ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / fs / xfs / xfs_attr_leaf.c
blobf49ecf2e7d36943fce53811268ab80d11e33c27b
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_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_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_inode_item.h"
38 #include "xfs_bmap.h"
39 #include "xfs_attr.h"
40 #include "xfs_attr_leaf.h"
41 #include "xfs_error.h"
42 #include "xfs_trace.h"
45 * xfs_attr_leaf.c
47 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
50 /*========================================================================
51 * Function prototypes for the kernel.
52 *========================================================================*/
55 * Routines used for growing the Btree.
57 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
58 xfs_dabuf_t **bpp);
59 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
60 int freemap_index);
61 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
62 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
63 xfs_da_state_blk_t *blk1,
64 xfs_da_state_blk_t *blk2);
65 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
66 xfs_da_state_blk_t *leaf_blk_1,
67 xfs_da_state_blk_t *leaf_blk_2,
68 int *number_entries_in_blk1,
69 int *number_usedbytes_in_blk1);
72 * Routines used for shrinking the Btree.
74 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
75 xfs_dabuf_t *bp, int level);
76 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
77 xfs_dabuf_t *bp);
78 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
79 xfs_dablk_t blkno, int blkcnt);
82 * Utility routines.
84 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
85 int src_start,
86 xfs_attr_leafblock_t *dst_leaf,
87 int dst_start, int move_count,
88 xfs_mount_t *mp);
89 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
91 /*========================================================================
92 * Namespace helper routines
93 *========================================================================*/
96 * If namespace bits don't match return 0.
97 * If all match then return 1.
99 STATIC int
100 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
102 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
106 /*========================================================================
107 * External routines when attribute fork size < XFS_LITINO(mp).
108 *========================================================================*/
111 * Query whether the requested number of additional bytes of extended
112 * attribute space will be able to fit inline.
114 * Returns zero if not, else the di_forkoff fork offset to be used in the
115 * literal area for attribute data once the new bytes have been added.
117 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
118 * special case for dev/uuid inodes, they have fixed size data forks.
121 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
123 int offset;
124 int minforkoff; /* lower limit on valid forkoff locations */
125 int maxforkoff; /* upper limit on valid forkoff locations */
126 int dsize;
127 xfs_mount_t *mp = dp->i_mount;
129 offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
131 switch (dp->i_d.di_format) {
132 case XFS_DINODE_FMT_DEV:
133 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
134 return (offset >= minforkoff) ? minforkoff : 0;
135 case XFS_DINODE_FMT_UUID:
136 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
137 return (offset >= minforkoff) ? minforkoff : 0;
141 * If the requested numbers of bytes is smaller or equal to the
142 * current attribute fork size we can always proceed.
144 * Note that if_bytes in the data fork might actually be larger than
145 * the current data fork size is due to delalloc extents. In that
146 * case either the extent count will go down when they are converted
147 * to real extents, or the delalloc conversion will take care of the
148 * literal area rebalancing.
150 if (bytes <= XFS_IFORK_ASIZE(dp))
151 return dp->i_d.di_forkoff;
154 * For attr2 we can try to move the forkoff if there is space in the
155 * literal area, but for the old format we are done if there is no
156 * space in the fixed attribute fork.
158 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
159 return 0;
161 dsize = dp->i_df.if_bytes;
163 switch (dp->i_d.di_format) {
164 case XFS_DINODE_FMT_EXTENTS:
166 * If there is no attr fork and the data fork is extents,
167 * determine if creating the default attr fork will result
168 * in the extents form migrating to btree. If so, the
169 * minimum offset only needs to be the space required for
170 * the btree root.
172 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
173 xfs_default_attroffset(dp))
174 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
175 break;
176 case XFS_DINODE_FMT_BTREE:
178 * If we have a data btree then keep forkoff if we have one,
179 * otherwise we are adding a new attr, so then we set
180 * minforkoff to where the btree root can finish so we have
181 * plenty of room for attrs
183 if (dp->i_d.di_forkoff) {
184 if (offset < dp->i_d.di_forkoff)
185 return 0;
186 return dp->i_d.di_forkoff;
188 dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
189 break;
193 * A data fork btree root must have space for at least
194 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
196 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
197 minforkoff = roundup(minforkoff, 8) >> 3;
199 /* attr fork btree root can have at least this many key/ptr pairs */
200 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
201 maxforkoff = maxforkoff >> 3; /* rounded down */
203 if (offset >= maxforkoff)
204 return maxforkoff;
205 if (offset >= minforkoff)
206 return offset;
207 return 0;
211 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
213 STATIC void
214 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
216 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
217 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
218 spin_lock(&mp->m_sb_lock);
219 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
220 xfs_sb_version_addattr2(&mp->m_sb);
221 spin_unlock(&mp->m_sb_lock);
222 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
223 } else
224 spin_unlock(&mp->m_sb_lock);
229 * Create the initial contents of a shortform attribute list.
231 void
232 xfs_attr_shortform_create(xfs_da_args_t *args)
234 xfs_attr_sf_hdr_t *hdr;
235 xfs_inode_t *dp;
236 xfs_ifork_t *ifp;
238 dp = args->dp;
239 ASSERT(dp != NULL);
240 ifp = dp->i_afp;
241 ASSERT(ifp != NULL);
242 ASSERT(ifp->if_bytes == 0);
243 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
244 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
245 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
246 ifp->if_flags |= XFS_IFINLINE;
247 } else {
248 ASSERT(ifp->if_flags & XFS_IFINLINE);
250 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
251 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
252 hdr->count = 0;
253 hdr->totsize = cpu_to_be16(sizeof(*hdr));
254 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
258 * Add a name/value pair to the shortform attribute list.
