Merge branch 'lsk-3.10-armlt-fixes' of git://git.linaro.org/landing-teams/working...
[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         struct xfs_attr_leafblock *leaf;
938         struct xfs_attr_leaf_entry *entry;
939         xfs_attr_leaf_name_local_t *name_loc;
940         struct xfs_attr3_icleaf_hdr leafhdr;
941         int                     bytes;
942         int                     i;
943
944         leaf = bp->b_addr;
945         xfs_attr3_leaf_hdr_from_disk(&leafhdr, leaf);
946         entry = xfs_attr3_leaf_entryp(leaf);
947
948         bytes = sizeof(struct xfs_attr_sf_hdr);
949         for (i = 0; i < leafhdr.count; entry++, i++) {
950                 if (entry->flags & XFS_ATTR_INCOMPLETE)
951                         continue;               /* don't copy partial entries */
952                 if (!(entry->flags & XFS_ATTR_LOCAL))
953                         return(0);
954                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
955                 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
956                         return(0);
957                 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
958                         return(0);
959                 bytes += sizeof(struct xfs_attr_sf_entry) - 1
960                                 + name_loc->namelen
961                                 + be16_to_cpu(name_loc->valuelen);
962         }
963         if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
964             (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
965             (bytes == sizeof(struct xfs_attr_sf_hdr)))
966                 return -1;
967         return xfs_attr_shortform_bytesfit(dp, bytes);
968 }
969
970 /*
971  * Convert a leaf attribute list to shortform attribute list
972  */
973 int
974 xfs_attr3_leaf_to_shortform(
975         struct xfs_buf          *bp,
976         struct xfs_da_args      *args,
977         int                     forkoff)
978 {
979         struct xfs_attr_leafblock *leaf;
980         struct xfs_attr3_icleaf_hdr ichdr;
981         struct xfs_attr_leaf_entry *entry;
982         struct xfs_attr_leaf_name_local *name_loc;
983         struct xfs_da_args      nargs;
984         struct xfs_inode        *dp = args->dp;
985         char                    *tmpbuffer;
986         int                     error;
987         int                     i;
988
989         trace_xfs_attr_leaf_to_sf(args);
990
991         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
992         if (!tmpbuffer)
993                 return ENOMEM;
994
995         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(dp->i_mount));
996
997         leaf = (xfs_attr_leafblock_t *)tmpbuffer;
998         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
999         entry = xfs_attr3_leaf_entryp(leaf);
1000
1001         /* XXX (dgc): buffer is about to be marked stale - why zero it? */
1002         memset(bp->b_addr, 0, XFS_LBSIZE(dp->i_mount));
1003
1004         /*
1005          * Clean out the prior contents of the attribute list.
1006          */
1007         error = xfs_da_shrink_inode(args, 0, bp);
1008         if (error)
1009                 goto out;
1010
1011         if (forkoff == -1) {
1012                 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
1013                 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
1014                 xfs_attr_fork_reset(dp, args->trans);
1015                 goto out;
1016         }
1017
1018         xfs_attr_shortform_create(args);
1019
1020         /*
1021          * Copy the attributes
1022          */
1023         memset((char *)&nargs, 0, sizeof(nargs));
1024         nargs.dp = dp;
1025         nargs.firstblock = args->firstblock;
1026         nargs.flist = args->flist;
1027         nargs.total = args->total;
1028         nargs.whichfork = XFS_ATTR_FORK;
1029         nargs.trans = args->trans;
1030         nargs.op_flags = XFS_DA_OP_OKNOENT;
1031
1032         for (i = 0; i < ichdr.count; entry++, i++) {
1033                 if (entry->flags & XFS_ATTR_INCOMPLETE)
1034                         continue;       /* don't copy partial entries */
1035                 if (!entry->nameidx)
1036                         continue;
1037                 ASSERT(entry->flags & XFS_ATTR_LOCAL);
1038                 name_loc = xfs_attr3_leaf_name_local(leaf, i);
1039                 nargs.name = name_loc->nameval;
1040                 nargs.namelen = name_loc->namelen;
1041                 nargs.value = &name_loc->nameval[nargs.namelen];
1042                 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
1043                 nargs.hashval = be32_to_cpu(entry->hashval);
1044                 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
1045                 xfs_attr_shortform_add(&nargs, forkoff);
1046         }
1047         error = 0;
1048
1049 out:
1050         kmem_free(tmpbuffer);
1051         return error;
1052 }
1053
1054 /*
1055  * Convert from using a single leaf to a root node and a leaf.
1056  */
1057 int
1058 xfs_attr3_leaf_to_node(
1059         struct xfs_da_args      *args)
1060 {
1061         struct xfs_attr_leafblock *leaf;
1062         struct xfs_attr3_icleaf_hdr icleafhdr;
1063         struct xfs_attr_leaf_entry *entries;
1064         struct xfs_da_node_entry *btree;
1065         struct xfs_da3_icnode_hdr icnodehdr;
1066         struct xfs_da_intnode   *node;
1067         struct xfs_inode        *dp = args->dp;
1068         struct xfs_mount        *mp = dp->i_mount;
1069         struct xfs_buf          *bp1 = NULL;
1070         struct xfs_buf          *bp2 = NULL;
1071         xfs_dablk_t             blkno;
1072         int                     error;
1073
1074         trace_xfs_attr_leaf_to_node(args);
1075
1076         error = xfs_da_grow_inode(args, &blkno);
1077         if (error)
1078                 goto out;
1079         error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
1080         if (error)
1081                 goto out;
1082
1083         error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
1084         if (error)
1085                 goto out;
1086
1087         /* copy leaf to new buffer, update identifiers */
1088         xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
1089         bp2->b_ops = bp1->b_ops;
1090         memcpy(bp2->b_addr, bp1->b_addr, XFS_LBSIZE(mp));
1091         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1092                 struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
1093                 hdr3->blkno = cpu_to_be64(bp2->b_bn);
1094         }
1095         xfs_trans_log_buf(args->trans, bp2, 0, XFS_LBSIZE(mp) - 1);
1096
1097         /*
1098          * Set up the new root node.
1099          */
1100         error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1101         if (error)
1102                 goto out;
1103         node = bp1->b_addr;
1104         xfs_da3_node_hdr_from_disk(&icnodehdr, node);
1105         btree = xfs_da3_node_tree_p(node);
1106
1107         leaf = bp2->b_addr;
1108         xfs_attr3_leaf_hdr_from_disk(&icleafhdr, leaf);
1109         entries = xfs_attr3_leaf_entryp(leaf);
1110
1111         /* both on-disk, don't endian-flip twice */
1112         btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1113         btree[0].before = cpu_to_be32(blkno);
1114         icnodehdr.count = 1;
1115         xfs_da3_node_hdr_to_disk(node, &icnodehdr);
1116         xfs_trans_log_buf(args->trans, bp1, 0, XFS_LBSIZE(mp) - 1);
1117         error = 0;
1118 out:
1119         return error;
1120 }
1121
1122
1123 /*========================================================================
1124  * Routines used for growing the Btree.
1125  *========================================================================*/
1126
1127 /*
1128  * Create the initial contents of a leaf attribute list
1129  * or a leaf in a node attribute list.
1130  */
1131 STATIC int
1132 xfs_attr3_leaf_create(
1133         struct xfs_da_args      *args,
1134         xfs_dablk_t             blkno,
1135         struct xfs_buf          **bpp)
1136 {
1137         struct xfs_attr_leafblock *leaf;
1138         struct xfs_attr3_icleaf_hdr ichdr;
1139         struct xfs_inode        *dp = args->dp;
1140         struct xfs_mount        *mp = dp->i_mount;
1141         struct xfs_buf          *bp;
1142         int                     error;
1143
1144         trace_xfs_attr_leaf_create(args);
1145
1146         error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1147                                             XFS_ATTR_FORK);
1148         if (error)
1149                 return error;
1150         bp->b_ops = &xfs_attr3_leaf_buf_ops;
1151         xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1152         leaf = bp->b_addr;
1153         memset(leaf, 0, XFS_LBSIZE(mp));
1154
1155         memset(&ichdr, 0, sizeof(ichdr));
1156         ichdr.firstused = XFS_LBSIZE(mp);
1157
1158         if (xfs_sb_version_hascrc(&mp->m_sb)) {
1159                 struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1160
1161                 ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1162
1163                 hdr3->blkno = cpu_to_be64(bp->b_bn);
1164                 hdr3->owner = cpu_to_be64(dp->i_ino);
1165                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
1166
1167                 ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1168         } else {
1169                 ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1170                 ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1171         }
1172         ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1173
1174         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1175         xfs_trans_log_buf(args->trans, bp, 0, XFS_LBSIZE(mp) - 1);
1176
1177         *bpp = bp;
1178         return 0;
1179 }
1180
1181 /*
1182  * Split the leaf node, rebalance, then add the new entry.
1183  */
1184 int
1185 xfs_attr3_leaf_split(
1186         struct xfs_da_state     *state,
1187         struct xfs_da_state_blk *oldblk,
1188         struct xfs_da_state_blk *newblk)
1189 {
1190         xfs_dablk_t blkno;
1191         int error;
1192
1193         trace_xfs_attr_leaf_split(state->args);
1194
1195         /*
1196          * Allocate space for a new leaf node.
1197          */
1198         ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1199         error = xfs_da_grow_inode(state->args, &blkno);
1200         if (error)
1201                 return(error);
1202         error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1203         if (error)
1204                 return(error);
1205         newblk->blkno = blkno;
1206         newblk->magic = XFS_ATTR_LEAF_MAGIC;
1207
1208         /*
1209          * Rebalance the entries across the two leaves.
1210          * NOTE: rebalance() currently depends on the 2nd block being empty.
1211          */
1212         xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1213         error = xfs_da3_blk_link(state, oldblk, newblk);
1214         if (error)
1215                 return(error);
1216
1217         /*
1218          * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1219          * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1220          * "new" attrs info.  Will need the "old" info to remove it later.
1221          *
1222          * Insert the "new" entry in the correct block.
1223          */
1224         if (state->inleaf) {
1225                 trace_xfs_attr_leaf_add_old(state->args);
1226                 error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1227         } else {
1228                 trace_xfs_attr_leaf_add_new(state->args);
1229                 error = xfs_attr3_leaf_add(newblk->bp, state->args);
1230         }
1231
1232         /*
1233          * Update last hashval in each block since we added the name.
