xfs: xfs_attr_shortform_allfit() does not handle attr3 format.
[cascardo/linux.git] / fs / xfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_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         bp = NULL;
694         error = xfs_da_grow_inode(args, &blkno);
695         if (error) {
696                 /*
697                  * If we hit an IO error middle of the transaction inside
698                  * grow_inode(), we may have inconsistent data. Bail out.
699                  */
700                 if (error == EIO)
701                         goto out;
702                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
703                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
704                 goto out;
705         }
706
707         ASSERT(blkno == 0);
708         error = xfs_attr3_leaf_create(args, blkno, &bp);
709         if (error) {
710                 error = xfs_da_shrink_inode(args, 0, bp);
711                 bp = NULL;
712                 if (error)
713                         goto out;
714                 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
715                 memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
716                 goto out;
717         }
718
719         memset((char *)&nargs, 0, sizeof(nargs));
720         nargs.dp = dp;
721         nargs.firstblock = args->firstblock;
722         nargs.flist = args->flist;
723         nargs.total = args->total;
724         nargs.whichfork = XFS_ATTR_FORK;
725         nargs.trans = args->trans;
726         nargs.op_flags = XFS_DA_OP_OKNOENT;
727
728         sfe = &sf->list[0];
729         for (i = 0; i < sf->hdr.count; i++) {
730                 nargs.name = sfe->nameval;
731                 nargs.namelen = sfe->namelen;
732                 nargs.value = &sfe->nameval[nargs.namelen];
733                 nargs.valuelen = sfe->valuelen;
734                 nargs.hashval = xfs_da_hashname(sfe->nameval,
735                                                 sfe->namelen);
736                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
737                 error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
738                 ASSERT(error == ENOATTR);
739                 error = xfs_attr3_leaf_add(bp, &nargs);
740                 ASSERT(error != ENOSPC);
741                 if (error)
742                         goto out;
743                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
744         }
745         error = 0;
746
747 out:
748         kmem_free(tmpbuffer);
749         return(error);
750 }
751
752 STATIC int
753 xfs_attr_shortform_compare(const void *a, const void *b)
754 {
755         xfs_attr_sf_sort_t *sa, *sb;
756
757         sa = (xfs_attr_sf_sort_t *)a;
758         sb = (xfs_attr_sf_sort_t *)b;
759         if (sa->hash < sb->hash) {
760                 return(-1);
761         } else if (sa->hash > sb->hash) {
762                 return(1);
763         } else {
764                 return(sa->entno - sb->entno);
765         }
766 }
767
768
769 #define XFS_ISRESET_CURSOR(cursor) \
770         (!((cursor)->initted) && !((cursor)->hashval) && \
771          !((cursor)->blkno) && !((cursor)->offset))
772 /*
773  * Copy out entries of shortform attribute lists for attr_list().
774  * Shortform attribute lists are not stored in hashval sorted order.
775  * If the output buffer is not large enough to hold them all, then we
776  * we have to calculate each entries' hashvalue and sort them before
777  * we can begin returning them to the user.
778  */
779 /*ARGSUSED*/
780 int
781 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
782 {
783         attrlist_cursor_kern_t *cursor;
784         xfs_attr_sf_sort_t *sbuf, *sbp;
785         xfs_attr_shortform_t *sf;
786         xfs_attr_sf_entry_t *sfe;
787         xfs_inode_t *dp;
788         int sbsize, nsbuf, count, i;
789         int error;
790
791         ASSERT(context != NULL);
792         dp = context->dp;
793         ASSERT(dp != NULL);
794         ASSERT(dp->i_afp != NULL);
795         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
796         ASSERT(sf != NULL);
797         if (!sf->hdr.count)
798                 return(0);
799         cursor = context->cursor;
800         ASSERT(cursor != NULL);
801
802         trace_xfs_attr_list_sf(context);
803
804         /*
805          * If the buffer is large enough and the cursor is at the start,
806          * do not bother with sorting since we will return everything in
807          * one buffer and another call using the cursor won't need to be
808          * made.
809          * Note the generous fudge factor of 16 overhead bytes per entry.
810          * If bufsize is zero then put_listent must be a search function
811          * and can just scan through what we have.
812          */
813         if (context->bufsize == 0 ||
814             (XFS_ISRESET_CURSOR(cursor) &&
815              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
816                 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
817                         error = context->put_listent(context,
818                                            sfe->flags,
819                                            sfe->nameval,
820                                            (int)sfe->namelen,
821                                            (int)sfe->valuelen,
822                                            &sfe->nameval[sfe->namelen]);
823
824                         /*
825                          * Either search callback finished early or
826                          * didn't fit it all in the buffer after all.
827                          */
828                         if (context->seen_enough)
829                                 break;
830
831                         if (error)
832                                 return error;
833                         sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
834                 }
835                 trace_xfs_attr_list_sf_all(context);
836                 return(0);
837         }
838
839         /* do no more for a search callback */
840         if (context->bufsize == 0)
841                 return 0;
842
843         /*
844          * It didn't all fit, so we have to sort everything on hashval.
845          */
846         sbsize = sf->hdr.count * sizeof(*sbuf);
847         sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
848
849         /*
850          * Scan the attribute list for the rest of the entries, storing
851          * the relevant info from only those that match into a buffer.
852          */
853         nsbuf = 0;
854         for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
855                 if (unlikely(
856                     ((char *)sfe < (char *)sf) ||
857                     ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
858                         XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
859                                              XFS_ERRLEVEL_LOW,
860                                              context->dp->i_mount, sfe);
861                         kmem_free(sbuf);
862                         return XFS_ERROR(EFSCORRUPTED);
863                 }
864
865                 sbp->entno = i;
866                 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
867                 sbp->name = sfe->nameval;
868                 sbp->namelen = sfe->namelen;
869                 /* These are bytes, and both on-disk, don't endian-flip */
870                 sbp->valuelen = sfe->valuelen;
871                 sbp->flags = sfe->flags;
872                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
873                 sbp++;
874                 nsbuf++;
875         }
876
877         /*
878          * Sort the entries on hash then entno.
879          */
880         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
881
882         /*
883          * Re-find our place IN THE SORTED LIST.
884          */
885         count = 0;
886         cursor->initted = 1;
887         cursor->blkno = 0;
888         for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
889                 if (sbp->hash == cursor->hashval) {
890                         if (cursor->offset == count) {
891                                 break;
892                         }
893                         count++;
894                 } else if (sbp->hash > cursor->hashval) {
895                         break;
896                 }
897         }
898         if (i == nsbuf) {
899                 kmem_free(sbuf);
900                 return(0);
901         }
902
903         /*
904          * Loop putting entries into the user buffer.
905          */
906         for ( ; i < nsbuf; i++, sbp++) {
907                 if (cursor->hashval != sbp->hash) {
908                         cursor->hashval = sbp->hash;
909                         cursor->offset = 0;
910                 }
911                 error = context->put_listent(context,
912                                         sbp->flags,
913                                         sbp->name,
914                                         sbp->namelen,
915                                         sbp->valuelen,
916                                         &sbp->name[sbp->namelen]);
917                 if (error)
918                         return error;
919                 if (context->seen_enough)
920                         break;
921                 cursor->offset++;
922         }
923
924         kmem_free(sbuf);
925         return(0);
926 }
927
928 /*
929  * Check a leaf attribute block to see if all the entries would fit into
930  * a shortform attribute list.
931  */
932 int
933 xfs_attr_shortform_allfit(
934         struct xfs_buf          *bp,
935         struct xfs_inode        *dp)
936 {
937         struct xfs_attr_leafblock *leaf;
938         struct xfs_attr_leaf_entry *entry;
939         xfs_attr_leaf_name_local_t *name_loc;
940         struct xfs_attr3_icleaf_hdr leafhdr;
941         int                     bytes;
942         int                     i;
943
944         leaf = bp->b_addr;
945         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
946         entry = xfs_attr3_leaf_entryp(leaf);
947
948         bytes = sizeof(struct xfs_attr_sf_hdr);
949         for (i = 0; i < leafhdr.count; entry++, i++) {
950                 if (entry->flags & XFS_ATTR_INCOMPLETE)
951                         continue;               /* don't copy partial entries */
952                 if (!(entry->flags & XFS_ATTR_LOCAL))
953                         return(0);
954                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
955                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
956                         return(0);
957                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
958                         return(0);
959                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
960                                 + name_loc->namelen
961                                 + be16_to_cpu(name_loc->valuelen);
962         }
963         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
964             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
965             (bytes == sizeof(struct xfs_attr_sf_hdr)))
966                 return -1;
967         return xfs_attr_shortform_bytesfit(dp, bytes);
968 }
969
970 /*
971  * Convert a leaf attribute list to shortform attribute list
972  */
973 int
974 xfs_attr3_leaf_to_shortform(
975         struct xfs_buf          *bp,
976         struct xfs_da_args      *args,
977         int                     forkoff)
978 {
979         struct xfs_attr_leafblock *leaf;
980         struct xfs_attr3_icleaf_hdr ichdr;
981         struct xfs_attr_leaf_entry *entry;
982         struct xfs_attr_leaf_name_local *name_loc;
983         struct xfs_da_args      nargs;
984         struct xfs_inode        *dp = args->dp;
985         char                    *tmpbuffer;
986         int                     error;
987         int                     i;
988
989         trace_xfs_attr_leaf_to_sf(args);
990
991         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
992         if (!tmpbuffer)
993                 return ENOMEM;
994
995         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
996
997         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
998         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
999         entry = xfs_attr3_leaf_entryp(leaf);
1000
1001         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1002         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
1003
1004         /*
1005          * Clean out the prior contents of the attribute list.
