Linux 2.6.28-rc5
[cris-mirror.git] / fs / xfs / xfs_attr_leaf.c
blob79da6b2ea99efa35dee2f72f12785598995daf5a
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_alloc.h"
35 #include "xfs_btree.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_bmap.h"
42 #include "xfs_attr.h"
43 #include "xfs_attr_leaf.h"
44 #include "xfs_error.h"
47 * xfs_attr_leaf.c
49 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
52 /*========================================================================
53 * Function prototypes for the kernel.
54 *========================================================================*/
57 * Routines used for growing the Btree.
59 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
60 xfs_dabuf_t **bpp);
61 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
62 int freemap_index);
63 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
64 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
65 xfs_da_state_blk_t *blk1,
66 xfs_da_state_blk_t *blk2);
67 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
68 xfs_da_state_blk_t *leaf_blk_1,
69 xfs_da_state_blk_t *leaf_blk_2,
70 int *number_entries_in_blk1,
71 int *number_usedbytes_in_blk1);
74 * Routines used for shrinking the Btree.
76 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
77 xfs_dabuf_t *bp, int level);
78 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
79 xfs_dabuf_t *bp);
80 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
81 xfs_dablk_t blkno, int blkcnt);
84 * Utility routines.
86 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
87 int src_start,
88 xfs_attr_leafblock_t *dst_leaf,
89 int dst_start, int move_count,
90 xfs_mount_t *mp);
91 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
93 /*========================================================================
94 * Namespace helper routines
95 *========================================================================*/
98 * If namespace bits don't match return 0.
99 * If all match then return 1.
101 STATIC_INLINE int
102 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
104 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
108 /*========================================================================
109 * External routines when attribute fork size < XFS_LITINO(mp).
110 *========================================================================*/
113 * Query whether the requested number of additional bytes of extended
114 * attribute space will be able to fit inline.
115 * Returns zero if not, else the di_forkoff fork offset to be used in the
116 * literal area for attribute data once the new bytes have been added.
118 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
119 * special case for dev/uuid inodes, they have fixed size data forks.
122 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
124 int offset;
125 int minforkoff; /* lower limit on valid forkoff locations */
126 int maxforkoff; /* upper limit on valid forkoff locations */
127 int dsize;
128 xfs_mount_t *mp = dp->i_mount;
130 offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
132 switch (dp->i_d.di_format) {
133 case XFS_DINODE_FMT_DEV:
134 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
135 return (offset >= minforkoff) ? minforkoff : 0;
136 case XFS_DINODE_FMT_UUID:
137 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
138 return (offset >= minforkoff) ? minforkoff : 0;
141 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) {
142 if (bytes <= XFS_IFORK_ASIZE(dp))
143 return dp->i_d.di_forkoff;
144 return 0;
147 dsize = dp->i_df.if_bytes;
149 switch (dp->i_d.di_format) {
150 case XFS_DINODE_FMT_EXTENTS:
152 * If there is no attr fork and the data fork is extents,
153 * determine if creating the default attr fork will result
154 * in the extents form migrating to btree. If so, the
155 * minimum offset only needs to be the space required for
156 * the btree root.
158 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes > mp->m_attroffset)
159 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
160 break;
162 case XFS_DINODE_FMT_BTREE:
164 * If have data btree then keep forkoff if we have one,
165 * otherwise we are adding a new attr, so then we set
166 * minforkoff to where the btree root can finish so we have
167 * plenty of room for attrs
169 if (dp->i_d.di_forkoff) {
170 if (offset < dp->i_d.di_forkoff)
171 return 0;
172 else
173 return dp->i_d.di_forkoff;
174 } else
175 dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
176 break;
180 * A data fork btree root must have space for at least
181 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
183 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
184 minforkoff = roundup(minforkoff, 8) >> 3;
186 /* attr fork btree root can have at least this many key/ptr pairs */
187 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
188 maxforkoff = maxforkoff >> 3; /* rounded down */
190 if (offset >= minforkoff && offset < maxforkoff)
191 return offset;
192 if (offset >= maxforkoff)
193 return maxforkoff;
194 return 0;
198 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
200 STATIC void
201 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
203 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
204 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
205 spin_lock(&mp->m_sb_lock);
206 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
207 xfs_sb_version_addattr2(&mp->m_sb);
208 spin_unlock(&mp->m_sb_lock);
209 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
210 } else
211 spin_unlock(&mp->m_sb_lock);
216 * Create the initial contents of a shortform attribute list.
218 void
219 xfs_attr_shortform_create(xfs_da_args_t *args)
221 xfs_attr_sf_hdr_t *hdr;
222 xfs_inode_t *dp;
223 xfs_ifork_t *ifp;
225 dp = args->dp;
226 ASSERT(dp != NULL);
227 ifp = dp->i_afp;
228 ASSERT(ifp != NULL);
229 ASSERT(ifp->if_bytes == 0);
230 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
231 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
232 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
233 ifp->if_flags |= XFS_IFINLINE;
234 } else {
235 ASSERT(ifp->if_flags & XFS_IFINLINE);
237 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
238 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
239 hdr->count = 0;
240 hdr->totsize = cpu_to_be16(sizeof(*hdr));
241 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
245 * Add a name/value pair to the shortform attribute list.
246 * Overflow from the inode has already been checked for.
248 void
249 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
251 xfs_attr_shortform_t *sf;
252 xfs_attr_sf_entry_t *sfe;
253 int i, offset, size;
254 xfs_mount_t *mp;
255 xfs_inode_t *dp;
256 xfs_ifork_t *ifp;
258 dp = args->dp;
259 mp = dp->i_mount;
260 dp->i_d.di_forkoff = forkoff;
261 dp->i_df.if_ext_max =
262 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
263 dp->i_afp->if_ext_max =
264 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
266 ifp = dp->i_afp;
267 ASSERT(ifp->if_flags & XFS_IFINLINE);
268 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
269 sfe = &sf->list[0];
270 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
271 #ifdef DEBUG
272 if (sfe->namelen != args->namelen)
273 continue;
274 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
275 continue;
276 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
277 continue;
278 ASSERT(0);
279 #endif
282 offset = (char *)sfe - (char *)sf;
283 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
284 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
285 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
286 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
288 sfe->namelen = args->namelen;
289 sfe->valuelen = args->valuelen;
290 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
291 memcpy(sfe->nameval, args->name, args->namelen);
292 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
293 sf->hdr.count++;
294 be16_add_cpu(&sf->hdr.totsize, size);
295 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
297 xfs_sbversion_add_attr2(mp, args->trans);
301 * Remove an attribute from the shortform attribute list structure.
304 xfs_attr_shortform_remove(xfs_da_args_t *args)
306 xfs_attr_shortform_t *sf;
307 xfs_attr_sf_entry_t *sfe;
308 int base, size=0, end, totsize, i;
309 xfs_mount_t *mp;
310 xfs_inode_t *dp;
312 dp = args->dp;
313 mp = dp->i_mount;
314 base = sizeof(xfs_attr_sf_hdr_t);
315 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
316 sfe = &sf->list[0];
317 end = sf->hdr.count;
318 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
319 base += size, i++) {
320 size = XFS_ATTR_SF_ENTSIZE(sfe);
321 if (sfe->namelen != args->namelen)
322 continue;
323 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
324 continue;
325 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
326 continue;
327 break;
329 if (i == end)
330 return(XFS_ERROR(ENOATTR));
333 * Fix up the attribute fork data, covering the hole
335 end = base + size;
336 totsize = be16_to_cpu(sf->hdr.totsize);
337 if (end != totsize)
338 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
339 sf->hdr.count--;
340 be16_add_cpu(&sf->hdr.totsize, -size);
343 * Fix up the start offset of the attribute fork
345 totsize -= size;
346 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
347 !(args->op_flags & XFS_DA_OP_ADDNAME) &&
348 (mp->m_flags & XFS_MOUNT_ATTR2) &&
349 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE)) {
351 * Last attribute now removed, revert to original
352 * inode format making all literal area available
353 * to the data fork once more.
355 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
356 dp->i_d.di_forkoff = 0;
357 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
358 ASSERT(dp->i_d.di_anextents == 0);
359 ASSERT(dp->i_afp == NULL);
360 dp->i_df.if_ext_max =
361 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
362 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
363 } else {
364 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
365 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
366 ASSERT(dp->i_d.di_forkoff);
367 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
368 (args->op_flags & XFS_DA_OP_ADDNAME) ||
369 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
370 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
371 dp->i_afp->if_ext_max =
372 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
373 dp->i_df.if_ext_max =
374 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
375 xfs_trans_log_inode(args->trans, dp,
376 XFS_ILOG_CORE | XFS_ILOG_ADATA);
379 xfs_sbversion_add_attr2(mp, args->trans);
381 return(0);
385 * Look up a name in a shortform attribute list structure.
387 /*ARGSUSED*/
389 xfs_attr_shortform_lookup(xfs_da_args_t *args)
391 xfs_attr_shortform_t *sf;
392 xfs_attr_sf_entry_t *sfe;
393 int i;
394 xfs_ifork_t *ifp;
396 ifp = args->dp->i_afp;
397 ASSERT(ifp->if_flags & XFS_IFINLINE);
398 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
399 sfe = &sf->list[0];
400 for (i = 0; i < sf->hdr.count;
401 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
402 if (sfe->namelen != args->namelen)
403 continue;
404 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
405 continue;
406 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
407 continue;
408 return(XFS_ERROR(EEXIST));
410 return(XFS_ERROR(ENOATTR));
414 * Look up a name in a shortform attribute list structure.
416 /*ARGSUSED*/
418 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
420 xfs_attr_shortform_t *sf;
421 xfs_attr_sf_entry_t *sfe;
422 int i;
424 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
425 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
426 sfe = &sf->list[0];
427 for (i = 0; i < sf->hdr.count;
428 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
429 if (sfe->namelen != args->namelen)
430 continue;
431 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
432 continue;
433 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
434 continue;
435 if (args->flags & ATTR_KERNOVAL) {
436 args->valuelen = sfe->valuelen;
437 return(XFS_ERROR(EEXIST));
439 if (args->valuelen < sfe->valuelen) {
440 args->valuelen = sfe->valuelen;
441 return(XFS_ERROR(ERANGE));
443 args->valuelen = sfe->valuelen;
444 memcpy(args->value, &sfe->nameval[args->namelen],
445 args->valuelen);
446 return(XFS_ERROR(EEXIST));
448 return(XFS_ERROR(ENOATTR));
452 * Convert from using the shortform to the leaf.