1234          */
1235         oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1236         newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1237         return(error);
1238 }
1239
1240 /*
1241  * Add a name to the leaf attribute list structure.
1242  */
1243 int
1244 xfs_attr3_leaf_add(
1245         struct xfs_buf          *bp,
1246         struct xfs_da_args      *args)
1247 {
1248         struct xfs_attr_leafblock *leaf;
1249         struct xfs_attr3_icleaf_hdr ichdr;
1250         int                     tablesize;
1251         int                     entsize;
1252         int                     sum;
1253         int                     tmp;
1254         int                     i;
1255
1256         trace_xfs_attr_leaf_add(args);
1257
1258         leaf = bp->b_addr;
1259         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1260         ASSERT(args->index >= 0 && args->index <= ichdr.count);
1261         entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1262                            args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1263
1264         /*
1265          * Search through freemap for first-fit on new name length.
1266          * (may need to figure in size of entry struct too)
1267          */
1268         tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1269                                         + xfs_attr3_leaf_hdr_size(leaf);
1270         for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1271                 if (tablesize > ichdr.firstused) {
1272                         sum += ichdr.freemap[i].size;
1273                         continue;
1274                 }
1275                 if (!ichdr.freemap[i].size)
1276                         continue;       /* no space in this map */
1277                 tmp = entsize;
1278                 if (ichdr.freemap[i].base < ichdr.firstused)
1279                         tmp += sizeof(xfs_attr_leaf_entry_t);
1280                 if (ichdr.freemap[i].size >= tmp) {
1281                         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1282                         goto out_log_hdr;
1283                 }
1284                 sum += ichdr.freemap[i].size;
1285         }
1286
1287         /*
1288          * If there are no holes in the address space of the block,
1289          * and we don't have enough freespace, then compaction will do us
1290          * no good and we should just give up.
1291          */
1292         if (!ichdr.holes && sum < entsize)
1293                 return XFS_ERROR(ENOSPC);
1294
1295         /*
1296          * Compact the entries to coalesce free space.
1297          * This may change the hdr->count via dropping INCOMPLETE entries.
1298          */
1299         xfs_attr3_leaf_compact(args, &ichdr, bp);
1300
1301         /*
1302          * After compaction, the block is guaranteed to have only one
1303          * free region, in freemap[0].  If it is not big enough, give up.
1304          */
1305         if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1306                 tmp = ENOSPC;
1307                 goto out_log_hdr;
1308         }
1309
1310         tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1311
1312 out_log_hdr:
1313         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
1314         xfs_trans_log_buf(args->trans, bp,
1315                 XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1316                                 xfs_attr3_leaf_hdr_size(leaf)));
1317         return tmp;
1318 }
1319
1320 /*
1321  * Add a name to a leaf attribute list structure.
1322  */
1323 STATIC int
1324 xfs_attr3_leaf_add_work(
1325         struct xfs_buf          *bp,
1326         struct xfs_attr3_icleaf_hdr *ichdr,
1327         struct xfs_da_args      *args,
1328         int                     mapindex)
1329 {
1330         struct xfs_attr_leafblock *leaf;
1331         struct xfs_attr_leaf_entry *entry;
1332         struct xfs_attr_leaf_name_local *name_loc;
1333         struct xfs_attr_leaf_name_remote *name_rmt;
1334         struct xfs_mount        *mp;
1335         int                     tmp;
1336         int                     i;
1337
1338         trace_xfs_attr_leaf_add_work(args);
1339
1340         leaf = bp->b_addr;
1341         ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1342         ASSERT(args->index >= 0 && args->index <= ichdr->count);
1343
1344         /*
1345          * Force open some space in the entry array and fill it in.
1346          */
1347         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1348         if (args->index < ichdr->count) {
1349                 tmp  = ichdr->count - args->index;
1350                 tmp *= sizeof(xfs_attr_leaf_entry_t);
1351                 memmove(entry + 1, entry, tmp);
1352                 xfs_trans_log_buf(args->trans, bp,
1353                     XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1354         }
1355         ichdr->count++;
1356
1357         /*
1358          * Allocate space for the new string (at the end of the run).
1359          */
1360         mp = args->trans->t_mountp;
1361         ASSERT(ichdr->freemap[mapindex].base < XFS_LBSIZE(mp));
1362         ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1363         ASSERT(ichdr->freemap[mapindex].size >=
1364                 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1365                                          mp->m_sb.sb_blocksize, NULL));
1366         ASSERT(ichdr->freemap[mapindex].size < XFS_LBSIZE(mp));
1367         ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1368
1369         ichdr->freemap[mapindex].size -=
1370                         xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1371                                                  mp->m_sb.sb_blocksize, &tmp);
1372
1373         entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1374                                      ichdr->freemap[mapindex].size);
1375         entry->hashval = cpu_to_be32(args->hashval);
1376         entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1377         entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1378         if (args->op_flags & XFS_DA_OP_RENAME) {
1379                 entry->flags |= XFS_ATTR_INCOMPLETE;
1380                 if ((args->blkno2 == args->blkno) &&
1381                     (args->index2 <= args->index)) {
1382                         args->index2++;
1383                 }
1384         }
1385         xfs_trans_log_buf(args->trans, bp,
1386                           XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1387         ASSERT((args->index == 0) ||
1388                (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1389         ASSERT((args->index == ichdr->count - 1) ||
1390                (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1391
1392         /*
1393          * For "remote" attribute values, simply note that we need to
1394          * allocate space for the "remote" value.  We can't actually
1395          * allocate the extents in this transaction, and we can't decide
1396          * which blocks they should be as we might allocate more blocks
1397          * as part of this transaction (a split operation for example).
1398          */
1399         if (entry->flags & XFS_ATTR_LOCAL) {
1400                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1401                 name_loc->namelen = args->namelen;
1402                 name_loc->valuelen = cpu_to_be16(args->valuelen);
1403                 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1404                 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1405                                    be16_to_cpu(name_loc->valuelen));
1406         } else {
1407                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1408                 name_rmt->namelen = args->namelen;
1409                 memcpy((char *)name_rmt->name, args->name, args->namelen);
1410                 entry->flags |= XFS_ATTR_INCOMPLETE;
1411                 /* just in case */
1412                 name_rmt->valuelen = 0;
1413                 name_rmt->valueblk = 0;
1414                 args->rmtblkno = 1;
1415                 args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1416         }
1417         xfs_trans_log_buf(args->trans, bp,
1418              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1419                                    xfs_attr_leaf_entsize(leaf, args->index)));
1420
1421         /*
1422          * Update the control info for this leaf node
1423          */
1424         if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1425                 ichdr->firstused = be16_to_cpu(entry->nameidx);
1426
1427         ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1428                                         + xfs_attr3_leaf_hdr_size(leaf));
1429         tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1430                                         + xfs_attr3_leaf_hdr_size(leaf);
1431
1432         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1433                 if (ichdr->freemap[i].base == tmp) {
1434                         ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1435                         ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1436                 }
1437         }
1438         ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1439         return 0;
1440 }
1441
1442 /*
1443  * Garbage collect a leaf attribute list block by copying it to a new buffer.
1444  */
1445 STATIC void
1446 xfs_attr3_leaf_compact(
1447         struct xfs_da_args      *args,
1448         struct xfs_attr3_icleaf_hdr *ichdr_dst,
1449         struct xfs_buf          *bp)
1450 {
1451         struct xfs_attr_leafblock *leaf_src;
1452         struct xfs_attr_leafblock *leaf_dst;
1453         struct xfs_attr3_icleaf_hdr ichdr_src;
1454         struct xfs_trans        *trans = args->trans;
1455         struct xfs_mount        *mp = trans->t_mountp;
1456         char                    *tmpbuffer;
1457
1458         trace_xfs_attr_leaf_compact(args);
1459
1460         tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1461         memcpy(tmpbuffer, bp->b_addr, XFS_LBSIZE(mp));
1462         memset(bp->b_addr, 0, XFS_LBSIZE(mp));
1463         leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1464         leaf_dst = bp->b_addr;
1465
1466         /*
1467          * Copy the on-disk header back into the destination buffer to ensure
1468          * all the information in the header that is not part of the incore
1469          * header structure is preserved.
1470          */
1471         memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1472
1473         /* Initialise the incore headers */
1474         ichdr_src = *ichdr_dst; /* struct copy */
1475         ichdr_dst->firstused = XFS_LBSIZE(mp);
1476         ichdr_dst->usedbytes = 0;
1477         ichdr_dst->count = 0;
1478         ichdr_dst->holes = 0;
1479         ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1480         ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1481                                                 ichdr_dst->freemap[0].base;
1482
1483
1484         /* write the header back to initialise the underlying buffer */
1485         xfs_attr3_leaf_hdr_to_disk(leaf_dst, ichdr_dst);
1486
1487         /*
1488          * Copy all entry's in the same (sorted) order,
1489          * but allocate name/value pairs packed and in sequence.
1490          */
1491         xfs_attr3_leaf_moveents(leaf_src, &ichdr_src, 0, leaf_dst, ichdr_dst, 0,
1492                                 ichdr_src.count, mp);
1493         /*
1494          * this logs the entire buffer, but the caller must write the header
1495          * back to the buffer when it is finished modifying it.
1496          */
1497         xfs_trans_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1498
1499         kmem_free(tmpbuffer);
1500 }
1501
1502 /*
1503  * Compare two leaf blocks "order".
1504  * Return 0 unless leaf2 should go before leaf1.
1505  */
1506 static int
1507 xfs_attr3_leaf_order(
1508         struct xfs_buf  *leaf1_bp,
1509         struct xfs_attr3_icleaf_hdr *leaf1hdr,
1510         struct xfs_buf  *leaf2_bp,
1511         struct xfs_attr3_icleaf_hdr *leaf2hdr)
1512 {
1513         struct xfs_attr_leaf_entry *entries1;
1514         struct xfs_attr_leaf_entry *entries2;
1515
1516         entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1517         entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1518         if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1519             ((be32_to_cpu(entries2[0].hashval) <
1520               be32_to_cpu(entries1[0].hashval)) ||
1521              (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1522               be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1523                 return 1;
1524         }
1525         return 0;
1526 }
1527
1528 int
1529 xfs_attr_leaf_order(
1530         struct xfs_buf  *leaf1_bp,
1531         struct xfs_buf  *leaf2_bp)
1532 {
1533         struct xfs_attr3_icleaf_hdr ichdr1;
1534         struct xfs_attr3_icleaf_hdr ichdr2;
1535
1536         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1_bp->b_addr);
1537         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2_bp->b_addr);
1538         return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1539 }
1540
1541 /*
1542  * Redistribute the attribute list entries between two leaf nodes,
1543  * taking into account the size of the new entry.