259 * Overflow from the inode has already been checked for.
261 void
262 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
264 xfs_attr_shortform_t *sf;
265 xfs_attr_sf_entry_t *sfe;
266 int i, offset, size;
267 xfs_mount_t *mp;
268 xfs_inode_t *dp;
269 xfs_ifork_t *ifp;
271 dp = args->dp;
272 mp = dp->i_mount;
273 dp->i_d.di_forkoff = forkoff;
274 dp->i_df.if_ext_max =
275 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
276 dp->i_afp->if_ext_max =
277 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
279 ifp = dp->i_afp;
280 ASSERT(ifp->if_flags & XFS_IFINLINE);
281 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
282 sfe = &sf->list[0];
283 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
284 #ifdef DEBUG
285 if (sfe->namelen != args->namelen)
286 continue;
287 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
288 continue;
289 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
290 continue;
291 ASSERT(0);
292 #endif
295 offset = (char *)sfe - (char *)sf;
296 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
297 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
298 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
299 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
301 sfe->namelen = args->namelen;
302 sfe->valuelen = args->valuelen;
303 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
304 memcpy(sfe->nameval, args->name, args->namelen);
305 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
306 sf->hdr.count++;
307 be16_add_cpu(&sf->hdr.totsize, size);
308 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
310 xfs_sbversion_add_attr2(mp, args->trans);
314 * After the last attribute is removed revert to original inode format,
315 * making all literal area available to the data fork once more.
317 STATIC void
318 xfs_attr_fork_reset(
319 struct xfs_inode *ip,
320 struct xfs_trans *tp)
322 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
323 ip->i_d.di_forkoff = 0;
324 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
326 ASSERT(ip->i_d.di_anextents == 0);
327 ASSERT(ip->i_afp == NULL);
329 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
330 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
334 * Remove an attribute from the shortform attribute list structure.
337 xfs_attr_shortform_remove(xfs_da_args_t *args)
339 xfs_attr_shortform_t *sf;
340 xfs_attr_sf_entry_t *sfe;
341 int base, size=0, end, totsize, i;
342 xfs_mount_t *mp;
343 xfs_inode_t *dp;
345 dp = args->dp;
346 mp = dp->i_mount;
347 base = sizeof(xfs_attr_sf_hdr_t);
348 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
349 sfe = &sf->list[0];
350 end = sf->hdr.count;
351 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
352 base += size, i++) {
353 size = XFS_ATTR_SF_ENTSIZE(sfe);
354 if (sfe->namelen != args->namelen)
355 continue;
356 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
357 continue;
358 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
359 continue;
360 break;
362 if (i == end)
363 return(XFS_ERROR(ENOATTR));
366 * Fix up the attribute fork data, covering the hole
368 end = base + size;
369 totsize = be16_to_cpu(sf->hdr.totsize);
370 if (end != totsize)
371 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
372 sf->hdr.count--;
373 be16_add_cpu(&sf->hdr.totsize, -size);
376 * Fix up the start offset of the attribute fork
378 totsize -= size;
379 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
380 (mp->m_flags & XFS_MOUNT_ATTR2) &&
381 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
382 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
383 xfs_attr_fork_reset(dp, args->trans);
384 } else {
385 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
386 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
387 ASSERT(dp->i_d.di_forkoff);
388 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
389 (args->op_flags & XFS_DA_OP_ADDNAME) ||
390 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
391 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
392 dp->i_afp->if_ext_max =
393 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
394 dp->i_df.if_ext_max =
395 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
396 xfs_trans_log_inode(args->trans, dp,
397 XFS_ILOG_CORE | XFS_ILOG_ADATA);
400 xfs_sbversion_add_attr2(mp, args->trans);
402 return(0);
406 * Look up a name in a shortform attribute list structure.
408 /*ARGSUSED*/
410 xfs_attr_shortform_lookup(xfs_da_args_t *args)
412 xfs_attr_shortform_t *sf;
413 xfs_attr_sf_entry_t *sfe;
414 int i;
415 xfs_ifork_t *ifp;
417 ifp = args->dp->i_afp;
418 ASSERT(ifp->if_flags & XFS_IFINLINE);
419 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
420 sfe = &sf->list[0];
421 for (i = 0; i < sf->hdr.count;
422 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
423 if (sfe->namelen != args->namelen)
424 continue;
425 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
426 continue;
427 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
428 continue;
429 return(XFS_ERROR(EEXIST));
431 return(XFS_ERROR(ENOATTR));
435 * Look up a name in a shortform attribute list structure.
437 /*ARGSUSED*/
439 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
441 xfs_attr_shortform_t *sf;
442 xfs_attr_sf_entry_t *sfe;
443 int i;
445 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
446 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
447 sfe = &sf->list[0];
448 for (i = 0; i < sf->hdr.count;
449 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
450 if (sfe->namelen != args->namelen)
451 continue;
452 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
453 continue;
454 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
455 continue;
456 if (args->flags & ATTR_KERNOVAL) {
457 args->valuelen = sfe->valuelen;
458 return(XFS_ERROR(EEXIST));
460 if (args->valuelen < sfe->valuelen) {
461 args->valuelen = sfe->valuelen;
462 return(XFS_ERROR(ERANGE));
464 args->valuelen = sfe->valuelen;
465 memcpy(args->value, &sfe->nameval[args->namelen],
466 args->valuelen);
467 return(XFS_ERROR(EEXIST));
469 return(XFS_ERROR(ENOATTR));
473 * Convert from using the shortform to the leaf.