455 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
457 xfs_inode_t *dp;
458 xfs_attr_shortform_t *sf;
459 xfs_attr_sf_entry_t *sfe;
460 xfs_da_args_t nargs;
461 char *tmpbuffer;
462 int error, i, size;
463 xfs_dablk_t blkno;
464 xfs_dabuf_t *bp;
465 xfs_ifork_t *ifp;
467 dp = args->dp;
468 ifp = dp->i_afp;
469 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
470 size = be16_to_cpu(sf->hdr.totsize);
471 tmpbuffer = kmem_alloc(size, KM_SLEEP);
472 ASSERT(tmpbuffer != NULL);
473 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
474 sf = (xfs_attr_shortform_t *)tmpbuffer;
476 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
477 bp = NULL;
478 error = xfs_da_grow_inode(args, &blkno);
479 if (error) {
481 * If we hit an IO error middle of the transaction inside
482 * grow_inode(), we may have inconsistent data. Bail out.
484 if (error == EIO)
485 goto out;
486 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
487 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
488 goto out;
491 ASSERT(blkno == 0);
492 error = xfs_attr_leaf_create(args, blkno, &bp);
493 if (error) {
494 error = xfs_da_shrink_inode(args, 0, bp);
495 bp = NULL;
496 if (error)
497 goto out;
498 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
499 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
500 goto out;
503 memset((char *)&nargs, 0, sizeof(nargs));
504 nargs.dp = dp;
505 nargs.firstblock = args->firstblock;
506 nargs.flist = args->flist;
507 nargs.total = args->total;
508 nargs.whichfork = XFS_ATTR_FORK;
509 nargs.trans = args->trans;
510 nargs.op_flags = XFS_DA_OP_OKNOENT;
512 sfe = &sf->list[0];
513 for (i = 0; i < sf->hdr.count; i++) {
514 nargs.name = (char *)sfe->nameval;
515 nargs.namelen = sfe->namelen;
516 nargs.value = (char *)&sfe->nameval[nargs.namelen];
517 nargs.valuelen = sfe->valuelen;
518 nargs.hashval = xfs_da_hashname((char *)sfe->nameval,
519 sfe->namelen);
520 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
521 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
522 ASSERT(error == ENOATTR);
523 error = xfs_attr_leaf_add(bp, &nargs);
524 ASSERT(error != ENOSPC);
525 if (error)
526 goto out;
527 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
529 error = 0;
531 out:
532 if(bp)
533 xfs_da_buf_done(bp);
534 kmem_free(tmpbuffer);
535 return(error);
538 STATIC int
539 xfs_attr_shortform_compare(const void *a, const void *b)
541 xfs_attr_sf_sort_t *sa, *sb;
543 sa = (xfs_attr_sf_sort_t *)a;
544 sb = (xfs_attr_sf_sort_t *)b;
545 if (sa->hash < sb->hash) {
546 return(-1);
547 } else if (sa->hash > sb->hash) {
548 return(1);
549 } else {
550 return(sa->entno - sb->entno);
555 #define XFS_ISRESET_CURSOR(cursor) \
556 (!((cursor)->initted) && !((cursor)->hashval) && \
557 !((cursor)->blkno) && !((cursor)->offset))
559 * Copy out entries of shortform attribute lists for attr_list().
560 * Shortform attribute lists are not stored in hashval sorted order.
561 * If the output buffer is not large enough to hold them all, then we
562 * we have to calculate each entries' hashvalue and sort them before
563 * we can begin returning them to the user.
565 /*ARGSUSED*/
567 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
569 attrlist_cursor_kern_t *cursor;
570 xfs_attr_sf_sort_t *sbuf, *sbp;
571 xfs_attr_shortform_t *sf;
572 xfs_attr_sf_entry_t *sfe;
573 xfs_inode_t *dp;
574 int sbsize, nsbuf, count, i;
575 int error;
577 ASSERT(context != NULL);
578 dp = context->dp;
579 ASSERT(dp != NULL);
580 ASSERT(dp->i_afp != NULL);
581 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
582 ASSERT(sf != NULL);
583 if (!sf->hdr.count)
584 return(0);
585 cursor = context->cursor;
586 ASSERT(cursor != NULL);
588 xfs_attr_trace_l_c("sf start", context);
591 * If the buffer is large enough and the cursor is at the start,
592 * do not bother with sorting since we will return everything in
593 * one buffer and another call using the cursor won't need to be
594 * made.
595 * Note the generous fudge factor of 16 overhead bytes per entry.
596 * If bufsize is zero then put_listent must be a search function
597 * and can just scan through what we have.
599 if (context->bufsize == 0 ||
600 (XFS_ISRESET_CURSOR(cursor) &&
601 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
602 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
603 error = context->put_listent(context,
604 sfe->flags,
605 (char *)sfe->nameval,
606 (int)sfe->namelen,
607 (int)sfe->valuelen,
608 (char*)&sfe->nameval[sfe->namelen]);
611 * Either search callback finished early or
612 * didn't fit it all in the buffer after all.
614 if (context->seen_enough)
615 break;
617 if (error)
618 return error;
619 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
621 xfs_attr_trace_l_c("sf big-gulp", context);
622 return(0);
625 /* do no more for a search callback */
626 if (context->bufsize == 0)
627 return 0;
630 * It didn't all fit, so we have to sort everything on hashval.
632 sbsize = sf->hdr.count * sizeof(*sbuf);
633 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP);
636 * Scan the attribute list for the rest of the entries, storing
637 * the relevant info from only those that match into a buffer.
639 nsbuf = 0;
640 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
641 if (unlikely(
642 ((char *)sfe < (char *)sf) ||
643 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
644 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
645 XFS_ERRLEVEL_LOW,
646 context->dp->i_mount, sfe);
647 xfs_attr_trace_l_c("sf corrupted", context);
648 kmem_free(sbuf);
649 return XFS_ERROR(EFSCORRUPTED);
652 sbp->entno = i;
653 sbp->hash = xfs_da_hashname((char *)sfe->nameval, sfe->namelen);
654 sbp->name = (char *)sfe->nameval;
655 sbp->namelen = sfe->namelen;
656 /* These are bytes, and both on-disk, don't endian-flip */
657 sbp->valuelen = sfe->valuelen;
658 sbp->flags = sfe->flags;
659 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
660 sbp++;
661 nsbuf++;
665 * Sort the entries on hash then entno.
667 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
670 * Re-find our place IN THE SORTED LIST.
672 count = 0;
673 cursor->initted = 1;
674 cursor->blkno = 0;
675 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
676 if (sbp->hash == cursor->hashval) {
677 if (cursor->offset == count) {
678 break;
680 count++;
681 } else if (sbp->hash > cursor->hashval) {
682 break;
685 if (i == nsbuf) {
686 kmem_free(sbuf);
687 xfs_attr_trace_l_c("blk end", context);
688 return(0);
692 * Loop putting entries into the user buffer.
694 for ( ; i < nsbuf; i++, sbp++) {
695 if (cursor->hashval != sbp->hash) {
696 cursor->hashval = sbp->hash;
697 cursor->offset = 0;
699 error = context->put_listent(context,
700 sbp->flags,
701 sbp->name,
702 sbp->namelen,
703 sbp->valuelen,
704 &sbp->name[sbp->namelen]);
705 if (error)
706 return error;
707 if (context->seen_enough)
708 break;
709 cursor->offset++;
712 kmem_free(sbuf);
713 xfs_attr_trace_l_c("sf E-O-F", context);
714 return(0);
718 * Check a leaf attribute block to see if all the entries would fit into
719 * a shortform attribute list.
722 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
724 xfs_attr_leafblock_t *leaf;
725 xfs_attr_leaf_entry_t *entry;
726 xfs_attr_leaf_name_local_t *name_loc;
727 int bytes, i;
729 leaf = bp->data;
730 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
732 entry = &leaf->entries[0];
733 bytes = sizeof(struct xfs_attr_sf_hdr);
734 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
735 if (entry->flags & XFS_ATTR_INCOMPLETE)
736 continue; /* don't copy partial entries */
737 if (!(entry->flags & XFS_ATTR_LOCAL))
738 return(0);
739 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
740 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
741 return(0);
742 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
743 return(0);
744 bytes += sizeof(struct xfs_attr_sf_entry)-1
745 + name_loc->namelen
746 + be16_to_cpu(name_loc->valuelen);
748 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
749 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
750 (bytes == sizeof(struct xfs_attr_sf_hdr)))
751 return(-1);
752 return(xfs_attr_shortform_bytesfit(dp, bytes));
756 * Convert a leaf attribute list to shortform attribute list
759 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
761 xfs_attr_leafblock_t *leaf;
762 xfs_attr_leaf_entry_t *entry;
763 xfs_attr_leaf_name_local_t *name_loc;
764 xfs_da_args_t nargs;
765 xfs_inode_t *dp;
766 char *tmpbuffer;
767 int error, i;
769 dp = args->dp;
770 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
771 ASSERT(tmpbuffer != NULL);
773 ASSERT(bp != NULL);
774 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
775 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
776 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
777 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
780 * Clean out the prior contents of the attribute list.
782 error = xfs_da_shrink_inode(args, 0, bp);
783 if (error)
784 goto out;
786 if (forkoff == -1) {
787 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
788 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
791 * Last attribute was removed, revert to original
792 * inode format making all literal area available
793 * to the data fork once more.
