ASoC: soc-compress: Send correct stream event for capture start
[firefly-linux-kernel-4.4.55.git] / fs / xfs / xfs_attr_leaf.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_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         xfs_attr_leafblock_t *leaf;
938         xfs_attr_leaf_entry_t *entry;
939         xfs_attr_leaf_name_local_t *name_loc;
940         int bytes, i;
941
942         leaf = bp->b_addr;
943         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
944
945         entry = &leaf->entries[0];
946         bytes = sizeof(struct xfs_attr_sf_hdr);
947         for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
948                 if (entry->flags & XFS_ATTR_INCOMPLETE)
949                         continue;               /* don't copy partial entries */
950                 if (!(entry->flags & XFS_ATTR_LOCAL))
951                         return(0);
952                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
953                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
954                         return(0);
955                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
956                         return(0);
957                 bytes += sizeof(struct xfs_attr_sf_entry)-1
958                                 + name_loc->namelen
959                                 + be16_to_cpu(name_loc->valuelen);
960         }
961         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
962             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
963             (bytes == sizeof(struct xfs_attr_sf_hdr)))
964                 return(-1);
965         return(xfs_attr_shortform_bytesfit(dp, bytes));
966 }
967
968 /*
969  * Convert a leaf attribute list to shortform attribute list
970  */
971 int
972 xfs_attr3_leaf_to_shortform(
973         struct xfs_buf          *bp,
974         struct xfs_da_args      *args,
975         int                     forkoff)
976 {
977         struct xfs_attr_leafblock *leaf;
978         struct xfs_attr3_icleaf_hdr ichdr;
979         struct xfs_attr_leaf_entry *entry;
980         struct xfs_attr_leaf_name_local *name_loc;
981         struct xfs_da_args      nargs;
982         struct xfs_inode        *dp = args->dp;
983         char                    *tmpbuffer;
984         int                     error;
985         int                     i;
986
987         trace_xfs_attr_leaf_to_sf(args);
988
989         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
990         if (!tmpbuffer)
991                 return ENOMEM;
992
993         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
994
995         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
996         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
997         entry = xfs_attr3_leaf_entryp(leaf);
998
999         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1000         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
1001
1002         /*
1003          * Clean out the prior contents of the attribute list.
1004          */
1005         error = xfs_da_shrink_inode(args, 0, bp);
1006         if (error)
1007                 goto out;
1008
1009         if (forkoff == -1) {
1010                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1011                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1012                 xfs_attr_fork_reset(dp, args->trans);
1013                 goto out;
1014         }
1015
1016         xfs_attr_shortform_create(args);
1017
1018         /*
1019          * Copy the attributes
1020          */
1021         memset((char *)&nargs, 0, sizeof(nargs));
1022         nargs.dp = dp;
1023         nargs.firstblock = args->firstblock;
1024         nargs.flist = args->flist;
1025         nargs.total = args->total;
1026         nargs.whichfork = XFS_ATTR_FORK;
1027         nargs.trans = args->trans;
1028         nargs.op_flags = XFS_DA_OP_OKNOENT;
1029
1030         for (i = 0; i < ichdr.count; entry++, i++) {
1031                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1032                         continue;       /* don't copy partial entries */
1033                 if (!entry->nameidx)
1034                         continue;
1035                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1036                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1037                 nargs.name = name_loc->nameval;
1038                 nargs.namelen = name_loc->namelen;
1039                 nargs.value = &name_loc->nameval[nargs.namelen];
1040                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1041                 nargs.hashval = be32_to_cpu(entry->hashval);
1042                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1043                 xfs_attr_shortform_add(&nargs, forkoff);
1044         }
1045         error = 0;
1046
1047 out:
1048         kmem_free(tmpbuffer);
1049         return error;
1050 }
1051
1052 /*
1053  * Convert from using a single leaf to a root node and a leaf.
1054  */
1055 int
1056 xfs_attr3_leaf_to_node(
1057         struct xfs_da_args      *args)
1058 {
1059         struct xfs_attr_leafblock *leaf;
1060         struct xfs_attr3_icleaf_hdr icleafhdr;
1061         struct xfs_attr_leaf_entry *entries;
1062         struct xfs_da_node_entry *btree;
1063         struct xfs_da3_icnode_hdr icnodehdr;
1064         struct xfs_da_intnode   *node;
1065         struct xfs_inode        *dp = args->dp;
1066         struct xfs_mount        *mp = dp->i_mount;
1067         struct xfs_buf          *bp1 = NULL;
1068         struct xfs_buf          *bp2 = NULL;
1069         xfs_dablk_t             blkno;
1070         int                     error;
1071
1072         trace_xfs_attr_leaf_to_node(args);
1073
1074         error = xfs_da_grow_inode(args, &blkno);
1075         if (error)
1076                 goto out;
1077         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1078         if (error)
1079                 goto out;
1080
1081         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1082         if (error)
1083                 goto out;
1084
1085         /* copy leaf to new buffer, update identifiers */
1086         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1087         bp2->b_ops = bp1->b_ops;
1088         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
1089         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1090                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1091                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1092         }
1093         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
1094
1095         /*
1096          * Set up the new root node.
1097          */
1098         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1099         if (error)
1100                 goto out;
1101         node = bp1->b_addr;
1102         xfs_da3_node_hdr_from_disk(&icnodehdr, node);
1103         btree = xfs_da3_node_tree_p(node);
1104
1105         leaf = bp2->b_addr;
1106         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
1107         entries = xfs_attr3_leaf_entryp(leaf);
1108
1109         /* both on-disk, don't endian-flip twice */
1110         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1111         btree[0].before = cpu_to_be32(blkno);
1112         icnodehdr.count = 1;
1113         xfs_da3_node_hdr_to_disk(node, &icnodehdr);
1114         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
1115         error = 0;
1116 out:
1117         return error;
1118 }
1119
1120
1121 /*========================================================================
1122  * Routines used for growing the Btree.
1123  *========================================================================*/
1124
1125 /*
1126  * Create the initial contents of a leaf attribute list
1127  * or a leaf in a node attribute list.
1128  */
1129 STATIC int
1130 xfs_attr3_leaf_create(
1131         struct xfs_da_args      *args,
1132         xfs_dablk_t             blkno,
1133         struct xfs_buf          **bpp)
1134 {
1135         struct xfs_attr_leafblock *leaf;
1136         struct xfs_attr3_icleaf_hdr ichdr;
1137         struct xfs_inode        *dp = args->dp;
1138         struct xfs_mount        *mp = dp->i_mount;
1139         struct xfs_buf          *bp;
1140         int                     error;
1141
1142         trace_xfs_attr_leaf_create(args);
1143
1144         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1145                                             XFS_ATTR_FORK);
1146         if (error)
1147                 return error;
1148         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1149         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1150         leaf = bp->b_addr;
1151         memset(leaf, 0, XFS_LBSIZE(mp));
1152
1153         memset(&ichdr, 0, sizeof(ichdr));
1154         ichdr.firstused = XFS_LBSIZE(mp);
1155
1156         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1157                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1158
1159                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1160
1161                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1162                 hdr3->owner = cpu_to_be64(dp->i_ino);
1163                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
1164
1165                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1166         } else {
1167                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1168                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1169         }
1170         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1171
1172         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1173         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
1174
1175         *bpp = bp;
1176         return 0;
1177 }
1178
1179 /*
1180  * Split the leaf node, rebalance, then add the new entry.
1181  */
1182 int
1183 xfs_attr3_leaf_split(
1184         struct xfs_da_state     *state,
1185         struct xfs_da_state_blk *oldblk,
1186         struct xfs_da_state_blk *newblk)
1187 {
1188         xfs_dablk_t blkno;
1189         int error;
1190
1191         trace_xfs_attr_leaf_split(state->args);
1192
1193         /*
1194          * Allocate space for a new leaf node.
1195          */
1196         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1197         error = xfs_da_grow_inode(state->args, &blkno);
1198         if (error)
1199                 return(error);
1200         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1201         if (error)
1202                 return(error);
1203         newblk->blkno = blkno;
1204         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1205
1206         /*
1207          * Rebalance the entries across the two leaves.
1208          * NOTE: rebalance() currently depends on the 2nd block being empty.
1209          */
1210         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1211         error = xfs_da3_blk_link(state, oldblk, newblk);
1212         if (error)
1213                 return(error);
1214
1215         /*
1216          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1217          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1218          * "new" attrs info.  Will need the "old" info to remove it later.
1219          *
1220          * Insert the "new" entry in the correct block.
1221          */
1222         if (state->inleaf) {
1223                 trace_xfs_attr_leaf_add_old(state->args);
1224                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1225         } else {
1226                 trace_xfs_attr_leaf_add_new(state->args);
1227                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1228         }
1229
1230         /*
1231          * Update last hashval in each block since we added the name.
1232          */
1233         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1234         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1235         return(error);
1236 }
1237
1238 /*
1239  * Add a name to the leaf attribute list structure.
1240  */
1241 int
1242 xfs_attr3_leaf_add(
1243         struct xfs_buf          *bp,
1244         struct xfs_da_args      *args)
1245 {
1246         struct xfs_attr_leafblock *leaf;
1247         struct xfs_attr3_icleaf_hdr ichdr;
1248         int                     tablesize;
1249         int                     entsize;
1250         int                     sum;
1251         int                     tmp;
1252         int                     i;
1253
1254         trace_xfs_attr_leaf_add(args);
1255
1256         leaf = bp->b_addr;
1257         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1258         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1259         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1260                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1261
1262         /*
1263          * Search through freemap for first-fit on new name length.