1544  *
1545  * NOTE: if new block is empty, then it will get the upper half of the
1546  * old block.  At present, all (one) callers pass in an empty second block.
1547  *
1548  * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1549  * to match what it is doing in splitting the attribute leaf block.  Those
1550  * values are used in "atomic rename" operations on attributes.  Note that
1551  * the "new" and "old" values can end up in different blocks.
1552  */
1553 STATIC void
1554 xfs_attr3_leaf_rebalance(
1555         struct xfs_da_state     *state,
1556         struct xfs_da_state_blk *blk1,
1557         struct xfs_da_state_blk *blk2)
1558 {
1559         struct xfs_da_args      *args;
1560         struct xfs_attr_leafblock *leaf1;
1561         struct xfs_attr_leafblock *leaf2;
1562         struct xfs_attr3_icleaf_hdr ichdr1;
1563         struct xfs_attr3_icleaf_hdr ichdr2;
1564         struct xfs_attr_leaf_entry *entries1;
1565         struct xfs_attr_leaf_entry *entries2;
1566         int                     count;
1567         int                     totallen;
1568         int                     max;
1569         int                     space;
1570         int                     swap;
1571
1572         /*
1573          * Set up environment.
1574          */
1575         ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1576         ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1577         leaf1 = blk1->bp->b_addr;
1578         leaf2 = blk2->bp->b_addr;
1579         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
1580         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
1581         ASSERT(ichdr2.count == 0);
1582         args = state->args;
1583
1584         trace_xfs_attr_leaf_rebalance(args);
1585
1586         /*
1587          * Check ordering of blocks, reverse if it makes things simpler.
1588          *
1589          * NOTE: Given that all (current) callers pass in an empty
1590          * second block, this code should never set "swap".
1591          */
1592         swap = 0;
1593         if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1594                 struct xfs_da_state_blk *tmp_blk;
1595                 struct xfs_attr3_icleaf_hdr tmp_ichdr;
1596
1597                 tmp_blk = blk1;
1598                 blk1 = blk2;
1599                 blk2 = tmp_blk;
1600
1601                 /* struct copies to swap them rather than reconverting */
1602                 tmp_ichdr = ichdr1;
1603                 ichdr1 = ichdr2;
1604                 ichdr2 = tmp_ichdr;
1605
1606                 leaf1 = blk1->bp->b_addr;
1607                 leaf2 = blk2->bp->b_addr;
1608                 swap = 1;
1609         }
1610
1611         /*
1612          * Examine entries until we reduce the absolute difference in
1613          * byte usage between the two blocks to a minimum.  Then get
1614          * the direction to copy and the number of elements to move.
1615          *
1616          * "inleaf" is true if the new entry should be inserted into blk1.
1617          * If "swap" is also true, then reverse the sense of "inleaf".
1618          */
1619         state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1620                                                       blk2, &ichdr2,
1621                                                       &count, &totallen);
1622         if (swap)
1623                 state->inleaf = !state->inleaf;
1624
1625         /*
1626          * Move any entries required from leaf to leaf:
1627          */
1628         if (count < ichdr1.count) {
1629                 /*
1630                  * Figure the total bytes to be added to the destination leaf.
1631                  */
1632                 /* number entries being moved */
1633                 count = ichdr1.count - count;
1634                 space  = ichdr1.usedbytes - totallen;
1635                 space += count * sizeof(xfs_attr_leaf_entry_t);
1636
1637                 /*
1638                  * leaf2 is the destination, compact it if it looks tight.
1639                  */
1640                 max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1641                 max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1642                 if (space > max)
1643                         xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1644
1645                 /*
1646                  * Move high entries from leaf1 to low end of leaf2.
1647                  */
1648                 xfs_attr3_leaf_moveents(leaf1, &ichdr1, ichdr1.count - count,
1649                                 leaf2, &ichdr2, 0, count, state->mp);
1650
1651         } else if (count > ichdr1.count) {
1652                 /*
1653                  * I assert that since all callers pass in an empty
1654                  * second buffer, this code should never execute.
1655                  */
1656                 ASSERT(0);
1657
1658                 /*
1659                  * Figure the total bytes to be added to the destination leaf.
1660                  */
1661                 /* number entries being moved */
1662                 count -= ichdr1.count;
1663                 space  = totallen - ichdr1.usedbytes;
1664                 space += count * sizeof(xfs_attr_leaf_entry_t);
1665
1666                 /*
1667                  * leaf1 is the destination, compact it if it looks tight.
1668                  */
1669                 max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1670                 max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1671                 if (space > max)
1672                         xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1673
1674                 /*
1675                  * Move low entries from leaf2 to high end of leaf1.
1676                  */
1677                 xfs_attr3_leaf_moveents(leaf2, &ichdr2, 0, leaf1, &ichdr1,
1678                                         ichdr1.count, count, state->mp);
1679         }
1680
1681         xfs_attr3_leaf_hdr_to_disk(leaf1, &ichdr1);
1682         xfs_attr3_leaf_hdr_to_disk(leaf2, &ichdr2);
1683         xfs_trans_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1684         xfs_trans_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1685
1686         /*
1687          * Copy out last hashval in each block for B-tree code.
1688          */
1689         entries1 = xfs_attr3_leaf_entryp(leaf1);
1690         entries2 = xfs_attr3_leaf_entryp(leaf2);
1691         blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1692         blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1693
1694         /*
1695          * Adjust the expected index for insertion.
1696          * NOTE: this code depends on the (current) situation that the
1697          * second block was originally empty.
1698          *
1699          * If the insertion point moved to the 2nd block, we must adjust
1700          * the index.  We must also track the entry just following the
1701          * new entry for use in an "atomic rename" operation, that entry
1702          * is always the "old" entry and the "new" entry is what we are
1703          * inserting.  The index/blkno fields refer to the "old" entry,
1704          * while the index2/blkno2 fields refer to the "new" entry.
1705          */
1706         if (blk1->index > ichdr1.count) {
1707                 ASSERT(state->inleaf == 0);
1708                 blk2->index = blk1->index - ichdr1.count;
1709                 args->index = args->index2 = blk2->index;
1710                 args->blkno = args->blkno2 = blk2->blkno;
1711         } else if (blk1->index == ichdr1.count) {
1712                 if (state->inleaf) {
1713                         args->index = blk1->index;
1714                         args->blkno = blk1->blkno;
1715                         args->index2 = 0;
1716                         args->blkno2 = blk2->blkno;
1717                 } else {
1718                         /*
1719                          * On a double leaf split, the original attr location
1720                          * is already stored in blkno2/index2, so don't
1721                          * overwrite it overwise we corrupt the tree.
1722                          */
1723                         blk2->index = blk1->index - ichdr1.count;
1724                         args->index = blk2->index;
1725                         args->blkno = blk2->blkno;
1726                         if (!state->extravalid) {
1727                                 /*
1728                                  * set the new attr location to match the old
1729                                  * one and let the higher level split code
1730                                  * decide where in the leaf to place it.
1731                                  */
1732                                 args->index2 = blk2->index;
1733                                 args->blkno2 = blk2->blkno;
1734                         }
1735                 }
1736         } else {
1737                 ASSERT(state->inleaf == 1);
1738                 args->index = args->index2 = blk1->index;
1739                 args->blkno = args->blkno2 = blk1->blkno;
1740         }
1741 }
1742
1743 /*
1744  * Examine entries until we reduce the absolute difference in
1745  * byte usage between the two blocks to a minimum.
1746  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1747  * GROT: there will always be enough room in either block for a new entry.
1748  * GROT: Do a double-split for this case?
1749  */
1750 STATIC int
1751 xfs_attr3_leaf_figure_balance(
1752         struct xfs_da_state             *state,
1753         struct xfs_da_state_blk         *blk1,
1754         struct xfs_attr3_icleaf_hdr     *ichdr1,
1755         struct xfs_da_state_blk         *blk2,
1756         struct xfs_attr3_icleaf_hdr     *ichdr2,
1757         int                             *countarg,
1758         int                             *usedbytesarg)
1759 {
1760         struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1761         struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1762         struct xfs_attr_leaf_entry      *entry;
1763         int                             count;
1764         int                             max;
1765         int                             index;
1766         int                             totallen = 0;
1767         int                             half;
1768         int                             lastdelta;
1769         int                             foundit = 0;
1770         int                             tmp;
1771
1772         /*
1773          * Examine entries until we reduce the absolute difference in
1774          * byte usage between the two blocks to a minimum.
1775          */
1776         max = ichdr1->count + ichdr2->count;
1777         half = (max + 1) * sizeof(*entry);
1778         half += ichdr1->usedbytes + ichdr2->usedbytes +
1779                         xfs_attr_leaf_newentsize(state->args->namelen,
1780                                                  state->args->valuelen,
1781                                                  state->blocksize, NULL);
1782         half /= 2;
1783         lastdelta = state->blocksize;
1784         entry = xfs_attr3_leaf_entryp(leaf1);
1785         for (count = index = 0; count < max; entry++, index++, count++) {
1786
1787 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1788                 /*
1789                  * The new entry is in the first block, account for it.
1790                  */
1791                 if (count == blk1->index) {
1792                         tmp = totallen + sizeof(*entry) +
1793                                 xfs_attr_leaf_newentsize(
1794                                                 state->args->namelen,
1795                                                 state->args->valuelen,
1796                                                 state->blocksize, NULL);
1797                         if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1798                                 break;
1799                         lastdelta = XFS_ATTR_ABS(half - tmp);
1800                         totallen = tmp;
1801                         foundit = 1;
1802                 }
1803
1804                 /*
1805                  * Wrap around into the second block if necessary.
1806                  */
1807                 if (count == ichdr1->count) {
1808                         leaf1 = leaf2;
1809                         entry = xfs_attr3_leaf_entryp(leaf1);
1810                         index = 0;
1811                 }
1812
1813                 /*
1814                  * Figure out if next leaf entry would be too much.