795 xfs_idestroy_fork(dp, XFS_ATTR_FORK);
796 dp->i_d.di_forkoff = 0;
797 dp->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
798 ASSERT(dp->i_d.di_anextents == 0);
799 ASSERT(dp->i_afp == NULL);
800 dp->i_df.if_ext_max =
801 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
802 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
803 goto out;
806 xfs_attr_shortform_create(args);
809 * Copy the attributes
811 memset((char *)&nargs, 0, sizeof(nargs));
812 nargs.dp = dp;
813 nargs.firstblock = args->firstblock;
814 nargs.flist = args->flist;
815 nargs.total = args->total;
816 nargs.whichfork = XFS_ATTR_FORK;
817 nargs.trans = args->trans;
818 nargs.op_flags = XFS_DA_OP_OKNOENT;
819 entry = &leaf->entries[0];
820 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
821 if (entry->flags & XFS_ATTR_INCOMPLETE)
822 continue; /* don't copy partial entries */
823 if (!entry->nameidx)
824 continue;
825 ASSERT(entry->flags & XFS_ATTR_LOCAL);
826 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
827 nargs.name = (char *)name_loc->nameval;
828 nargs.namelen = name_loc->namelen;
829 nargs.value = (char *)&name_loc->nameval[nargs.namelen];
830 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
831 nargs.hashval = be32_to_cpu(entry->hashval);
832 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
833 xfs_attr_shortform_add(&nargs, forkoff);
835 error = 0;
837 out:
838 kmem_free(tmpbuffer);
839 return(error);
843 * Convert from using a single leaf to a root node and a leaf.
846 xfs_attr_leaf_to_node(xfs_da_args_t *args)
848 xfs_attr_leafblock_t *leaf;
849 xfs_da_intnode_t *node;
850 xfs_inode_t *dp;
851 xfs_dabuf_t *bp1, *bp2;
852 xfs_dablk_t blkno;
853 int error;
855 dp = args->dp;
856 bp1 = bp2 = NULL;
857 error = xfs_da_grow_inode(args, &blkno);
858 if (error)
859 goto out;
860 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
861 XFS_ATTR_FORK);
862 if (error)
863 goto out;
864 ASSERT(bp1 != NULL);
865 bp2 = NULL;
866 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
867 XFS_ATTR_FORK);
868 if (error)
869 goto out;
870 ASSERT(bp2 != NULL);
871 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
872 xfs_da_buf_done(bp1);
873 bp1 = NULL;
874 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
877 * Set up the new root node.
879 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
880 if (error)
881 goto out;
882 node = bp1->data;
883 leaf = bp2->data;
884 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
885 /* both on-disk, don't endian-flip twice */
886 node->btree[0].hashval =
887 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
888 node->btree[0].before = cpu_to_be32(blkno);
889 node->hdr.count = cpu_to_be16(1);
890 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
891 error = 0;
892 out:
893 if (bp1)
894 xfs_da_buf_done(bp1);
895 if (bp2)
896 xfs_da_buf_done(bp2);
897 return(error);
901 /*========================================================================
902 * Routines used for growing the Btree.
903 *========================================================================*/
906 * Create the initial contents of a leaf attribute list
907 * or a leaf in a node attribute list.
909 STATIC int
910 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
912 xfs_attr_leafblock_t *leaf;
913 xfs_attr_leaf_hdr_t *hdr;
914 xfs_inode_t *dp;
915 xfs_dabuf_t *bp;
916 int error;
918 dp = args->dp;
919 ASSERT(dp != NULL);
920 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
921 XFS_ATTR_FORK);
922 if (error)
923 return(error);
924 ASSERT(bp != NULL);
925 leaf = bp->data;
926 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
927 hdr = &leaf->hdr;
928 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
929 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
930 if (!hdr->firstused) {
931 hdr->firstused = cpu_to_be16(
932 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
935 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
936 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
937 sizeof(xfs_attr_leaf_hdr_t));
939 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
941 *bpp = bp;
942 return(0);
946 * Split the leaf node, rebalance, then add the new entry.
949 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
950 xfs_da_state_blk_t *newblk)
952 xfs_dablk_t blkno;
953 int error;
956 * Allocate space for a new leaf node.
958 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
959 error = xfs_da_grow_inode(state->args, &blkno);
960 if (error)
961 return(error);
962 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
963 if (error)
964 return(error);
965 newblk->blkno = blkno;
966 newblk->magic = XFS_ATTR_LEAF_MAGIC;
969 * Rebalance the entries across the two leaves.
970 * NOTE: rebalance() currently depends on the 2nd block being empty.
972 xfs_attr_leaf_rebalance(state, oldblk, newblk);
973 error = xfs_da_blk_link(state, oldblk, newblk);
974 if (error)
975 return(error);
978 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
979 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
980 * "new" attrs info. Will need the "old" info to remove it later.
982 * Insert the "new" entry in the correct block.
984 if (state->inleaf)
985 error = xfs_attr_leaf_add(oldblk->bp, state->args);
986 else
987 error = xfs_attr_leaf_add(newblk->bp, state->args);
990 * Update last hashval in each block since we added the name.
992 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
993 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
994 return(error);
998 * Add a name to the leaf attribute list structure.
1001 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
1003 xfs_attr_leafblock_t *leaf;
1004 xfs_attr_leaf_hdr_t *hdr;
1005 xfs_attr_leaf_map_t *map;
1006 int tablesize, entsize, sum, tmp, i;
1008 leaf = bp->data;
1009 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1010 ASSERT((args->index >= 0)
1011 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1012 hdr = &leaf->hdr;
1013 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1014 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1017 * Search through freemap for first-fit on new name length.
1018 * (may need to figure in size of entry struct too)
1020 tablesize = (be16_to_cpu(hdr->count) + 1)
1021 * sizeof(xfs_attr_leaf_entry_t)
1022 + sizeof(xfs_attr_leaf_hdr_t);
1023 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1024 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1025 if (tablesize > be16_to_cpu(hdr->firstused)) {
1026 sum += be16_to_cpu(map->size);
1027 continue;
1029 if (!map->size)
1030 continue; /* no space in this map */
1031 tmp = entsize;
1032 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1033 tmp += sizeof(xfs_attr_leaf_entry_t);
1034 if (be16_to_cpu(map->size) >= tmp) {
1035 tmp = xfs_attr_leaf_add_work(bp, args, i);
1036 return(tmp);
1038 sum += be16_to_cpu(map->size);
1042 * If there are no holes in the address space of the block,
1043 * and we don't have enough freespace, then compaction will do us
1044 * no good and we should just give up.
1046 if (!hdr->holes && (sum < entsize))
1047 return(XFS_ERROR(ENOSPC));
1050 * Compact the entries to coalesce free space.
1051 * This may change the hdr->count via dropping INCOMPLETE entries.
1053 xfs_attr_leaf_compact(args->trans, bp);
1056 * After compaction, the block is guaranteed to have only one
1057 * free region, in freemap[0]. If it is not big enough, give up.
1059 if (be16_to_cpu(hdr->freemap[0].size)
1060 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1061 return(XFS_ERROR(ENOSPC));
1063 return(xfs_attr_leaf_add_work(bp, args, 0));
1067 * Add a name to a leaf attribute list structure.
1069 STATIC int
1070 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1072 xfs_attr_leafblock_t *leaf;
1073 xfs_attr_leaf_hdr_t *hdr;
1074 xfs_attr_leaf_entry_t *entry;
1075 xfs_attr_leaf_name_local_t *name_loc;
1076 xfs_attr_leaf_name_remote_t *name_rmt;
1077 xfs_attr_leaf_map_t *map;
1078 xfs_mount_t *mp;
1079 int tmp, i;
1081 leaf = bp->data;
1082 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1083 hdr = &leaf->hdr;
1084 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1085 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1088 * Force open some space in the entry array and fill it in.
1090 entry = &leaf->entries[args->index];
1091 if (args->index < be16_to_cpu(hdr->count)) {
1092 tmp = be16_to_cpu(hdr->count) - args->index;
1093 tmp *= sizeof(xfs_attr_leaf_entry_t);
1094 memmove((char *)(entry+1), (char *)entry, tmp);
1095 xfs_da_log_buf(args->trans, bp,
1096 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1098 be16_add_cpu(&hdr->count, 1);
1101 * Allocate space for the new string (at the end of the run).
1103 map = &hdr->freemap[mapindex];
1104 mp = args->trans->t_mountp;
1105 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1106 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1107 ASSERT(be16_to_cpu(map->size) >=
1108 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1109 mp->m_sb.sb_blocksize, NULL));
1110 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1111 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1112 be16_add_cpu(&map->size,
1113 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1114 mp->m_sb.sb_blocksize, &tmp));
1115 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1116 be16_to_cpu(map->size));
1117 entry->hashval = cpu_to_be32(args->hashval);
1118 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1119 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1120 if (args->op_flags & XFS_DA_OP_RENAME) {
1121 entry->flags |= XFS_ATTR_INCOMPLETE;
1122 if ((args->blkno2 == args->blkno) &&
1123 (args->index2 <= args->index)) {
1124 args->index2++;
1127 xfs_da_log_buf(args->trans, bp,
1128 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1129 ASSERT((args->index == 0) ||
1130 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1131 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1132 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1135 * Copy the attribute name and value into the new space.
1137 * For "remote" attribute values, simply note that we need to
1138 * allocate space for the "remote" value. We can't actually
1139 * allocate the extents in this transaction, and we can't decide
1140 * which blocks they should be as we might allocate more blocks
1141 * as part of this transaction (a split operation for example).