1006          */
1007         error = xfs_da_shrink_inode(args, 0, bp);
1008         if (error)
1009                 goto out;
1010
1011         if (forkoff == -1) {
1012                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1013                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1014                 xfs_attr_fork_reset(dp, args->trans);
1015                 goto out;
1016         }
1017
1018         xfs_attr_shortform_create(args);
1019
1020         /*
1021          * Copy the attributes
1022          */
1023         memset((char *)&nargs, 0, sizeof(nargs));
1024         nargs.dp = dp;
1025         nargs.firstblock = args->firstblock;
1026         nargs.flist = args->flist;
1027         nargs.total = args->total;
1028         nargs.whichfork = XFS_ATTR_FORK;
1029         nargs.trans = args->trans;
1030         nargs.op_flags = XFS_DA_OP_OKNOENT;
1031
1032         for (i = 0; i < ichdr.count; entry++, i++) {
1033                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1034                         continue;       /* don't copy partial entries */
1035                 if (!entry->nameidx)
1036                         continue;
1037                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1038                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1039                 nargs.name = name_loc->nameval;
1040                 nargs.namelen = name_loc->namelen;
1041                 nargs.value = &name_loc->nameval[nargs.namelen];
1042                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1043                 nargs.hashval = be32_to_cpu(entry->hashval);
1044                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1045                 xfs_attr_shortform_add(&nargs, forkoff);
1046         }
1047         error = 0;
1048
1049 out:
1050         kmem_free(tmpbuffer);
1051         return error;
1052 }
1053
1054 /*
1055  * Convert from using a single leaf to a root node and a leaf.
1056  */
1057 int
1058 xfs_attr3_leaf_to_node(
1059         struct xfs_da_args      *args)
1060 {
1061         struct xfs_attr_leafblock *leaf;
1062         struct xfs_attr3_icleaf_hdr icleafhdr;
1063         struct xfs_attr_leaf_entry *entries;
1064         struct xfs_da_node_entry *btree;
1065         struct xfs_da3_icnode_hdr icnodehdr;
1066         struct xfs_da_intnode   *node;
1067         struct xfs_inode        *dp = args->dp;
1068         struct xfs_mount        *mp = dp->i_mount;
1069         struct xfs_buf          *bp1 = NULL;
1070         struct xfs_buf          *bp2 = NULL;
1071         xfs_dablk_t             blkno;
1072         int                     error;
1073
1074         trace_xfs_attr_leaf_to_node(args);
1075
1076         error = xfs_da_grow_inode(args, &blkno);
1077         if (error)
1078                 goto out;
1079         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1080         if (error)
1081                 goto out;
1082
1083         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1084         if (error)
1085                 goto out;
1086
1087         /* copy leaf to new buffer, update identifiers */
1088         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1089         bp2->b_ops = bp1->b_ops;
1090         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
1091         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1092                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1093                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1094         }
1095         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
1096
1097         /*
1098          * Set up the new root node.
1099          */
1100         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1101         if (error)
1102                 goto out;
1103         node = bp1->b_addr;
1104         xfs_da3_node_hdr_from_disk(&icnodehdr, node);
1105         btree = xfs_da3_node_tree_p(node);
1106
1107         leaf = bp2->b_addr;
1108         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
1109         entries = xfs_attr3_leaf_entryp(leaf);
1110
1111         /* both on-disk, don't endian-flip twice */
1112         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1113         btree[0].before = cpu_to_be32(blkno);
1114         icnodehdr.count = 1;
1115         xfs_da3_node_hdr_to_disk(node, &icnodehdr);
1116         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
1117         error = 0;
1118 out:
1119         return error;
1120 }
1121
1122
1123 /*========================================================================
1124  * Routines used for growing the Btree.
1125  *========================================================================*/
1126
1127 /*
1128  * Create the initial contents of a leaf attribute list
1129  * or a leaf in a node attribute list.
1130  */
1131 STATIC int
1132 xfs_attr3_leaf_create(
1133         struct xfs_da_args      *args,
1134         xfs_dablk_t             blkno,
1135         struct xfs_buf          **bpp)
1136 {
1137         struct xfs_attr_leafblock *leaf;
1138         struct xfs_attr3_icleaf_hdr ichdr;
1139         struct xfs_inode        *dp = args->dp;
1140         struct xfs_mount        *mp = dp->i_mount;
1141         struct xfs_buf          *bp;
1142         int                     error;
1143
1144         trace_xfs_attr_leaf_create(args);
1145
1146         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1147                                             XFS_ATTR_FORK);
1148         if (error)
1149                 return error;
1150         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1151         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1152         leaf = bp->b_addr;
1153         memset(leaf, 0, XFS_LBSIZE(mp));
1154
1155         memset(&ichdr, 0, sizeof(ichdr));
1156         ichdr.firstused = XFS_LBSIZE(mp);
1157
1158         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1159                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1160
1161                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1162
1163                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1164                 hdr3->owner = cpu_to_be64(dp->i_ino);
1165                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
1166
1167                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1168         } else {
1169                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1170                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1171         }
1172         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1173
1174         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1175         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
1176
1177         *bpp = bp;
1178         return 0;
1179 }
1180
1181 /*
1182  * Split the leaf node, rebalance, then add the new entry.
1183  */
1184 int
1185 xfs_attr3_leaf_split(
1186         struct xfs_da_state     *state,
1187         struct xfs_da_state_blk *oldblk,
1188         struct xfs_da_state_blk *newblk)
1189 {
1190         xfs_dablk_t blkno;
1191         int error;
1192
1193         trace_xfs_attr_leaf_split(state->args);
1194
1195         /*
1196          * Allocate space for a new leaf node.
1197          */
1198         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1199         error = xfs_da_grow_inode(state->args, &blkno);
1200         if (error)
1201                 return(error);
1202         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1203         if (error)
1204                 return(error);
1205         newblk->blkno = blkno;
1206         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1207
1208         /*
1209          * Rebalance the entries across the two leaves.
1210          * NOTE: rebalance() currently depends on the 2nd block being empty.
1211          */
1212         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1213         error = xfs_da3_blk_link(state, oldblk, newblk);
1214         if (error)
1215                 return(error);
1216
1217         /*
1218          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1219          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1220          * "new" attrs info.  Will need the "old" info to remove it later.
1221          *
1222          * Insert the "new" entry in the correct block.
1223          */
1224         if (state->inleaf) {
1225                 trace_xfs_attr_leaf_add_old(state->args);
1226                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1227         } else {
1228                 trace_xfs_attr_leaf_add_new(state->args);
1229                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1230         }
1231
1232         /*
1233          * Update last hashval in each block since we added the name.
1234          */
1235         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1236         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1237         return(error);
1238 }
1239
1240 /*
1241  * Add a name to the leaf attribute list structure.
1242  */
1243 int
1244 xfs_attr3_leaf_add(
1245         struct xfs_buf          *bp,
1246         struct xfs_da_args      *args)
1247 {
1248         struct xfs_attr_leafblock *leaf;
1249         struct xfs_attr3_icleaf_hdr ichdr;
1250         int                     tablesize;
1251         int                     entsize;
1252         int                     sum;
1253         int                     tmp;
1254         int                     i;
1255
1256         trace_xfs_attr_leaf_add(args);
1257
1258         leaf = bp->b_addr;
1259         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1260         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1261         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1262                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1263
1264         /*
1265          * Search through freemap for first-fit on new name length.
1266          * (may need to figure in size of entry struct too)
1267          */
1268         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1269                                         + xfs_attr3_leaf_hdr_size(leaf);
1270         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1271                 if (tablesize > ichdr.firstused) {
1272                         sum += ichdr.freemap[i].size;
1273                         continue;
1274                 }
1275                 if (!ichdr.freemap[i].size)
1276                         continue;       /* no space in this map */
1277                 tmp = entsize;
1278                 if (ichdr.freemap[i].base < ichdr.firstused)
1279                         tmp += sizeof(xfs_attr_leaf_entry_t);
1280                 if (ichdr.freemap[i].size >= tmp) {
1281                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1282                         goto out_log_hdr;
1283                 }
1284                 sum += ichdr.freemap[i].size;
1285         }
1286
1287         /*
1288          * If there are no holes in the address space of the block,
1289          * and we don't have enough freespace, then compaction will do us
1290          * no good and we should just give up.
1291          */
1292         if (!ichdr.holes && sum < entsize)
1293                 return XFS_ERROR(ENOSPC);
1294
1295         /*
1296          * Compact the entries to coalesce free space.
1297          * This may change the hdr->count via dropping INCOMPLETE entries.
1298          */
1299         xfs_attr3_leaf_compact(args, &ichdr, bp);
1300
1301         /*
1302          * After compaction, the block is guaranteed to have only one
1303          * free region, in freemap[0].  If it is not big enough, give up.
1304          */
1305         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1306                 tmp = ENOSPC;
1307                 goto out_log_hdr;
1308         }
1309
1310         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1311
1312 out_log_hdr:
1313         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1314         xfs_trans_log_buf(args->trans, bp,
1315                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1316                                 xfs_attr3_leaf_hdr_size(leaf)));
1317         return tmp;
1318 }
1319
1320 /*
1321  * Add a name to a leaf attribute list structure.
1322  */
1323 STATIC int
1324 xfs_attr3_leaf_add_work(
1325         struct xfs_buf          *bp,
1326         struct xfs_attr3_icleaf_hdr *ichdr,
1327         struct xfs_da_args      *args,
1328         int                     mapindex)
1329 {
1330         struct xfs_attr_leafblock *leaf;
1331         struct xfs_attr_leaf_entry *entry;
1332         struct xfs_attr_leaf_name_local *name_loc;
1333         struct xfs_attr_leaf_name_remote *name_rmt;
1334         struct xfs_mount        *mp;
1335         int                     tmp;
1336         int                     i;
1337
1338         trace_xfs_attr_leaf_add_work(args);
1339
1340         leaf = bp->b_addr;
1341         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1342         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1343
1344         /*
1345          * Force open some space in the entry array and fill it in.
1346          */
1347         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1348         if (args->index < ichdr->count) {
1349                 tmp  = ichdr->count - args->index;
1350                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1351                 memmove(entry + 1, entry, tmp);
1352                 xfs_trans_log_buf(args->trans, bp,
1353                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1354         }
1355         ichdr->count++;
1356
1357         /*
1358          * Allocate space for the new string (at the end of the run).