1815                  */
1816                 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1817                                                                         index);
1818                 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1819                         break;
1820                 lastdelta = XFS_ATTR_ABS(half - tmp);
1821                 totallen = tmp;
1822 #undef XFS_ATTR_ABS
1823         }
1824
1825         /*
1826          * Calculate the number of usedbytes that will end up in lower block.
1827          * If new entry not in lower block, fix up the count.
1828          */
1829         totallen -= count * sizeof(*entry);
1830         if (foundit) {
1831                 totallen -= sizeof(*entry) +
1832                                 xfs_attr_leaf_newentsize(
1833                                                 state->args->namelen,
1834                                                 state->args->valuelen,
1835                                                 state->blocksize, NULL);
1836         }
1837
1838         *countarg = count;
1839         *usedbytesarg = totallen;
1840         return foundit;
1841 }
1842
1843 /*========================================================================
1844  * Routines used for shrinking the Btree.
1845  *========================================================================*/
1846
1847 /*
1848  * Check a leaf block and its neighbors to see if the block should be
1849  * collapsed into one or the other neighbor.  Always keep the block
1850  * with the smaller block number.
1851  * If the current block is over 50% full, don't try to join it, return 0.
1852  * If the block is empty, fill in the state structure and return 2.
1853  * If it can be collapsed, fill in the state structure and return 1.
1854  * If nothing can be done, return 0.
1855  *
1856  * GROT: allow for INCOMPLETE entries in calculation.
1857  */
1858 int
1859 xfs_attr3_leaf_toosmall(
1860         struct xfs_da_state     *state,
1861         int                     *action)
1862 {
1863         struct xfs_attr_leafblock *leaf;
1864         struct xfs_da_state_blk *blk;
1865         struct xfs_attr3_icleaf_hdr ichdr;
1866         struct xfs_buf          *bp;
1867         xfs_dablk_t             blkno;
1868         int                     bytes;
1869         int                     forward;
1870         int                     error;
1871         int                     retval;
1872         int                     i;
1873
1874         trace_xfs_attr_leaf_toosmall(state->args);
1875
1876         /*
1877          * Check for the degenerate case of the block being over 50% full.
1878          * If so, it's not worth even looking to see if we might be able
1879          * to coalesce with a sibling.
1880          */
1881         blk = &state->path.blk[ state->path.active-1 ];
1882         leaf = blk->bp->b_addr;
1883         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
1884         bytes = xfs_attr3_leaf_hdr_size(leaf) +
1885                 ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1886                 ichdr.usedbytes;
1887         if (bytes > (state->blocksize >> 1)) {
1888                 *action = 0;    /* blk over 50%, don't try to join */
1889                 return(0);
1890         }
1891
1892         /*
1893          * Check for the degenerate case of the block being empty.
1894          * If the block is empty, we'll simply delete it, no need to
1895          * coalesce it with a sibling block.  We choose (arbitrarily)
1896          * to merge with the forward block unless it is NULL.
1897          */
1898         if (ichdr.count == 0) {
1899                 /*
1900                  * Make altpath point to the block we want to keep and
1901                  * path point to the block we want to drop (this one).
1902                  */
1903                 forward = (ichdr.forw != 0);
1904                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1905                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1906                                                  0, &retval);
1907                 if (error)
1908                         return(error);
1909                 if (retval) {
1910                         *action = 0;
1911                 } else {
1912                         *action = 2;
1913                 }
1914                 return 0;
1915         }
1916
1917         /*
1918          * Examine each sibling block to see if we can coalesce with
1919          * at least 25% free space to spare.  We need to figure out
1920          * whether to merge with the forward or the backward block.
1921          * We prefer coalescing with the lower numbered sibling so as
1922          * to shrink an attribute list over time.
1923          */
1924         /* start with smaller blk num */
1925         forward = ichdr.forw < ichdr.back;
1926         for (i = 0; i < 2; forward = !forward, i++) {
1927                 struct xfs_attr3_icleaf_hdr ichdr2;
1928                 if (forward)
1929                         blkno = ichdr.forw;
1930                 else
1931                         blkno = ichdr.back;
1932                 if (blkno == 0)
1933                         continue;
1934                 error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1935                                         blkno, -1, &bp);
1936                 if (error)
1937                         return(error);
1938
1939                 xfs_attr3_leaf_hdr_from_disk(&ichdr2, bp->b_addr);
1940
1941                 bytes = state->blocksize - (state->blocksize >> 2) -
1942                         ichdr.usedbytes - ichdr2.usedbytes -
1943                         ((ichdr.count + ichdr2.count) *
1944                                         sizeof(xfs_attr_leaf_entry_t)) -
1945                         xfs_attr3_leaf_hdr_size(leaf);
1946
1947                 xfs_trans_brelse(state->args->trans, bp);
1948                 if (bytes >= 0)
1949                         break;  /* fits with at least 25% to spare */
1950         }
1951         if (i >= 2) {
1952                 *action = 0;
1953                 return(0);
1954         }
1955
1956         /*
1957          * Make altpath point to the block we want to keep (the lower
1958          * numbered block) and path point to the block we want to drop.
1959          */
1960         memcpy(&state->altpath, &state->path, sizeof(state->path));
1961         if (blkno < blk->blkno) {
1962                 error = xfs_da3_path_shift(state, &state->altpath, forward,
1963                                                  0, &retval);
1964         } else {
1965                 error = xfs_da3_path_shift(state, &state->path, forward,
1966                                                  0, &retval);
1967         }
1968         if (error)
1969                 return(error);
1970         if (retval) {
1971                 *action = 0;
1972         } else {
1973                 *action = 1;
1974         }
1975         return(0);
1976 }
1977
1978 /*
1979  * Remove a name from the leaf attribute list structure.
1980  *
1981  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1982  * If two leaves are 37% full, when combined they will leave 25% free.
1983  */
1984 int
1985 xfs_attr3_leaf_remove(
1986         struct xfs_buf          *bp,
1987         struct xfs_da_args      *args)
1988 {
1989         struct xfs_attr_leafblock *leaf;
1990         struct xfs_attr3_icleaf_hdr ichdr;
1991         struct xfs_attr_leaf_entry *entry;
1992         struct xfs_mount        *mp = args->trans->t_mountp;
1993         int                     before;
1994         int                     after;
1995         int                     smallest;
1996         int                     entsize;
1997         int                     tablesize;
1998         int                     tmp;
1999         int                     i;
2000
2001         trace_xfs_attr_leaf_remove(args);
2002
2003         leaf = bp->b_addr;
2004         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2005
2006         ASSERT(ichdr.count > 0 && ichdr.count < XFS_LBSIZE(mp) / 8);
2007         ASSERT(args->index >= 0 && args->index < ichdr.count);
2008         ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
2009                                         xfs_attr3_leaf_hdr_size(leaf));
2010
2011         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2012
2013         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2014         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2015
2016         /*
2017          * Scan through free region table:
2018          *    check for adjacency of free'd entry with an existing one,
2019          *    find smallest free region in case we need to replace it,
2020          *    adjust any map that borders the entry table,
2021          */
2022         tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
2023                                         + xfs_attr3_leaf_hdr_size(leaf);
2024         tmp = ichdr.freemap[0].size;
2025         before = after = -1;
2026         smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
2027         entsize = xfs_attr_leaf_entsize(leaf, args->index);
2028         for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
2029                 ASSERT(ichdr.freemap[i].base < XFS_LBSIZE(mp));
2030                 ASSERT(ichdr.freemap[i].size < XFS_LBSIZE(mp));
2031                 if (ichdr.freemap[i].base == tablesize) {
2032                         ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
2033                         ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
2034                 }
2035
2036                 if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
2037                                 be16_to_cpu(entry->nameidx)) {
2038                         before = i;
2039                 } else if (ichdr.freemap[i].base ==
2040                                 (be16_to_cpu(entry->nameidx) + entsize)) {
2041                         after = i;
2042                 } else if (ichdr.freemap[i].size < tmp) {
2043                         tmp = ichdr.freemap[i].size;
2044                         smallest = i;
2045                 }
2046         }
2047
2048         /*
2049          * Coalesce adjacent freemap regions,
2050          * or replace the smallest region.
2051          */
2052         if ((before >= 0) || (after >= 0)) {
2053                 if ((before >= 0) && (after >= 0)) {
2054                         ichdr.freemap[before].size += entsize;
2055                         ichdr.freemap[before].size += ichdr.freemap[after].size;
2056                         ichdr.freemap[after].base = 0;
2057                         ichdr.freemap[after].size = 0;
2058                 } else if (before >= 0) {
2059                         ichdr.freemap[before].size += entsize;
2060                 } else {
2061                         ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
2062                         ichdr.freemap[after].size += entsize;
2063                 }
2064         } else {
2065                 /*
2066                  * Replace smallest region (if it is smaller than free'd entry)
2067                  */
2068                 if (ichdr.freemap[smallest].size < entsize) {
2069                         ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
2070                         ichdr.freemap[smallest].size = entsize;
2071                 }
2072         }
2073
2074         /*
2075          * Did we remove the first entry?
2076          */
2077         if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
2078                 smallest = 1;
2079         else
2080                 smallest = 0;
2081
2082         /*
2083          * Compress the remaining entries and zero out the removed stuff.
2084          */
2085         memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
2086         ichdr.usedbytes -= entsize;
2087         xfs_trans_log_buf(args->trans, bp,
2088              XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
2089                                    entsize));
2090
2091         tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
2092         memmove(entry, entry + 1, tmp);
2093         ichdr.count--;
2094         xfs_trans_log_buf(args->trans, bp,
2095             XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
2096
2097         entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
2098         memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
2099
2100         /*
2101          * If we removed the first entry, re-find the first used byte
2102          * in the name area.  Note that if the entry was the "firstused",
2103          * then we don't have a "hole" in our block resulting from
2104          * removing the name.