1143 if (entry->flags & XFS_ATTR_LOCAL) {
1144 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
1145 name_loc->namelen = args->namelen;
1146 name_loc->valuelen = cpu_to_be16(args->valuelen);
1147 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1148 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1149 be16_to_cpu(name_loc->valuelen));
1150 } else {
1151 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
1152 name_rmt->namelen = args->namelen;
1153 memcpy((char *)name_rmt->name, args->name, args->namelen);
1154 entry->flags |= XFS_ATTR_INCOMPLETE;
1155 /* just in case */
1156 name_rmt->valuelen = 0;
1157 name_rmt->valueblk = 0;
1158 args->rmtblkno = 1;
1159 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1161 xfs_da_log_buf(args->trans, bp,
1162 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1163 xfs_attr_leaf_entsize(leaf, args->index)));
1166 * Update the control info for this leaf node
1168 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1169 /* both on-disk, don't endian-flip twice */
1170 hdr->firstused = entry->nameidx;
1172 ASSERT(be16_to_cpu(hdr->firstused) >=
1173 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1174 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1175 + sizeof(xfs_attr_leaf_hdr_t);
1176 map = &hdr->freemap[0];
1177 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1178 if (be16_to_cpu(map->base) == tmp) {
1179 be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1180 be16_add_cpu(&map->size,
1181 -((int)sizeof(xfs_attr_leaf_entry_t)));
1184 be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1185 xfs_da_log_buf(args->trans, bp,
1186 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1187 return(0);
1191 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1193 STATIC void
1194 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1196 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1197 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1198 xfs_mount_t *mp;
1199 char *tmpbuffer;
1201 mp = trans->t_mountp;
1202 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1203 ASSERT(tmpbuffer != NULL);
1204 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1205 memset(bp->data, 0, XFS_LBSIZE(mp));
1208 * Copy basic information
1210 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1211 leaf_d = bp->data;
1212 hdr_s = &leaf_s->hdr;
1213 hdr_d = &leaf_d->hdr;
1214 hdr_d->info = hdr_s->info; /* struct copy */
1215 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1216 /* handle truncation gracefully */
1217 if (!hdr_d->firstused) {
1218 hdr_d->firstused = cpu_to_be16(
1219 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1221 hdr_d->usedbytes = 0;
1222 hdr_d->count = 0;
1223 hdr_d->holes = 0;
1224 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1225 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1226 sizeof(xfs_attr_leaf_hdr_t));
1229 * Copy all entry's in the same (sorted) order,
1230 * but allocate name/value pairs packed and in sequence.
1232 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1233 be16_to_cpu(hdr_s->count), mp);
1234 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1236 kmem_free(tmpbuffer);
1240 * Redistribute the attribute list entries between two leaf nodes,
1241 * taking into account the size of the new entry.
1243 * NOTE: if new block is empty, then it will get the upper half of the
1244 * old block. At present, all (one) callers pass in an empty second block.
1246 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1247 * to match what it is doing in splitting the attribute leaf block. Those
1248 * values are used in "atomic rename" operations on attributes. Note that
1249 * the "new" and "old" values can end up in different blocks.
1251 STATIC void
1252 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1253 xfs_da_state_blk_t *blk2)
1255 xfs_da_args_t *args;
1256 xfs_da_state_blk_t *tmp_blk;
1257 xfs_attr_leafblock_t *leaf1, *leaf2;
1258 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1259 int count, totallen, max, space, swap;
1262 * Set up environment.
1264 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1265 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1266 leaf1 = blk1->bp->data;
1267 leaf2 = blk2->bp->data;
1268 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1269 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1270 args = state->args;
1273 * Check ordering of blocks, reverse if it makes things simpler.
1275 * NOTE: Given that all (current) callers pass in an empty
1276 * second block, this code should never set "swap".
1278 swap = 0;
1279 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1280 tmp_blk = blk1;
1281 blk1 = blk2;
1282 blk2 = tmp_blk;
1283 leaf1 = blk1->bp->data;
1284 leaf2 = blk2->bp->data;
1285 swap = 1;
1287 hdr1 = &leaf1->hdr;
1288 hdr2 = &leaf2->hdr;
1291 * Examine entries until we reduce the absolute difference in
1292 * byte usage between the two blocks to a minimum. Then get
1293 * the direction to copy and the number of elements to move.
1295 * "inleaf" is true if the new entry should be inserted into blk1.
1296 * If "swap" is also true, then reverse the sense of "inleaf".
1298 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1299 &count, &totallen);
1300 if (swap)
1301 state->inleaf = !state->inleaf;
1304 * Move any entries required from leaf to leaf:
1306 if (count < be16_to_cpu(hdr1->count)) {
1308 * Figure the total bytes to be added to the destination leaf.
1310 /* number entries being moved */
1311 count = be16_to_cpu(hdr1->count) - count;
1312 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1313 space += count * sizeof(xfs_attr_leaf_entry_t);
1316 * leaf2 is the destination, compact it if it looks tight.
1318 max = be16_to_cpu(hdr2->firstused)
1319 - sizeof(xfs_attr_leaf_hdr_t);
1320 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1321 if (space > max) {
1322 xfs_attr_leaf_compact(args->trans, blk2->bp);
1326 * Move high entries from leaf1 to low end of leaf2.
1328 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1329 leaf2, 0, count, state->mp);
1331 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1332 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1333 } else if (count > be16_to_cpu(hdr1->count)) {
1335 * I assert that since all callers pass in an empty
1336 * second buffer, this code should never execute.
1340 * Figure the total bytes to be added to the destination leaf.
1342 /* number entries being moved */
1343 count -= be16_to_cpu(hdr1->count);
1344 space = totallen - be16_to_cpu(hdr1->usedbytes);
1345 space += count * sizeof(xfs_attr_leaf_entry_t);
1348 * leaf1 is the destination, compact it if it looks tight.
1350 max = be16_to_cpu(hdr1->firstused)
1351 - sizeof(xfs_attr_leaf_hdr_t);
1352 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1353 if (space > max) {
1354 xfs_attr_leaf_compact(args->trans, blk1->bp);
1358 * Move low entries from leaf2 to high end of leaf1.
1360 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1361 be16_to_cpu(hdr1->count), count, state->mp);
1363 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1364 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1368 * Copy out last hashval in each block for B-tree code.
1370 blk1->hashval = be32_to_cpu(
1371 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1372 blk2->hashval = be32_to_cpu(
1373 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1376 * Adjust the expected index for insertion.
1377 * NOTE: this code depends on the (current) situation that the
1378 * second block was originally empty.
1380 * If the insertion point moved to the 2nd block, we must adjust
1381 * the index. We must also track the entry just following the
1382 * new entry for use in an "atomic rename" operation, that entry
1383 * is always the "old" entry and the "new" entry is what we are
1384 * inserting. The index/blkno fields refer to the "old" entry,
1385 * while the index2/blkno2 fields refer to the "new" entry.
1387 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1388 ASSERT(state->inleaf == 0);
1389 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1390 args->index = args->index2 = blk2->index;
1391 args->blkno = args->blkno2 = blk2->blkno;
1392 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1393 if (state->inleaf) {
1394 args->index = blk1->index;
1395 args->blkno = blk1->blkno;
1396 args->index2 = 0;
1397 args->blkno2 = blk2->blkno;
1398 } else {
1399 blk2->index = blk1->index
1400 - be16_to_cpu(leaf1->hdr.count);
1401 args->index = args->index2 = blk2->index;
1402 args->blkno = args->blkno2 = blk2->blkno;
1404 } else {
1405 ASSERT(state->inleaf == 1);
1406 args->index = args->index2 = blk1->index;
1407 args->blkno = args->blkno2 = blk1->blkno;
1412 * Examine entries until we reduce the absolute difference in
1413 * byte usage between the two blocks to a minimum.
1414 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1415 * GROT: there will always be enough room in either block for a new entry.
1416 * GROT: Do a double-split for this case?
1418 STATIC int
1419 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1420 xfs_da_state_blk_t *blk1,
1421 xfs_da_state_blk_t *blk2,
1422 int *countarg, int *usedbytesarg)
1424 xfs_attr_leafblock_t *leaf1, *leaf2;
1425 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1426 xfs_attr_leaf_entry_t *entry;
1427 int count, max, index, totallen, half;
1428 int lastdelta, foundit, tmp;
1431 * Set up environment.
1433 leaf1 = blk1->bp->data;
1434 leaf2 = blk2->bp->data;
1435 hdr1 = &leaf1->hdr;
1436 hdr2 = &leaf2->hdr;
1437 foundit = 0;
1438 totallen = 0;
1441 * Examine entries until we reduce the absolute difference in
1442 * byte usage between the two blocks to a minimum.
1444 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1445 half = (max+1) * sizeof(*entry);
1446 half += be16_to_cpu(hdr1->usedbytes) +
1447 be16_to_cpu(hdr2->usedbytes) +
1448 xfs_attr_leaf_newentsize(
1449 state->args->namelen,
1450 state->args->valuelen,
1451 state->blocksize, NULL);
1452 half /= 2;
1453 lastdelta = state->blocksize;
1454 entry = &leaf1->entries[0];
1455 for (count = index = 0; count < max; entry++, index++, count++) {
1457 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1459 * The new entry is in the first block, account for it.
1461 if (count == blk1->index) {
1462 tmp = totallen + sizeof(*entry) +
1463 xfs_attr_leaf_newentsize(
1464 state->args->namelen,
1465 state->args->valuelen,
1466 state->blocksize, NULL);
1467 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1468 break;
1469 lastdelta = XFS_ATTR_ABS(half - tmp);
1470 totallen = tmp;
1471 foundit = 1;
1475 * Wrap around into the second block if necessary.
1477 if (count == be16_to_cpu(hdr1->count)) {
1478 leaf1 = leaf2;
1479 entry = &leaf1->entries[0];
1480 index = 0;
1484 * Figure out if next leaf entry would be too much.
1486 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1487 index);
1488 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1489 break;
1490 lastdelta = XFS_ATTR_ABS(half - tmp);
1491 totallen = tmp;
1492 #undef XFS_ATTR_ABS
1496 * Calculate the number of usedbytes that will end up in lower block.
1497 * If new entry not in lower block, fix up the count.
1499 totallen -= count * sizeof(*entry);
1500 if (foundit) {
1501 totallen -= sizeof(*entry) +
1502 xfs_attr_leaf_newentsize(
1503 state->args->namelen,
1504 state->args->valuelen,
1505 state->blocksize, NULL);
1508 *countarg = count;
1509 *usedbytesarg = totallen;
1510 return(foundit);
1513 /*========================================================================
1514 * Routines used for shrinking the Btree.
1515 *========================================================================*/
1518 * Check a leaf block and its neighbors to see if the block should be
1519 * collapsed into one or the other neighbor. Always keep the block
1520 * with the smaller block number.
1521 * If the current block is over 50% full, don't try to join it, return 0.
1522 * If the block is empty, fill in the state structure and return 2.
1523 * If it can be collapsed, fill in the state structure and return 1.
1524 * If nothing can be done, return 0.
1526 * GROT: allow for INCOMPLETE entries in calculation.