1264          * (may need to figure in size of entry struct too)
1265          */
1266         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1267                                         + xfs_attr3_leaf_hdr_size(leaf);
1268         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1269                 if (tablesize > ichdr.firstused) {
1270                         sum += ichdr.freemap[i].size;
1271                         continue;
1272                 }
1273                 if (!ichdr.freemap[i].size)
1274                         continue;       /* no space in this map */
1275                 tmp = entsize;
1276                 if (ichdr.freemap[i].base < ichdr.firstused)
1277                         tmp += sizeof(xfs_attr_leaf_entry_t);
1278                 if (ichdr.freemap[i].size >= tmp) {
1279                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1280                         goto out_log_hdr;
1281                 }
1282                 sum += ichdr.freemap[i].size;
1283         }
1284
1285         /*
1286          * If there are no holes in the address space of the block,
1287          * and we don't have enough freespace, then compaction will do us
1288          * no good and we should just give up.
1289          */
1290         if (!ichdr.holes && sum < entsize)
1291                 return XFS_ERROR(ENOSPC);
1292
1293         /*
1294          * Compact the entries to coalesce free space.
1295          * This may change the hdr->count via dropping INCOMPLETE entries.
1296          */
1297         xfs_attr3_leaf_compact(args, &ichdr, bp);
1298
1299         /*
1300          * After compaction, the block is guaranteed to have only one
1301          * free region, in freemap[0].  If it is not big enough, give up.
1302          */
1303         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1304                 tmp = ENOSPC;
1305                 goto out_log_hdr;
1306         }
1307
1308         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1309
1310 out_log_hdr:
1311         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1312         xfs_trans_log_buf(args->trans, bp,
1313                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1314                                 xfs_attr3_leaf_hdr_size(leaf)));
1315         return tmp;
1316 }
1317
1318 /*
1319  * Add a name to a leaf attribute list structure.
1320  */
1321 STATIC int
1322 xfs_attr3_leaf_add_work(
1323         struct xfs_buf          *bp,
1324         struct xfs_attr3_icleaf_hdr *ichdr,
1325         struct xfs_da_args      *args,
1326         int                     mapindex)
1327 {
1328         struct xfs_attr_leafblock *leaf;
1329         struct xfs_attr_leaf_entry *entry;
1330         struct xfs_attr_leaf_name_local *name_loc;
1331         struct xfs_attr_leaf_name_remote *name_rmt;
1332         struct xfs_mount        *mp;
1333         int                     tmp;
1334         int                     i;
1335
1336         trace_xfs_attr_leaf_add_work(args);
1337
1338         leaf = bp->b_addr;
1339         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1340         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1341
1342         /*
1343          * Force open some space in the entry array and fill it in.
1344          */
1345         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1346         if (args->index < ichdr->count) {
1347                 tmp  = ichdr->count - args->index;
1348                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1349                 memmove(entry + 1, entry, tmp);
1350                 xfs_trans_log_buf(args->trans, bp,
1351                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1352         }
1353         ichdr->count++;
1354
1355         /*
1356          * Allocate space for the new string (at the end of the run).
1357          */
1358         mp = args->trans->t_mountp;
1359         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1360         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1361         ASSERT(ichdr->freemap[mapindex].size >=
1362                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1363                                          mp->m_sb.sb_blocksize, NULL));
1364         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1365         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1366
1367         ichdr->freemap[mapindex].size -=
1368                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1369                                                  mp->m_sb.sb_blocksize, &tmp);
1370
1371         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1372                                      ichdr->freemap[mapindex].size);
1373         entry->hashval = cpu_to_be32(args->hashval);
1374         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1375         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1376         if (args->op_flags & XFS_DA_OP_RENAME) {
1377                 entry->flags |= XFS_ATTR_INCOMPLETE;
1378                 if ((args->blkno2 == args->blkno) &&
1379                     (args->index2 <= args->index)) {
1380                         args->index2++;
1381                 }
1382         }
1383         xfs_trans_log_buf(args->trans, bp,
1384                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1385         ASSERT((args->index == 0) ||
1386                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1387         ASSERT((args->index == ichdr->count - 1) ||
1388                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1389
1390         /*
1391          * For "remote" attribute values, simply note that we need to
1392          * allocate space for the "remote" value.  We can't actually
1393          * allocate the extents in this transaction, and we can't decide
1394          * which blocks they should be as we might allocate more blocks
1395          * as part of this transaction (a split operation for example).
1396          */
1397         if (entry->flags & XFS_ATTR_LOCAL) {
1398                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1399                 name_loc->namelen = args->namelen;
1400                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1401                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1402                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1403                                    be16_to_cpu(name_loc->valuelen));
1404         } else {
1405                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1406                 name_rmt->namelen = args->namelen;
1407                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1408                 entry->flags |= XFS_ATTR_INCOMPLETE;
1409                 /* just in case */
1410                 name_rmt->valuelen = 0;
1411                 name_rmt->valueblk = 0;
1412                 args->rmtblkno = 1;
1413                 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1414         }
1415         xfs_trans_log_buf(args->trans, bp,
1416              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1417                                    xfs_attr_leaf_entsize(leaf, args->index)));
1418
1419         /*
1420          * Update the control info for this leaf node
1421          */
1422         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1423                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1424
1425         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1426                                         + xfs_attr3_leaf_hdr_size(leaf));
1427         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1428                                         + xfs_attr3_leaf_hdr_size(leaf);
1429
1430         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1431                 if (ichdr->freemap[i].base == tmp) {
1432                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1433                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1434                 }
1435         }
1436         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1437         return 0;
1438 }
1439
1440 /*
1441  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1442  */
1443 STATIC void
1444 xfs_attr3_leaf_compact(
1445         struct xfs_da_args      *args,
1446         struct xfs_attr3_icleaf_hdr *ichdr_d,
1447         struct xfs_buf          *bp)
1448 {
1449         xfs_attr_leafblock_t    *leaf_s, *leaf_d;
1450         struct xfs_attr3_icleaf_hdr ichdr_s;
1451         struct xfs_trans        *trans = args->trans;
1452         struct xfs_mount        *mp = trans->t_mountp;
1453         char                    *tmpbuffer;
1454
1455         trace_xfs_attr_leaf_compact(args);
1456
1457         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1458         ASSERT(tmpbuffer != NULL);
1459         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1460         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1461
1462         /*
1463          * Copy basic information
1464          */
1465         leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1466         leaf_d = bp->b_addr;
1467         ichdr_s = *ichdr_d;     /* struct copy */
1468         ichdr_d->firstused = XFS_LBSIZE(mp);
1469         ichdr_d->usedbytes = 0;
1470         ichdr_d->count = 0;
1471         ichdr_d->holes = 0;
1472         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_s);
1473         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
1474
1475         /*
1476          * Copy all entry's in the same (sorted) order,
1477          * but allocate name/value pairs packed and in sequence.
1478          */
1479         xfs_attr3_leaf_moveents(leaf_s, &ichdr_s, 0, leaf_d, ichdr_d, 0,
1480                                 ichdr_s.count, mp);
1481         /*
1482          * this logs the entire buffer, but the caller must write the header
1483          * back to the buffer when it is finished modifying it.
1484          */
1485         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1486
1487         kmem_free(tmpbuffer);
1488 }
1489
1490 /*
1491  * Compare two leaf blocks "order".
1492  * Return 0 unless leaf2 should go before leaf1.
1493  */
1494 static int
1495 xfs_attr3_leaf_order(
1496         struct xfs_buf  *leaf1_bp,
1497         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1498         struct xfs_buf  *leaf2_bp,
1499         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1500 {
1501         struct xfs_attr_leaf_entry *entries1;
1502         struct xfs_attr_leaf_entry *entries2;
1503
1504         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1505         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1506         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1507             ((be32_to_cpu(entries2[0].hashval) <
1508               be32_to_cpu(entries1[0].hashval)) ||
1509              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1510               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1511                 return 1;
1512         }
1513         return 0;
1514 }
1515
1516 int
1517 xfs_attr_leaf_order(
1518         struct xfs_buf  *leaf1_bp,
1519         struct xfs_buf  *leaf2_bp)
1520 {
1521         struct xfs_attr3_icleaf_hdr ichdr1;
1522         struct xfs_attr3_icleaf_hdr ichdr2;
1523
1524         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1525         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1526         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1527 }
1528
1529 /*
1530  * Redistribute the attribute list entries between two leaf nodes,
1531  * taking into account the size of the new entry.
1532  *
1533  * NOTE: if new block is empty, then it will get the upper half of the
1534  * old block.  At present, all (one) callers pass in an empty second block.
1535  *
1536  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1537  * to match what it is doing in splitting the attribute leaf block.  Those
1538  * values are used in "atomic rename" operations on attributes.  Note that
1539  * the "new" and "old" values can end up in different blocks.