476 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
478 xfs_inode_t *dp;
479 xfs_attr_shortform_t *sf;
480 xfs_attr_sf_entry_t *sfe;
481 xfs_da_args_t nargs;
482 char *tmpbuffer;
483 int error, i, size;
484 xfs_dablk_t blkno;
485 xfs_dabuf_t *bp;
486 xfs_ifork_t *ifp;
488 dp = args->dp;
489 ifp = dp->i_afp;
490 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
491 size = be16_to_cpu(sf->hdr.totsize);
492 tmpbuffer = kmem_alloc(size, KM_SLEEP);
493 ASSERT(tmpbuffer != NULL);
494 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
495 sf = (xfs_attr_shortform_t *)tmpbuffer;
497 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
498 bp = NULL;
499 error = xfs_da_grow_inode(args, &blkno);
500 if (error) {
502 * If we hit an IO error middle of the transaction inside
503 * grow_inode(), we may have inconsistent data. Bail out.
505 if (error == EIO)
506 goto out;
507 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
508 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
509 goto out;
512 ASSERT(blkno == 0);
513 error = xfs_attr_leaf_create(args, blkno, &bp);
514 if (error) {
515 error = xfs_da_shrink_inode(args, 0, bp);
516 bp = NULL;
517 if (error)
518 goto out;
519 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
520 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
521 goto out;
524 memset((char *)&nargs, 0, sizeof(nargs));
525 nargs.dp = dp;
526 nargs.firstblock = args->firstblock;
527 nargs.flist = args->flist;
528 nargs.total = args->total;
529 nargs.whichfork = XFS_ATTR_FORK;
530 nargs.trans = args->trans;
531 nargs.op_flags = XFS_DA_OP_OKNOENT;
533 sfe = &sf->list[0];
534 for (i = 0; i < sf->hdr.count; i++) {
535 nargs.name = sfe->nameval;
536 nargs.namelen = sfe->namelen;
537 nargs.value = &sfe->nameval[nargs.namelen];
538 nargs.valuelen = sfe->valuelen;
539 nargs.hashval = xfs_da_hashname(sfe->nameval,
540 sfe->namelen);
541 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
542 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
543 ASSERT(error == ENOATTR);
544 error = xfs_attr_leaf_add(bp, &nargs);
545 ASSERT(error != ENOSPC);
546 if (error)
547 goto out;
548 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
550 error = 0;
552 out:
553 if(bp)
554 xfs_da_buf_done(bp);
555 kmem_free(tmpbuffer);
556 return(error);
559 STATIC int
560 xfs_attr_shortform_compare(const void *a, const void *b)
562 xfs_attr_sf_sort_t *sa, *sb;
564 sa = (xfs_attr_sf_sort_t *)a;
565 sb = (xfs_attr_sf_sort_t *)b;
566 if (sa->hash < sb->hash) {
567 return(-1);
568 } else if (sa->hash > sb->hash) {
569 return(1);
570 } else {
571 return(sa->entno - sb->entno);
576 #define XFS_ISRESET_CURSOR(cursor) \
577 (!((cursor)->initted) && !((cursor)->hashval) && \
578 !((cursor)->blkno) && !((cursor)->offset))
580 * Copy out entries of shortform attribute lists for attr_list().
581 * Shortform attribute lists are not stored in hashval sorted order.
582 * If the output buffer is not large enough to hold them all, then we
583 * we have to calculate each entries' hashvalue and sort them before
584 * we can begin returning them to the user.
586 /*ARGSUSED*/
588 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
590 attrlist_cursor_kern_t *cursor;
591 xfs_attr_sf_sort_t *sbuf, *sbp;
592 xfs_attr_shortform_t *sf;
593 xfs_attr_sf_entry_t *sfe;
594 xfs_inode_t *dp;
595 int sbsize, nsbuf, count, i;
596 int error;
598 ASSERT(context != NULL);
599 dp = context->dp;
600 ASSERT(dp != NULL);
601 ASSERT(dp->i_afp != NULL);
602 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
603 ASSERT(sf != NULL);
604 if (!sf->hdr.count)
605 return(0);
606 cursor = context->cursor;
607 ASSERT(cursor != NULL);
609 trace_xfs_attr_list_sf(context);
612 * If the buffer is large enough and the cursor is at the start,
613 * do not bother with sorting since we will return everything in
614 * one buffer and another call using the cursor won't need to be
615 * made.
616 * Note the generous fudge factor of 16 overhead bytes per entry.
617 * If bufsize is zero then put_listent must be a search function
618 * and can just scan through what we have.
620 if (context->bufsize == 0 ||
621 (XFS_ISRESET_CURSOR(cursor) &&
622 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
623 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
624 error = context->put_listent(context,
625 sfe->flags,
626 sfe->nameval,
627 (int)sfe->namelen,
628 (int)sfe->valuelen,
629 &sfe->nameval[sfe->namelen]);
632 * Either search callback finished early or
633 * didn't fit it all in the buffer after all.
635 if (context->seen_enough)
636 break;
638 if (error)
639 return error;
640 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
642 trace_xfs_attr_list_sf_all(context);
643 return(0);
646 /* do no more for a search callback */
647 if (context->bufsize == 0)
648 return 0;
651 * It didn't all fit, so we have to sort everything on hashval.