1529 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1531 xfs_attr_leafblock_t *leaf;
1532 xfs_da_state_blk_t *blk;
1533 xfs_da_blkinfo_t *info;
1534 int count, bytes, forward, error, retval, i;
1535 xfs_dablk_t blkno;
1536 xfs_dabuf_t *bp;
1539 * Check for the degenerate case of the block being over 50% full.
1540 * If so, it's not worth even looking to see if we might be able
1541 * to coalesce with a sibling.
1543 blk = &state->path.blk[ state->path.active-1 ];
1544 info = blk->bp->data;
1545 ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1546 leaf = (xfs_attr_leafblock_t *)info;
1547 count = be16_to_cpu(leaf->hdr.count);
1548 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1549 count * sizeof(xfs_attr_leaf_entry_t) +
1550 be16_to_cpu(leaf->hdr.usedbytes);
1551 if (bytes > (state->blocksize >> 1)) {
1552 *action = 0; /* blk over 50%, don't try to join */
1553 return(0);
1557 * Check for the degenerate case of the block being empty.
1558 * If the block is empty, we'll simply delete it, no need to
1559 * coalesce it with a sibling block. We choose (arbitrarily)
1560 * to merge with the forward block unless it is NULL.
1562 if (count == 0) {
1564 * Make altpath point to the block we want to keep and
1565 * path point to the block we want to drop (this one).
1567 forward = (info->forw != 0);
1568 memcpy(&state->altpath, &state->path, sizeof(state->path));
1569 error = xfs_da_path_shift(state, &state->altpath, forward,
1570 0, &retval);
1571 if (error)
1572 return(error);
1573 if (retval) {
1574 *action = 0;
1575 } else {
1576 *action = 2;
1578 return(0);
1582 * Examine each sibling block to see if we can coalesce with
1583 * at least 25% free space to spare. We need to figure out
1584 * whether to merge with the forward or the backward block.
1585 * We prefer coalescing with the lower numbered sibling so as
1586 * to shrink an attribute list over time.
1588 /* start with smaller blk num */
1589 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1590 for (i = 0; i < 2; forward = !forward, i++) {
1591 if (forward)
1592 blkno = be32_to_cpu(info->forw);
1593 else
1594 blkno = be32_to_cpu(info->back);
1595 if (blkno == 0)
1596 continue;
1597 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1598 blkno, -1, &bp, XFS_ATTR_FORK);
1599 if (error)
1600 return(error);
1601 ASSERT(bp != NULL);
1603 leaf = (xfs_attr_leafblock_t *)info;
1604 count = be16_to_cpu(leaf->hdr.count);
1605 bytes = state->blocksize - (state->blocksize>>2);
1606 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1607 leaf = bp->data;
1608 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1609 count += be16_to_cpu(leaf->hdr.count);
1610 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1611 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1612 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1613 xfs_da_brelse(state->args->trans, bp);
1614 if (bytes >= 0)
1615 break; /* fits with at least 25% to spare */
1617 if (i >= 2) {
1618 *action = 0;
1619 return(0);
1623 * Make altpath point to the block we want to keep (the lower
1624 * numbered block) and path point to the block we want to drop.
1626 memcpy(&state->altpath, &state->path, sizeof(state->path));
1627 if (blkno < blk->blkno) {
1628 error = xfs_da_path_shift(state, &state->altpath, forward,
1629 0, &retval);
1630 } else {
1631 error = xfs_da_path_shift(state, &state->path, forward,
1632 0, &retval);
1634 if (error)
1635 return(error);
1636 if (retval) {
1637 *action = 0;
1638 } else {
1639 *action = 1;
1641 return(0);
1645 * Remove a name from the leaf attribute list structure.
1647 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1648 * If two leaves are 37% full, when combined they will leave 25% free.
1651 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1653 xfs_attr_leafblock_t *leaf;
1654 xfs_attr_leaf_hdr_t *hdr;
1655 xfs_attr_leaf_map_t *map;
1656 xfs_attr_leaf_entry_t *entry;
1657 int before, after, smallest, entsize;
1658 int tablesize, tmp, i;
1659 xfs_mount_t *mp;
1661 leaf = bp->data;
1662 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1663 hdr = &leaf->hdr;
1664 mp = args->trans->t_mountp;
1665 ASSERT((be16_to_cpu(hdr->count) > 0)
1666 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1667 ASSERT((args->index >= 0)
1668 && (args->index < be16_to_cpu(hdr->count)));
1669 ASSERT(be16_to_cpu(hdr->firstused) >=
1670 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1671 entry = &leaf->entries[args->index];
1672 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1673 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1676 * Scan through free region table:
1677 * check for adjacency of free'd entry with an existing one,
1678 * find smallest free region in case we need to replace it,
1679 * adjust any map that borders the entry table,
1681 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1682 + sizeof(xfs_attr_leaf_hdr_t);
1683 map = &hdr->freemap[0];
1684 tmp = be16_to_cpu(map->size);
1685 before = after = -1;
1686 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1687 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1688 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1689 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1690 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1691 if (be16_to_cpu(map->base) == tablesize) {
1692 be16_add_cpu(&map->base,
1693 -((int)sizeof(xfs_attr_leaf_entry_t)));
1694 be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1697 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1698 == be16_to_cpu(entry->nameidx)) {
1699 before = i;
1700 } else if (be16_to_cpu(map->base)
1701 == (be16_to_cpu(entry->nameidx) + entsize)) {
1702 after = i;
1703 } else if (be16_to_cpu(map->size) < tmp) {
1704 tmp = be16_to_cpu(map->size);
1705 smallest = i;
1710 * Coalesce adjacent freemap regions,
1711 * or replace the smallest region.
1713 if ((before >= 0) || (after >= 0)) {
1714 if ((before >= 0) && (after >= 0)) {
1715 map = &hdr->freemap[before];
1716 be16_add_cpu(&map->size, entsize);
1717 be16_add_cpu(&map->size,
1718 be16_to_cpu(hdr->freemap[after].size));
1719 hdr->freemap[after].base = 0;
1720 hdr->freemap[after].size = 0;
1721 } else if (before >= 0) {
1722 map = &hdr->freemap[before];
1723 be16_add_cpu(&map->size, entsize);
1724 } else {
1725 map = &hdr->freemap[after];
1726 /* both on-disk, don't endian flip twice */
1727 map->base = entry->nameidx;
1728 be16_add_cpu(&map->size, entsize);
1730 } else {
1732 * Replace smallest region (if it is smaller than free'd entry)
1734 map = &hdr->freemap[smallest];
1735 if (be16_to_cpu(map->size) < entsize) {
1736 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1737 map->size = cpu_to_be16(entsize);
1742 * Did we remove the first entry?
1744 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1745 smallest = 1;
1746 else
1747 smallest = 0;
1750 * Compress the remaining entries and zero out the removed stuff.
1752 memset(XFS_ATTR_LEAF_NAME(leaf, args->index), 0, entsize);
1753 be16_add_cpu(&hdr->usedbytes, -entsize);
1754 xfs_da_log_buf(args->trans, bp,
1755 XFS_DA_LOGRANGE(leaf, XFS_ATTR_LEAF_NAME(leaf, args->index),
1756 entsize));
1758 tmp = (be16_to_cpu(hdr->count) - args->index)
1759 * sizeof(xfs_attr_leaf_entry_t);
1760 memmove((char *)entry, (char *)(entry+1), tmp);
1761 be16_add_cpu(&hdr->count, -1);
1762 xfs_da_log_buf(args->trans, bp,
1763 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1764 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1765 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1768 * If we removed the first entry, re-find the first used byte
1769 * in the name area. Note that if the entry was the "firstused",
1770 * then we don't have a "hole" in our block resulting from
1771 * removing the name.
1773 if (smallest) {
1774 tmp = XFS_LBSIZE(mp);
1775 entry = &leaf->entries[0];
1776 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1777 ASSERT(be16_to_cpu(entry->nameidx) >=
1778 be16_to_cpu(hdr->firstused));
1779 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1781 if (be16_to_cpu(entry->nameidx) < tmp)
1782 tmp = be16_to_cpu(entry->nameidx);
1784 hdr->firstused = cpu_to_be16(tmp);
1785 if (!hdr->firstused) {
1786 hdr->firstused = cpu_to_be16(
1787 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1789 } else {
1790 hdr->holes = 1; /* mark as needing compaction */
1792 xfs_da_log_buf(args->trans, bp,
1793 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1796 * Check if leaf is less than 50% full, caller may want to
1797 * "join" the leaf with a sibling if so.
1799 tmp = sizeof(xfs_attr_leaf_hdr_t);
1800 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1801 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1802 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1806 * Move all the attribute list entries from drop_leaf into save_leaf.
1808 void
1809 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1810 xfs_da_state_blk_t *save_blk)
1812 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1813 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1814 xfs_mount_t *mp;
1815 char *tmpbuffer;
1818 * Set up environment.
1820 mp = state->mp;
1821 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1822 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1823 drop_leaf = drop_blk->bp->data;
1824 save_leaf = save_blk->bp->data;
1825 ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1826 ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1827 drop_hdr = &drop_leaf->hdr;
1828 save_hdr = &save_leaf->hdr;
1831 * Save last hashval from dying block for later Btree fixup.
1833 drop_blk->hashval = be32_to_cpu(
1834 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1837 * Check if we need a temp buffer, or can we do it in place.
1838 * Note that we don't check "leaf" for holes because we will
1839 * always be dropping it, toosmall() decided that for us already.
1841 if (save_hdr->holes == 0) {
1843 * dest leaf has no holes, so we add there. May need
1844 * to make some room in the entry array.
1846 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1847 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1848 be16_to_cpu(drop_hdr->count), mp);
1849 } else {
1850 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1851 be16_to_cpu(save_hdr->count),
1852 be16_to_cpu(drop_hdr->count), mp);
1854 } else {
1856 * Destination has holes, so we make a temporary copy
1857 * of the leaf and add them both to that.