1359          */
1360         mp = args->trans->t_mountp;
1361         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1362         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1363         ASSERT(ichdr->freemap[mapindex].size >=
1364                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1365                                          mp->m_sb.sb_blocksize, NULL));
1366         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1367         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1368
1369         ichdr->freemap[mapindex].size -=
1370                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1371                                                  mp->m_sb.sb_blocksize, &tmp);
1372
1373         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1374                                      ichdr->freemap[mapindex].size);
1375         entry->hashval = cpu_to_be32(args->hashval);
1376         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1377         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1378         if (args->op_flags & XFS_DA_OP_RENAME) {
1379                 entry->flags |= XFS_ATTR_INCOMPLETE;
1380                 if ((args->blkno2 == args->blkno) &&
1381                     (args->index2 <= args->index)) {
1382                         args->index2++;
1383                 }
1384         }
1385         xfs_trans_log_buf(args->trans, bp,
1386                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1387         ASSERT((args->index == 0) ||
1388                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1389         ASSERT((args->index == ichdr->count - 1) ||
1390                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1391
1392         /*
1393          * For "remote" attribute values, simply note that we need to
1394          * allocate space for the "remote" value.  We can't actually
1395          * allocate the extents in this transaction, and we can't decide
1396          * which blocks they should be as we might allocate more blocks
1397          * as part of this transaction (a split operation for example).
1398          */
1399         if (entry->flags & XFS_ATTR_LOCAL) {
1400                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1401                 name_loc->namelen = args->namelen;
1402                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1403                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1404                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1405                                    be16_to_cpu(name_loc->valuelen));
1406         } else {
1407                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1408                 name_rmt->namelen = args->namelen;
1409                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1410                 entry->flags |= XFS_ATTR_INCOMPLETE;
1411                 /* just in case */
1412                 name_rmt->valuelen = 0;
1413                 name_rmt->valueblk = 0;
1414                 args->rmtblkno = 1;
1415                 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1416         }
1417         xfs_trans_log_buf(args->trans, bp,
1418              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1419                                    xfs_attr_leaf_entsize(leaf, args->index)));
1420
1421         /*
1422          * Update the control info for this leaf node
1423          */
1424         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1425                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1426
1427         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1428                                         + xfs_attr3_leaf_hdr_size(leaf));
1429         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1430                                         + xfs_attr3_leaf_hdr_size(leaf);
1431
1432         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1433                 if (ichdr->freemap[i].base == tmp) {
1434                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1435                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1436                 }
1437         }
1438         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1439         return 0;
1440 }
1441
1442 /*
1443  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1444  */
1445 STATIC void
1446 xfs_attr3_leaf_compact(
1447         struct xfs_da_args      *args,
1448         struct xfs_attr3_icleaf_hdr *ichdr_d,
1449         struct xfs_buf          *bp)
1450 {
1451         xfs_attr_leafblock_t    *leaf_s, *leaf_d;
1452         struct xfs_attr3_icleaf_hdr ichdr_s;
1453         struct xfs_trans        *trans = args->trans;
1454         struct xfs_mount        *mp = trans->t_mountp;
1455         char                    *tmpbuffer;
1456
1457         trace_xfs_attr_leaf_compact(args);
1458
1459         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1460         ASSERT(tmpbuffer != NULL);
1461         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1462         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1463
1464         /*
1465          * Copy basic information
1466          */
1467         leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1468         leaf_d = bp->b_addr;
1469         ichdr_s = *ichdr_d;     /* struct copy */
1470         ichdr_d->firstused = XFS_LBSIZE(mp);
1471         ichdr_d->usedbytes = 0;
1472         ichdr_d->count = 0;
1473         ichdr_d->holes = 0;
1474         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_s);
1475         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
1476
1477         /*
1478          * Copy all entry's in the same (sorted) order,
1479          * but allocate name/value pairs packed and in sequence.
1480          */
1481         xfs_attr3_leaf_moveents(leaf_s, &ichdr_s, 0, leaf_d, ichdr_d, 0,
1482                                 ichdr_s.count, mp);
1483         /*
1484          * this logs the entire buffer, but the caller must write the header
1485          * back to the buffer when it is finished modifying it.
1486          */
1487         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1488
1489         kmem_free(tmpbuffer);
1490 }
1491
1492 /*
1493  * Compare two leaf blocks "order".
1494  * Return 0 unless leaf2 should go before leaf1.
1495  */
1496 static int
1497 xfs_attr3_leaf_order(
1498         struct xfs_buf  *leaf1_bp,
1499         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1500         struct xfs_buf  *leaf2_bp,
1501         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1502 {
1503         struct xfs_attr_leaf_entry *entries1;
1504         struct xfs_attr_leaf_entry *entries2;
1505
1506         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1507         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1508         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1509             ((be32_to_cpu(entries2[0].hashval) <
1510               be32_to_cpu(entries1[0].hashval)) ||
1511              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1512               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1513                 return 1;
1514         }
1515         return 0;
1516 }
1517
1518 int
1519 xfs_attr_leaf_order(
1520         struct xfs_buf  *leaf1_bp,
1521         struct xfs_buf  *leaf2_bp)
1522 {
1523         struct xfs_attr3_icleaf_hdr ichdr1;
1524         struct xfs_attr3_icleaf_hdr ichdr2;
1525
1526         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1527         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1528         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1529 }
1530
1531 /*
1532  * Redistribute the attribute list entries between two leaf nodes,
1533  * taking into account the size of the new entry.
1534  *
1535  * NOTE: if new block is empty, then it will get the upper half of the
1536  * old block.  At present, all (one) callers pass in an empty second block.
1537  *
1538  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1539  * to match what it is doing in splitting the attribute leaf block.  Those
1540  * values are used in "atomic rename" operations on attributes.  Note that
1541  * the "new" and "old" values can end up in different blocks.
1542  */
1543 STATIC void
1544 xfs_attr3_leaf_rebalance(
1545         struct xfs_da_state     *state,
1546         struct xfs_da_state_blk *blk1,
1547         struct xfs_da_state_blk *blk2)
1548 {
1549         struct xfs_da_args      *args;
1550         struct xfs_attr_leafblock *leaf1;
1551         struct xfs_attr_leafblock *leaf2;
1552         struct xfs_attr3_icleaf_hdr ichdr1;
1553         struct xfs_attr3_icleaf_hdr ichdr2;
1554         struct xfs_attr_leaf_entry *entries1;
1555         struct xfs_attr_leaf_entry *entries2;
1556         int                     count;
1557         int                     totallen;
1558         int                     max;
1559         int                     space;
1560         int                     swap;
1561
1562         /*
1563          * Set up environment.
1564          */
1565         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1566         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1567         leaf1 = blk1->bp->b_addr;
1568         leaf2 = blk2->bp->b_addr;
1569         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1570         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1571         ASSERT(ichdr2.count == 0);
1572         args = state->args;
1573
1574         trace_xfs_attr_leaf_rebalance(args);
1575
1576         /*
1577          * Check ordering of blocks, reverse if it makes things simpler.
1578          *
1579          * NOTE: Given that all (current) callers pass in an empty
1580          * second block, this code should never set "swap".
1581          */
1582         swap = 0;
1583         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1584                 struct xfs_da_state_blk *tmp_blk;
1585                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1586
1587                 tmp_blk = blk1;
1588                 blk1 = blk2;
1589                 blk2 = tmp_blk;
1590
1591                 /* struct copies to swap them rather than reconverting */
1592                 tmp_ichdr = ichdr1;
1593                 ichdr1 = ichdr2;
1594                 ichdr2 = tmp_ichdr;
1595
1596                 leaf1 = blk1->bp->b_addr;
1597                 leaf2 = blk2->bp->b_addr;
1598                 swap = 1;
1599         }
1600
1601         /*
1602          * Examine entries until we reduce the absolute difference in
1603          * byte usage between the two blocks to a minimum.  Then get
1604          * the direction to copy and the number of elements to move.
1605          *
1606          * "inleaf" is true if the new entry should be inserted into blk1.
1607          * If "swap" is also true, then reverse the sense of "inleaf".
1608          */
1609         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1610                                                       blk2, &ichdr2,
1611                                                       &count, &totallen);
1612         if (swap)
1613                 state->inleaf = !state->inleaf;
1614
1615         /*
1616          * Move any entries required from leaf to leaf:
1617          */
1618         if (count < ichdr1.count) {
1619                 /*
1620                  * Figure the total bytes to be added to the destination leaf.
1621                  */
1622                 /* number entries being moved */
1623                 count = ichdr1.count - count;
1624                 space  = ichdr1.usedbytes - totallen;
1625                 space += count * sizeof(xfs_attr_leaf_entry_t);
1626
1627                 /*
1628                  * leaf2 is the destination, compact it if it looks tight.
1629                  */
1630                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1631                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1632                 if (space > max)
1633                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1634
1635                 /*
1636                  * Move high entries from leaf1 to low end of leaf2.
1637                  */
1638                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1639                                 leaf2, &ichdr2, 0, count, state->mp);
1640
1641         } else if (count > ichdr1.count) {
1642                 /*
1643                  * I assert that since all callers pass in an empty
1644                  * second buffer, this code should never execute.
1645                  */
1646                 ASSERT(0);
1647
1648                 /*
1649                  * Figure the total bytes to be added to the destination leaf.
1650                  */
1651                 /* number entries being moved */
1652                 count -= ichdr1.count;
1653                 space  = totallen - ichdr1.usedbytes;
1654                 space += count * sizeof(xfs_attr_leaf_entry_t);
1655
1656                 /*
1657                  * leaf1 is the destination, compact it if it looks tight.
1658                  */
1659                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1660                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1661                 if (space > max)
1662                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1663
1664                 /*
1665                  * Move low entries from leaf2 to high end of leaf1.
1666                  */
1667                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1668                                         ichdr1.count, count, state->mp);
1669         }
1670
1671         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1672         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1673         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1674         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1675
1676         /*
1677          * Copy out last hashval in each block for B-tree code.
1678          */
1679         entries1 = xfs_attr3_leaf_entryp(leaf1);
1680         entries2 = xfs_attr3_leaf_entryp(leaf2);
1681         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1682         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1683
1684         /*
1685          * Adjust the expected index for insertion.
1686          * NOTE: this code depends on the (current) situation that the
1687          * second block was originally empty.
1688          *
1689          * If the insertion point moved to the 2nd block, we must adjust
1690          * the index.  We must also track the entry just following the
1691          * new entry for use in an "atomic rename" operation, that entry
1692          * is always the "old" entry and the "new" entry is what we are
1693          * inserting.  The index/blkno fields refer to the "old" entry,
1694          * while the index2/blkno2 fields refer to the "new" entry.