653 sbsize = sf->hdr.count * sizeof(*sbuf);
654 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
657 * Scan the attribute list for the rest of the entries, storing
658 * the relevant info from only those that match into a buffer.
660 nsbuf = 0;
661 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
662 if (unlikely(
663 ((char *)sfe < (char *)sf) ||
664 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
665 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
666 XFS_ERRLEVEL_LOW,
667 context->dp->i_mount, sfe);
668 kmem_free(sbuf);
669 return XFS_ERROR(EFSCORRUPTED);
672 sbp->entno = i;
673 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
674 sbp->name = sfe->nameval;
675 sbp->namelen = sfe->namelen;
676 /* These are bytes, and both on-disk, don't endian-flip */
677 sbp->valuelen = sfe->valuelen;
678 sbp->flags = sfe->flags;
679 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
680 sbp++;
681 nsbuf++;
685 * Sort the entries on hash then entno.
687 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
690 * Re-find our place IN THE SORTED LIST.
692 count = 0;
693 cursor->initted = 1;
694 cursor->blkno = 0;
695 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
696 if (sbp->hash == cursor->hashval) {
697 if (cursor->offset == count) {
698 break;
700 count++;
701 } else if (sbp->hash > cursor->hashval) {
702 break;
705 if (i == nsbuf) {
706 kmem_free(sbuf);
707 return(0);
711 * Loop putting entries into the user buffer.
713 for ( ; i < nsbuf; i++, sbp++) {
714 if (cursor->hashval != sbp->hash) {
715 cursor->hashval = sbp->hash;
716 cursor->offset = 0;
718 error = context->put_listent(context,
719 sbp->flags,
720 sbp->name,
721 sbp->namelen,
722 sbp->valuelen,
723 &sbp->name[sbp->namelen]);
724 if (error)
725 return error;
726 if (context->seen_enough)
727 break;
728 cursor->offset++;
731 kmem_free(sbuf);
732 return(0);
736 * Check a leaf attribute block to see if all the entries would fit into
737 * a shortform attribute list.
740 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
742 xfs_attr_leafblock_t *leaf;
743 xfs_attr_leaf_entry_t *entry;
744 xfs_attr_leaf_name_local_t *name_loc;
745 int bytes, i;
747 leaf = bp->data;
748 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
750 entry = &leaf->entries[0];
751 bytes = sizeof(struct xfs_attr_sf_hdr);
752 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
753 if (entry->flags & XFS_ATTR_INCOMPLETE)
754 continue; /* don't copy partial entries */
755 if (!(entry->flags & XFS_ATTR_LOCAL))
756 return(0);
757 name_loc = xfs_attr_leaf_name_local(leaf, i);
758 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
759 return(0);
760 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
761 return(0);
762 bytes += sizeof(struct xfs_attr_sf_entry)-1
763 + name_loc->namelen
764 + be16_to_cpu(name_loc->valuelen);
766 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
767 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
768 (bytes == sizeof(struct xfs_attr_sf_hdr)))
769 return(-1);
770 return(xfs_attr_shortform_bytesfit(dp, bytes));
774 * Convert a leaf attribute list to shortform attribute list
777 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
779 xfs_attr_leafblock_t *leaf;
780 xfs_attr_leaf_entry_t *entry;
781 xfs_attr_leaf_name_local_t *name_loc;
782 xfs_da_args_t nargs;
783 xfs_inode_t *dp;
784 char *tmpbuffer;
785 int error, i;
787 dp = args->dp;
788 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
789 ASSERT(tmpbuffer != NULL);
791 ASSERT(bp != NULL);
792 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
793 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
794 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
795 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
798 * Clean out the prior contents of the attribute list.
800 error = xfs_da_shrink_inode(args, 0, bp);
801 if (error)
802 goto out;
804 if (forkoff == -1) {
805 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
806 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
807 xfs_attr_fork_reset(dp, args->trans);
808 goto out;
811 xfs_attr_shortform_create(args);
814 * Copy the attributes
816 memset((char *)&nargs, 0, sizeof(nargs));
817 nargs.dp = dp;
818 nargs.firstblock = args->firstblock;
819 nargs.flist = args->flist;
820 nargs.total = args->total;
821 nargs.whichfork = XFS_ATTR_FORK;
822 nargs.trans = args->trans;
823 nargs.op_flags = XFS_DA_OP_OKNOENT;
824 entry = &leaf->entries[0];
825 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
826 if (entry->flags & XFS_ATTR_INCOMPLETE)
827 continue; /* don't copy partial entries */
828 if (!entry->nameidx)
829 continue;
830 ASSERT(entry->flags & XFS_ATTR_LOCAL);
831 name_loc = xfs_attr_leaf_name_local(leaf, i);
832 nargs.name = name_loc->nameval;
833 nargs.namelen = name_loc->namelen;
834 nargs.value = &name_loc->nameval[nargs.namelen];
835 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
836 nargs.hashval = be32_to_cpu(entry->hashval);
837 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
838 xfs_attr_shortform_add(&nargs, forkoff);
840 error = 0;
842 out:
843 kmem_free(tmpbuffer);
844 return(error);
848 * Convert from using a single leaf to a root node and a leaf.