1540  */
1541 STATIC void
1542 xfs_attr3_leaf_rebalance(
1543         struct xfs_da_state     *state,
1544         struct xfs_da_state_blk *blk1,
1545         struct xfs_da_state_blk *blk2)
1546 {
1547         struct xfs_da_args      *args;
1548         struct xfs_attr_leafblock *leaf1;
1549         struct xfs_attr_leafblock *leaf2;
1550         struct xfs_attr3_icleaf_hdr ichdr1;
1551         struct xfs_attr3_icleaf_hdr ichdr2;
1552         struct xfs_attr_leaf_entry *entries1;
1553         struct xfs_attr_leaf_entry *entries2;
1554         int                     count;
1555         int                     totallen;
1556         int                     max;
1557         int                     space;
1558         int                     swap;
1559
1560         /*
1561          * Set up environment.
1562          */
1563         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1564         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1565         leaf1 = blk1->bp->b_addr;
1566         leaf2 = blk2->bp->b_addr;
1567         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1568         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1569         ASSERT(ichdr2.count == 0);
1570         args = state->args;
1571
1572         trace_xfs_attr_leaf_rebalance(args);
1573
1574         /*
1575          * Check ordering of blocks, reverse if it makes things simpler.
1576          *
1577          * NOTE: Given that all (current) callers pass in an empty
1578          * second block, this code should never set "swap".
1579          */
1580         swap = 0;
1581         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1582                 struct xfs_da_state_blk *tmp_blk;
1583                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1584
1585                 tmp_blk = blk1;
1586                 blk1 = blk2;
1587                 blk2 = tmp_blk;
1588
1589                 /* struct copies to swap them rather than reconverting */
1590                 tmp_ichdr = ichdr1;
1591                 ichdr1 = ichdr2;
1592                 ichdr2 = tmp_ichdr;
1593
1594                 leaf1 = blk1->bp->b_addr;
1595                 leaf2 = blk2->bp->b_addr;
1596                 swap = 1;
1597         }
1598
1599         /*
1600          * Examine entries until we reduce the absolute difference in
1601          * byte usage between the two blocks to a minimum.  Then get
1602          * the direction to copy and the number of elements to move.
1603          *
1604          * "inleaf" is true if the new entry should be inserted into blk1.
1605          * If "swap" is also true, then reverse the sense of "inleaf".
1606          */
1607         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1608                                                       blk2, &ichdr2,
1609                                                       &count, &totallen);
1610         if (swap)
1611                 state->inleaf = !state->inleaf;
1612
1613         /*
1614          * Move any entries required from leaf to leaf:
1615          */
1616         if (count < ichdr1.count) {
1617                 /*
1618                  * Figure the total bytes to be added to the destination leaf.
1619                  */
1620                 /* number entries being moved */
1621                 count = ichdr1.count - count;
1622                 space  = ichdr1.usedbytes - totallen;
1623                 space += count * sizeof(xfs_attr_leaf_entry_t);
1624
1625                 /*
1626                  * leaf2 is the destination, compact it if it looks tight.
1627                  */
1628                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1629                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1630                 if (space > max)
1631                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1632
1633                 /*
1634                  * Move high entries from leaf1 to low end of leaf2.
1635                  */
1636                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1637                                 leaf2, &ichdr2, 0, count, state->mp);
1638
1639         } else if (count > ichdr1.count) {
1640                 /*
1641                  * I assert that since all callers pass in an empty
1642                  * second buffer, this code should never execute.
1643                  */
1644                 ASSERT(0);
1645
1646                 /*
1647                  * Figure the total bytes to be added to the destination leaf.
1648                  */
1649                 /* number entries being moved */
1650                 count -= ichdr1.count;
1651                 space  = totallen - ichdr1.usedbytes;
1652                 space += count * sizeof(xfs_attr_leaf_entry_t);
1653
1654                 /*
1655                  * leaf1 is the destination, compact it if it looks tight.
1656                  */
1657                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1658                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1659                 if (space > max)
1660                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1661
1662                 /*
1663                  * Move low entries from leaf2 to high end of leaf1.
1664                  */
1665                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1666                                         ichdr1.count, count, state->mp);
1667         }
1668
1669         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1670         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1671         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1672         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1673
1674         /*
1675          * Copy out last hashval in each block for B-tree code.
1676          */
1677         entries1 = xfs_attr3_leaf_entryp(leaf1);
1678         entries2 = xfs_attr3_leaf_entryp(leaf2);
1679         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1680         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1681
1682         /*
1683          * Adjust the expected index for insertion.
1684          * NOTE: this code depends on the (current) situation that the
1685          * second block was originally empty.
1686          *
1687          * If the insertion point moved to the 2nd block, we must adjust
1688          * the index.  We must also track the entry just following the
1689          * new entry for use in an "atomic rename" operation, that entry
1690          * is always the "old" entry and the "new" entry is what we are
1691          * inserting.  The index/blkno fields refer to the "old" entry,
1692          * while the index2/blkno2 fields refer to the "new" entry.
1693          */
1694         if (blk1->index > ichdr1.count) {
1695                 ASSERT(state->inleaf == 0);
1696                 blk2->index = blk1->index - ichdr1.count;
1697                 args->index = args->index2 = blk2->index;
1698                 args->blkno = args->blkno2 = blk2->blkno;
1699         } else if (blk1->index == ichdr1.count) {
1700                 if (state->inleaf) {
1701                         args->index = blk1->index;
1702                         args->blkno = blk1->blkno;
1703                         args->index2 = 0;
1704                         args->blkno2 = blk2->blkno;
1705                 } else {
1706                         /*
1707                          * On a double leaf split, the original attr location
1708                          * is already stored in blkno2/index2, so don't
1709                          * overwrite it overwise we corrupt the tree.
1710                          */
1711                         blk2->index = blk1->index - ichdr1.count;
1712                         args->index = blk2->index;
1713                         args->blkno = blk2->blkno;
1714                         if (!state->extravalid) {
1715                                 /*
1716                                  * set the new attr location to match the old
1717                                  * one and let the higher level split code
1718                                  * decide where in the leaf to place it.
1719                                  */
1720                                 args->index2 = blk2->index;
1721                                 args->blkno2 = blk2->blkno;
1722                         }
1723                 }
1724         } else {
1725                 ASSERT(state->inleaf == 1);
1726                 args->index = args->index2 = blk1->index;
1727                 args->blkno = args->blkno2 = blk1->blkno;
1728         }
1729 }
1730
1731 /*
1732  * Examine entries until we reduce the absolute difference in
1733  * byte usage between the two blocks to a minimum.
1734  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1735  * GROT: there will always be enough room in either block for a new entry.
1736  * GROT: Do a double-split for this case?
1737  */
1738 STATIC int
1739 xfs_attr3_leaf_figure_balance(
1740         struct xfs_da_state             *state,
1741         struct xfs_da_state_blk         *blk1,
1742         struct xfs_attr3_icleaf_hdr     *ichdr1,
1743         struct xfs_da_state_blk         *blk2,
1744         struct xfs_attr3_icleaf_hdr     *ichdr2,
1745         int                             *countarg,
1746         int                             *usedbytesarg)
1747 {
1748         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1749         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1750         struct xfs_attr_leaf_entry      *entry;
1751         int                             count;
1752         int                             max;
1753         int                             index;
1754         int                             totallen = 0;
1755         int                             half;
1756         int                             lastdelta;
1757         int                             foundit = 0;
1758         int                             tmp;
1759
1760         /*
1761          * Examine entries until we reduce the absolute difference in
1762          * byte usage between the two blocks to a minimum.
1763          */
1764         max = ichdr1->count + ichdr2->count;
1765         half = (max + 1) * sizeof(*entry);
1766         half += ichdr1->usedbytes + ichdr2->usedbytes +
1767                         xfs_attr_leaf_newentsize(state->args->namelen,
1768                                                  state->args->valuelen,
1769                                                  state->blocksize, NULL);
1770         half /= 2;
1771         lastdelta = state->blocksize;
1772         entry = xfs_attr3_leaf_entryp(leaf1);
1773         for (count = index = 0; count < max; entry++, index++, count++) {
1774
1775 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1776                 /*
1777                  * The new entry is in the first block, account for it.
1778                  */
1779                 if (count == blk1->index) {
1780                         tmp = totallen + sizeof(*entry) +
1781                                 xfs_attr_leaf_newentsize(
1782                                                 state->args->namelen,
1783                                                 state->args->valuelen,
1784                                                 state->blocksize, NULL);
1785                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1786                                 break;
1787                         lastdelta = XFS_ATTR_ABS(half - tmp);
1788                         totallen = tmp;
1789                         foundit = 1;
1790                 }
1791
1792                 /*
1793                  * Wrap around into the second block if necessary.
1794                  */
1795                 if (count == ichdr1->count) {
1796                         leaf1 = leaf2;
1797                         entry = xfs_attr3_leaf_entryp(leaf1);
1798                         index = 0;
1799                 }
1800
1801                 /*
1802                  * Figure out if next leaf entry would be too much.
1803                  */
1804                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1805                                                                         index);
1806                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1807                         break;
1808                 lastdelta = XFS_ATTR_ABS(half - tmp);
1809                 totallen = tmp;
1810 #undef XFS_ATTR_ABS
1811         }
1812
1813         /*
1814          * Calculate the number of usedbytes that will end up in lower block.
1815          * If new entry not in lower block, fix up the count.
1816          */
1817         totallen -= count * sizeof(*entry);
1818         if (foundit) {
1819                 totallen -= sizeof(*entry) +
1820                                 xfs_attr_leaf_newentsize(
1821                                                 state->args->namelen,
1822                                                 state->args->valuelen,
1823                                                 state->blocksize, NULL);
1824         }
1825
1826         *countarg = count;
1827         *usedbytesarg = totallen;
1828         return foundit;
1829 }
1830
1831 /*========================================================================
1832  * Routines used for shrinking the Btree.