1695          */
1696         if (blk1->index > ichdr1.count) {
1697                 ASSERT(state->inleaf == 0);
1698                 blk2->index = blk1->index - ichdr1.count;
1699                 args->index = args->index2 = blk2->index;
1700                 args->blkno = args->blkno2 = blk2->blkno;
1701         } else if (blk1->index == ichdr1.count) {
1702                 if (state->inleaf) {
1703                         args->index = blk1->index;
1704                         args->blkno = blk1->blkno;
1705                         args->index2 = 0;
1706                         args->blkno2 = blk2->blkno;
1707                 } else {
1708                         /*
1709                          * On a double leaf split, the original attr location
1710                          * is already stored in blkno2/index2, so don't
1711                          * overwrite it overwise we corrupt the tree.
1712                          */
1713                         blk2->index = blk1->index - ichdr1.count;
1714                         args->index = blk2->index;
1715                         args->blkno = blk2->blkno;
1716                         if (!state->extravalid) {
1717                                 /*
1718                                  * set the new attr location to match the old
1719                                  * one and let the higher level split code
1720                                  * decide where in the leaf to place it.
1721                                  */
1722                                 args->index2 = blk2->index;
1723                                 args->blkno2 = blk2->blkno;
1724                         }
1725                 }
1726         } else {
1727                 ASSERT(state->inleaf == 1);
1728                 args->index = args->index2 = blk1->index;
1729                 args->blkno = args->blkno2 = blk1->blkno;
1730         }
1731 }
1732
1733 /*
1734  * Examine entries until we reduce the absolute difference in
1735  * byte usage between the two blocks to a minimum.
1736  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1737  * GROT: there will always be enough room in either block for a new entry.
1738  * GROT: Do a double-split for this case?
1739  */
1740 STATIC int
1741 xfs_attr3_leaf_figure_balance(
1742         struct xfs_da_state             *state,
1743         struct xfs_da_state_blk         *blk1,
1744         struct xfs_attr3_icleaf_hdr     *ichdr1,
1745         struct xfs_da_state_blk         *blk2,
1746         struct xfs_attr3_icleaf_hdr     *ichdr2,
1747         int                             *countarg,
1748         int                             *usedbytesarg)
1749 {
1750         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1751         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1752         struct xfs_attr_leaf_entry      *entry;
1753         int                             count;
1754         int                             max;
1755         int                             index;
1756         int                             totallen = 0;
1757         int                             half;
1758         int                             lastdelta;
1759         int                             foundit = 0;
1760         int                             tmp;
1761
1762         /*
1763          * Examine entries until we reduce the absolute difference in
1764          * byte usage between the two blocks to a minimum.
1765          */
1766         max = ichdr1->count + ichdr2->count;
1767         half = (max + 1) * sizeof(*entry);
1768         half += ichdr1->usedbytes + ichdr2->usedbytes +
1769                         xfs_attr_leaf_newentsize(state->args->namelen,
1770                                                  state->args->valuelen,
1771                                                  state->blocksize, NULL);
1772         half /= 2;
1773         lastdelta = state->blocksize;
1774         entry = xfs_attr3_leaf_entryp(leaf1);
1775         for (count = index = 0; count < max; entry++, index++, count++) {
1776
1777 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1778                 /*
1779                  * The new entry is in the first block, account for it.
1780                  */
1781                 if (count == blk1->index) {
1782                         tmp = totallen + sizeof(*entry) +
1783                                 xfs_attr_leaf_newentsize(
1784                                                 state->args->namelen,
1785                                                 state->args->valuelen,
1786                                                 state->blocksize, NULL);
1787                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1788                                 break;
1789                         lastdelta = XFS_ATTR_ABS(half - tmp);
1790                         totallen = tmp;
1791                         foundit = 1;
1792                 }
1793
1794                 /*
1795                  * Wrap around into the second block if necessary.
1796                  */
1797                 if (count == ichdr1->count) {
1798                         leaf1 = leaf2;
1799                         entry = xfs_attr3_leaf_entryp(leaf1);
1800                         index = 0;
1801                 }
1802
1803                 /*
1804                  * Figure out if next leaf entry would be too much.
1805                  */
1806                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1807                                                                         index);
1808                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1809                         break;
1810                 lastdelta = XFS_ATTR_ABS(half - tmp);
1811                 totallen = tmp;
1812 #undef XFS_ATTR_ABS
1813         }
1814
1815         /*
1816          * Calculate the number of usedbytes that will end up in lower block.
1817          * If new entry not in lower block, fix up the count.
1818          */
1819         totallen -= count * sizeof(*entry);
1820         if (foundit) {
1821                 totallen -= sizeof(*entry) +
1822                                 xfs_attr_leaf_newentsize(
1823                                                 state->args->namelen,
1824                                                 state->args->valuelen,
1825                                                 state->blocksize, NULL);
1826         }
1827
1828         *countarg = count;
1829         *usedbytesarg = totallen;
1830         return foundit;
1831 }
1832
1833 /*========================================================================
1834  * Routines used for shrinking the Btree.
1835  *========================================================================*/
1836
1837 /*
1838  * Check a leaf block and its neighbors to see if the block should be
1839  * collapsed into one or the other neighbor.  Always keep the block
1840  * with the smaller block number.
1841  * If the current block is over 50% full, don't try to join it, return 0.
1842  * If the block is empty, fill in the state structure and return 2.
1843  * If it can be collapsed, fill in the state structure and return 1.
1844  * If nothing can be done, return 0.
1845  *
1846  * GROT: allow for INCOMPLETE entries in calculation.
1847  */
1848 int
1849 xfs_attr3_leaf_toosmall(
1850         struct xfs_da_state     *state,
1851         int                     *action)
1852 {
1853         struct xfs_attr_leafblock *leaf;
1854         struct xfs_da_state_blk *blk;
1855         struct xfs_attr3_icleaf_hdr ichdr;
1856         struct xfs_buf          *bp;
1857         xfs_dablk_t             blkno;
1858         int                     bytes;
1859         int                     forward;
1860         int                     error;
1861         int                     retval;
1862         int                     i;
1863
1864         trace_xfs_attr_leaf_toosmall(state->args);
1865
1866         /*
1867          * Check for the degenerate case of the block being over 50% full.
1868          * If so, it's not worth even looking to see if we might be able
1869          * to coalesce with a sibling.
1870          */
1871         blk = &state->path.blk[ state->path.active-1 ];
1872         leaf = blk->bp->b_addr;
1873         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1874         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1875                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1876                 ichdr.usedbytes;
1877         if (bytes > (state->blocksize >> 1)) {
1878                 *action = 0;    /* blk over 50%, don't try to join */
1879                 return(0);
1880         }
1881
1882         /*
1883          * Check for the degenerate case of the block being empty.
1884          * If the block is empty, we'll simply delete it, no need to
1885          * coalesce it with a sibling block.  We choose (arbitrarily)
1886          * to merge with the forward block unless it is NULL.
1887          */
1888         if (ichdr.count == 0) {
1889                 /*
1890                  * Make altpath point to the block we want to keep and
1891                  * path point to the block we want to drop (this one).
1892                  */
1893                 forward = (ichdr.forw != 0);
1894                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1895                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1896                                                  0, &retval);
1897                 if (error)
1898                         return(error);
1899                 if (retval) {
1900                         *action = 0;
1901                 } else {
1902                         *action = 2;
1903                 }
1904                 return 0;
1905         }
1906
1907         /*
1908          * Examine each sibling block to see if we can coalesce with
1909          * at least 25% free space to spare.  We need to figure out
1910          * whether to merge with the forward or the backward block.
1911          * We prefer coalescing with the lower numbered sibling so as
1912          * to shrink an attribute list over time.
1913          */
1914         /* start with smaller blk num */
1915         forward = ichdr.forw < ichdr.back;
1916         for (i = 0; i < 2; forward = !forward, i++) {
1917                 struct xfs_attr3_icleaf_hdr ichdr2;
1918                 if (forward)
1919                         blkno = ichdr.forw;
1920                 else
1921                         blkno = ichdr.back;
1922                 if (blkno == 0)
1923                         continue;
1924                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1925                                         blkno, -1, &bp);
1926                 if (error)
1927                         return(error);
1928
1929                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1930
1931                 bytes = state->blocksize - (state->blocksize >> 2) -
1932                         ichdr.usedbytes - ichdr2.usedbytes -
1933                         ((ichdr.count + ichdr2.count) *
1934                                         sizeof(xfs_attr_leaf_entry_t)) -
1935                         xfs_attr3_leaf_hdr_size(leaf);
1936
1937                 xfs_trans_brelse(state->args->trans, bp);
1938                 if (bytes >= 0)
1939                         break;  /* fits with at least 25% to spare */
1940         }
1941         if (i >= 2) {
1942                 *action = 0;
1943                 return(0);
1944         }
1945
1946         /*
1947          * Make altpath point to the block we want to keep (the lower
1948          * numbered block) and path point to the block we want to drop.
1949          */
1950         memcpy(&state->altpath, &state->path, sizeof(state->path));
1951         if (blkno < blk->blkno) {
1952                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1953                                                  0, &retval);
1954         } else {
1955                 error = xfs_da3_path_shift(state, &state->path, forward,
1956                                                  0, &retval);
1957         }
1958         if (error)
1959                 return(error);
1960         if (retval) {
1961                 *action = 0;
1962         } else {
1963                 *action = 1;
1964         }
1965         return(0);
1966 }
1967
1968 /*
1969  * Remove a name from the leaf attribute list structure.
1970  *
1971  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1972  * If two leaves are 37% full, when combined they will leave 25% free.