851 xfs_attr_leaf_to_node(xfs_da_args_t *args)
853 xfs_attr_leafblock_t *leaf;
854 xfs_da_intnode_t *node;
855 xfs_inode_t *dp;
856 xfs_dabuf_t *bp1, *bp2;
857 xfs_dablk_t blkno;
858 int error;
860 dp = args->dp;
861 bp1 = bp2 = NULL;
862 error = xfs_da_grow_inode(args, &blkno);
863 if (error)
864 goto out;
865 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
866 XFS_ATTR_FORK);
867 if (error)
868 goto out;
869 ASSERT(bp1 != NULL);
870 bp2 = NULL;
871 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
872 XFS_ATTR_FORK);
873 if (error)
874 goto out;
875 ASSERT(bp2 != NULL);
876 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
877 xfs_da_buf_done(bp1);
878 bp1 = NULL;
879 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
882 * Set up the new root node.
884 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
885 if (error)
886 goto out;
887 node = bp1->data;
888 leaf = bp2->data;
889 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
890 /* both on-disk, don't endian-flip twice */
891 node->btree[0].hashval =
892 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
893 node->btree[0].before = cpu_to_be32(blkno);
894 node->hdr.count = cpu_to_be16(1);
895 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
896 error = 0;
897 out:
898 if (bp1)
899 xfs_da_buf_done(bp1);
900 if (bp2)
901 xfs_da_buf_done(bp2);
902 return(error);
906 /*========================================================================
907 * Routines used for growing the Btree.
908 *========================================================================*/
911 * Create the initial contents of a leaf attribute list
912 * or a leaf in a node attribute list.
914 STATIC int
915 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
917 xfs_attr_leafblock_t *leaf;
918 xfs_attr_leaf_hdr_t *hdr;
919 xfs_inode_t *dp;
920 xfs_dabuf_t *bp;
921 int error;
923 dp = args->dp;
924 ASSERT(dp != NULL);
925 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
926 XFS_ATTR_FORK);
927 if (error)
928 return(error);
929 ASSERT(bp != NULL);
930 leaf = bp->data;
931 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
932 hdr = &leaf->hdr;
933 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
934 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
935 if (!hdr->firstused) {
936 hdr->firstused = cpu_to_be16(
937 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
940 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
941 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
942 sizeof(xfs_attr_leaf_hdr_t));
944 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
946 *bpp = bp;
947 return(0);
951 * Split the leaf node, rebalance, then add the new entry.
954 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
955 xfs_da_state_blk_t *newblk)
957 xfs_dablk_t blkno;
958 int error;
961 * Allocate space for a new leaf node.
963 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
964 error = xfs_da_grow_inode(state->args, &blkno);
965 if (error)
966 return(error);
967 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
968 if (error)
969 return(error);
970 newblk->blkno = blkno;
971 newblk->magic = XFS_ATTR_LEAF_MAGIC;
974 * Rebalance the entries across the two leaves.
975 * NOTE: rebalance() currently depends on the 2nd block being empty.
977 xfs_attr_leaf_rebalance(state, oldblk, newblk);
978 error = xfs_da_blk_link(state, oldblk, newblk);
979 if (error)
980 return(error);
983 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
984 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
985 * "new" attrs info. Will need the "old" info to remove it later.
987 * Insert the "new" entry in the correct block.
989 if (state->inleaf)
990 error = xfs_attr_leaf_add(oldblk->bp, state->args);
991 else
992 error = xfs_attr_leaf_add(newblk->bp, state->args);
995 * Update last hashval in each block since we added the name.
997 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
998 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
999 return(error);
1003 * Add a name to the leaf attribute list structure.
1006 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
1008 xfs_attr_leafblock_t *leaf;
1009 xfs_attr_leaf_hdr_t *hdr;
1010 xfs_attr_leaf_map_t *map;
1011 int tablesize, entsize, sum, tmp, i;
1013 leaf = bp->data;
1014 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1015 ASSERT((args->index >= 0)
1016 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1017 hdr = &leaf->hdr;
1018 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1019 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1022 * Search through freemap for first-fit on new name length.
1023 * (may need to figure in size of entry struct too)
1025 tablesize = (be16_to_cpu(hdr->count) + 1)
1026 * sizeof(xfs_attr_leaf_entry_t)
1027 + sizeof(xfs_attr_leaf_hdr_t);
1028 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1029 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1030 if (tablesize > be16_to_cpu(hdr->firstused)) {
1031 sum += be16_to_cpu(map->size);
1032 continue;
1034 if (!map->size)
1035 continue; /* no space in this map */
1036 tmp = entsize;
1037 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1038 tmp += sizeof(xfs_attr_leaf_entry_t);
1039 if (be16_to_cpu(map->size) >= tmp) {
1040 tmp = xfs_attr_leaf_add_work(bp, args, i);
1041 return(tmp);
1043 sum += be16_to_cpu(map->size);
1047 * If there are no holes in the address space of the block,
1048 * and we don't have enough freespace, then compaction will do us
1049 * no good and we should just give up.
1051 if (!hdr->holes && (sum < entsize))
1052 return(XFS_ERROR(ENOSPC));
1055 * Compact the entries to coalesce free space.
1056 * This may change the hdr->count via dropping INCOMPLETE entries.
1058 xfs_attr_leaf_compact(args->trans, bp);
1061 * After compaction, the block is guaranteed to have only one
1062 * free region, in freemap[0]. If it is not big enough, give up.
1064 if (be16_to_cpu(hdr->freemap[0].size)
1065 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1066 return(XFS_ERROR(ENOSPC));
1068 return(xfs_attr_leaf_add_work(bp, args, 0));
1072 * Add a name to a leaf attribute list structure.