1859 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1860 ASSERT(tmpbuffer != NULL);
1861 memset(tmpbuffer, 0, state->blocksize);
1862 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1863 tmp_hdr = &tmp_leaf->hdr;
1864 tmp_hdr->info = save_hdr->info; /* struct copy */
1865 tmp_hdr->count = 0;
1866 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1867 if (!tmp_hdr->firstused) {
1868 tmp_hdr->firstused = cpu_to_be16(
1869 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1871 tmp_hdr->usedbytes = 0;
1872 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1873 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1874 be16_to_cpu(drop_hdr->count), mp);
1875 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1876 be16_to_cpu(tmp_leaf->hdr.count),
1877 be16_to_cpu(save_hdr->count), mp);
1878 } else {
1879 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1880 be16_to_cpu(save_hdr->count), mp);
1881 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1882 be16_to_cpu(tmp_leaf->hdr.count),
1883 be16_to_cpu(drop_hdr->count), mp);
1885 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1886 kmem_free(tmpbuffer);
1889 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1890 state->blocksize - 1);
1893 * Copy out last hashval in each block for B-tree code.
1895 save_blk->hashval = be32_to_cpu(
1896 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1899 /*========================================================================
1900 * Routines used for finding things in the Btree.
1901 *========================================================================*/
1904 * Look up a name in a leaf attribute list structure.
1905 * This is the internal routine, it uses the caller's buffer.
1907 * Note that duplicate keys are allowed, but only check within the
1908 * current leaf node. The Btree code must check in adjacent leaf nodes.
1910 * Return in args->index the index into the entry[] array of either
1911 * the found entry, or where the entry should have been (insert before
1912 * that entry).
1914 * Don't change the args->value unless we find the attribute.
1917 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1919 xfs_attr_leafblock_t *leaf;
1920 xfs_attr_leaf_entry_t *entry;
1921 xfs_attr_leaf_name_local_t *name_loc;
1922 xfs_attr_leaf_name_remote_t *name_rmt;
1923 int probe, span;
1924 xfs_dahash_t hashval;
1926 leaf = bp->data;
1927 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1928 ASSERT(be16_to_cpu(leaf->hdr.count)
1929 < (XFS_LBSIZE(args->dp->i_mount)/8));
1932 * Binary search. (note: small blocks will skip this loop)
1934 hashval = args->hashval;
1935 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1936 for (entry = &leaf->entries[probe]; span > 4;
1937 entry = &leaf->entries[probe]) {
1938 span /= 2;
1939 if (be32_to_cpu(entry->hashval) < hashval)
1940 probe += span;
1941 else if (be32_to_cpu(entry->hashval) > hashval)
1942 probe -= span;
1943 else
1944 break;
1946 ASSERT((probe >= 0) &&
1947 (!leaf->hdr.count
1948 || (probe < be16_to_cpu(leaf->hdr.count))));
1949 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1952 * Since we may have duplicate hashval's, find the first matching
1953 * hashval in the leaf.
1955 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1956 entry--;
1957 probe--;
1959 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1960 (be32_to_cpu(entry->hashval) < hashval)) {
1961 entry++;
1962 probe++;
1964 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1965 (be32_to_cpu(entry->hashval) != hashval)) {
1966 args->index = probe;
1967 return(XFS_ERROR(ENOATTR));
1971 * Duplicate keys may be present, so search all of them for a match.
1973 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1974 (be32_to_cpu(entry->hashval) == hashval);
1975 entry++, probe++) {
1977 * GROT: Add code to remove incomplete entries.
1980 * If we are looking for INCOMPLETE entries, show only those.
1981 * If we are looking for complete entries, show only those.
1983 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
1984 (entry->flags & XFS_ATTR_INCOMPLETE)) {
1985 continue;
1987 if (entry->flags & XFS_ATTR_LOCAL) {
1988 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, probe);
1989 if (name_loc->namelen != args->namelen)
1990 continue;
1991 if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
1992 continue;
1993 if (!xfs_attr_namesp_match(args->flags, entry->flags))
1994 continue;
1995 args->index = probe;
1996 return(XFS_ERROR(EEXIST));
1997 } else {
1998 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, probe);
1999 if (name_rmt->namelen != args->namelen)
2000 continue;
2001 if (memcmp(args->name, (char *)name_rmt->name,
2002 args->namelen) != 0)
2003 continue;
2004 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2005 continue;
2006 args->index = probe;
2007 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2008 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2009 be32_to_cpu(name_rmt->valuelen));
2010 return(XFS_ERROR(EEXIST));
2013 args->index = probe;
2014 return(XFS_ERROR(ENOATTR));
2018 * Get the value associated with an attribute name from a leaf attribute
2019 * list structure.
2022 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2024 int valuelen;
2025 xfs_attr_leafblock_t *leaf;
2026 xfs_attr_leaf_entry_t *entry;
2027 xfs_attr_leaf_name_local_t *name_loc;
2028 xfs_attr_leaf_name_remote_t *name_rmt;
2030 leaf = bp->data;
2031 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2032 ASSERT(be16_to_cpu(leaf->hdr.count)
2033 < (XFS_LBSIZE(args->dp->i_mount)/8));
2034 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2036 entry = &leaf->entries[args->index];
2037 if (entry->flags & XFS_ATTR_LOCAL) {
2038 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2039 ASSERT(name_loc->namelen == args->namelen);
2040 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2041 valuelen = be16_to_cpu(name_loc->valuelen);
2042 if (args->flags & ATTR_KERNOVAL) {
2043 args->valuelen = valuelen;
2044 return(0);
2046 if (args->valuelen < valuelen) {
2047 args->valuelen = valuelen;
2048 return(XFS_ERROR(ERANGE));
2050 args->valuelen = valuelen;
2051 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2052 } else {
2053 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2054 ASSERT(name_rmt->namelen == args->namelen);
2055 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2056 valuelen = be32_to_cpu(name_rmt->valuelen);
2057 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2058 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2059 if (args->flags & ATTR_KERNOVAL) {
2060 args->valuelen = valuelen;
2061 return(0);
2063 if (args->valuelen < valuelen) {
2064 args->valuelen = valuelen;
2065 return(XFS_ERROR(ERANGE));
2067 args->valuelen = valuelen;
2069 return(0);
2072 /*========================================================================
2073 * Utility routines.
2074 *========================================================================*/
2077 * Move the indicated entries from one leaf to another.
2078 * NOTE: this routine modifies both source and destination leaves.
2080 /*ARGSUSED*/
2081 STATIC void
2082 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2083 xfs_attr_leafblock_t *leaf_d, int start_d,
2084 int count, xfs_mount_t *mp)
2086 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2087 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2088 int desti, tmp, i;
2091 * Check for nothing to do.
2093 if (count == 0)
2094 return;
2097 * Set up environment.
2099 ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2100 ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2101 hdr_s = &leaf_s->hdr;
2102 hdr_d = &leaf_d->hdr;
2103 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2104 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2105 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2106 ((be16_to_cpu(hdr_s->count)
2107 * sizeof(*entry_s))+sizeof(*hdr_s)));
2108 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2109 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2110 ((be16_to_cpu(hdr_d->count)
2111 * sizeof(*entry_d))+sizeof(*hdr_d)));
2113 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2114 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2115 ASSERT(count <= be16_to_cpu(hdr_s->count));
2118 * Move the entries in the destination leaf up to make a hole?
2120 if (start_d < be16_to_cpu(hdr_d->count)) {
2121 tmp = be16_to_cpu(hdr_d->count) - start_d;
2122 tmp *= sizeof(xfs_attr_leaf_entry_t);
2123 entry_s = &leaf_d->entries[start_d];
2124 entry_d = &leaf_d->entries[start_d + count];
2125 memmove((char *)entry_d, (char *)entry_s, tmp);
2129 * Copy all entry's in the same (sorted) order,
2130 * but allocate attribute info packed and in sequence.
2132 entry_s = &leaf_s->entries[start_s];
2133 entry_d = &leaf_d->entries[start_d];
2134 desti = start_d;
2135 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2136 ASSERT(be16_to_cpu(entry_s->nameidx)
2137 >= be16_to_cpu(hdr_s->firstused));
2138 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2139 #ifdef GROT
2141 * Code to drop INCOMPLETE entries. Difficult to use as we
2142 * may also need to change the insertion index. Code turned
2143 * off for 6.2, should be revisited later.
2145 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2146 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2147 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2148 be16_add_cpu(&hdr_s->count, -1);
2149 entry_d--; /* to compensate for ++ in loop hdr */
2150 desti--;
2151 if ((start_s + i) < offset)
2152 result++; /* insertion index adjustment */
2153 } else {
2154 #endif /* GROT */
2155 be16_add_cpu(&hdr_d->firstused, -tmp);
2156 /* both on-disk, don't endian flip twice */
2157 entry_d->hashval = entry_s->hashval;
2158 /* both on-disk, don't endian flip twice */
2159 entry_d->nameidx = hdr_d->firstused;
2160 entry_d->flags = entry_s->flags;
2161 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2162 <= XFS_LBSIZE(mp));
2163 memmove(XFS_ATTR_LEAF_NAME(leaf_d, desti),
2164 XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), tmp);
2165 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2166 <= XFS_LBSIZE(mp));
2167 memset(XFS_ATTR_LEAF_NAME(leaf_s, start_s + i), 0, tmp);
2168 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2169 be16_add_cpu(&hdr_d->usedbytes, tmp);
2170 be16_add_cpu(&hdr_s->count, -1);
2171 be16_add_cpu(&hdr_d->count, 1);
2172 tmp = be16_to_cpu(hdr_d->count)
2173 * sizeof(xfs_attr_leaf_entry_t)
2174 + sizeof(xfs_attr_leaf_hdr_t);
2175 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2176 #ifdef GROT
2178 #endif /* GROT */
2182 * Zero out the entries we just copied.
2184 if (start_s == be16_to_cpu(hdr_s->count)) {
2185 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2186 entry_s = &leaf_s->entries[start_s];
2187 ASSERT(((char *)entry_s + tmp) <=
2188 ((char *)leaf_s + XFS_LBSIZE(mp)));
2189 memset((char *)entry_s, 0, tmp);
2190 } else {
2192 * Move the remaining entries down to fill the hole,
2193 * then zero the entries at the top.