1833  *========================================================================*/
1834
1835 /*
1836  * Check a leaf block and its neighbors to see if the block should be
1837  * collapsed into one or the other neighbor.  Always keep the block
1838  * with the smaller block number.
1839  * If the current block is over 50% full, don't try to join it, return 0.
1840  * If the block is empty, fill in the state structure and return 2.
1841  * If it can be collapsed, fill in the state structure and return 1.
1842  * If nothing can be done, return 0.
1843  *
1844  * GROT: allow for INCOMPLETE entries in calculation.
1845  */
1846 int
1847 xfs_attr3_leaf_toosmall(
1848         struct xfs_da_state     *state,
1849         int                     *action)
1850 {
1851         struct xfs_attr_leafblock *leaf;
1852         struct xfs_da_state_blk *blk;
1853         struct xfs_attr3_icleaf_hdr ichdr;
1854         struct xfs_buf          *bp;
1855         xfs_dablk_t             blkno;
1856         int                     bytes;
1857         int                     forward;
1858         int                     error;
1859         int                     retval;
1860         int                     i;
1861
1862         trace_xfs_attr_leaf_toosmall(state->args);
1863
1864         /*
1865          * Check for the degenerate case of the block being over 50% full.
1866          * If so, it's not worth even looking to see if we might be able
1867          * to coalesce with a sibling.
1868          */
1869         blk = &state->path.blk[ state->path.active-1 ];
1870         leaf = blk->bp->b_addr;
1871         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1872         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1873                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1874                 ichdr.usedbytes;
1875         if (bytes > (state->blocksize >> 1)) {
1876                 *action = 0;    /* blk over 50%, don't try to join */
1877                 return(0);
1878         }
1879
1880         /*
1881          * Check for the degenerate case of the block being empty.
1882          * If the block is empty, we'll simply delete it, no need to
1883          * coalesce it with a sibling block.  We choose (arbitrarily)
1884          * to merge with the forward block unless it is NULL.
1885          */
1886         if (ichdr.count == 0) {
1887                 /*
1888                  * Make altpath point to the block we want to keep and
1889                  * path point to the block we want to drop (this one).
1890                  */
1891                 forward = (ichdr.forw != 0);
1892                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1893                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1894                                                  0, &retval);
1895                 if (error)
1896                         return(error);
1897                 if (retval) {
1898                         *action = 0;
1899                 } else {
1900                         *action = 2;
1901                 }
1902                 return 0;
1903         }
1904
1905         /*
1906          * Examine each sibling block to see if we can coalesce with
1907          * at least 25% free space to spare.  We need to figure out
1908          * whether to merge with the forward or the backward block.
1909          * We prefer coalescing with the lower numbered sibling so as
1910          * to shrink an attribute list over time.
1911          */
1912         /* start with smaller blk num */
1913         forward = ichdr.forw < ichdr.back;
1914         for (i = 0; i < 2; forward = !forward, i++) {
1915                 struct xfs_attr3_icleaf_hdr ichdr2;
1916                 if (forward)
1917                         blkno = ichdr.forw;
1918                 else
1919                         blkno = ichdr.back;
1920                 if (blkno == 0)
1921                         continue;
1922                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1923                                         blkno, -1, &bp);
1924                 if (error)
1925                         return(error);
1926
1927                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1928
1929                 bytes = state->blocksize - (state->blocksize >> 2) -
1930                         ichdr.usedbytes - ichdr2.usedbytes -
1931                         ((ichdr.count + ichdr2.count) *
1932                                         sizeof(xfs_attr_leaf_entry_t)) -
1933                         xfs_attr3_leaf_hdr_size(leaf);
1934
1935                 xfs_trans_brelse(state->args->trans, bp);
1936                 if (bytes >= 0)
1937                         break;  /* fits with at least 25% to spare */
1938         }
1939         if (i >= 2) {
1940                 *action = 0;
1941                 return(0);
1942         }
1943
1944         /*
1945          * Make altpath point to the block we want to keep (the lower
1946          * numbered block) and path point to the block we want to drop.
1947          */
1948         memcpy(&state->altpath, &state->path, sizeof(state->path));
1949         if (blkno < blk->blkno) {
1950                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1951                                                  0, &retval);
1952         } else {
1953                 error = xfs_da3_path_shift(state, &state->path, forward,
1954                                                  0, &retval);
1955         }
1956         if (error)
1957                 return(error);
1958         if (retval) {
1959                 *action = 0;
1960         } else {
1961                 *action = 1;
1962         }
1963         return(0);
1964 }
1965
1966 /*
1967  * Remove a name from the leaf attribute list structure.
1968  *
1969  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1970  * If two leaves are 37% full, when combined they will leave 25% free.
1971  */
1972 int
1973 xfs_attr3_leaf_remove(
1974         struct xfs_buf          *bp,
1975         struct xfs_da_args      *args)
1976 {
1977         struct xfs_attr_leafblock *leaf;
1978         struct xfs_attr3_icleaf_hdr ichdr;
1979         struct xfs_attr_leaf_entry *entry;
1980         struct xfs_mount        *mp = args->trans->t_mountp;
1981         int                     before;
1982         int                     after;
1983         int                     smallest;
1984         int                     entsize;
1985         int                     tablesize;
1986         int                     tmp;
1987         int                     i;
1988
1989         trace_xfs_attr_leaf_remove(args);
1990
1991         leaf = bp->b_addr;
1992         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1993
1994         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
1995         ASSERT(args->index >= 0 && args->index < ichdr.count);
1996         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1997                                         xfs_attr3_leaf_hdr_size(leaf));
1998
1999         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2000
2001         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2002         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2003
2004         /*
2005          * Scan through free region table:
2006          *    check for adjacency of free'd entry with an existing one,
2007          *    find smallest free region in case we need to replace it,
2008          *    adjust any map that borders the entry table,
2009          */
2010         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2011                                         + xfs_attr3_leaf_hdr_size(leaf);
2012         tmp = ichdr.freemap[0].size;
2013         before = after = -1;
2014         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2015         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2016         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2017                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
2018                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
2019                 if (ichdr.freemap[i].base == tablesize) {
2020                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2021                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2022                 }
2023
2024                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2025                                 be16_to_cpu(entry->nameidx)) {
2026                         before = i;
2027                 } else if (ichdr.freemap[i].base ==
2028                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2029                         after = i;
2030                 } else if (ichdr.freemap[i].size < tmp) {
2031                         tmp = ichdr.freemap[i].size;
2032                         smallest = i;
2033                 }
2034         }
2035
2036         /*
2037          * Coalesce adjacent freemap regions,
2038          * or replace the smallest region.
2039          */
2040         if ((before >= 0) || (after >= 0)) {
2041                 if ((before >= 0) && (after >= 0)) {
2042                         ichdr.freemap[before].size += entsize;
2043                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2044                         ichdr.freemap[after].base = 0;
2045                         ichdr.freemap[after].size = 0;
2046                 } else if (before >= 0) {
2047                         ichdr.freemap[before].size += entsize;
2048                 } else {
2049                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2050                         ichdr.freemap[after].size += entsize;
2051                 }
2052         } else {
2053                 /*
2054                  * Replace smallest region (if it is smaller than free'd entry)
2055                  */
2056                 if (ichdr.freemap[smallest].size < entsize) {
2057                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2058                         ichdr.freemap[smallest].size = entsize;
2059                 }
2060         }
2061
2062         /*
2063          * Did we remove the first entry?
2064          */
2065         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2066                 smallest = 1;
2067         else
2068                 smallest = 0;
2069
2070         /*
2071          * Compress the remaining entries and zero out the removed stuff.
2072          */
2073         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2074         ichdr.usedbytes -= entsize;
2075         xfs_trans_log_buf(args->trans, bp,
2076              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2077                                    entsize));
2078
2079         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2080         memmove(entry, entry + 1, tmp);
2081         ichdr.count--;
2082         xfs_trans_log_buf(args->trans, bp,
2083             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2084
2085         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2086         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2087
2088         /*
2089          * If we removed the first entry, re-find the first used byte
2090          * in the name area.  Note that if the entry was the "firstused",
2091          * then we don't have a "hole" in our block resulting from
2092          * removing the name.
2093          */
2094         if (smallest) {
2095                 tmp = XFS_LBSIZE(mp);
2096                 entry = xfs_attr3_leaf_entryp(leaf);
2097                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2098                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2099                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2100
2101                         if (be16_to_cpu(entry->nameidx) < tmp)
2102                                 tmp = be16_to_cpu(entry->nameidx);
2103                 }
2104                 ichdr.firstused = tmp;
2105                 if (!ichdr.firstused)
2106                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
2107         } else {
2108                 ichdr.holes = 1;        /* mark as needing compaction */
2109         }
2110         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
2111         xfs_trans_log_buf(args->trans, bp,
2112                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2113                                           xfs_attr3_leaf_hdr_size(leaf)));
2114
2115         /*
2116          * Check if leaf is less than 50% full, caller may want to
2117          * "join" the leaf with a sibling if so.
2118          */
2119         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2120               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2121
2122         return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
2123 }
2124
2125 /*
2126  * Move all the attribute list entries from drop_leaf into save_leaf.