1074 STATIC int
1075 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1077 xfs_attr_leafblock_t *leaf;
1078 xfs_attr_leaf_hdr_t *hdr;
1079 xfs_attr_leaf_entry_t *entry;
1080 xfs_attr_leaf_name_local_t *name_loc;
1081 xfs_attr_leaf_name_remote_t *name_rmt;
1082 xfs_attr_leaf_map_t *map;
1083 xfs_mount_t *mp;
1084 int tmp, i;
1086 leaf = bp->data;
1087 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1088 hdr = &leaf->hdr;
1089 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1090 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1093 * Force open some space in the entry array and fill it in.
1095 entry = &leaf->entries[args->index];
1096 if (args->index < be16_to_cpu(hdr->count)) {
1097 tmp = be16_to_cpu(hdr->count) - args->index;
1098 tmp *= sizeof(xfs_attr_leaf_entry_t);
1099 memmove((char *)(entry+1), (char *)entry, tmp);
1100 xfs_da_log_buf(args->trans, bp,
1101 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1103 be16_add_cpu(&hdr->count, 1);
1106 * Allocate space for the new string (at the end of the run).
1108 map = &hdr->freemap[mapindex];
1109 mp = args->trans->t_mountp;
1110 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1111 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1112 ASSERT(be16_to_cpu(map->size) >=
1113 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1114 mp->m_sb.sb_blocksize, NULL));
1115 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1116 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1117 be16_add_cpu(&map->size,
1118 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1119 mp->m_sb.sb_blocksize, &tmp));
1120 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1121 be16_to_cpu(map->size));
1122 entry->hashval = cpu_to_be32(args->hashval);
1123 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1124 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1125 if (args->op_flags & XFS_DA_OP_RENAME) {
1126 entry->flags |= XFS_ATTR_INCOMPLETE;
1127 if ((args->blkno2 == args->blkno) &&
1128 (args->index2 <= args->index)) {
1129 args->index2++;
1132 xfs_da_log_buf(args->trans, bp,
1133 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1134 ASSERT((args->index == 0) ||
1135 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1136 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1137 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1140 * Copy the attribute name and value into the new space.
1142 * For "remote" attribute values, simply note that we need to
1143 * allocate space for the "remote" value. We can't actually
1144 * allocate the extents in this transaction, and we can't decide
1145 * which blocks they should be as we might allocate more blocks
1146 * as part of this transaction (a split operation for example).
1148 if (entry->flags & XFS_ATTR_LOCAL) {
1149 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
1150 name_loc->namelen = args->namelen;
1151 name_loc->valuelen = cpu_to_be16(args->valuelen);
1152 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1153 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1154 be16_to_cpu(name_loc->valuelen));
1155 } else {
1156 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
1157 name_rmt->namelen = args->namelen;
1158 memcpy((char *)name_rmt->name, args->name, args->namelen);
1159 entry->flags |= XFS_ATTR_INCOMPLETE;
1160 /* just in case */
1161 name_rmt->valuelen = 0;
1162 name_rmt->valueblk = 0;
1163 args->rmtblkno = 1;
1164 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1166 xfs_da_log_buf(args->trans, bp,
1167 XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1168 xfs_attr_leaf_entsize(leaf, args->index)));
1171 * Update the control info for this leaf node
1173 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1174 /* both on-disk, don't endian-flip twice */
1175 hdr->firstused = entry->nameidx;
1177 ASSERT(be16_to_cpu(hdr->firstused) >=
1178 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1179 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1180 + sizeof(xfs_attr_leaf_hdr_t);
1181 map = &hdr->freemap[0];
1182 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1183 if (be16_to_cpu(map->base) == tmp) {
1184 be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1185 be16_add_cpu(&map->size,
1186 -((int)sizeof(xfs_attr_leaf_entry_t)));
1189 be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1190 xfs_da_log_buf(args->trans, bp,
1191 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1192 return(0);
1196 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1198 STATIC void
1199 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1201 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1202 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1203 xfs_mount_t *mp;
1204 char *tmpbuffer;
1206 mp = trans->t_mountp;
1207 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1208 ASSERT(tmpbuffer != NULL);
1209 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1210 memset(bp->data, 0, XFS_LBSIZE(mp));
1213 * Copy basic information
1215 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1216 leaf_d = bp->data;
1217 hdr_s = &leaf_s->hdr;
1218 hdr_d = &leaf_d->hdr;
1219 hdr_d->info = hdr_s->info; /* struct copy */
1220 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1221 /* handle truncation gracefully */
1222 if (!hdr_d->firstused) {
1223 hdr_d->firstused = cpu_to_be16(
1224 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1226 hdr_d->usedbytes = 0;
1227 hdr_d->count = 0;
1228 hdr_d->holes = 0;
1229 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1230 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1231 sizeof(xfs_attr_leaf_hdr_t));
1234 * Copy all entry's in the same (sorted) order,
1235 * but allocate name/value pairs packed and in sequence.
1237 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1238 be16_to_cpu(hdr_s->count), mp);
1239 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1241 kmem_free(tmpbuffer);
1245 * Redistribute the attribute list entries between two leaf nodes,
1246 * taking into account the size of the new entry.
1248 * NOTE: if new block is empty, then it will get the upper half of the
1249 * old block. At present, all (one) callers pass in an empty second block.