2195 tmp = be16_to_cpu(hdr_s->count) - count;
2196 tmp *= sizeof(xfs_attr_leaf_entry_t);
2197 entry_s = &leaf_s->entries[start_s + count];
2198 entry_d = &leaf_s->entries[start_s];
2199 memmove((char *)entry_d, (char *)entry_s, tmp);
2201 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2202 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2203 ASSERT(((char *)entry_s + tmp) <=
2204 ((char *)leaf_s + XFS_LBSIZE(mp)));
2205 memset((char *)entry_s, 0, tmp);
2209 * Fill in the freemap information
2211 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2212 be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2213 sizeof(xfs_attr_leaf_entry_t));
2214 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2215 - be16_to_cpu(hdr_d->freemap[0].base));
2216 hdr_d->freemap[1].base = 0;
2217 hdr_d->freemap[2].base = 0;
2218 hdr_d->freemap[1].size = 0;
2219 hdr_d->freemap[2].size = 0;
2220 hdr_s->holes = 1; /* leaf may not be compact */
2224 * Compare two leaf blocks "order".
2225 * Return 0 unless leaf2 should go before leaf1.
2228 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2230 xfs_attr_leafblock_t *leaf1, *leaf2;
2232 leaf1 = leaf1_bp->data;
2233 leaf2 = leaf2_bp->data;
2234 ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2235 (be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC));
2236 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2237 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2238 ((be32_to_cpu(leaf2->entries[0].hashval) <
2239 be32_to_cpu(leaf1->entries[0].hashval)) ||
2240 (be32_to_cpu(leaf2->entries[
2241 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2242 be32_to_cpu(leaf1->entries[
2243 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2244 return(1);
2246 return(0);
2250 * Pick up the last hashvalue from a leaf block.
2252 xfs_dahash_t
2253 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2255 xfs_attr_leafblock_t *leaf;
2257 leaf = bp->data;
2258 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2259 if (count)
2260 *count = be16_to_cpu(leaf->hdr.count);
2261 if (!leaf->hdr.count)
2262 return(0);
2263 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2267 * Calculate the number of bytes used to store the indicated attribute
2268 * (whether local or remote only calculate bytes in this block).
2270 STATIC int
2271 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2273 xfs_attr_leaf_name_local_t *name_loc;
2274 xfs_attr_leaf_name_remote_t *name_rmt;
2275 int size;
2277 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2278 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2279 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, index);
2280 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(name_loc->namelen,
2281 be16_to_cpu(name_loc->valuelen));
2282 } else {
2283 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, index);
2284 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(name_rmt->namelen);
2286 return(size);
2290 * Calculate the number of bytes that would be required to store the new
2291 * attribute (whether local or remote only calculate bytes in this block).
2292 * This routine decides as a side effect whether the attribute will be
2293 * a "local" or a "remote" attribute.
2296 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2298 int size;
2300 size = XFS_ATTR_LEAF_ENTSIZE_LOCAL(namelen, valuelen);
2301 if (size < XFS_ATTR_LEAF_ENTSIZE_LOCAL_MAX(blocksize)) {
2302 if (local) {
2303 *local = 1;
2305 } else {
2306 size = XFS_ATTR_LEAF_ENTSIZE_REMOTE(namelen);
2307 if (local) {
2308 *local = 0;
2311 return(size);
2315 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2318 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2320 attrlist_cursor_kern_t *cursor;
2321 xfs_attr_leafblock_t *leaf;
2322 xfs_attr_leaf_entry_t *entry;
2323 int retval, i;
2325 ASSERT(bp != NULL);
2326 leaf = bp->data;
2327 cursor = context->cursor;
2328 cursor->initted = 1;
2330 xfs_attr_trace_l_cl("blk start", context, leaf);
2333 * Re-find our place in the leaf block if this is a new syscall.
2335 if (context->resynch) {
2336 entry = &leaf->entries[0];
2337 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2338 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2339 if (cursor->offset == context->dupcnt) {
2340 context->dupcnt = 0;
2341 break;
2343 context->dupcnt++;
2344 } else if (be32_to_cpu(entry->hashval) >
2345 cursor->hashval) {
2346 context->dupcnt = 0;
2347 break;
2350 if (i == be16_to_cpu(leaf->hdr.count)) {
2351 xfs_attr_trace_l_c("not found", context);
2352 return(0);
2354 } else {
2355 entry = &leaf->entries[0];
2356 i = 0;
2358 context->resynch = 0;
2361 * We have found our place, start copying out the new attributes.
2363 retval = 0;
2364 for ( ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2365 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2366 cursor->hashval = be32_to_cpu(entry->hashval);
2367 cursor->offset = 0;
2370 if (entry->flags & XFS_ATTR_INCOMPLETE)
2371 continue; /* skip incomplete entries */
2373 if (entry->flags & XFS_ATTR_LOCAL) {
2374 xfs_attr_leaf_name_local_t *name_loc =
2375 XFS_ATTR_LEAF_NAME_LOCAL(leaf, i);
2377 retval = context->put_listent(context,
2378 entry->flags,
2379 (char *)name_loc->nameval,
2380 (int)name_loc->namelen,
2381 be16_to_cpu(name_loc->valuelen),
2382 (char *)&name_loc->nameval[name_loc->namelen]);
2383 if (retval)
2384 return retval;
2385 } else {
2386 xfs_attr_leaf_name_remote_t *name_rmt =
2387 XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2389 int valuelen = be32_to_cpu(name_rmt->valuelen);
2391 if (context->put_value) {
2392 xfs_da_args_t args;
2394 memset((char *)&args, 0, sizeof(args));
2395 args.dp = context->dp;
2396 args.whichfork = XFS_ATTR_FORK;
2397 args.valuelen = valuelen;
2398 args.value = kmem_alloc(valuelen, KM_SLEEP);
2399 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2400 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2401 retval = xfs_attr_rmtval_get(&args);
2402 if (retval)
2403 return retval;
2404 retval = context->put_listent(context,
2405 entry->flags,
2406 (char *)name_rmt->name,
2407 (int)name_rmt->namelen,
2408 valuelen,
2409 (char*)args.value);
2410 kmem_free(args.value);
2411 } else {
2412 retval = context->put_listent(context,
2413 entry->flags,
2414 (char *)name_rmt->name,
2415 (int)name_rmt->namelen,
2416 valuelen,
2417 NULL);
2419 if (retval)
2420 return retval;
2422 if (context->seen_enough)
2423 break;
2424 cursor->offset++;
2426 xfs_attr_trace_l_cl("blk end", context, leaf);
2427 return(retval);
2431 /*========================================================================
2432 * Manage the INCOMPLETE flag in a leaf entry
2433 *========================================================================*/
2436 * Clear the INCOMPLETE flag on an entry in a leaf block.
2439 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2441 xfs_attr_leafblock_t *leaf;
2442 xfs_attr_leaf_entry_t *entry;
2443 xfs_attr_leaf_name_remote_t *name_rmt;
2444 xfs_dabuf_t *bp;
2445 int error;
2446 #ifdef DEBUG
2447 xfs_attr_leaf_name_local_t *name_loc;
2448 int namelen;
2449 char *name;
2450 #endif /* DEBUG */
2453 * Set up the operation.
2455 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2456 XFS_ATTR_FORK);
2457 if (error) {
2458 return(error);
2460 ASSERT(bp != NULL);
2462 leaf = bp->data;
2463 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2464 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2465 ASSERT(args->index >= 0);
2466 entry = &leaf->entries[ args->index ];
2467 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2469 #ifdef DEBUG
2470 if (entry->flags & XFS_ATTR_LOCAL) {
2471 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf, args->index);
2472 namelen = name_loc->namelen;
2473 name = (char *)name_loc->nameval;
2474 } else {
2475 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2476 namelen = name_rmt->namelen;
2477 name = (char *)name_rmt->name;
2479 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2480 ASSERT(namelen == args->namelen);
2481 ASSERT(memcmp(name, args->name, namelen) == 0);
2482 #endif /* DEBUG */
2484 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2485 xfs_da_log_buf(args->trans, bp,
2486 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2488 if (args->rmtblkno) {
2489 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2490 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2491 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2492 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2493 xfs_da_log_buf(args->trans, bp,
2494 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2496 xfs_da_buf_done(bp);
2499 * Commit the flag value change and start the next trans in series.
2501 return xfs_trans_roll(&args->trans, args->dp);
2505 * Set the INCOMPLETE flag on an entry in a leaf block.
2508 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2510 xfs_attr_leafblock_t *leaf;
2511 xfs_attr_leaf_entry_t *entry;
2512 xfs_attr_leaf_name_remote_t *name_rmt;
2513 xfs_dabuf_t *bp;
2514 int error;
2517 * Set up the operation.
2519 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2520 XFS_ATTR_FORK);
2521 if (error) {
2522 return(error);
2524 ASSERT(bp != NULL);
2526 leaf = bp->data;
2527 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2528 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2529 ASSERT(args->index >= 0);
2530 entry = &leaf->entries[ args->index ];
2532 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2533 entry->flags |= XFS_ATTR_INCOMPLETE;
2534 xfs_da_log_buf(args->trans, bp,
2535 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2536 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2537 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, args->index);
2538 name_rmt->valueblk = 0;
2539 name_rmt->valuelen = 0;
2540 xfs_da_log_buf(args->trans, bp,
2541 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2543 xfs_da_buf_done(bp);
2546 * Commit the flag value change and start the next trans in series.
2548 return xfs_trans_roll(&args->trans, args->dp);
2552 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2553 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2554 * entry given by args->blkno2/index2.
2556 * Note that they could be in different blocks, or in the same block.