2127  */
2128 void
2129 xfs_attr3_leaf_unbalance(
2130         struct xfs_da_state     *state,
2131         struct xfs_da_state_blk *drop_blk,
2132         struct xfs_da_state_blk *save_blk)
2133 {
2134         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2135         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2136         struct xfs_attr3_icleaf_hdr drophdr;
2137         struct xfs_attr3_icleaf_hdr savehdr;
2138         struct xfs_attr_leaf_entry *entry;
2139         struct xfs_mount        *mp = state->mp;
2140
2141         trace_xfs_attr_leaf_unbalance(state->args);
2142
2143         drop_leaf = drop_blk->bp->b_addr;
2144         save_leaf = save_blk->bp->b_addr;
2145         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
2146         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
2147         entry = xfs_attr3_leaf_entryp(drop_leaf);
2148
2149         /*
2150          * Save last hashval from dying block for later Btree fixup.
2151          */
2152         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2153
2154         /*
2155          * Check if we need a temp buffer, or can we do it in place.
2156          * Note that we don't check "leaf" for holes because we will
2157          * always be dropping it, toosmall() decided that for us already.
2158          */
2159         if (savehdr.holes == 0) {
2160                 /*
2161                  * dest leaf has no holes, so we add there.  May need
2162                  * to make some room in the entry array.
2163                  */
2164                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2165                                          drop_blk->bp, &drophdr)) {
2166                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2167                                                 save_leaf, &savehdr, 0,
2168                                                 drophdr.count, mp);
2169                 } else {
2170                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2171                                                 save_leaf, &savehdr,
2172                                                 savehdr.count, drophdr.count, mp);
2173                 }
2174         } else {
2175                 /*
2176                  * Destination has holes, so we make a temporary copy
2177                  * of the leaf and add them both to that.
2178                  */
2179                 struct xfs_attr_leafblock *tmp_leaf;
2180                 struct xfs_attr3_icleaf_hdr tmphdr;
2181
2182                 tmp_leaf = kmem_alloc(state->blocksize, KM_SLEEP);
2183                 memset(tmp_leaf, 0, state->blocksize);
2184                 memset(&tmphdr, 0, sizeof(tmphdr));
2185
2186                 tmphdr.magic = savehdr.magic;
2187                 tmphdr.forw = savehdr.forw;
2188                 tmphdr.back = savehdr.back;
2189                 tmphdr.firstused = state->blocksize;
2190                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2191                                          drop_blk->bp, &drophdr)) {
2192                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2193                                                 tmp_leaf, &tmphdr, 0,
2194                                                 drophdr.count, mp);
2195                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2196                                                 tmp_leaf, &tmphdr, tmphdr.count,
2197                                                 savehdr.count, mp);
2198                 } else {
2199                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2200                                                 tmp_leaf, &tmphdr, 0,
2201                                                 savehdr.count, mp);
2202                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2203                                                 tmp_leaf, &tmphdr, tmphdr.count,
2204                                                 drophdr.count, mp);
2205                 }
2206                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2207                 savehdr = tmphdr; /* struct copy */
2208                 kmem_free(tmp_leaf);
2209         }
2210
2211         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2212         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2213                                            state->blocksize - 1);
2214
2215         /*
2216          * Copy out last hashval in each block for B-tree code.
2217          */
2218         entry = xfs_attr3_leaf_entryp(save_leaf);
2219         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2220 }
2221
2222 /*========================================================================
2223  * Routines used for finding things in the Btree.
2224  *========================================================================*/
2225
2226 /*
2227  * Look up a name in a leaf attribute list structure.
2228  * This is the internal routine, it uses the caller's buffer.
2229  *
2230  * Note that duplicate keys are allowed, but only check within the
2231  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2232  *
2233  * Return in args->index the index into the entry[] array of either
2234  * the found entry, or where the entry should have been (insert before
2235  * that entry).
2236  *
2237  * Don't change the args->value unless we find the attribute.
2238  */
2239 int
2240 xfs_attr3_leaf_lookup_int(
2241         struct xfs_buf          *bp,
2242         struct xfs_da_args      *args)
2243 {
2244         struct xfs_attr_leafblock *leaf;
2245         struct xfs_attr3_icleaf_hdr ichdr;
2246         struct xfs_attr_leaf_entry *entry;
2247         struct xfs_attr_leaf_entry *entries;
2248         struct xfs_attr_leaf_name_local *name_loc;
2249         struct xfs_attr_leaf_name_remote *name_rmt;
2250         xfs_dahash_t            hashval;
2251         int                     probe;
2252         int                     span;
2253
2254         trace_xfs_attr_leaf_lookup(args);
2255
2256         leaf = bp->b_addr;
2257         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2258         entries = xfs_attr3_leaf_entryp(leaf);
2259         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2260
2261         /*
2262          * Binary search.  (note: small blocks will skip this loop)
2263          */
2264         hashval = args->hashval;
2265         probe = span = ichdr.count / 2;
2266         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2267                 span /= 2;
2268                 if (be32_to_cpu(entry->hashval) < hashval)
2269                         probe += span;
2270                 else if (be32_to_cpu(entry->hashval) > hashval)
2271                         probe -= span;
2272                 else
2273                         break;
2274         }
2275         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2276         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2277
2278         /*
2279          * Since we may have duplicate hashval's, find the first matching
2280          * hashval in the leaf.
2281          */
2282         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2283                 entry--;
2284                 probe--;
2285         }
2286         while (probe < ichdr.count &&
2287                be32_to_cpu(entry->hashval) < hashval) {
2288                 entry++;
2289                 probe++;
2290         }
2291         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2292                 args->index = probe;
2293                 return XFS_ERROR(ENOATTR);
2294         }
2295
2296         /*
2297          * Duplicate keys may be present, so search all of them for a match.
2298          */
2299         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2300                         entry++, probe++) {
2301 /*
2302  * GROT: Add code to remove incomplete entries.
2303  */
2304                 /*
2305                  * If we are looking for INCOMPLETE entries, show only those.
2306                  * If we are looking for complete entries, show only those.
2307                  */
2308                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2309                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2310                         continue;
2311                 }
2312                 if (entry->flags & XFS_ATTR_LOCAL) {
2313                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2314                         if (name_loc->namelen != args->namelen)
2315                                 continue;
2316                         if (memcmp(args->name, name_loc->nameval,
2317                                                         args->namelen) != 0)
2318                                 continue;
2319                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2320                                 continue;
2321                         args->index = probe;
2322                         return XFS_ERROR(EEXIST);
2323                 } else {
2324                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2325                         if (name_rmt->namelen != args->namelen)
2326                                 continue;
2327                         if (memcmp(args->name, name_rmt->name,
2328                                                         args->namelen) != 0)
2329                                 continue;
2330                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2331                                 continue;
2332                         args->index = probe;
2333                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2334                         args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2335                                                    be32_to_cpu(name_rmt->valuelen));
2336                         return XFS_ERROR(EEXIST);
2337                 }
2338         }
2339         args->index = probe;
2340         return XFS_ERROR(ENOATTR);
2341 }
2342
2343 /*
2344  * Get the value associated with an attribute name from a leaf attribute
2345  * list structure.
2346  */
2347 int
2348 xfs_attr3_leaf_getvalue(
2349         struct xfs_buf          *bp,
2350         struct xfs_da_args      *args)
2351 {
2352         struct xfs_attr_leafblock *leaf;
2353         struct xfs_attr3_icleaf_hdr ichdr;
2354         struct xfs_attr_leaf_entry *entry;
2355         struct xfs_attr_leaf_name_local *name_loc;
2356         struct xfs_attr_leaf_name_remote *name_rmt;
2357         int                     valuelen;
2358
2359         leaf = bp->b_addr;
2360         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2361         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2362         ASSERT(args->index < ichdr.count);
2363
2364         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2365         if (entry->flags & XFS_ATTR_LOCAL) {
2366                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2367                 ASSERT(name_loc->namelen == args->namelen);
2368                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2369                 valuelen = be16_to_cpu(name_loc->valuelen);
2370                 if (args->flags & ATTR_KERNOVAL) {
2371                         args->valuelen = valuelen;
2372                         return 0;
2373                 }
2374                 if (args->valuelen < valuelen) {
2375                         args->valuelen = valuelen;
2376                         return XFS_ERROR(ERANGE);
2377                 }
2378                 args->valuelen = valuelen;
2379                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2380         } else {
2381                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2382                 ASSERT(name_rmt->namelen == args->namelen);
2383                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2384                 valuelen = be32_to_cpu(name_rmt->valuelen);
2385                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2386                 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2387                 if (args->flags & ATTR_KERNOVAL) {
2388                         args->valuelen = valuelen;
2389                         return 0;
2390                 }
2391                 if (args->valuelen < valuelen) {
2392                         args->valuelen = valuelen;
2393                         return XFS_ERROR(ERANGE);
2394                 }
2395                 args->valuelen = valuelen;
2396         }
2397         return 0;
2398 }
2399
2400 /*========================================================================
2401  * Utility routines.
2402  *========================================================================*/
2403
2404 /*
2405  * Move the indicated entries from one leaf to another.
2406  * NOTE: this routine modifies both source and destination leaves.