1251 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1252 * to match what it is doing in splitting the attribute leaf block. Those
1253 * values are used in "atomic rename" operations on attributes. Note that
1254 * the "new" and "old" values can end up in different blocks.
1256 STATIC void
1257 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1258 xfs_da_state_blk_t *blk2)
1260 xfs_da_args_t *args;
1261 xfs_da_state_blk_t *tmp_blk;
1262 xfs_attr_leafblock_t *leaf1, *leaf2;
1263 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1264 int count, totallen, max, space, swap;
1267 * Set up environment.
1269 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1270 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1271 leaf1 = blk1->bp->data;
1272 leaf2 = blk2->bp->data;
1273 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1274 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1275 args = state->args;
1278 * Check ordering of blocks, reverse if it makes things simpler.
1280 * NOTE: Given that all (current) callers pass in an empty
1281 * second block, this code should never set "swap".
1283 swap = 0;
1284 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1285 tmp_blk = blk1;
1286 blk1 = blk2;
1287 blk2 = tmp_blk;
1288 leaf1 = blk1->bp->data;
1289 leaf2 = blk2->bp->data;
1290 swap = 1;
1292 hdr1 = &leaf1->hdr;
1293 hdr2 = &leaf2->hdr;
1296 * Examine entries until we reduce the absolute difference in
1297 * byte usage between the two blocks to a minimum. Then get
1298 * the direction to copy and the number of elements to move.
1300 * "inleaf" is true if the new entry should be inserted into blk1.
1301 * If "swap" is also true, then reverse the sense of "inleaf".
1303 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1304 &count, &totallen);
1305 if (swap)
1306 state->inleaf = !state->inleaf;
1309 * Move any entries required from leaf to leaf:
1311 if (count < be16_to_cpu(hdr1->count)) {
1313 * Figure the total bytes to be added to the destination leaf.
1315 /* number entries being moved */
1316 count = be16_to_cpu(hdr1->count) - count;
1317 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1318 space += count * sizeof(xfs_attr_leaf_entry_t);
1321 * leaf2 is the destination, compact it if it looks tight.
1323 max = be16_to_cpu(hdr2->firstused)
1324 - sizeof(xfs_attr_leaf_hdr_t);
1325 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1326 if (space > max) {
1327 xfs_attr_leaf_compact(args->trans, blk2->bp);
1331 * Move high entries from leaf1 to low end of leaf2.
1333 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1334 leaf2, 0, count, state->mp);
1336 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1337 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1338 } else if (count > be16_to_cpu(hdr1->count)) {
1340 * I assert that since all callers pass in an empty
1341 * second buffer, this code should never execute.
1345 * Figure the total bytes to be added to the destination leaf.
1347 /* number entries being moved */
1348 count -= be16_to_cpu(hdr1->count);
1349 space = totallen - be16_to_cpu(hdr1->usedbytes);
1350 space += count * sizeof(xfs_attr_leaf_entry_t);
1353 * leaf1 is the destination, compact it if it looks tight.
1355 max = be16_to_cpu(hdr1->firstused)
1356 - sizeof(xfs_attr_leaf_hdr_t);
1357 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1358 if (space > max) {
1359 xfs_attr_leaf_compact(args->trans, blk1->bp);
1363 * Move low entries from leaf2 to high end of leaf1.
1365 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1366 be16_to_cpu(hdr1->count), count, state->mp);
1368 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1369 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1373 * Copy out last hashval in each block for B-tree code.
1375 blk1->hashval = be32_to_cpu(
1376 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1377 blk2->hashval = be32_to_cpu(
1378 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1381 * Adjust the expected index for insertion.
1382 * NOTE: this code depends on the (current) situation that the
1383 * second block was originally empty.
1385 * If the insertion point moved to the 2nd block, we must adjust
1386 * the index. We must also track the entry just following the
1387 * new entry for use in an "atomic rename" operation, that entry
1388 * is always the "old" entry and the "new" entry is what we are
1389 * inserting. The index/blkno fields refer to the "old" entry,
1390 * while the index2/blkno2 fields refer to the "new" entry.
1392 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1393 ASSERT(state->inleaf == 0);
1394 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1395 args->index = args->index2 = blk2->index;
1396 args->blkno = args->blkno2 = blk2->blkno;
1397 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1398 if (state->inleaf) {
1399 args->index = blk1->index;
1400 args->blkno = blk1->blkno;
1401 args->index2 = 0;
1402 args->blkno2 = blk2->blkno;
1403 } else {
1404 blk2->index = blk1->index
1405 - be16_to_cpu(leaf1->hdr.count);
1406 args->index = args->index2 = blk2->index;
1407 args->blkno = args->blkno2 = blk2->blkno;
1409 } else {
1410 ASSERT(state->inleaf == 1);
1411 args->index = args->index2 = blk1->index;
1412 args->blkno = args->blkno2 = blk1->blkno;
1417 * Examine entries until we reduce the absolute difference in
1418 * byte usage between the two blocks to a minimum.
1419 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1420 * GROT: there will always be enough room in either block for a new entry.
1421 * GROT: Do a double-split for this case?
1423 STATIC int
1424 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1425 xfs_da_state_blk_t *blk1,
1426 xfs_da_state_blk_t *blk2,
1427 int *countarg, int *usedbytesarg)
1429 xfs_attr_leafblock_t *leaf1, *leaf2;
1430 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1431 xfs_attr_leaf_entry_t *entry;
1432 int count, max, index, totallen, half;
1433 int lastdelta, foundit, tmp;
1436 * Set up environment.