1973  */
1974 int
1975 xfs_attr3_leaf_remove(
1976         struct xfs_buf          *bp,
1977         struct xfs_da_args      *args)
1978 {
1979         struct xfs_attr_leafblock *leaf;
1980         struct xfs_attr3_icleaf_hdr ichdr;
1981         struct xfs_attr_leaf_entry *entry;
1982         struct xfs_mount        *mp = args->trans->t_mountp;
1983         int                     before;
1984         int                     after;
1985         int                     smallest;
1986         int                     entsize;
1987         int                     tablesize;
1988         int                     tmp;
1989         int                     i;
1990
1991         trace_xfs_attr_leaf_remove(args);
1992
1993         leaf = bp->b_addr;
1994         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1995
1996         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
1997         ASSERT(args->index >= 0 && args->index < ichdr.count);
1998         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1999                                         xfs_attr3_leaf_hdr_size(leaf));
2000
2001         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2002
2003         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2004         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2005
2006         /*
2007          * Scan through free region table:
2008          *    check for adjacency of free'd entry with an existing one,
2009          *    find smallest free region in case we need to replace it,
2010          *    adjust any map that borders the entry table,
2011          */
2012         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2013                                         + xfs_attr3_leaf_hdr_size(leaf);
2014         tmp = ichdr.freemap[0].size;
2015         before = after = -1;
2016         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2017         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2018         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2019                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
2020                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
2021                 if (ichdr.freemap[i].base == tablesize) {
2022                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2023                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2024                 }
2025
2026                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2027                                 be16_to_cpu(entry->nameidx)) {
2028                         before = i;
2029                 } else if (ichdr.freemap[i].base ==
2030                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2031                         after = i;
2032                 } else if (ichdr.freemap[i].size < tmp) {
2033                         tmp = ichdr.freemap[i].size;
2034                         smallest = i;
2035                 }
2036         }
2037
2038         /*
2039          * Coalesce adjacent freemap regions,
2040          * or replace the smallest region.
2041          */
2042         if ((before >= 0) || (after >= 0)) {
2043                 if ((before >= 0) && (after >= 0)) {
2044                         ichdr.freemap[before].size += entsize;
2045                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2046                         ichdr.freemap[after].base = 0;
2047                         ichdr.freemap[after].size = 0;
2048                 } else if (before >= 0) {
2049                         ichdr.freemap[before].size += entsize;
2050                 } else {
2051                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2052                         ichdr.freemap[after].size += entsize;
2053                 }
2054         } else {
2055                 /*
2056                  * Replace smallest region (if it is smaller than free'd entry)
2057                  */
2058                 if (ichdr.freemap[smallest].size < entsize) {
2059                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2060                         ichdr.freemap[smallest].size = entsize;
2061                 }
2062         }
2063
2064         /*
2065          * Did we remove the first entry?
2066          */
2067         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2068                 smallest = 1;
2069         else
2070                 smallest = 0;
2071
2072         /*
2073          * Compress the remaining entries and zero out the removed stuff.
2074          */
2075         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2076         ichdr.usedbytes -= entsize;
2077         xfs_trans_log_buf(args->trans, bp,
2078              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2079                                    entsize));
2080
2081         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2082         memmove(entry, entry + 1, tmp);
2083         ichdr.count--;
2084         xfs_trans_log_buf(args->trans, bp,
2085             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2086
2087         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2088         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2089
2090         /*
2091          * If we removed the first entry, re-find the first used byte
2092          * in the name area.  Note that if the entry was the "firstused",
2093          * then we don't have a "hole" in our block resulting from
2094          * removing the name.
2095          */
2096         if (smallest) {
2097                 tmp = XFS_LBSIZE(mp);
2098                 entry = xfs_attr3_leaf_entryp(leaf);
2099                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2100                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2101                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2102
2103                         if (be16_to_cpu(entry->nameidx) < tmp)
2104                                 tmp = be16_to_cpu(entry->nameidx);
2105                 }
2106                 ichdr.firstused = tmp;
2107                 if (!ichdr.firstused)
2108                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
2109         } else {
2110                 ichdr.holes = 1;        /* mark as needing compaction */
2111         }
2112         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
2113         xfs_trans_log_buf(args->trans, bp,
2114                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2115                                           xfs_attr3_leaf_hdr_size(leaf)));
2116
2117         /*
2118          * Check if leaf is less than 50% full, caller may want to
2119          * "join" the leaf with a sibling if so.
2120          */
2121         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2122               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2123
2124         return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
2125 }
2126
2127 /*
2128  * Move all the attribute list entries from drop_leaf into save_leaf.
2129  */
2130 void
2131 xfs_attr3_leaf_unbalance(
2132         struct xfs_da_state     *state,
2133         struct xfs_da_state_blk *drop_blk,
2134         struct xfs_da_state_blk *save_blk)
2135 {
2136         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2137         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2138         struct xfs_attr3_icleaf_hdr drophdr;
2139         struct xfs_attr3_icleaf_hdr savehdr;
2140         struct xfs_attr_leaf_entry *entry;
2141         struct xfs_mount        *mp = state->mp;
2142
2143         trace_xfs_attr_leaf_unbalance(state->args);
2144
2145         drop_leaf = drop_blk->bp->b_addr;
2146         save_leaf = save_blk->bp->b_addr;
2147         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
2148         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
2149         entry = xfs_attr3_leaf_entryp(drop_leaf);
2150
2151         /*
2152          * Save last hashval from dying block for later Btree fixup.
2153          */
2154         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2155
2156         /*
2157          * Check if we need a temp buffer, or can we do it in place.
2158          * Note that we don't check "leaf" for holes because we will
2159          * always be dropping it, toosmall() decided that for us already.
2160          */
2161         if (savehdr.holes == 0) {
2162                 /*
2163                  * dest leaf has no holes, so we add there.  May need
2164                  * to make some room in the entry array.
2165                  */
2166                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2167                                          drop_blk->bp, &drophdr)) {
2168                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2169                                                 save_leaf, &savehdr, 0,
2170                                                 drophdr.count, mp);
2171                 } else {
2172                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2173                                                 save_leaf, &savehdr,
2174                                                 savehdr.count, drophdr.count, mp);
2175                 }
2176         } else {
2177                 /*
2178                  * Destination has holes, so we make a temporary copy
2179                  * of the leaf and add them both to that.
2180                  */
2181                 struct xfs_attr_leafblock *tmp_leaf;
2182                 struct xfs_attr3_icleaf_hdr tmphdr;
2183
2184                 tmp_leaf = kmem_alloc(state->blocksize, KM_SLEEP);
2185                 memset(tmp_leaf, 0, state->blocksize);
2186                 memset(&tmphdr, 0, sizeof(tmphdr));
2187
2188                 tmphdr.magic = savehdr.magic;
2189                 tmphdr.forw = savehdr.forw;
2190                 tmphdr.back = savehdr.back;
2191                 tmphdr.firstused = state->blocksize;
2192                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2193                                          drop_blk->bp, &drophdr)) {
2194                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2195                                                 tmp_leaf, &tmphdr, 0,
2196                                                 drophdr.count, mp);
2197                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2198                                                 tmp_leaf, &tmphdr, tmphdr.count,
2199                                                 savehdr.count, mp);
2200                 } else {
2201                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2202                                                 tmp_leaf, &tmphdr, 0,
2203                                                 savehdr.count, mp);
2204                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2205                                                 tmp_leaf, &tmphdr, tmphdr.count,
2206                                                 drophdr.count, mp);
2207                 }
2208                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2209                 savehdr = tmphdr; /* struct copy */
2210                 kmem_free(tmp_leaf);
2211         }
2212
2213         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2214         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2215                                            state->blocksize - 1);
2216
2217         /*
2218          * Copy out last hashval in each block for B-tree code.
2219          */
2220         entry = xfs_attr3_leaf_entryp(save_leaf);
2221         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2222 }
2223
2224 /*========================================================================
2225  * Routines used for finding things in the Btree.
2226  *========================================================================*/
2227
2228 /*
2229  * Look up a name in a leaf attribute list structure.
2230  * This is the internal routine, it uses the caller's buffer.
2231  *
2232  * Note that duplicate keys are allowed, but only check within the
2233  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2234  *
2235  * Return in args->index the index into the entry[] array of either
2236  * the found entry, or where the entry should have been (insert before
2237  * that entry).
2238  *
2239  * Don't change the args->value unless we find the attribute.
2240  */
2241 int
2242 xfs_attr3_leaf_lookup_int(
2243         struct xfs_buf          *bp,
2244         struct xfs_da_args      *args)
2245 {
2246         struct xfs_attr_leafblock *leaf;
2247         struct xfs_attr3_icleaf_hdr ichdr;
2248         struct xfs_attr_leaf_entry *entry;
2249         struct xfs_attr_leaf_entry *entries;
2250         struct xfs_attr_leaf_name_local *name_loc;
2251         struct xfs_attr_leaf_name_remote *name_rmt;
2252         xfs_dahash_t            hashval;
2253         int                     probe;
2254         int                     span;
2255
2256         trace_xfs_attr_leaf_lookup(args);
2257
2258         leaf = bp->b_addr;
2259         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2260         entries = xfs_attr3_leaf_entryp(leaf);
2261         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2262
2263         /*
2264          * Binary search.  (note: small blocks will skip this loop)
2265          */
2266         hashval = args->hashval;
2267         probe = span = ichdr.count / 2;
2268         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2269                 span /= 2;
2270                 if (be32_to_cpu(entry->hashval) < hashval)
2271                         probe += span;
2272                 else if (be32_to_cpu(entry->hashval) > hashval)
2273                         probe -= span;
2274                 else
2275                         break;
2276         }
2277         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2278         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2279
2280         /*
2281          * Since we may have duplicate hashval's, find the first matching
2282          * hashval in the leaf.
2283          */
2284         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2285                 entry--;
2286                 probe--;
2287         }
2288         while (probe < ichdr.count &&
2289                be32_to_cpu(entry->hashval) < hashval) {
2290                 entry++;
2291                 probe++;
2292         }
2293         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2294                 args->index = probe;
2295                 return XFS_ERROR(ENOATTR);
2296         }
2297
2298         /*
2299          * Duplicate keys may be present, so search all of them for a match.
2300          */
2301         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2302                         entry++, probe++) {
2303 /*
2304  * GROT: Add code to remove incomplete entries.
2305  */
2306                 /*
2307                  * If we are looking for INCOMPLETE entries, show only those.
2308                  * If we are looking for complete entries, show only those.