2407  */
2408 /*ARGSUSED*/
2409 STATIC void
2410 xfs_attr3_leaf_moveents(
2411         struct xfs_attr_leafblock       *leaf_s,
2412         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2413         int                             start_s,
2414         struct xfs_attr_leafblock       *leaf_d,
2415         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2416         int                             start_d,
2417         int                             count,
2418         struct xfs_mount                *mp)
2419 {
2420         struct xfs_attr_leaf_entry      *entry_s;
2421         struct xfs_attr_leaf_entry      *entry_d;
2422         int                             desti;
2423         int                             tmp;
2424         int                             i;
2425
2426         /*
2427          * Check for nothing to do.
2428          */
2429         if (count == 0)
2430                 return;
2431
2432         /*
2433          * Set up environment.
2434          */
2435         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2436                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2437         ASSERT(ichdr_s->magic == ichdr_d->magic);
2438         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2439         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2440                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2441         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2442         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2443                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2444
2445         ASSERT(start_s < ichdr_s->count);
2446         ASSERT(start_d <= ichdr_d->count);
2447         ASSERT(count <= ichdr_s->count);
2448
2449
2450         /*
2451          * Move the entries in the destination leaf up to make a hole?
2452          */
2453         if (start_d < ichdr_d->count) {
2454                 tmp  = ichdr_d->count - start_d;
2455                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2456                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2457                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2458                 memmove(entry_d, entry_s, tmp);
2459         }
2460
2461         /*
2462          * Copy all entry's in the same (sorted) order,
2463          * but allocate attribute info packed and in sequence.
2464          */
2465         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2466         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2467         desti = start_d;
2468         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2469                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2470                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2471 #ifdef GROT
2472                 /*
2473                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2474                  * may also need to change the insertion index.  Code turned
2475                  * off for 6.2, should be revisited later.
2476                  */
2477                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2478                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2479                         ichdr_s->usedbytes -= tmp;
2480                         ichdr_s->count -= 1;
2481                         entry_d--;      /* to compensate for ++ in loop hdr */
2482                         desti--;
2483                         if ((start_s + i) < offset)
2484                                 result++;       /* insertion index adjustment */
2485                 } else {
2486 #endif /* GROT */
2487                         ichdr_d->firstused -= tmp;
2488                         /* both on-disk, don't endian flip twice */
2489                         entry_d->hashval = entry_s->hashval;
2490                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2491                         entry_d->flags = entry_s->flags;
2492                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2493                                                         <= XFS_LBSIZE(mp));
2494                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2495                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2496                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2497                                                         <= XFS_LBSIZE(mp));
2498                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2499                         ichdr_s->usedbytes -= tmp;
2500                         ichdr_d->usedbytes += tmp;
2501                         ichdr_s->count -= 1;
2502                         ichdr_d->count += 1;
2503                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2504                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2505                         ASSERT(ichdr_d->firstused >= tmp);
2506 #ifdef GROT
2507                 }
2508 #endif /* GROT */
2509         }
2510
2511         /*
2512          * Zero out the entries we just copied.
2513          */
2514         if (start_s == ichdr_s->count) {
2515                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2516                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2517                 ASSERT(((char *)entry_s + tmp) <=
2518                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2519                 memset(entry_s, 0, tmp);
2520         } else {
2521                 /*
2522                  * Move the remaining entries down to fill the hole,
2523                  * then zero the entries at the top.
2524                  */
2525                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2526                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2527                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2528                 memmove(entry_d, entry_s, tmp);
2529
2530                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2531                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2532                 ASSERT(((char *)entry_s + tmp) <=
2533                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2534                 memset(entry_s, 0, tmp);
2535         }
2536
2537         /*
2538          * Fill in the freemap information
2539          */
2540         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2541         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2542         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2543         ichdr_d->freemap[1].base = 0;
2544         ichdr_d->freemap[2].base = 0;
2545         ichdr_d->freemap[1].size = 0;
2546         ichdr_d->freemap[2].size = 0;
2547         ichdr_s->holes = 1;     /* leaf may not be compact */
2548 }
2549
2550 /*
2551  * Pick up the last hashvalue from a leaf block.
2552  */
2553 xfs_dahash_t
2554 xfs_attr_leaf_lasthash(
2555         struct xfs_buf  *bp,
2556         int             *count)
2557 {
2558         struct xfs_attr3_icleaf_hdr ichdr;
2559         struct xfs_attr_leaf_entry *entries;
2560
2561         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2562         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2563         if (count)
2564                 *count = ichdr.count;
2565         if (!ichdr.count)
2566                 return 0;
2567         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2568 }
2569
2570 /*
2571  * Calculate the number of bytes used to store the indicated attribute
2572  * (whether local or remote only calculate bytes in this block).
2573  */
2574 STATIC int
2575 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2576 {
2577         struct xfs_attr_leaf_entry *entries;
2578         xfs_attr_leaf_name_local_t *name_loc;
2579         xfs_attr_leaf_name_remote_t *name_rmt;
2580         int size;
2581
2582         entries = xfs_attr3_leaf_entryp(leaf);
2583         if (entries[index].flags & XFS_ATTR_LOCAL) {
2584                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2585                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2586                                                    be16_to_cpu(name_loc->valuelen));
2587         } else {
2588                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2589                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2590         }
2591         return size;
2592 }
2593
2594 /*
2595  * Calculate the number of bytes that would be required to store the new
2596  * attribute (whether local or remote only calculate bytes in this block).
2597  * This routine decides as a side effect whether the attribute will be
2598  * a "local" or a "remote" attribute.
2599  */
2600 int
2601 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2602 {
2603         int size;
2604
2605         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2606         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2607                 if (local) {
2608                         *local = 1;
2609                 }
2610         } else {
2611                 size = xfs_attr_leaf_entsize_remote(namelen);
2612                 if (local) {
2613                         *local = 0;
2614                 }
2615         }
2616         return size;
2617 }
2618
2619 /*
2620  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2621  */
2622 int
2623 xfs_attr3_leaf_list_int(
2624         struct xfs_buf                  *bp,
2625         struct xfs_attr_list_context    *context)
2626 {
2627         struct attrlist_cursor_kern     *cursor;
2628         struct xfs_attr_leafblock       *leaf;
2629         struct xfs_attr3_icleaf_hdr     ichdr;
2630         struct xfs_attr_leaf_entry      *entries;
2631         struct xfs_attr_leaf_entry      *entry;
2632         int                             retval;
2633         int                             i;
2634
2635         trace_xfs_attr_list_leaf(context);
2636
2637         leaf = bp->b_addr;
2638         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2639         entries = xfs_attr3_leaf_entryp(leaf);
2640
2641         cursor = context->cursor;
2642         cursor->initted = 1;
2643
2644         /*
2645          * Re-find our place in the leaf block if this is a new syscall.
2646          */
2647         if (context->resynch) {
2648                 entry = &entries[0];
2649                 for (i = 0; i < ichdr.count; entry++, i++) {
2650                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2651                                 if (cursor->offset == context->dupcnt) {
2652                                         context->dupcnt = 0;
2653                                         break;
2654                                 }
2655                                 context->dupcnt++;
2656                         } else if (be32_to_cpu(entry->hashval) >
2657                                         cursor->hashval) {
2658                                 context->dupcnt = 0;
2659                                 break;
2660                         }
2661                 }
2662                 if (i == ichdr.count) {
2663                         trace_xfs_attr_list_notfound(context);
2664                         return 0;
2665                 }
2666         } else {
2667                 entry = &entries[0];
2668                 i = 0;
2669         }
2670         context->resynch = 0;
2671
2672         /*
2673          * We have found our place, start copying out the new attributes.
2674          */
2675         retval = 0;
2676         for (; i < ichdr.count; entry++, i++) {
2677                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2678                         cursor->hashval = be32_to_cpu(entry->hashval);
2679                         cursor->offset = 0;
2680                 }
2681
2682                 if (entry->flags & XFS_ATTR_INCOMPLETE)
2683                         continue;               /* skip incomplete entries */
2684
2685                 if (entry->flags & XFS_ATTR_LOCAL) {
2686                         xfs_attr_leaf_name_local_t *name_loc =
2687                                 xfs_attr3_leaf_name_local(leaf, i);
2688
2689                         retval = context->put_listent(context,
2690                                                 entry->flags,
2691                                                 name_loc->nameval,
2692                                                 (int)name_loc->namelen,
2693                                                 be16_to_cpu(name_loc->valuelen),
2694                                                 &name_loc->nameval[name_loc->namelen]);
2695                         if (retval)
2696                                 return retval;
2697                 } else {
2698                         xfs_attr_leaf_name_remote_t *name_rmt =
2699                                 xfs_attr3_leaf_name_remote(leaf, i);
2700
2701                         int valuelen = be32_to_cpu(name_rmt->valuelen);
2702
2703                         if (context->put_value) {
2704                                 xfs_da_args_t args;
2705
2706                                 memset((char *)&args, 0, sizeof(args));
2707                                 args.dp = context->dp;
2708                                 args.whichfork = XFS_ATTR_FORK;
2709                                 args.valuelen = valuelen;
2710                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2711                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2712                                 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2713                                 retval = xfs_attr_rmtval_get(&args);
2714                                 if (retval)
2715                                         return retval;
2716                                 retval = context->put_listent(context,
2717                                                 entry->flags,
2718                                                 name_rmt->name,
2719                                                 (int)name_rmt->namelen,
2720                                                 valuelen,
2721                                                 args.value);
2722                                 kmem_free(args.value);
2723                         } else {
2724                                 retval = context->put_listent(context,
2725                                                 entry->flags,
2726                                                 name_rmt->name,
2727                                                 (int)name_rmt->namelen,
2728                                                 valuelen,
2729                                                 NULL);
2730                         }
2731                         if (retval)
2732                                 return retval;
2733                 }
2734                 if (context->seen_enough)
2735                         break;
2736                 cursor->offset++;
2737         }
2738         trace_xfs_attr_list_leaf_end(context);
2739         return retval;
2740 }
2741
2742
2743 /*========================================================================
2744  * Manage the INCOMPLETE flag in a leaf entry
2745  *========================================================================*/
2746
2747 /*
2748  * Clear the INCOMPLETE flag on an entry in a leaf block.
