 /*
3327                  * If it's a hole, these are already unmapped
3328                  * so there's nothing to invalidate.
3329                  */
3330                 if (map.br_startblock != HOLESTARTBLOCK) {
3331
3332                         dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3333                                                   map.br_startblock);
3334                         dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3335                                                 map.br_blockcount);
3336                         bp = xfs_trans_get_buf(*trans,
3337                                         dp->i_mount->m_ddev_targp,
3338                                         dblkno, dblkcnt, 0);
3339                         if (!bp)
3340                                 return ENOMEM;
3341                         xfs_trans_binval(*trans, bp);
3342                         /*
3343                          * Roll to next transaction.
3344                          */
3345                         error = xfs_trans_roll(trans, dp);
3346                         if (error)
3347                                 return (error);
3348                 }
3349
3350                 tblkno += map.br_blockcount;
3351                 tblkcnt -= map.br_blockcount;
3352         }
3353
3354         return(0);
3355 }