2309                  */
2310                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2311                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2312                         continue;
2313                 }
2314                 if (entry->flags & XFS_ATTR_LOCAL) {
2315                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2316                         if (name_loc->namelen != args->namelen)
2317                                 continue;
2318                         if (memcmp(args->name, name_loc->nameval,
2319                                                         args->namelen) != 0)
2320                                 continue;
2321                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2322                                 continue;
2323                         args->index = probe;
2324                         return XFS_ERROR(EEXIST);
2325                 } else {
2326                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2327                         if (name_rmt->namelen != args->namelen)
2328                                 continue;
2329                         if (memcmp(args->name, name_rmt->name,
2330                                                         args->namelen) != 0)
2331                                 continue;
2332                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2333                                 continue;
2334                         args->index = probe;
2335                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2336                         args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2337                                                    be32_to_cpu(name_rmt->valuelen));
2338                         return XFS_ERROR(EEXIST);
2339                 }
2340         }
2341         args->index = probe;
2342         return XFS_ERROR(ENOATTR);
2343 }
2344
2345 /*
2346  * Get the value associated with an attribute name from a leaf attribute
2347  * list structure.
2348  */
2349 int
2350 xfs_attr3_leaf_getvalue(
2351         struct xfs_buf          *bp,
2352         struct xfs_da_args      *args)
2353 {
2354         struct xfs_attr_leafblock *leaf;
2355         struct xfs_attr3_icleaf_hdr ichdr;
2356         struct xfs_attr_leaf_entry *entry;
2357         struct xfs_attr_leaf_name_local *name_loc;
2358         struct xfs_attr_leaf_name_remote *name_rmt;
2359         int                     valuelen;
2360
2361         leaf = bp->b_addr;
2362         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2363         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2364         ASSERT(args->index < ichdr.count);
2365
2366         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2367         if (entry->flags & XFS_ATTR_LOCAL) {
2368                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2369                 ASSERT(name_loc->namelen == args->namelen);
2370                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2371                 valuelen = be16_to_cpu(name_loc->valuelen);
2372                 if (args->flags & ATTR_KERNOVAL) {
2373                         args->valuelen = valuelen;
2374                         return 0;
2375                 }
2376                 if (args->valuelen < valuelen) {
2377                         args->valuelen = valuelen;
2378                         return XFS_ERROR(ERANGE);
2379                 }
2380                 args->valuelen = valuelen;
2381                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2382         } else {
2383                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2384                 ASSERT(name_rmt->namelen == args->namelen);
2385                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2386                 valuelen = be32_to_cpu(name_rmt->valuelen);
2387                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2388                 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2389                 if (args->flags & ATTR_KERNOVAL) {
2390                         args->valuelen = valuelen;
2391                         return 0;
2392                 }
2393                 if (args->valuelen < valuelen) {
2394                         args->valuelen = valuelen;
2395                         return XFS_ERROR(ERANGE);
2396                 }
2397                 args->valuelen = valuelen;
2398         }
2399         return 0;
2400 }
2401
2402 /*========================================================================
2403  * Utility routines.
2404  *========================================================================*/
2405
2406 /*
2407  * Move the indicated entries from one leaf to another.
2408  * NOTE: this routine modifies both source and destination leaves.
2409  */
2410 /*ARGSUSED*/
2411 STATIC void
2412 xfs_attr3_leaf_moveents(
2413         struct xfs_attr_leafblock       *leaf_s,
2414         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2415         int                             start_s,
2416         struct xfs_attr_leafblock       *leaf_d,
2417         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2418         int                             start_d,
2419         int                             count,
2420         struct xfs_mount                *mp)
2421 {
2422         struct xfs_attr_leaf_entry      *entry_s;
2423         struct xfs_attr_leaf_entry      *entry_d;
2424         int                             desti;
2425         int                             tmp;
2426         int                             i;
2427
2428         /*
2429          * Check for nothing to do.
2430          */
2431         if (count == 0)
2432                 return;
2433
2434         /*
2435          * Set up environment.
2436          */
2437         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2438                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2439         ASSERT(ichdr_s->magic == ichdr_d->magic);
2440         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2441         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2442                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2443         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2444         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2445                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2446
2447         ASSERT(start_s < ichdr_s->count);
2448         ASSERT(start_d <= ichdr_d->count);
2449         ASSERT(count <= ichdr_s->count);
2450
2451
2452         /*
2453          * Move the entries in the destination leaf up to make a hole?
2454          */
2455         if (start_d < ichdr_d->count) {
2456                 tmp  = ichdr_d->count - start_d;
2457                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2458                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2459                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2460                 memmove(entry_d, entry_s, tmp);
2461         }
2462
2463         /*
2464          * Copy all entry's in the same (sorted) order,
2465          * but allocate attribute info packed and in sequence.
2466          */
2467         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2468         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2469         desti = start_d;
2470         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2471                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2472                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2473 #ifdef GROT
2474                 /*
2475                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2476                  * may also need to change the insertion index.  Code turned
2477                  * off for 6.2, should be revisited later.
2478                  */
2479                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2480                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2481                         ichdr_s->usedbytes -= tmp;
2482                         ichdr_s->count -= 1;
2483                         entry_d--;      /* to compensate for ++ in loop hdr */
2484                         desti--;
2485                         if ((start_s + i) < offset)
2486                                 result++;       /* insertion index adjustment */
2487                 } else {
2488 #endif /* GROT */
2489                         ichdr_d->firstused -= tmp;
2490                         /* both on-disk, don't endian flip twice */
2491                         entry_d->hashval = entry_s->hashval;
2492                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2493                         entry_d->flags = entry_s->flags;
2494                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2495                                                         <= XFS_LBSIZE(mp));
2496                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2497                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2498                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2499                                                         <= XFS_LBSIZE(mp));
2500                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2501                         ichdr_s->usedbytes -= tmp;
2502                         ichdr_d->usedbytes += tmp;
2503                         ichdr_s->count -= 1;
2504                         ichdr_d->count += 1;
2505                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2506                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2507                         ASSERT(ichdr_d->firstused >= tmp);
2508 #ifdef GROT
2509                 }
2510 #endif /* GROT */
2511         }
2512
2513         /*
2514          * Zero out the entries we just copied.
2515          */
2516         if (start_s == ichdr_s->count) {
2517                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2518                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2519                 ASSERT(((char *)entry_s + tmp) <=
2520                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2521                 memset(entry_s, 0, tmp);
2522         } else {
2523                 /*
2524                  * Move the remaining entries down to fill the hole,
2525                  * then zero the entries at the top.
2526                  */
2527                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2528                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2529                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2530                 memmove(entry_d, entry_s, tmp);
2531
2532                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2533                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2534                 ASSERT(((char *)entry_s + tmp) <=
2535                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2536                 memset(entry_s, 0, tmp);
2537         }
2538
2539         /*
2540          * Fill in the freemap information
2541          */
2542         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2543         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2544         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2545         ichdr_d->freemap[1].base = 0;
2546         ichdr_d->freemap[2].base = 0;
2547         ichdr_d->freemap[1].size = 0;
2548         ichdr_d->freemap[2].size = 0;
2549         ichdr_s->holes = 1;     /* leaf may not be compact */
2550 }
2551
2552 /*
2553  * Pick up the last hashvalue from a leaf block.
2554  */
2555 xfs_dahash_t
2556 xfs_attr_leaf_lasthash(
2557         struct xfs_buf  *bp,
2558         int             *count)
2559 {
2560         struct xfs_attr3_icleaf_hdr ichdr;
2561         struct xfs_attr_leaf_entry *entries;
2562
2563         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2564         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2565         if (count)
2566                 *count = ichdr.count;
2567         if (!ichdr.count)
2568                 return 0;
2569         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2570 }
2571
2572 /*
2573  * Calculate the number of bytes used to store the indicated attribute
2574  * (whether local or remote only calculate bytes in this block).
2575  */
2576 STATIC int
2577 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2578 {
2579         struct xfs_attr_leaf_entry *entries;
2580         xfs_attr_leaf_name_local_t *name_loc;
2581         xfs_attr_leaf_name_remote_t *name_rmt;
2582         int size;
2583
2584         entries = xfs_attr3_leaf_entryp(leaf);
2585         if (entries[index].flags & XFS_ATTR_LOCAL) {
2586                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2587                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2588                                                    be16_to_cpu(name_loc->valuelen));
2589         } else {
2590                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2591                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2592         }
2593         return size;
2594 }
2595
2596 /*
2597  * Calculate the number of bytes that would be required to store the new
2598  * attribute (whether local or remote only calculate bytes in this block).
2599  * This routine decides as a side effect whether the attribute will be
2600  * a "local" or a "remote" attribute.
2601  */
2602 int
2603 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2604 {
2605         int size;
2606
2607         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2608         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2609                 if (local) {
2610                         *local = 1;
2611                 }
2612         } else {
2613                 size = xfs_attr_leaf_entsize_remote(namelen);
2614                 if (local) {
2615                         *local = 0;
2616                 }
2617         }
2618         return size;
2619 }
2620
2621 /*
2622  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2623  */
2624 int
2625 xfs_attr3_leaf_list_int(
2626         struct xfs_buf                  *bp,
2627         struct xfs_attr_list_context    *context)
2628 {
2629         struct attrlist_cursor_kern     *cursor;
2630         struct xfs_attr_leafblock       *leaf;
2631         struct xfs_attr3_icleaf_hdr     ichdr;
2632         struct xfs_attr_leaf_entry      *entries;
2633         struct xfs_attr_leaf_entry      *entry;
2634         int                             retval;
2635         int                             i;
2636
2637         trace_xfs_attr_list_leaf(context);
2638
2639         leaf = bp->b_addr;
2640         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2641         entries = xfs_attr3_leaf_entryp(leaf);
2642
2643         cursor = context->cursor;
2644         cursor->initted = 1;
2645
2646         /*
2647          * Re-find our place in the leaf block if this is a new syscall.
2648          */
2649         if (context->resynch) {
2650                 entry = &entries[0];
2651                 for (i = 0; i < ichdr.count; entry++, i++) {
2652                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2653                                 if (cursor->offset == context->dupcnt) {
2654                                         context->dupcnt = 0;
2655                                         break;
2656                                 }
2657                                 context->dupcnt++;
2658                         } else if (be32_to_cpu(entry->hashval) >
2659                                         cursor->hashval) {
2660                                 context->dupcnt = 0;
2661                                 break;
2662                         }
2663                 }
2664                 if (i == ichdr.count) {
2665                         trace_xfs_attr_list_notfound(context);
2666                         return 0;
2667                 }
2668         } else {
2669                 entry = &entries[0];
2670                 i = 0;
2671         }
2672         context->resynch = 0;
2673
2674         /*
2675          * We have found our place, start copying out the new attributes.