2749  */
2750 int
2751 xfs_attr3_leaf_clearflag(
2752         struct xfs_da_args      *args)
2753 {
2754         struct xfs_attr_leafblock *leaf;
2755         struct xfs_attr_leaf_entry *entry;
2756         struct xfs_attr_leaf_name_remote *name_rmt;
2757         struct xfs_buf          *bp;
2758         int                     error;
2759 #ifdef DEBUG
2760         struct xfs_attr3_icleaf_hdr ichdr;
2761         xfs_attr_leaf_name_local_t *name_loc;
2762         int namelen;
2763         char *name;
2764 #endif /* DEBUG */
2765
2766         trace_xfs_attr_leaf_clearflag(args);
2767         /*
2768          * Set up the operation.
2769          */
2770         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2771         if (error)
2772                 return(error);
2773
2774         leaf = bp->b_addr;
2775         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2776         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2777
2778 #ifdef DEBUG
2779         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2780         ASSERT(args->index < ichdr.count);
2781         ASSERT(args->index >= 0);
2782
2783         if (entry->flags & XFS_ATTR_LOCAL) {
2784                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2785                 namelen = name_loc->namelen;
2786                 name = (char *)name_loc->nameval;
2787         } else {
2788                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2789                 namelen = name_rmt->namelen;
2790                 name = (char *)name_rmt->name;
2791         }
2792         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2793         ASSERT(namelen == args->namelen);
2794         ASSERT(memcmp(name, args->name, namelen) == 0);
2795 #endif /* DEBUG */
2796
2797         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2798         xfs_trans_log_buf(args->trans, bp,
2799                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2800
2801         if (args->rmtblkno) {
2802                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2803                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2804                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2805                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2806                 xfs_trans_log_buf(args->trans, bp,
2807                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2808         }
2809
2810         /*
2811          * Commit the flag value change and start the next trans in series.
2812          */
2813         return xfs_trans_roll(&args->trans, args->dp);
2814 }
2815
2816 /*
2817  * Set the INCOMPLETE flag on an entry in a leaf block.
2818  */
2819 int
2820 xfs_attr3_leaf_setflag(
2821         struct xfs_da_args      *args)
2822 {
2823         struct xfs_attr_leafblock *leaf;
2824         struct xfs_attr_leaf_entry *entry;
2825         struct xfs_attr_leaf_name_remote *name_rmt;
2826         struct xfs_buf          *bp;
2827         int error;
2828 #ifdef DEBUG
2829         struct xfs_attr3_icleaf_hdr ichdr;
2830 #endif
2831
2832         trace_xfs_attr_leaf_setflag(args);
2833
2834         /*
2835          * Set up the operation.
2836          */
2837         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2838         if (error)
2839                 return(error);
2840
2841         leaf = bp->b_addr;
2842 #ifdef DEBUG
2843         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2844         ASSERT(args->index < ichdr.count);
2845         ASSERT(args->index >= 0);
2846 #endif
2847         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2848
2849         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2850         entry->flags |= XFS_ATTR_INCOMPLETE;
2851         xfs_trans_log_buf(args->trans, bp,
2852                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2853         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2854                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2855                 name_rmt->valueblk = 0;
2856                 name_rmt->valuelen = 0;
2857                 xfs_trans_log_buf(args->trans, bp,
2858                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2859         }
2860
2861         /*
2862          * Commit the flag value change and start the next trans in series.
2863          */
2864         return xfs_trans_roll(&args->trans, args->dp);
2865 }
2866
2867 /*
2868  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2869  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2870  * entry given by args->blkno2/index2.
2871  *
2872  * Note that they could be in different blocks, or in the same block.
2873  */
2874 int
2875 xfs_attr3_leaf_flipflags(
2876         struct xfs_da_args      *args)
2877 {
2878         struct xfs_attr_leafblock *leaf1;
2879         struct xfs_attr_leafblock *leaf2;
2880         struct xfs_attr_leaf_entry *entry1;
2881         struct xfs_attr_leaf_entry *entry2;
2882         struct xfs_attr_leaf_name_remote *name_rmt;
2883         struct xfs_buf          *bp1;
2884         struct xfs_buf          *bp2;
2885         int error;
2886 #ifdef DEBUG
2887         struct xfs_attr3_icleaf_hdr ichdr1;
2888         struct xfs_attr3_icleaf_hdr ichdr2;
2889         xfs_attr_leaf_name_local_t *name_loc;
2890         int namelen1, namelen2;
2891         char *name1, *name2;
2892 #endif /* DEBUG */
2893
2894         trace_xfs_attr_leaf_flipflags(args);
2895
2896         /*
2897          * Read the block containing the "old" attr
2898          */
2899         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2900         if (error)
2901                 return error;
2902
2903         /*
2904          * Read the block containing the "new" attr, if it is different
2905          */
2906         if (args->blkno2 != args->blkno) {
2907                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2908                                            -1, &bp2);
2909                 if (error)
2910                         return error;
2911         } else {
2912                 bp2 = bp1;
2913         }
2914
2915         leaf1 = bp1->b_addr;
2916         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2917
2918         leaf2 = bp2->b_addr;
2919         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2920
2921 #ifdef DEBUG
2922         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2923         ASSERT(args->index < ichdr1.count);
2924         ASSERT(args->index >= 0);
2925
2926         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2927         ASSERT(args->index2 < ichdr2.count);
2928         ASSERT(args->index2 >= 0);
2929
2930         if (entry1->flags & XFS_ATTR_LOCAL) {
2931                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2932                 namelen1 = name_loc->namelen;
2933                 name1 = (char *)name_loc->nameval;
2934         } else {
2935                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2936                 namelen1 = name_rmt->namelen;
2937                 name1 = (char *)name_rmt->name;
2938         }
2939         if (entry2->flags & XFS_ATTR_LOCAL) {
2940                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2941                 namelen2 = name_loc->namelen;
2942                 name2 = (char *)name_loc->nameval;
2943         } else {
2944                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2945                 namelen2 = name_rmt->namelen;
2946                 name2 = (char *)name_rmt->name;
2947         }
2948         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2949         ASSERT(namelen1 == namelen2);
2950         ASSERT(memcmp(name1, name2, namelen1) == 0);
2951 #endif /* DEBUG */
2952
2953         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2954         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2955
2956         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2957         xfs_trans_log_buf(args->trans, bp1,
2958                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2959         if (args->rmtblkno) {
2960                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2961                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2962                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2963                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2964                 xfs_trans_log_buf(args->trans, bp1,
2965                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2966         }
2967
2968         entry2->flags |= XFS_ATTR_INCOMPLETE;
2969         xfs_trans_log_buf(args->trans, bp2,
2970                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2971         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2972                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2973                 name_rmt->valueblk = 0;
2974                 name_rmt->valuelen = 0;
2975                 xfs_trans_log_buf(args->trans, bp2,
2976                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2977         }
2978
2979         /*
2980          * Commit the flag value change and start the next trans in series.
2981          */
2982         error = xfs_trans_roll(&args->trans, args->dp);
2983
2984         return error;
2985 }
2986
2987 /*========================================================================
2988  * Indiscriminately delete the entire attribute fork
2989  *========================================================================*/
2990
2991 /*
2992  * Recurse (gasp!) through the attribute nodes until we find leaves.
2993  * We're doing a depth-first traversal in order to invalidate everything.
2994  */
2995 int
2996 xfs_attr3_root_inactive(
2997         struct xfs_trans        **trans,
2998         struct xfs_inode        *dp)
2999 {
3000         struct xfs_da_blkinfo   *info;
3001         struct xfs_buf          *bp;
3002         xfs_daddr_t             blkno;
3003         int                     error;
3004
3005         /*
3006          * Read block 0 to see what we have to work with.
3007          * We only get here if we have extents, since we remove
3008          * the extents in reverse order the extent containing
3009          * block 0 must still be there.
3010          */
3011         error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
3012         if (error)
3013                 return error;
3014         blkno = bp->b_bn;
3015
3016         /*
3017          * Invalidate the tree, even if the "tree" is only a single leaf block.
3018          * This is a depth-first traversal!
3019          */
3020         info = bp->b_addr;
3021         switch (info->magic) {
3022         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3023         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3024                 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
3025                 break;
3026         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3027         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3028                 error = xfs_attr3_leaf_inactive(trans, dp, bp);
3029                 break;
3030         default:
3031                 error = XFS_ERROR(EIO);
3032                 xfs_trans_brelse(*trans, bp);
3033                 break;
3034         }
3035         if (error)
3036                 return error;
3037
3038         /*
3039          * Invalidate the incore copy of the root block.