2105          */
2106         if (smallest) {
2107                 tmp = XFS_LBSIZE(mp);
2108                 entry = xfs_attr3_leaf_entryp(leaf);
2109                 for (i = ichdr.count - 1; i >= 0; entry++, i--) {
2110                         ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
2111                         ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
2112
2113                         if (be16_to_cpu(entry->nameidx) < tmp)
2114                                 tmp = be16_to_cpu(entry->nameidx);
2115                 }
2116                 ichdr.firstused = tmp;
2117                 if (!ichdr.firstused)
2118                         ichdr.firstused = tmp - XFS_ATTR_LEAF_NAME_ALIGN;
2119         } else {
2120                 ichdr.holes = 1;        /* mark as needing compaction */
2121         }
2122         xfs_attr3_leaf_hdr_to_disk(leaf, &ichdr);
2123         xfs_trans_log_buf(args->trans, bp,
2124                           XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2125                                           xfs_attr3_leaf_hdr_size(leaf)));
2126
2127         /*
2128          * Check if leaf is less than 50% full, caller may want to
2129          * "join" the leaf with a sibling if so.
2130          */
2131         tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2132               ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2133
2134         return tmp < mp->m_attr_magicpct; /* leaf is < 37% full */
2135 }
2136
2137 /*
2138  * Move all the attribute list entries from drop_leaf into save_leaf.
2139  */
2140 void
2141 xfs_attr3_leaf_unbalance(
2142         struct xfs_da_state     *state,
2143         struct xfs_da_state_blk *drop_blk,
2144         struct xfs_da_state_blk *save_blk)
2145 {
2146         struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2147         struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2148         struct xfs_attr3_icleaf_hdr drophdr;
2149         struct xfs_attr3_icleaf_hdr savehdr;
2150         struct xfs_attr_leaf_entry *entry;
2151         struct xfs_mount        *mp = state->mp;
2152
2153         trace_xfs_attr_leaf_unbalance(state->args);
2154
2155         drop_leaf = drop_blk->bp->b_addr;
2156         save_leaf = save_blk->bp->b_addr;
2157         xfs_attr3_leaf_hdr_from_disk(&drophdr, drop_leaf);
2158         xfs_attr3_leaf_hdr_from_disk(&savehdr, save_leaf);
2159         entry = xfs_attr3_leaf_entryp(drop_leaf);
2160
2161         /*
2162          * Save last hashval from dying block for later Btree fixup.
2163          */
2164         drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2165
2166         /*
2167          * Check if we need a temp buffer, or can we do it in place.
2168          * Note that we don't check "leaf" for holes because we will
2169          * always be dropping it, toosmall() decided that for us already.
2170          */
2171         if (savehdr.holes == 0) {
2172                 /*
2173                  * dest leaf has no holes, so we add there.  May need
2174                  * to make some room in the entry array.
2175                  */
2176                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2177                                          drop_blk->bp, &drophdr)) {
2178                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2179                                                 save_leaf, &savehdr, 0,
2180                                                 drophdr.count, mp);
2181                 } else {
2182                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2183                                                 save_leaf, &savehdr,
2184                                                 savehdr.count, drophdr.count, mp);
2185                 }
2186         } else {
2187                 /*
2188                  * Destination has holes, so we make a temporary copy
2189                  * of the leaf and add them both to that.
2190                  */
2191                 struct xfs_attr_leafblock *tmp_leaf;
2192                 struct xfs_attr3_icleaf_hdr tmphdr;
2193
2194                 tmp_leaf = kmem_zalloc(state->blocksize, KM_SLEEP);
2195
2196                 /*
2197                  * Copy the header into the temp leaf so that all the stuff
2198                  * not in the incore header is present and gets copied back in
2199                  * once we've moved all the entries.
2200                  */
2201                 memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2202
2203                 memset(&tmphdr, 0, sizeof(tmphdr));
2204                 tmphdr.magic = savehdr.magic;
2205                 tmphdr.forw = savehdr.forw;
2206                 tmphdr.back = savehdr.back;
2207                 tmphdr.firstused = state->blocksize;
2208
2209                 /* write the header to the temp buffer to initialise it */
2210                 xfs_attr3_leaf_hdr_to_disk(tmp_leaf, &tmphdr);
2211
2212                 if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2213                                          drop_blk->bp, &drophdr)) {
2214                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2215                                                 tmp_leaf, &tmphdr, 0,
2216                                                 drophdr.count, mp);
2217                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2218                                                 tmp_leaf, &tmphdr, tmphdr.count,
2219                                                 savehdr.count, mp);
2220                 } else {
2221                         xfs_attr3_leaf_moveents(save_leaf, &savehdr, 0,
2222                                                 tmp_leaf, &tmphdr, 0,
2223                                                 savehdr.count, mp);
2224                         xfs_attr3_leaf_moveents(drop_leaf, &drophdr, 0,
2225                                                 tmp_leaf, &tmphdr, tmphdr.count,
2226                                                 drophdr.count, mp);
2227                 }
2228                 memcpy(save_leaf, tmp_leaf, state->blocksize);
2229                 savehdr = tmphdr; /* struct copy */
2230                 kmem_free(tmp_leaf);
2231         }
2232
2233         xfs_attr3_leaf_hdr_to_disk(save_leaf, &savehdr);
2234         xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2235                                            state->blocksize - 1);
2236
2237         /*
2238          * Copy out last hashval in each block for B-tree code.
2239          */
2240         entry = xfs_attr3_leaf_entryp(save_leaf);
2241         save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2242 }
2243
2244 /*========================================================================
2245  * Routines used for finding things in the Btree.
2246  *========================================================================*/
2247
2248 /*
2249  * Look up a name in a leaf attribute list structure.
2250  * This is the internal routine, it uses the caller's buffer.
2251  *
2252  * Note that duplicate keys are allowed, but only check within the
2253  * current leaf node.  The Btree code must check in adjacent leaf nodes.
2254  *
2255  * Return in args->index the index into the entry[] array of either
2256  * the found entry, or where the entry should have been (insert before
2257  * that entry).
2258  *
2259  * Don't change the args->value unless we find the attribute.
2260  */
2261 int
2262 xfs_attr3_leaf_lookup_int(
2263         struct xfs_buf          *bp,
2264         struct xfs_da_args      *args)
2265 {
2266         struct xfs_attr_leafblock *leaf;
2267         struct xfs_attr3_icleaf_hdr ichdr;
2268         struct xfs_attr_leaf_entry *entry;
2269         struct xfs_attr_leaf_entry *entries;
2270         struct xfs_attr_leaf_name_local *name_loc;
2271         struct xfs_attr_leaf_name_remote *name_rmt;
2272         xfs_dahash_t            hashval;
2273         int                     probe;
2274         int                     span;
2275
2276         trace_xfs_attr_leaf_lookup(args);
2277
2278         leaf = bp->b_addr;
2279         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2280         entries = xfs_attr3_leaf_entryp(leaf);
2281         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2282
2283         /*
2284          * Binary search.  (note: small blocks will skip this loop)
2285          */
2286         hashval = args->hashval;
2287         probe = span = ichdr.count / 2;
2288         for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2289                 span /= 2;
2290                 if (be32_to_cpu(entry->hashval) < hashval)
2291                         probe += span;
2292                 else if (be32_to_cpu(entry->hashval) > hashval)
2293                         probe -= span;
2294                 else
2295                         break;
2296         }
2297         ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2298         ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2299
2300         /*
2301          * Since we may have duplicate hashval's, find the first matching
2302          * hashval in the leaf.
2303          */
2304         while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2305                 entry--;
2306                 probe--;
2307         }
2308         while (probe < ichdr.count &&
2309                be32_to_cpu(entry->hashval) < hashval) {
2310                 entry++;
2311                 probe++;
2312         }
2313         if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2314                 args->index = probe;
2315                 return XFS_ERROR(ENOATTR);
2316         }
2317
2318         /*
2319          * Duplicate keys may be present, so search all of them for a match.
2320          */
2321         for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2322                         entry++, probe++) {
2323 /*
2324  * GROT: Add code to remove incomplete entries.
2325  */
2326                 /*
2327                  * If we are looking for INCOMPLETE entries, show only those.
2328                  * If we are looking for complete entries, show only those.
2329                  */
2330                 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2331                     (entry->flags & XFS_ATTR_INCOMPLETE)) {
2332                         continue;
2333                 }
2334                 if (entry->flags & XFS_ATTR_LOCAL) {
2335                         name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2336                         if (name_loc->namelen != args->namelen)
2337                                 continue;
2338                         if (memcmp(args->name, name_loc->nameval,
2339                                                         args->namelen) != 0)
2340                                 continue;
2341                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2342                                 continue;
2343                         args->index = probe;
2344                         return XFS_ERROR(EEXIST);
2345                 } else {
2346                         name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2347                         if (name_rmt->namelen != args->namelen)
2348                                 continue;
2349                         if (memcmp(args->name, name_rmt->name,
2350                                                         args->namelen) != 0)
2351                                 continue;
2352                         if (!xfs_attr_namesp_match(args->flags, entry->flags))
2353                                 continue;
2354                         args->index = probe;
2355                         args->valuelen = be32_to_cpu(name_rmt->valuelen);
2356                         args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2357                         args->rmtblkcnt = xfs_attr3_rmt_blocks(
2358                                                         args->dp->i_mount,
2359                                                         args->valuelen);
2360                         return XFS_ERROR(EEXIST);
2361                 }
2362         }
2363         args->index = probe;
2364         return XFS_ERROR(ENOATTR);
2365 }
2366
2367 /*
2368  * Get the value associated with an attribute name from a leaf attribute
2369  * list structure.
2370  */
2371 int
2372 xfs_attr3_leaf_getvalue(
2373         struct xfs_buf          *bp,
2374         struct xfs_da_args      *args)
2375 {
2376         struct xfs_attr_leafblock *leaf;
2377         struct xfs_attr3_icleaf_hdr ichdr;
2378         struct xfs_attr_leaf_entry *entry;
2379         struct xfs_attr_leaf_name_local *name_loc;
2380         struct xfs_attr_leaf_name_remote *name_rmt;
2381         int                     valuelen;
2382
2383         leaf = bp->b_addr;
2384         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2385         ASSERT(ichdr.count < XFS_LBSIZE(args->dp->i_mount) / 8);
2386         ASSERT(args->index < ichdr.count);
2387
2388         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2389         if (entry->flags & XFS_ATTR_LOCAL) {
2390                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2391                 ASSERT(name_loc->namelen == args->namelen);
2392                 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2393                 valuelen = be16_to_cpu(name_loc->valuelen);
2394                 if (args->flags & ATTR_KERNOVAL) {
2395                         args->valuelen = valuelen;
2396                         return 0;
2397                 }
2398                 if (args->valuelen < valuelen) {
2399                         args->valuelen = valuelen;
2400                         return XFS_ERROR(ERANGE);
2401                 }
2402                 args->valuelen = valuelen;
2403                 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2404         } else {
2405                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2406                 ASSERT(name_rmt->namelen == args->namelen);
2407                 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2408                 valuelen = be32_to_cpu(name_rmt->valuelen);
2409                 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2410                 args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2411                                                        valuelen);
2412                 if (args->flags & ATTR_KERNOVAL) {
2413                         args->valuelen = valuelen;
2414                         return 0;
2415                 }
2416                 if (args->valuelen < valuelen) {
2417                         args->valuelen = valuelen;
2418                         return XFS_ERROR(ERANGE);
2419                 }
2420                 args->valuelen = valuelen;
2421         }
2422         return 0;
2423 }
2424
2425 /*========================================================================
2426  * Utility routines.