2676          */
2677         retval = 0;
2678         for (; i < ichdr.count; entry++, i++) {
2679                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2680                         cursor->hashval = be32_to_cpu(entry->hashval);
2681                         cursor->offset = 0;
2682                 }
2683
2684                 if (entry->flags & XFS_ATTR_INCOMPLETE)
2685                         continue;               /* skip incomplete entries */
2686
2687                 if (entry->flags & XFS_ATTR_LOCAL) {
2688                         xfs_attr_leaf_name_local_t *name_loc =
2689                                 xfs_attr3_leaf_name_local(leaf, i);
2690
2691                         retval = context->put_listent(context,
2692                                                 entry->flags,
2693                                                 name_loc->nameval,
2694                                                 (int)name_loc->namelen,
2695                                                 be16_to_cpu(name_loc->valuelen),
2696                                                 &name_loc->nameval[name_loc->namelen]);
2697                         if (retval)
2698                                 return retval;
2699                 } else {
2700                         xfs_attr_leaf_name_remote_t *name_rmt =
2701                                 xfs_attr3_leaf_name_remote(leaf, i);
2702
2703                         int valuelen = be32_to_cpu(name_rmt->valuelen);
2704
2705                         if (context->put_value) {
2706                                 xfs_da_args_t args;
2707
2708                                 memset((char *)&args, 0, sizeof(args));
2709                                 args.dp = context->dp;
2710                                 args.whichfork = XFS_ATTR_FORK;
2711                                 args.valuelen = valuelen;
2712                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2713                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2714                                 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2715                                 retval = xfs_attr_rmtval_get(&args);
2716                                 if (retval)
2717                                         return retval;
2718                                 retval = context->put_listent(context,
2719                                                 entry->flags,
2720                                                 name_rmt->name,
2721                                                 (int)name_rmt->namelen,
2722                                                 valuelen,
2723                                                 args.value);
2724                                 kmem_free(args.value);
2725                         } else {
2726                                 retval = context->put_listent(context,
2727                                                 entry->flags,
2728                                                 name_rmt->name,
2729                                                 (int)name_rmt->namelen,
2730                                                 valuelen,
2731                                                 NULL);
2732                         }
2733                         if (retval)
2734                                 return retval;
2735                 }
2736                 if (context->seen_enough)
2737                         break;
2738                 cursor->offset++;
2739         }
2740         trace_xfs_attr_list_leaf_end(context);
2741         return retval;
2742 }
2743
2744
2745 /*========================================================================
2746  * Manage the INCOMPLETE flag in a leaf entry
2747  *========================================================================*/
2748
2749 /*
2750  * Clear the INCOMPLETE flag on an entry in a leaf block.
2751  */
2752 int
2753 xfs_attr3_leaf_clearflag(
2754         struct xfs_da_args      *args)
2755 {
2756         struct xfs_attr_leafblock *leaf;
2757         struct xfs_attr_leaf_entry *entry;
2758         struct xfs_attr_leaf_name_remote *name_rmt;
2759         struct xfs_buf          *bp;
2760         int                     error;
2761 #ifdef DEBUG
2762         struct xfs_attr3_icleaf_hdr ichdr;
2763         xfs_attr_leaf_name_local_t *name_loc;
2764         int namelen;
2765         char *name;
2766 #endif /* DEBUG */
2767
2768         trace_xfs_attr_leaf_clearflag(args);
2769         /*
2770          * Set up the operation.
2771          */
2772         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2773         if (error)
2774                 return(error);
2775
2776         leaf = bp->b_addr;
2777         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2778         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2779
2780 #ifdef DEBUG
2781         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2782         ASSERT(args->index < ichdr.count);
2783         ASSERT(args->index >= 0);
2784
2785         if (entry->flags & XFS_ATTR_LOCAL) {
2786                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2787                 namelen = name_loc->namelen;
2788                 name = (char *)name_loc->nameval;
2789         } else {
2790                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2791                 namelen = name_rmt->namelen;
2792                 name = (char *)name_rmt->name;
2793         }
2794         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2795         ASSERT(namelen == args->namelen);
2796         ASSERT(memcmp(name, args->name, namelen) == 0);
2797 #endif /* DEBUG */
2798
2799         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2800         xfs_trans_log_buf(args->trans, bp,
2801                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2802
2803         if (args->rmtblkno) {
2804                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2805                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2806                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2807                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2808                 xfs_trans_log_buf(args->trans, bp,
2809                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2810         }
2811
2812         /*
2813          * Commit the flag value change and start the next trans in series.
2814          */
2815         return xfs_trans_roll(&args->trans, args->dp);
2816 }
2817
2818 /*
2819  * Set the INCOMPLETE flag on an entry in a leaf block.
2820  */
2821 int
2822 xfs_attr3_leaf_setflag(
2823         struct xfs_da_args      *args)
2824 {
2825         struct xfs_attr_leafblock *leaf;
2826         struct xfs_attr_leaf_entry *entry;
2827         struct xfs_attr_leaf_name_remote *name_rmt;
2828         struct xfs_buf          *bp;
2829         int error;
2830 #ifdef DEBUG
2831         struct xfs_attr3_icleaf_hdr ichdr;
2832 #endif
2833
2834         trace_xfs_attr_leaf_setflag(args);
2835
2836         /*
2837          * Set up the operation.
2838          */
2839         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2840         if (error)
2841                 return(error);
2842
2843         leaf = bp->b_addr;
2844 #ifdef DEBUG
2845         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2846         ASSERT(args->index < ichdr.count);
2847         ASSERT(args->index >= 0);
2848 #endif
2849         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2850
2851         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2852         entry->flags |= XFS_ATTR_INCOMPLETE;
2853         xfs_trans_log_buf(args->trans, bp,
2854                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2855         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2856                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2857                 name_rmt->valueblk = 0;
2858                 name_rmt->valuelen = 0;
2859                 xfs_trans_log_buf(args->trans, bp,
2860                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2861         }
2862
2863         /*
2864          * Commit the flag value change and start the next trans in series.
2865          */
2866         return xfs_trans_roll(&args->trans, args->dp);
2867 }
2868
2869 /*
2870  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2871  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2872  * entry given by args->blkno2/index2.
2873  *
2874  * Note that they could be in different blocks, or in the same block.
2875  */
2876 int
2877 xfs_attr3_leaf_flipflags(
2878         struct xfs_da_args      *args)
2879 {
2880         struct xfs_attr_leafblock *leaf1;
2881         struct xfs_attr_leafblock *leaf2;
2882         struct xfs_attr_leaf_entry *entry1;
2883         struct xfs_attr_leaf_entry *entry2;
2884         struct xfs_attr_leaf_name_remote *name_rmt;
2885         struct xfs_buf          *bp1;
2886         struct xfs_buf          *bp2;
2887         int error;
2888 #ifdef DEBUG
2889         struct xfs_attr3_icleaf_hdr ichdr1;
2890         struct xfs_attr3_icleaf_hdr ichdr2;
2891         xfs_attr_leaf_name_local_t *name_loc;
2892         int namelen1, namelen2;
2893         char *name1, *name2;
2894 #endif /* DEBUG */
2895
2896         trace_xfs_attr_leaf_flipflags(args);
2897
2898         /*
2899          * Read the block containing the "old" attr
2900          */
2901         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2902         if (error)
2903                 return error;
2904
2905         /*
2906          * Read the block containing the "new" attr, if it is different
2907          */
2908         if (args->blkno2 != args->blkno) {
2909                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2910                                            -1, &bp2);
2911                 if (error)
2912                         return error;
2913         } else {
2914                 bp2 = bp1;
2915         }
2916
2917         leaf1 = bp1->b_addr;
2918         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2919
2920         leaf2 = bp2->b_addr;
2921         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2922
2923 #ifdef DEBUG
2924         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2925         ASSERT(args->index < ichdr1.count);
2926         ASSERT(args->index >= 0);
2927
2928         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2929         ASSERT(args->index2 < ichdr2.count);
2930         ASSERT(args->index2 >= 0);
2931
2932         if (entry1->flags & XFS_ATTR_LOCAL) {
2933                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2934                 namelen1 = name_loc->namelen;
2935                 name1 = (char *)name_loc->nameval;
2936         } else {
2937                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2938                 namelen1 = name_rmt->namelen;
2939                 name1 = (char *)name_rmt->name;
2940         }
2941         if (entry2->flags & XFS_ATTR_LOCAL) {
2942                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2943                 namelen2 = name_loc->namelen;
2944                 name2 = (char *)name_loc->nameval;
2945         } else {
2946                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2947                 namelen2 = name_rmt->namelen;
2948                 name2 = (char *)name_rmt->name;
2949         }
2950         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2951         ASSERT(namelen1 == namelen2);
2952         ASSERT(memcmp(name1, name2, namelen1) == 0);
2953 #endif /* DEBUG */
2954
2955         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2956         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2957
2958         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2959         xfs_trans_log_buf(args->trans, bp1,
2960                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2961         if (args->rmtblkno) {
2962                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2963                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2964                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2965                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2966                 xfs_trans_log_buf(args->trans, bp1,
2967                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2968         }
2969
2970         entry2->flags |= XFS_ATTR_INCOMPLETE;
2971         xfs_trans_log_buf(args->trans, bp2,
2972                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2973         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2974                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2975                 name_rmt->valueblk = 0;
2976                 name_rmt->valuelen = 0;
2977                 xfs_trans_log_buf(args->trans, bp2,
2978                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2979         }
2980
2981         /*
2982          * Commit the flag value change and start the next trans in series.
2983          */
2984         error = xfs_trans_roll(&args->trans, args->dp);
2985
2986         return error;
2987 }
2988
2989 /*========================================================================
2990  * Indiscriminately delete the entire attribute fork
2991  *========================================================================*/
2992
2993 /*
2994  * Recurse (gasp!) through the attribute nodes until we find leaves.
2995  * We're doing a depth-first traversal in order to invalidate everything.
2996  */
2997 int
2998 xfs_attr3_root_inactive(
2999         struct xfs_trans        **trans,
3000         struct xfs_inode        *dp)
3001 {
3002         struct xfs_da_blkinfo   *info;
3003         struct xfs_buf          *bp;
3004         xfs_daddr_t             blkno;
3005         int                     error;
3006
3007         /*
3008          * Read block 0 to see what we have to work with.