3040          */
3041         error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
3042         if (error)
3043                 return error;
3044         xfs_trans_binval(*trans, bp);   /* remove from cache */
3045         /*
3046          * Commit the invalidate and start the next transaction.
3047          */
3048         error = xfs_trans_roll(trans, dp);
3049
3050         return error;
3051 }
3052
3053 /*
3054  * Recurse (gasp!) through the attribute nodes until we find leaves.
3055  * We're doing a depth-first traversal in order to invalidate everything.
3056  */
3057 STATIC int
3058 xfs_attr3_node_inactive(
3059         struct xfs_trans **trans,
3060         struct xfs_inode *dp,
3061         struct xfs_buf  *bp,
3062         int             level)
3063 {
3064         xfs_da_blkinfo_t *info;
3065         xfs_da_intnode_t *node;
3066         xfs_dablk_t child_fsb;
3067         xfs_daddr_t parent_blkno, child_blkno;
3068         int error, i;
3069         struct xfs_buf *child_bp;
3070         struct xfs_da_node_entry *btree;
3071         struct xfs_da3_icnode_hdr ichdr;
3072
3073         /*
3074          * Since this code is recursive (gasp!) we must protect ourselves.
3075          */
3076         if (level > XFS_DA_NODE_MAXDEPTH) {
3077                 xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3078                 return XFS_ERROR(EIO);
3079         }
3080
3081         node = bp->b_addr;
3082         xfs_da3_node_hdr_from_disk(&ichdr, node);
3083         parent_blkno = bp->b_bn;
3084         if (!ichdr.count) {
3085                 xfs_trans_brelse(*trans, bp);
3086                 return 0;
3087         }
3088         btree = xfs_da3_node_tree_p(node);
3089         child_fsb = be32_to_cpu(btree[0].before);
3090         xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3091
3092         /*
3093          * If this is the node level just above the leaves, simply loop
3094          * over the leaves removing all of them.  If this is higher up
3095          * in the tree, recurse downward.
3096          */
3097         for (i = 0; i < ichdr.count; i++) {
3098                 /*
3099                  * Read the subsidiary block to see what we have to work with.
3100                  * Don't do this in a transaction.  This is a depth-first
3101                  * traversal of the tree so we may deal with many blocks
3102                  * before we come back to this one.
3103                  */
3104                 error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp,
3105                                                 XFS_ATTR_FORK);
3106                 if (error)
3107                         return(error);
3108                 if (child_bp) {
3109                                                 /* save for re-read later */
3110                         child_blkno = XFS_BUF_ADDR(child_bp);
3111
3112                         /*
3113                          * Invalidate the subtree, however we have to.
3114                          */
3115                         info = child_bp->b_addr;
3116                         switch (info->magic) {
3117                         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3118                         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3119                                 error = xfs_attr3_node_inactive(trans, dp,
3120                                                         child_bp, level + 1);
3121                                 break;
3122                         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3123                         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3124                                 error = xfs_attr3_leaf_inactive(trans, dp,
3125                                                         child_bp);
3126                                 break;
3127                         default:
3128                                 error = XFS_ERROR(EIO);
3129                                 xfs_trans_brelse(*trans, child_bp);
3130                                 break;
3131                         }
3132                         if (error)
3133                                 return error;
3134
3135                         /*
3136                          * Remove the subsidiary block from the cache
3137                          * and from the log.
3138                          */
3139                         error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
3140                                 &child_bp, XFS_ATTR_FORK);
3141                         if (error)
3142                                 return error;
3143                         xfs_trans_binval(*trans, child_bp);
3144                 }
3145
3146                 /*
3147                  * If we're not done, re-read the parent to get the next
3148                  * child block number.
3149                  */
3150                 if (i + 1 < ichdr.count) {
3151                         error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
3152                                                  &bp, XFS_ATTR_FORK);
3153                         if (error)
3154                                 return error;
3155                         child_fsb = be32_to_cpu(btree[i + 1].before);
3156                         xfs_trans_brelse(*trans, bp);
3157                 }
3158                 /*
3159                  * Atomically commit the whole invalidate stuff.
3160                  */
3161                 error = xfs_trans_roll(trans, dp);
3162                 if (error)
3163                         return  error;
3164         }
3165
3166         return 0;
3167 }
3168
3169 /*
3170  * Invalidate all of the "remote" value regions pointed to by a particular
3171  * leaf block.
3172  * Note that we must release the lock on the buffer so that we are not
3173  * caught holding something that the logging code wants to flush to disk.
3174  */
3175 STATIC int
3176 xfs_attr3_leaf_inactive(
3177         struct xfs_trans        **trans,
3178         struct xfs_inode        *dp,
3179         struct xfs_buf          *bp)
3180 {
3181         struct xfs_attr_leafblock *leaf;
3182         struct xfs_attr3_icleaf_hdr ichdr;
3183         struct xfs_attr_leaf_entry *entry;
3184         struct xfs_attr_leaf_name_remote *name_rmt;
3185         struct xfs_attr_inactive_list *list;
3186         struct xfs_attr_inactive_list *lp;
3187         int                     error;
3188         int                     count;
3189         int                     size;
3190         int                     tmp;
3191         int                     i;
3192
3193         leaf = bp->b_addr;
3194         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
3195
3196         /*
3197          * Count the number of "remote" value extents.
3198          */
3199         count = 0;
3200         entry = xfs_attr3_leaf_entryp(leaf);
3201         for (i = 0; i < ichdr.count; entry++, i++) {
3202                 if (be16_to_cpu(entry->nameidx) &&
3203                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3204                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3205                         if (name_rmt->valueblk)
3206                                 count++;
3207                 }
3208         }
3209
3210         /*
3211          * If there are no "remote" values, we're done.
3212          */
3213         if (count == 0) {
3214                 xfs_trans_brelse(*trans, bp);
3215                 return 0;
3216         }
3217
3218         /*
3219          * Allocate storage for a list of all the "remote" value extents.
3220          */
3221         size = count * sizeof(xfs_attr_inactive_list_t);
3222         list = kmem_alloc(size, KM_SLEEP);
3223
3224         /*
3225          * Identify each of the "remote" value extents.
3226          */
3227         lp = list;
3228         entry = xfs_attr3_leaf_entryp(leaf);
3229         for (i = 0; i < ichdr.count; entry++, i++) {
3230                 if (be16_to_cpu(entry->nameidx) &&
3231                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3232                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3233                         if (name_rmt->valueblk) {
3234                                 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
3235                                 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
3236                                                     be32_to_cpu(name_rmt->valuelen));
3237                                 lp++;
3238                         }
3239                 }
3240         }
3241         xfs_trans_brelse(*trans, bp);   /* unlock for trans. in freextent() */
3242
3243         /*
3244          * Invalidate each of the "remote" value extents.
3245          */
3246         error = 0;
3247         for (lp = list, i = 0; i < count; i++, lp++) {
3248                 tmp = xfs_attr3_leaf_freextent(trans, dp,
3249                                 lp->valueblk, lp->valuelen);
3250
3251                 if (error == 0)
3252                         error = tmp;    /* save only the 1st errno */
3253         }
3254
3255         kmem_free(list);
3256         return error;
3257 }
3258
3259 /*
3260  * Look at all the extents for this logical region,
3261  * invalidate any buffers that are incore/in transactions.
3262  */
3263 STATIC int
3264 xfs_attr3_leaf_freextent(
3265         struct xfs_trans        **trans,
3266         struct xfs_inode        *dp,
3267         xfs_dablk_t             blkno,
3268         int                     blkcnt)
3269 {
3270         struct xfs_bmbt_irec    map;
3271         struct xfs_buf          *bp;
3272         xfs_dablk_t             tblkno;
3273         xfs_daddr_t             dblkno;
3274         int                     tblkcnt;
3275         int                     dblkcnt;
3276         int                     nmap;
3277         int                     error;
3278
3279         /*
3280          * Roll through the "value", invalidating the attribute value's
3281          * blocks.
3282          */
3283         tblkno = blkno;
3284         tblkcnt = blkcnt;
3285         while (tblkcnt > 0) {
3286                 /*
3287                  * Try to remember where we decided to put the value.
3288                  */
3289                 nmap = 1;
3290                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
3291                                        &map, &nmap, XFS_BMAPI_ATTRFORK);
3292                 if (error) {
3293                         return(error);
3294                 }
3295                 ASSERT(nmap == 1);
3296                 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3297
3298                 /*
3299                  * If it's a hole, these are already unmapped
3300                  * so there's nothing to invalidate.
3301                  */
3302                 if (map.br_startblock != HOLESTARTBLOCK) {
3303
3304                         dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3305                                                   map.br_startblock);
3306                         dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3307                                                 map.br_blockcount);
3308                         bp = xfs_trans_get_buf(*trans,
3309                                         dp->i_mount->m_ddev_targp,
3310                                         dblkno, dblkcnt, 0);
3311                         if (!bp)
3312                                 return ENOMEM;
3313                         xfs_trans_binval(*trans, bp);
3314                         /*
3315                          * Roll to next transaction.
3316                          */
3317                         error = xfs_trans_roll(trans, dp);
3318                         if (error)
3319                                 return (error);
3320                 }
3321
3322                 tblkno += map.br_blockcount;
3323                 tblkcnt -= map.br_blockcount;
3324         }
3325
3326         return(0);
3327 }