2559 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2561 xfs_attr_leafblock_t *leaf1, *leaf2;
2562 xfs_attr_leaf_entry_t *entry1, *entry2;
2563 xfs_attr_leaf_name_remote_t *name_rmt;
2564 xfs_dabuf_t *bp1, *bp2;
2565 int error;
2566 #ifdef DEBUG
2567 xfs_attr_leaf_name_local_t *name_loc;
2568 int namelen1, namelen2;
2569 char *name1, *name2;
2570 #endif /* DEBUG */
2573 * Read the block containing the "old" attr
2575 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2576 XFS_ATTR_FORK);
2577 if (error) {
2578 return(error);
2580 ASSERT(bp1 != NULL);
2583 * Read the block containing the "new" attr, if it is different
2585 if (args->blkno2 != args->blkno) {
2586 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2587 -1, &bp2, XFS_ATTR_FORK);
2588 if (error) {
2589 return(error);
2591 ASSERT(bp2 != NULL);
2592 } else {
2593 bp2 = bp1;
2596 leaf1 = bp1->data;
2597 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2598 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2599 ASSERT(args->index >= 0);
2600 entry1 = &leaf1->entries[ args->index ];
2602 leaf2 = bp2->data;
2603 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2604 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2605 ASSERT(args->index2 >= 0);
2606 entry2 = &leaf2->entries[ args->index2 ];
2608 #ifdef DEBUG
2609 if (entry1->flags & XFS_ATTR_LOCAL) {
2610 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf1, args->index);
2611 namelen1 = name_loc->namelen;
2612 name1 = (char *)name_loc->nameval;
2613 } else {
2614 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2615 namelen1 = name_rmt->namelen;
2616 name1 = (char *)name_rmt->name;
2618 if (entry2->flags & XFS_ATTR_LOCAL) {
2619 name_loc = XFS_ATTR_LEAF_NAME_LOCAL(leaf2, args->index2);
2620 namelen2 = name_loc->namelen;
2621 name2 = (char *)name_loc->nameval;
2622 } else {
2623 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2624 namelen2 = name_rmt->namelen;
2625 name2 = (char *)name_rmt->name;
2627 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2628 ASSERT(namelen1 == namelen2);
2629 ASSERT(memcmp(name1, name2, namelen1) == 0);
2630 #endif /* DEBUG */
2632 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2633 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2635 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2636 xfs_da_log_buf(args->trans, bp1,
2637 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2638 if (args->rmtblkno) {
2639 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2640 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf1, args->index);
2641 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2642 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2643 xfs_da_log_buf(args->trans, bp1,
2644 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2647 entry2->flags |= XFS_ATTR_INCOMPLETE;
2648 xfs_da_log_buf(args->trans, bp2,
2649 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2650 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2651 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf2, args->index2);
2652 name_rmt->valueblk = 0;
2653 name_rmt->valuelen = 0;
2654 xfs_da_log_buf(args->trans, bp2,
2655 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2657 xfs_da_buf_done(bp1);
2658 if (bp1 != bp2)
2659 xfs_da_buf_done(bp2);
2662 * Commit the flag value change and start the next trans in series.
2664 error = xfs_trans_roll(&args->trans, args->dp);
2666 return(error);
2669 /*========================================================================
2670 * Indiscriminately delete the entire attribute fork
2671 *========================================================================*/
2674 * Recurse (gasp!) through the attribute nodes until we find leaves.
2675 * We're doing a depth-first traversal in order to invalidate everything.
2678 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2680 xfs_da_blkinfo_t *info;
2681 xfs_daddr_t blkno;
2682 xfs_dabuf_t *bp;
2683 int error;
2686 * Read block 0 to see what we have to work with.
2687 * We only get here if we have extents, since we remove
2688 * the extents in reverse order the extent containing
2689 * block 0 must still be there.
2691 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2692 if (error)
2693 return(error);
2694 blkno = xfs_da_blkno(bp);
2697 * Invalidate the tree, even if the "tree" is only a single leaf block.
2698 * This is a depth-first traversal!
2700 info = bp->data;
2701 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2702 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2703 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2704 error = xfs_attr_leaf_inactive(trans, dp, bp);
2705 } else {
2706 error = XFS_ERROR(EIO);
2707 xfs_da_brelse(*trans, bp);
2709 if (error)
2710 return(error);
2713 * Invalidate the incore copy of the root block.
2715 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2716 if (error)
2717 return(error);
2718 xfs_da_binval(*trans, bp); /* remove from cache */
2720 * Commit the invalidate and start the next transaction.
2722 error = xfs_trans_roll(trans, dp);
2724 return (error);
2728 * Recurse (gasp!) through the attribute nodes until we find leaves.
2729 * We're doing a depth-first traversal in order to invalidate everything.
2731 STATIC int
2732 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2733 int level)
2735 xfs_da_blkinfo_t *info;
2736 xfs_da_intnode_t *node;
2737 xfs_dablk_t child_fsb;
2738 xfs_daddr_t parent_blkno, child_blkno;
2739 int error, count, i;
2740 xfs_dabuf_t *child_bp;
2743 * Since this code is recursive (gasp!) we must protect ourselves.
2745 if (level > XFS_DA_NODE_MAXDEPTH) {
2746 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2747 return(XFS_ERROR(EIO));
2750 node = bp->data;
2751 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2752 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2753 count = be16_to_cpu(node->hdr.count);
2754 if (!count) {
2755 xfs_da_brelse(*trans, bp);
2756 return(0);
2758 child_fsb = be32_to_cpu(node->btree[0].before);
2759 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2762 * If this is the node level just above the leaves, simply loop
2763 * over the leaves removing all of them. If this is higher up
2764 * in the tree, recurse downward.
2766 for (i = 0; i < count; i++) {
2768 * Read the subsidiary block to see what we have to work with.
2769 * Don't do this in a transaction. This is a depth-first
2770 * traversal of the tree so we may deal with many blocks
2771 * before we come back to this one.
2773 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2774 XFS_ATTR_FORK);
2775 if (error)
2776 return(error);
2777 if (child_bp) {
2778 /* save for re-read later */
2779 child_blkno = xfs_da_blkno(child_bp);
2782 * Invalidate the subtree, however we have to.
2784 info = child_bp->data;
2785 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2786 error = xfs_attr_node_inactive(trans, dp,
2787 child_bp, level+1);
2788 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2789 error = xfs_attr_leaf_inactive(trans, dp,
2790 child_bp);
2791 } else {
2792 error = XFS_ERROR(EIO);
2793 xfs_da_brelse(*trans, child_bp);
2795 if (error)
2796 return(error);
2799 * Remove the subsidiary block from the cache
2800 * and from the log.
2802 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2803 &child_bp, XFS_ATTR_FORK);
2804 if (error)
2805 return(error);
2806 xfs_da_binval(*trans, child_bp);
2810 * If we're not done, re-read the parent to get the next
2811 * child block number.
2813 if ((i+1) < count) {
2814 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2815 &bp, XFS_ATTR_FORK);
2816 if (error)
2817 return(error);
2818 child_fsb = be32_to_cpu(node->btree[i+1].before);
2819 xfs_da_brelse(*trans, bp);
2822 * Atomically commit the whole invalidate stuff.
2824 error = xfs_trans_roll(trans, dp);
2825 if (error)
2826 return (error);
2829 return(0);
2833 * Invalidate all of the "remote" value regions pointed to by a particular
2834 * leaf block.
2835 * Note that we must release the lock on the buffer so that we are not
2836 * caught holding something that the logging code wants to flush to disk.
2838 STATIC int
2839 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2841 xfs_attr_leafblock_t *leaf;
2842 xfs_attr_leaf_entry_t *entry;
2843 xfs_attr_leaf_name_remote_t *name_rmt;
2844 xfs_attr_inactive_list_t *list, *lp;
2845 int error, count, size, tmp, i;
2847 leaf = bp->data;
2848 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2851 * Count the number of "remote" value extents.
2853 count = 0;
2854 entry = &leaf->entries[0];
2855 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2856 if (be16_to_cpu(entry->nameidx) &&
2857 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2858 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2859 if (name_rmt->valueblk)
2860 count++;
2865 * If there are no "remote" values, we're done.
2867 if (count == 0) {
2868 xfs_da_brelse(*trans, bp);
2869 return(0);
2873 * Allocate storage for a list of all the "remote" value extents.
2875 size = count * sizeof(xfs_attr_inactive_list_t);
2876 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2879 * Identify each of the "remote" value extents.
2881 lp = list;
2882 entry = &leaf->entries[0];
2883 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2884 if (be16_to_cpu(entry->nameidx) &&
2885 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2886 name_rmt = XFS_ATTR_LEAF_NAME_REMOTE(leaf, i);
2887 if (name_rmt->valueblk) {
2888 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2889 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2890 be32_to_cpu(name_rmt->valuelen));
2891 lp++;
2895 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2898 * Invalidate each of the "remote" value extents.
2900 error = 0;
2901 for (lp = list, i = 0; i < count; i++, lp++) {
2902 tmp = xfs_attr_leaf_freextent(trans, dp,
2903 lp->valueblk, lp->valuelen);
2905 if (error == 0)
2906 error = tmp; /* save only the 1st errno */
2909 kmem_free((xfs_caddr_t)list);
2910 return(error);
2914 * Look at all the extents for this logical region,
2915 * invalidate any buffers that are incore/in transactions.
2917 STATIC int
2918 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2919 xfs_dablk_t blkno, int blkcnt)
2921 xfs_bmbt_irec_t map;
2922 xfs_dablk_t tblkno;
2923 int tblkcnt, dblkcnt, nmap, error;
2924 xfs_daddr_t dblkno;
2925 xfs_buf_t *bp;
2928 * Roll through the "value", invalidating the attribute value's
2929 * blocks.
2931 tblkno = blkno;
2932 tblkcnt = blkcnt;
2933 while (tblkcnt > 0) {
2935 * Try to remember where we decided to put the value.
2937 nmap = 1;
2938 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
2939 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2940 NULL, 0, &map, &nmap, NULL, NULL);
2941 if (error) {
2942 return(error);
2944 ASSERT(nmap == 1);
2945 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2948 * If it's a hole, these are already unmapped
2949 * so there's nothing to invalidate.
2951 if (map.br_startblock != HOLESTARTBLOCK) {
2953 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
2954 map.br_startblock);
2955 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
2956 map.br_blockcount);
2957 bp = xfs_trans_get_buf(*trans,
2958 dp->i_mount->m_ddev_targp,
2959 dblkno, dblkcnt, XFS_BUF_LOCK);
2960 xfs_trans_binval(*trans, bp);
2962 * Roll to next transaction.
2964 error = xfs_trans_roll(trans, dp);
2965 if (error)
2966 return (error);
2969 tblkno += map.br_blockcount;
2970 tblkcnt -= map.br_blockcount;
2973 return(0);