2427  *========================================================================*/
2428
2429 /*
2430  * Move the indicated entries from one leaf to another.
2431  * NOTE: this routine modifies both source and destination leaves.
2432  */
2433 /*ARGSUSED*/
2434 STATIC void
2435 xfs_attr3_leaf_moveents(
2436         struct xfs_attr_leafblock       *leaf_s,
2437         struct xfs_attr3_icleaf_hdr     *ichdr_s,
2438         int                             start_s,
2439         struct xfs_attr_leafblock       *leaf_d,
2440         struct xfs_attr3_icleaf_hdr     *ichdr_d,
2441         int                             start_d,
2442         int                             count,
2443         struct xfs_mount                *mp)
2444 {
2445         struct xfs_attr_leaf_entry      *entry_s;
2446         struct xfs_attr_leaf_entry      *entry_d;
2447         int                             desti;
2448         int                             tmp;
2449         int                             i;
2450
2451         /*
2452          * Check for nothing to do.
2453          */
2454         if (count == 0)
2455                 return;
2456
2457         /*
2458          * Set up environment.
2459          */
2460         ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2461                ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2462         ASSERT(ichdr_s->magic == ichdr_d->magic);
2463         ASSERT(ichdr_s->count > 0 && ichdr_s->count < XFS_LBSIZE(mp) / 8);
2464         ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2465                                         + xfs_attr3_leaf_hdr_size(leaf_s));
2466         ASSERT(ichdr_d->count < XFS_LBSIZE(mp) / 8);
2467         ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2468                                         + xfs_attr3_leaf_hdr_size(leaf_d));
2469
2470         ASSERT(start_s < ichdr_s->count);
2471         ASSERT(start_d <= ichdr_d->count);
2472         ASSERT(count <= ichdr_s->count);
2473
2474
2475         /*
2476          * Move the entries in the destination leaf up to make a hole?
2477          */
2478         if (start_d < ichdr_d->count) {
2479                 tmp  = ichdr_d->count - start_d;
2480                 tmp *= sizeof(xfs_attr_leaf_entry_t);
2481                 entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2482                 entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2483                 memmove(entry_d, entry_s, tmp);
2484         }
2485
2486         /*
2487          * Copy all entry's in the same (sorted) order,
2488          * but allocate attribute info packed and in sequence.
2489          */
2490         entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2491         entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2492         desti = start_d;
2493         for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2494                 ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2495                 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2496 #ifdef GROT
2497                 /*
2498                  * Code to drop INCOMPLETE entries.  Difficult to use as we
2499                  * may also need to change the insertion index.  Code turned
2500                  * off for 6.2, should be revisited later.
2501                  */
2502                 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2503                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2504                         ichdr_s->usedbytes -= tmp;
2505                         ichdr_s->count -= 1;
2506                         entry_d--;      /* to compensate for ++ in loop hdr */
2507                         desti--;
2508                         if ((start_s + i) < offset)
2509                                 result++;       /* insertion index adjustment */
2510                 } else {
2511 #endif /* GROT */
2512                         ichdr_d->firstused -= tmp;
2513                         /* both on-disk, don't endian flip twice */
2514                         entry_d->hashval = entry_s->hashval;
2515                         entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2516                         entry_d->flags = entry_s->flags;
2517                         ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2518                                                         <= XFS_LBSIZE(mp));
2519                         memmove(xfs_attr3_leaf_name(leaf_d, desti),
2520                                 xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2521                         ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2522                                                         <= XFS_LBSIZE(mp));
2523                         memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2524                         ichdr_s->usedbytes -= tmp;
2525                         ichdr_d->usedbytes += tmp;
2526                         ichdr_s->count -= 1;
2527                         ichdr_d->count += 1;
2528                         tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2529                                         + xfs_attr3_leaf_hdr_size(leaf_d);
2530                         ASSERT(ichdr_d->firstused >= tmp);
2531 #ifdef GROT
2532                 }
2533 #endif /* GROT */
2534         }
2535
2536         /*
2537          * Zero out the entries we just copied.
2538          */
2539         if (start_s == ichdr_s->count) {
2540                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2541                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2542                 ASSERT(((char *)entry_s + tmp) <=
2543                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2544                 memset(entry_s, 0, tmp);
2545         } else {
2546                 /*
2547                  * Move the remaining entries down to fill the hole,
2548                  * then zero the entries at the top.
2549                  */
2550                 tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2551                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2552                 entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2553                 memmove(entry_d, entry_s, tmp);
2554
2555                 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2556                 entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2557                 ASSERT(((char *)entry_s + tmp) <=
2558                        ((char *)leaf_s + XFS_LBSIZE(mp)));
2559                 memset(entry_s, 0, tmp);
2560         }
2561
2562         /*
2563          * Fill in the freemap information
2564          */
2565         ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2566         ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2567         ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2568         ichdr_d->freemap[1].base = 0;
2569         ichdr_d->freemap[2].base = 0;
2570         ichdr_d->freemap[1].size = 0;
2571         ichdr_d->freemap[2].size = 0;
2572         ichdr_s->holes = 1;     /* leaf may not be compact */
2573 }
2574
2575 /*
2576  * Pick up the last hashvalue from a leaf block.
2577  */
2578 xfs_dahash_t
2579 xfs_attr_leaf_lasthash(
2580         struct xfs_buf  *bp,
2581         int             *count)
2582 {
2583         struct xfs_attr3_icleaf_hdr ichdr;
2584         struct xfs_attr_leaf_entry *entries;
2585
2586         xfs_attr3_leaf_hdr_from_disk(&ichdr, bp->b_addr);
2587         entries = xfs_attr3_leaf_entryp(bp->b_addr);
2588         if (count)
2589                 *count = ichdr.count;
2590         if (!ichdr.count)
2591                 return 0;
2592         return be32_to_cpu(entries[ichdr.count - 1].hashval);
2593 }
2594
2595 /*
2596  * Calculate the number of bytes used to store the indicated attribute
2597  * (whether local or remote only calculate bytes in this block).
2598  */
2599 STATIC int
2600 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2601 {
2602         struct xfs_attr_leaf_entry *entries;
2603         xfs_attr_leaf_name_local_t *name_loc;
2604         xfs_attr_leaf_name_remote_t *name_rmt;
2605         int size;
2606
2607         entries = xfs_attr3_leaf_entryp(leaf);
2608         if (entries[index].flags & XFS_ATTR_LOCAL) {
2609                 name_loc = xfs_attr3_leaf_name_local(leaf, index);
2610                 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2611                                                    be16_to_cpu(name_loc->valuelen));
2612         } else {
2613                 name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2614                 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2615         }
2616         return size;
2617 }
2618
2619 /*
2620  * Calculate the number of bytes that would be required to store the new
2621  * attribute (whether local or remote only calculate bytes in this block).
2622  * This routine decides as a side effect whether the attribute will be
2623  * a "local" or a "remote" attribute.
2624  */
2625 int
2626 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2627 {
2628         int size;
2629
2630         size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2631         if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2632                 if (local) {
2633                         *local = 1;
2634                 }
2635         } else {
2636                 size = xfs_attr_leaf_entsize_remote(namelen);
2637                 if (local) {
2638                         *local = 0;
2639                 }
2640         }
2641         return size;
2642 }
2643
2644 /*
2645  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2646  */
2647 int
2648 xfs_attr3_leaf_list_int(
2649         struct xfs_buf                  *bp,
2650         struct xfs_attr_list_context    *context)
2651 {
2652         struct attrlist_cursor_kern     *cursor;
2653         struct xfs_attr_leafblock       *leaf;
2654         struct xfs_attr3_icleaf_hdr     ichdr;
2655         struct xfs_attr_leaf_entry      *entries;
2656         struct xfs_attr_leaf_entry      *entry;
2657         int                             retval;
2658         int                             i;
2659
2660         trace_xfs_attr_list_leaf(context);
2661
2662         leaf = bp->b_addr;
2663         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2664         entries = xfs_attr3_leaf_entryp(leaf);
2665
2666         cursor = context->cursor;
2667         cursor->initted = 1;
2668
2669         /*
2670          * Re-find our place in the leaf block if this is a new syscall.
2671          */
2672         if (context->resynch) {
2673                 entry = &entries[0];
2674                 for (i = 0; i < ichdr.count; entry++, i++) {
2675                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2676                                 if (cursor->offset == context->dupcnt) {
2677                                         context->dupcnt = 0;
2678                                         break;
2679                                 }
2680                                 context->dupcnt++;
2681                         } else if (be32_to_cpu(entry->hashval) >
2682                                         cursor->hashval) {
2683                                 context->dupcnt = 0;
2684                                 break;
2685                         }
2686                 }
2687                 if (i == ichdr.count) {
2688                         trace_xfs_attr_list_notfound(context);
2689                         return 0;
2690                 }
2691         } else {
2692                 entry = &entries[0];
2693                 i = 0;
2694         }
2695         context->resynch = 0;
2696
2697         /*
2698          * We have found our place, start copying out the new attributes.