3009          * We only get here if we have extents, since we remove
3010          * the extents in reverse order the extent containing
3011          * block 0 must still be there.
3012          */
3013         error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
3014         if (error)
3015                 return error;
3016         blkno = bp->b_bn;
3017
3018         /*
3019          * Invalidate the tree, even if the "tree" is only a single leaf block.
3020          * This is a depth-first traversal!
3021          */
3022         info = bp->b_addr;
3023         switch (info->magic) {
3024         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3025         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3026                 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
3027                 break;
3028         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3029         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3030                 error = xfs_attr3_leaf_inactive(trans, dp, bp);
3031                 break;
3032         default:
3033                 error = XFS_ERROR(EIO);
3034                 xfs_trans_brelse(*trans, bp);
3035                 break;
3036         }
3037         if (error)
3038                 return error;
3039
3040         /*
3041          * Invalidate the incore copy of the root block.
3042          */
3043         error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
3044         if (error)
3045                 return error;
3046         xfs_trans_binval(*trans, bp);   /* remove from cache */
3047         /*
3048          * Commit the invalidate and start the next transaction.
3049          */
3050         error = xfs_trans_roll(trans, dp);
3051
3052         return error;
3053 }
3054
3055 /*
3056  * Recurse (gasp!) through the attribute nodes until we find leaves.
3057  * We're doing a depth-first traversal in order to invalidate everything.
3058  */
3059 STATIC int
3060 xfs_attr3_node_inactive(
3061         struct xfs_trans **trans,
3062         struct xfs_inode *dp,
3063         struct xfs_buf  *bp,
3064         int             level)
3065 {
3066         xfs_da_blkinfo_t *info;
3067         xfs_da_intnode_t *node;
3068         xfs_dablk_t child_fsb;
3069         xfs_daddr_t parent_blkno, child_blkno;
3070         int error, i;
3071         struct xfs_buf *child_bp;
3072         struct xfs_da_node_entry *btree;
3073         struct xfs_da3_icnode_hdr ichdr;
3074
3075         /*
3076          * Since this code is recursive (gasp!) we must protect ourselves.
3077          */
3078         if (level > XFS_DA_NODE_MAXDEPTH) {
3079                 xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3080                 return XFS_ERROR(EIO);
3081         }
3082
3083         node = bp->b_addr;
3084         xfs_da3_node_hdr_from_disk(&ichdr, node);
3085         parent_blkno = bp->b_bn;
3086         if (!ichdr.count) {
3087                 xfs_trans_brelse(*trans, bp);
3088                 return 0;
3089         }
3090         btree = xfs_da3_node_tree_p(node);
3091         child_fsb = be32_to_cpu(btree[0].before);
3092         xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3093
3094         /*
3095          * If this is the node level just above the leaves, simply loop
3096          * over the leaves removing all of them.  If this is higher up
3097          * in the tree, recurse downward.
3098          */
3099         for (i = 0; i < ichdr.count; i++) {
3100                 /*
3101                  * Read the subsidiary block to see what we have to work with.
3102                  * Don't do this in a transaction.  This is a depth-first
3103                  * traversal of the tree so we may deal with many blocks
3104                  * before we come back to this one.
3105                  */
3106                 error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp,
3107                                                 XFS_ATTR_FORK);
3108                 if (error)
3109                         return(error);
3110                 if (child_bp) {
3111                                                 /* save for re-read later */
3112                         child_blkno = XFS_BUF_ADDR(child_bp);
3113
3114                         /*
3115                          * Invalidate the subtree, however we have to.
3116                          */
3117                         info = child_bp->b_addr;
3118                         switch (info->magic) {
3119                         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3120                         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3121                                 error = xfs_attr3_node_inactive(trans, dp,
3122                                                         child_bp, level + 1);
3123                                 break;
3124                         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3125                         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3126                                 error = xfs_attr3_leaf_inactive(trans, dp,
3127                                                         child_bp);
3128                                 break;
3129                         default:
3130                                 error = XFS_ERROR(EIO);
3131                                 xfs_trans_brelse(*trans, child_bp);
3132                                 break;
3133                         }
3134                         if (error)
3135                                 return error;
3136
3137                         /*
3138                          * Remove the subsidiary block from the cache
3139                          * and from the log.
3140                          */
3141                         error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
3142                                 &child_bp, XFS_ATTR_FORK);
3143                         if (error)
3144                                 return error;
3145                         xfs_trans_binval(*trans, child_bp);
3146                 }
3147
3148                 /*
3149                  * If we're not done, re-read the parent to get the next
3150                  * child block number.
3151                  */
3152                 if (i + 1 < ichdr.count) {
3153                         error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
3154                                                  &bp, XFS_ATTR_FORK);
3155                         if (error)
3156                                 return error;
3157                         child_fsb = be32_to_cpu(btree[i + 1].before);
3158                         xfs_trans_brelse(*trans, bp);
3159                 }
3160                 /*
3161                  * Atomically commit the whole invalidate stuff.
3162                  */
3163                 error = xfs_trans_roll(trans, dp);
3164                 if (error)
3165                         return  error;
3166         }
3167
3168         return 0;
3169 }
3170
3171 /*
3172  * Invalidate all of the "remote" value regions pointed to by a particular
3173  * leaf block.
3174  * Note that we must release the lock on the buffer so that we are not
3175  * caught holding something that the logging code wants to flush to disk.
3176  */
3177 STATIC int
3178 xfs_attr3_leaf_inactive(
3179         struct xfs_trans        **trans,
3180         struct xfs_inode        *dp,
3181         struct xfs_buf          *bp)
3182 {
3183         struct xfs_attr_leafblock *leaf;
3184         struct xfs_attr3_icleaf_hdr ichdr;
3185         struct xfs_attr_leaf_entry *entry;
3186         struct xfs_attr_leaf_name_remote *name_rmt;
3187         struct xfs_attr_inactive_list *list;
3188         struct xfs_attr_inactive_list *lp;
3189         int                     error;
3190         int                     count;
3191         int                     size;
3192         int                     tmp;
3193         int                     i;
3194
3195         leaf = bp->b_addr;
3196         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
3197
3198         /*
3199          * Count the number of "remote" value extents.
3200          */
3201         count = 0;
3202         entry = xfs_attr3_leaf_entryp(leaf);
3203         for (i = 0; i < ichdr.count; entry++, i++) {
3204                 if (be16_to_cpu(entry->nameidx) &&
3205                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3206                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3207                         if (name_rmt->valueblk)
3208                                 count++;
3209                 }
3210         }
3211
3212         /*
3213          * If there are no "remote" values, we're done.
3214          */
3215         if (count == 0) {
3216                 xfs_trans_brelse(*trans, bp);
3217                 return 0;
3218         }
3219
3220         /*
3221          * Allocate storage for a list of all the "remote" value extents.
3222          */
3223         size = count * sizeof(xfs_attr_inactive_list_t);
3224         list = kmem_alloc(size, KM_SLEEP);
3225
3226         /*
3227          * Identify each of the "remote" value extents.
3228          */
3229         lp = list;
3230         entry = xfs_attr3_leaf_entryp(leaf);
3231         for (i = 0; i < ichdr.count; entry++, i++) {
3232                 if (be16_to_cpu(entry->nameidx) &&
3233                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3234                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3235                         if (name_rmt->valueblk) {
3236                                 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
3237                                 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
3238                                                     be32_to_cpu(name_rmt->valuelen));
3239                                 lp++;
3240                         }
3241                 }
3242         }
3243         xfs_trans_brelse(*trans, bp);   /* unlock for trans. in freextent() */
3244
3245         /*
3246          * Invalidate each of the "remote" value extents.
3247          */
3248         error = 0;
3249         for (lp = list, i = 0; i < count; i++, lp++) {
3250                 tmp = xfs_attr3_leaf_freextent(trans, dp,
3251                                 lp->valueblk, lp->valuelen);
3252
3253                 if (error == 0)
3254                         error = tmp;    /* save only the 1st errno */
3255         }
3256
3257         kmem_free(list);
3258         return error;
3259 }
3260
3261 /*
3262  * Look at all the extents for this logical region,
3263  * invalidate any buffers that are incore/in transactions.
3264  */
3265 STATIC int
3266 xfs_attr3_leaf_freextent(
3267         struct xfs_trans        **trans,
3268         struct xfs_inode        *dp,
3269         xfs_dablk_t             blkno,
3270         int                     blkcnt)
3271 {
3272         struct xfs_bmbt_irec    map;
3273         struct xfs_buf          *bp;
3274         xfs_dablk_t             tblkno;
3275         xfs_daddr_t             dblkno;
3276         int                     tblkcnt;
3277         int                     dblkcnt;
3278         int                     nmap;
3279         int                     error;
3280
3281         /*
3282          * Roll through the "value", invalidating the attribute value's
3283          * blocks.
3284          */
3285         tblkno = blkno;
3286         tblkcnt = blkcnt;
3287         while (tblkcnt > 0) {
3288                 /*
3289                  * Try to remember where we decided to put the value.
3290                  */
3291                 nmap = 1;
3292                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
3293                                        &map, &nmap, XFS_BMAPI_ATTRFORK);
3294                 if (error) {
3295                         return(error);
3296                 }
3297                 ASSERT(nmap == 1);
3298                 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3299
3300                 /*
3301                  * If it's a hole, these are already unmapped
3302                  * so there's nothing to invalidate.
3303                  */
3304                 if (map.br_startblock != HOLESTARTBLOCK) {
3305
3306                         dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3307                                                   map.br_startblock);
3308                         dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3309                                                 map.br_blockcount);
3310                         bp = xfs_trans_get_buf(*trans,
3311                                         dp->i_mount->m_ddev_targp,
3312                                         dblkno, dblkcnt, 0);
3313                         if (!bp)
3314                                 return ENOMEM;
3315                         xfs_trans_binval(*trans, bp);
3316                         /*
3317                          * Roll to next transaction.
3318                          */
3319                         error = xfs_trans_roll(trans, dp);
3320                         if (error)
3321                                 return (error);
3322                 }
3323
3324                 tblkno += map.br_blockcount;
3325                 tblkcnt -= map.br_blockcount;
3326         }
3327
3328         return(0);
3329 }