2699          */
2700         retval = 0;
2701         for (; i < ichdr.count; entry++, i++) {
2702                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2703                         cursor->hashval = be32_to_cpu(entry->hashval);
2704                         cursor->offset = 0;
2705                 }
2706
2707                 if (entry->flags & XFS_ATTR_INCOMPLETE)
2708                         continue;               /* skip incomplete entries */
2709
2710                 if (entry->flags & XFS_ATTR_LOCAL) {
2711                         xfs_attr_leaf_name_local_t *name_loc =
2712                                 xfs_attr3_leaf_name_local(leaf, i);
2713
2714                         retval = context->put_listent(context,
2715                                                 entry->flags,
2716                                                 name_loc->nameval,
2717                                                 (int)name_loc->namelen,
2718                                                 be16_to_cpu(name_loc->valuelen),
2719                                                 &name_loc->nameval[name_loc->namelen]);
2720                         if (retval)
2721                                 return retval;
2722                 } else {
2723                         xfs_attr_leaf_name_remote_t *name_rmt =
2724                                 xfs_attr3_leaf_name_remote(leaf, i);
2725
2726                         int valuelen = be32_to_cpu(name_rmt->valuelen);
2727
2728                         if (context->put_value) {
2729                                 xfs_da_args_t args;
2730
2731                                 memset((char *)&args, 0, sizeof(args));
2732                                 args.dp = context->dp;
2733                                 args.whichfork = XFS_ATTR_FORK;
2734                                 args.valuelen = valuelen;
2735                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2736                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2737                                 args.rmtblkcnt = xfs_attr3_rmt_blocks(
2738                                                         args.dp->i_mount, valuelen);
2739                                 retval = xfs_attr_rmtval_get(&args);
2740                                 if (retval)
2741                                         return retval;
2742                                 retval = context->put_listent(context,
2743                                                 entry->flags,
2744                                                 name_rmt->name,
2745                                                 (int)name_rmt->namelen,
2746                                                 valuelen,
2747                                                 args.value);
2748                                 kmem_free(args.value);
2749                         } else {
2750                                 retval = context->put_listent(context,
2751                                                 entry->flags,
2752                                                 name_rmt->name,
2753                                                 (int)name_rmt->namelen,
2754                                                 valuelen,
2755                                                 NULL);
2756                         }
2757                         if (retval)
2758                                 return retval;
2759                 }
2760                 if (context->seen_enough)
2761                         break;
2762                 cursor->offset++;
2763         }
2764         trace_xfs_attr_list_leaf_end(context);
2765         return retval;
2766 }
2767
2768
2769 /*========================================================================
2770  * Manage the INCOMPLETE flag in a leaf entry
2771  *========================================================================*/
2772
2773 /*
2774  * Clear the INCOMPLETE flag on an entry in a leaf block.
2775  */
2776 int
2777 xfs_attr3_leaf_clearflag(
2778         struct xfs_da_args      *args)
2779 {
2780         struct xfs_attr_leafblock *leaf;
2781         struct xfs_attr_leaf_entry *entry;
2782         struct xfs_attr_leaf_name_remote *name_rmt;
2783         struct xfs_buf          *bp;
2784         int                     error;
2785 #ifdef DEBUG
2786         struct xfs_attr3_icleaf_hdr ichdr;
2787         xfs_attr_leaf_name_local_t *name_loc;
2788         int namelen;
2789         char *name;
2790 #endif /* DEBUG */
2791
2792         trace_xfs_attr_leaf_clearflag(args);
2793         /*
2794          * Set up the operation.
2795          */
2796         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2797         if (error)
2798                 return(error);
2799
2800         leaf = bp->b_addr;
2801         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2802         ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2803
2804 #ifdef DEBUG
2805         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2806         ASSERT(args->index < ichdr.count);
2807         ASSERT(args->index >= 0);
2808
2809         if (entry->flags & XFS_ATTR_LOCAL) {
2810                 name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2811                 namelen = name_loc->namelen;
2812                 name = (char *)name_loc->nameval;
2813         } else {
2814                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2815                 namelen = name_rmt->namelen;
2816                 name = (char *)name_rmt->name;
2817         }
2818         ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2819         ASSERT(namelen == args->namelen);
2820         ASSERT(memcmp(name, args->name, namelen) == 0);
2821 #endif /* DEBUG */
2822
2823         entry->flags &= ~XFS_ATTR_INCOMPLETE;
2824         xfs_trans_log_buf(args->trans, bp,
2825                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2826
2827         if (args->rmtblkno) {
2828                 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2829                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2830                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2831                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2832                 xfs_trans_log_buf(args->trans, bp,
2833                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2834         }
2835
2836         /*
2837          * Commit the flag value change and start the next trans in series.
2838          */
2839         return xfs_trans_roll(&args->trans, args->dp);
2840 }
2841
2842 /*
2843  * Set the INCOMPLETE flag on an entry in a leaf block.
2844  */
2845 int
2846 xfs_attr3_leaf_setflag(
2847         struct xfs_da_args      *args)
2848 {
2849         struct xfs_attr_leafblock *leaf;
2850         struct xfs_attr_leaf_entry *entry;
2851         struct xfs_attr_leaf_name_remote *name_rmt;
2852         struct xfs_buf          *bp;
2853         int error;
2854 #ifdef DEBUG
2855         struct xfs_attr3_icleaf_hdr ichdr;
2856 #endif
2857
2858         trace_xfs_attr_leaf_setflag(args);
2859
2860         /*
2861          * Set up the operation.
2862          */
2863         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2864         if (error)
2865                 return(error);
2866
2867         leaf = bp->b_addr;
2868 #ifdef DEBUG
2869         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
2870         ASSERT(args->index < ichdr.count);
2871         ASSERT(args->index >= 0);
2872 #endif
2873         entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2874
2875         ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2876         entry->flags |= XFS_ATTR_INCOMPLETE;
2877         xfs_trans_log_buf(args->trans, bp,
2878                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2879         if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2880                 name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2881                 name_rmt->valueblk = 0;
2882                 name_rmt->valuelen = 0;
2883                 xfs_trans_log_buf(args->trans, bp,
2884                          XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2885         }
2886
2887         /*
2888          * Commit the flag value change and start the next trans in series.
2889          */
2890         return xfs_trans_roll(&args->trans, args->dp);
2891 }
2892
2893 /*
2894  * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2895  * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2896  * entry given by args->blkno2/index2.
2897  *
2898  * Note that they could be in different blocks, or in the same block.
2899  */
2900 int
2901 xfs_attr3_leaf_flipflags(
2902         struct xfs_da_args      *args)
2903 {
2904         struct xfs_attr_leafblock *leaf1;
2905         struct xfs_attr_leafblock *leaf2;
2906         struct xfs_attr_leaf_entry *entry1;
2907         struct xfs_attr_leaf_entry *entry2;
2908         struct xfs_attr_leaf_name_remote *name_rmt;
2909         struct xfs_buf          *bp1;
2910         struct xfs_buf          *bp2;
2911         int error;
2912 #ifdef DEBUG
2913         struct xfs_attr3_icleaf_hdr ichdr1;
2914         struct xfs_attr3_icleaf_hdr ichdr2;
2915         xfs_attr_leaf_name_local_t *name_loc;
2916         int namelen1, namelen2;
2917         char *name1, *name2;
2918 #endif /* DEBUG */
2919
2920         trace_xfs_attr_leaf_flipflags(args);
2921
2922         /*
2923          * Read the block containing the "old" attr
2924          */
2925         error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2926         if (error)
2927                 return error;
2928
2929         /*
2930          * Read the block containing the "new" attr, if it is different
2931          */
2932         if (args->blkno2 != args->blkno) {
2933                 error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2934                                            -1, &bp2);
2935                 if (error)
2936                         return error;
2937         } else {
2938                 bp2 = bp1;
2939         }
2940
2941         leaf1 = bp1->b_addr;
2942         entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2943
2944         leaf2 = bp2->b_addr;
2945         entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2946
2947 #ifdef DEBUG
2948         xfs_attr3_leaf_hdr_from_disk(&ichdr1, leaf1);
2949         ASSERT(args->index < ichdr1.count);
2950         ASSERT(args->index >= 0);
2951
2952         xfs_attr3_leaf_hdr_from_disk(&ichdr2, leaf2);
2953         ASSERT(args->index2 < ichdr2.count);
2954         ASSERT(args->index2 >= 0);
2955
2956         if (entry1->flags & XFS_ATTR_LOCAL) {
2957                 name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2958                 namelen1 = name_loc->namelen;
2959                 name1 = (char *)name_loc->nameval;
2960         } else {
2961                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2962                 namelen1 = name_rmt->namelen;
2963                 name1 = (char *)name_rmt->name;
2964         }
2965         if (entry2->flags & XFS_ATTR_LOCAL) {
2966                 name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2967                 namelen2 = name_loc->namelen;
2968                 name2 = (char *)name_loc->nameval;
2969         } else {
2970                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2971                 namelen2 = name_rmt->namelen;
2972                 name2 = (char *)name_rmt->name;
2973         }
2974         ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2975         ASSERT(namelen1 == namelen2);
2976         ASSERT(memcmp(name1, name2, namelen1) == 0);
2977 #endif /* DEBUG */
2978
2979         ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2980         ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2981
2982         entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2983         xfs_trans_log_buf(args->trans, bp1,
2984                           XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2985         if (args->rmtblkno) {
2986                 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2987                 name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2988                 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2989                 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2990                 xfs_trans_log_buf(args->trans, bp1,
2991                          XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2992         }
2993
2994         entry2->flags |= XFS_ATTR_INCOMPLETE;
2995         xfs_trans_log_buf(args->trans, bp2,
2996                           XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2997         if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2998                 name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2999                 name_rmt->valueblk = 0;
3000                 name_rmt->valuelen = 0;
3001                 xfs_trans_log_buf(args->trans, bp2,
3002                          XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
3003         }
3004
3005         /*
3006          * Commit the flag value change and start the next trans in series.
3007          */
3008         error = xfs_trans_roll(&args->trans, args->dp);
3009
3010         return error;
3011 }
3012
3013 /*========================================================================
3014  * Indiscriminately delete the entire attribute fork
3015  *========================================================================*/
3016
3017 /*
3018  * Recurse (gasp!) through the attribute nodes until we find leaves.
3019  * We're doing a depth-first traversal in order to invalidate everything.
3020  */
3021 int
3022 xfs_attr3_root_inactive(
3023         struct xfs_trans        **trans,
3024         struct xfs_inode        *dp)
3025 {
3026         struct xfs_da_blkinfo   *info;
3027         struct xfs_buf          *bp;
3028         xfs_daddr_t             blkno;
3029         int                     error;
3030
3031         /*
3032          * Read block 0 to see what we have to work with.
3033          * We only get here if we have extents, since we remove
3034          * the extents in reverse order the extent containing
3035          * block 0 must still be there.
3036          */
3037         error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
3038         if (error)
3039                 return error;
3040         blkno = bp->b_bn;
3041
3042         /*
3043          * Invalidate the tree, even if the "tree" is only a single leaf block.
3044          * This is a depth-first traversal!
3045          */
3046         info = bp->b_addr;
3047         switch (info->magic) {
3048         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3049         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3050                 error = xfs_attr3_node_inactive(trans, dp, bp, 1);
3051                 break;
3052         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3053         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3054                 error = xfs_attr3_leaf_inactive(trans, dp, bp);
3055                 break;
3056         default:
3057                 error = XFS_ERROR(EIO);
3058                 xfs_trans_brelse(*trans, bp);
3059                 break;
3060         }
3061         if (error)
3062                 return error;
3063
3064         /*
3065          * Invalidate the incore copy of the root block.
3066          */
3067         error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
3068         if (error)
3069                 return error;
3070         xfs_trans_binval(*trans, bp);   /* remove from cache */
3071         /*
3072          * Commit the invalidate and start the next transaction.
3073          */
3074         error = xfs_trans_roll(trans, dp);
3075
3076         return error;
3077 }
3078
3079 /*
3080  * Recurse (gasp!) through the attribute nodes until we find leaves.
3081  * We're doing a depth-first traversal in order to invalidate everything.
3082  */
3083 STATIC int
3084 xfs_attr3_node_inactive(
3085         struct xfs_trans **trans,
3086         struct xfs_inode *dp,
3087         struct xfs_buf  *bp,
3088         int             level)
3089 {
3090         xfs_da_blkinfo_t *info;
3091         xfs_da_intnode_t *node;
3092         xfs_dablk_t child_fsb;
3093         xfs_daddr_t parent_blkno, child_blkno;
3094         int error, i;
3095         struct xfs_buf *child_bp;
3096         struct xfs_da_node_entry *btree;
3097         struct xfs_da3_icnode_hdr ichdr;
3098
3099         /*
3100          * Since this code is recursive (gasp!) we must protect ourselves.
3101          */
3102         if (level > XFS_DA_NODE_MAXDEPTH) {
3103                 xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3104                 return XFS_ERROR(EIO);
3105         }
3106
3107         node = bp->b_addr;
3108         xfs_da3_node_hdr_from_disk(&ichdr, node);
3109         parent_blkno = bp->b_bn;
3110         if (!ichdr.count) {
3111                 xfs_trans_brelse(*trans, bp);
3112                 return 0;
3113         }
3114         btree = xfs_da3_node_tree_p(node);
3115         child_fsb = be32_to_cpu(btree[0].before);
3116         xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
3117
3118         /*
3119          * If this is the node level just above the leaves, simply loop
3120          * over the leaves removing all of them.  If this is higher up
3121          * in the tree, recurse downward.
3122          */
3123         for (i = 0; i < ichdr.count; i++) {
3124                 /*
3125                  * Read the subsidiary block to see what we have to work with.
3126                  * Don't do this in a transaction.  This is a depth-first
3127                  * traversal of the tree so we may deal with many blocks
3128                  * before we come back to this one.
3129                  */
3130                 error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp,
3131                                                 XFS_ATTR_FORK);
3132                 if (error)
3133                         return(error);
3134                 if (child_bp) {
3135                                                 /* save for re-read later */
3136                         child_blkno = XFS_BUF_ADDR(child_bp);
3137
3138                         /*
3139                          * Invalidate the subtree, however we have to.
3140                          */
3141                         info = child_bp->b_addr;
3142                         switch (info->magic) {
3143                         case cpu_to_be16(XFS_DA_NODE_MAGIC):
3144                         case cpu_to_be16(XFS_DA3_NODE_MAGIC):
3145                                 error = xfs_attr3_node_inactive(trans, dp,
3146                                                         child_bp, level + 1);
3147                                 break;
3148                         case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
3149                         case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
3150                                 error = xfs_attr3_leaf_inactive(trans, dp,
3151                                                         child_bp);
3152                                 break;
3153                         default:
3154                                 error = XFS_ERROR(EIO);
3155                                 xfs_trans_brelse(*trans, child_bp);
3156                                 break;
3157                         }
3158                         if (error)
3159                                 return error;
3160
3161                         /*
3162                          * Remove the subsidiary block from the cache
3163                          * and from the log.
3164                          */
3165                         error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
3166                                 &child_bp, XFS_ATTR_FORK);
3167                         if (error)
3168                                 return error;
3169                         xfs_trans_binval(*trans, child_bp);
3170                 }
3171
3172                 /*
3173                  * If we're not done, re-read the parent to get the next
3174                  * child block number.
3175                  */
3176                 if (i + 1 < ichdr.count) {
3177                         error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
3178                                                  &bp, XFS_ATTR_FORK);
3179                         if (error)
3180                                 return error;
3181                         child_fsb = be32_to_cpu(btree[i + 1].before);
3182                         xfs_trans_brelse(*trans, bp);
3183                 }
3184                 /*
3185                  * Atomically commit the whole invalidate stuff.
3186                  */
3187                 error = xfs_trans_roll(trans, dp);
3188                 if (error)
3189                         return  error;
3190         }
3191
3192         return 0;
3193 }
3194
3195 /*
3196  * Invalidate all of the "remote" value regions pointed to by a particular
3197  * leaf block.
3198  * Note that we must release the lock on the buffer so that we are not
3199  * caught holding something that the logging code wants to flush to disk.
3200  */
3201 STATIC int
3202 xfs_attr3_leaf_inactive(
3203         struct xfs_trans        **trans,
3204         struct xfs_inode        *dp,
3205         struct xfs_buf          *bp)
3206 {
3207         struct xfs_attr_leafblock *leaf;
3208         struct xfs_attr3_icleaf_hdr ichdr;
3209         struct xfs_attr_leaf_entry *entry;
3210         struct xfs_attr_leaf_name_remote *name_rmt;
3211         struct xfs_attr_inactive_list *list;
3212         struct xfs_attr_inactive_list *lp;
3213         int                     error;
3214         int                     count;
3215         int                     size;
3216         int                     tmp;
3217         int                     i;
3218
3219         leaf = bp->b_addr;
3220         xfs_attr3_leaf_hdr_from_disk(&ichdr, leaf);
3221
3222         /*
3223          * Count the number of "remote" value extents.
3224          */
3225         count = 0;
3226         entry = xfs_attr3_leaf_entryp(leaf);
3227         for (i = 0; i < ichdr.count; entry++, i++) {
3228                 if (be16_to_cpu(entry->nameidx) &&
3229                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3230                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3231                         if (name_rmt->valueblk)
3232                                 count++;
3233                 }
3234         }
3235
3236         /*
3237          * If there are no "remote" values, we're done.
3238          */
3239         if (count == 0) {
3240                 xfs_trans_brelse(*trans, bp);
3241                 return 0;
3242         }
3243
3244         /*
3245          * Allocate storage for a list of all the "remote" value extents.
3246          */
3247         size = count * sizeof(xfs_attr_inactive_list_t);
3248         list = kmem_alloc(size, KM_SLEEP);
3249
3250         /*
3251          * Identify each of the "remote" value extents.
3252          */
3253         lp = list;
3254         entry = xfs_attr3_leaf_entryp(leaf);
3255         for (i = 0; i < ichdr.count; entry++, i++) {
3256                 if (be16_to_cpu(entry->nameidx) &&
3257                     ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
3258                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
3259                         if (name_rmt->valueblk) {
3260                                 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
3261                                 lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount,
3262                                                     be32_to_cpu(name_rmt->valuelen));
3263                                 lp++;
3264                         }
3265                 }
3266         }
3267         xfs_trans_brelse(*trans, bp);   /* unlock for trans. in freextent() */
3268
3269         /*
3270          * Invalidate each of the "remote" value extents.
3271          */
3272         error = 0;
3273         for (lp = list, i = 0; i < count; i++, lp++) {
3274                 tmp = xfs_attr3_leaf_freextent(trans, dp,
3275                                 lp->valueblk, lp->valuelen);
3276
3277                 if (error == 0)
3278                         error = tmp;    /* save only the 1st errno */
3279         }
3280
3281         kmem_free(list);
3282         return error;
3283 }
3284
3285 /*
3286  * Look at all the extents for this logical region,
3287  * invalidate any buffers that are incore/in transactions.
3288  */
3289 STATIC int
3290 xfs_attr3_leaf_freextent(
3291         struct xfs_trans        **trans,
3292         struct xfs_inode        *dp,
3293         xfs_dablk_t             blkno,
3294         int                     blkcnt)
3295 {
3296         struct xfs_bmbt_irec    map;
3297         struct xfs_buf          *bp;
3298         xfs_dablk_t             tblkno;
3299         xfs_daddr_t             dblkno;
3300         int                     tblkcnt;
3301         int                     dblkcnt;
3302         int                     nmap;
3303         int                     error;
3304
3305         /*
3306          * Roll through the "value", invalidating the attribute value's
3307          * blocks.
3308          */
3309         tblkno = blkno;
3310         tblkcnt = blkcnt;
3311         while (tblkcnt > 0) {
3312                 /*
3313                  * Try to remember where we decided to put the value.
3314                  */
3315                 nmap = 1;
3316                 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
3317                                        &map, &nmap, XFS_BMAPI_ATTRFORK);
3318                 if (error) {
3319                         return(error);
3320                 }
3321                 ASSERT(nmap == 1);
3322                 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
3323
3324                 /*
3325                  * If it's a hole, these are already unmapped
3326                  * so there's nothing to invalidate.
3327                  */
3328                 if (map.br_startblock != HOLESTARTBLOCK) {
3329
3330                         dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
3331                                                   map.br_startblock);
3332                         dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
3333                                                 map.br_blockcount);
3334                         bp = xfs_trans_get_buf(*trans,
3335                                         dp->i_mount->m_ddev_targp,
3336                                         dblkno, dblkcnt, 0);
3337                         if (!bp)
3338                                 return ENOMEM;
3339                         xfs_trans_binval(*trans, bp);
3340                         /*
3341                          * Roll to next transaction.
3342                          */
3343                         error = xfs_trans_roll(trans, dp);
3344                         if (error)
3345                                 return (error);
3346                 }
3347
3348                 tblkno += map.br_blockcount;
3349                 tblkcnt -= map.br_blockcount;
3350         }
3351
3352         return(0);
3353 }