1438 leaf1 = blk1->bp->data;
1439 leaf2 = blk2->bp->data;
1440 hdr1 = &leaf1->hdr;
1441 hdr2 = &leaf2->hdr;
1442 foundit = 0;
1443 totallen = 0;
1446 * Examine entries until we reduce the absolute difference in
1447 * byte usage between the two blocks to a minimum.
1449 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1450 half = (max+1) * sizeof(*entry);
1451 half += be16_to_cpu(hdr1->usedbytes) +
1452 be16_to_cpu(hdr2->usedbytes) +
1453 xfs_attr_leaf_newentsize(
1454 state->args->namelen,
1455 state->args->valuelen,
1456 state->blocksize, NULL);
1457 half /= 2;
1458 lastdelta = state->blocksize;
1459 entry = &leaf1->entries[0];
1460 for (count = index = 0; count < max; entry++, index++, count++) {
1462 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1464 * The new entry is in the first block, account for it.
1466 if (count == blk1->index) {
1467 tmp = totallen + sizeof(*entry) +
1468 xfs_attr_leaf_newentsize(
1469 state->args->namelen,
1470 state->args->valuelen,
1471 state->blocksize, NULL);
1472 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1473 break;
1474 lastdelta = XFS_ATTR_ABS(half - tmp);
1475 totallen = tmp;
1476 foundit = 1;
1480 * Wrap around into the second block if necessary.
1482 if (count == be16_to_cpu(hdr1->count)) {
1483 leaf1 = leaf2;
1484 entry = &leaf1->entries[0];
1485 index = 0;
1489 * Figure out if next leaf entry would be too much.
1491 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1492 index);
1493 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1494 break;
1495 lastdelta = XFS_ATTR_ABS(half - tmp);
1496 totallen = tmp;
1497 #undef XFS_ATTR_ABS
1501 * Calculate the number of usedbytes that will end up in lower block.
1502 * If new entry not in lower block, fix up the count.
1504 totallen -= count * sizeof(*entry);
1505 if (foundit) {
1506 totallen -= sizeof(*entry) +
1507 xfs_attr_leaf_newentsize(
1508 state->args->namelen,
1509 state->args->valuelen,
1510 state->blocksize, NULL);
1513 *countarg = count;
1514 *usedbytesarg = totallen;
1515 return(foundit);
1518 /*========================================================================
1519 * Routines used for shrinking the Btree.
1520 *========================================================================*/
1523 * Check a leaf block and its neighbors to see if the block should be
1524 * collapsed into one or the other neighbor. Always keep the block
1525 * with the smaller block number.
1526 * If the current block is over 50% full, don't try to join it, return 0.
1527 * If the block is empty, fill in the state structure and return 2.
1528 * If it can be collapsed, fill in the state structure and return 1.
1529 * If nothing can be done, return 0.
1531 * GROT: allow for INCOMPLETE entries in calculation.
1534 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1536 xfs_attr_leafblock_t *leaf;
1537 xfs_da_state_blk_t *blk;
1538 xfs_da_blkinfo_t *info;
1539 int count, bytes, forward, error, retval, i;
1540 xfs_dablk_t blkno;
1541 xfs_dabuf_t *bp;
1544 * Check for the degenerate case of the block being over 50% full.
1545 * If so, it's not worth even looking to see if we might be able
1546 * to coalesce with a sibling.
1548 blk = &state->path.blk[ state->path.active-1 ];
1549 info = blk->bp->data;
1550 ASSERT(be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC);
1551 leaf = (xfs_attr_leafblock_t *)info;
1552 count = be16_to_cpu(leaf->hdr.count);
1553 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1554 count * sizeof(xfs_attr_leaf_entry_t) +
1555 be16_to_cpu(leaf->hdr.usedbytes);
1556 if (bytes > (state->blocksize >> 1)) {
1557 *action = 0; /* blk over 50%, don't try to join */
1558 return(0);
1562 * Check for the degenerate case of the block being empty.
1563 * If the block is empty, we'll simply delete it, no need to
1564 * coalesce it with a sibling block. We choose (arbitrarily)
1565 * to merge with the forward block unless it is NULL.
1567 if (count == 0) {
1569 * Make altpath point to the block we want to keep and
1570 * path point to the block we want to drop (this one).
1572 forward = (info->forw != 0);
1573 memcpy(&state->altpath, &state->path, sizeof(state->path));
1574 error = xfs_da_path_shift(state, &state->altpath, forward,
1575 0, &retval);
1576 if (error)
1577 return(error);
1578 if (retval) {
1579 *action = 0;
1580 } else {
1581 *action = 2;
1583 return(0);
1587 * Examine each sibling block to see if we can coalesce with
1588 * at least 25% free space to spare. We need to figure out
1589 * whether to merge with the forward or the backward block.
1590 * We prefer coalescing with the lower numbered sibling so as
1591 * to shrink an attribute list over time.
1593 /* start with smaller blk num */
1594 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1595 for (i = 0; i < 2; forward = !forward, i++) {
1596 if (forward)
1597 blkno = be32_to_cpu(info->forw);
1598 else
1599 blkno = be32_to_cpu(info->back);
1600 if (blkno == 0)
1601 continue;
1602 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1603 blkno, -1, &bp, XFS_ATTR_FORK);
1604 if (error)
1605 return(error);
1606 ASSERT(bp != NULL);
1608 leaf = (xfs_attr_leafblock_t *)info;
1609 count = be16_to_cpu(leaf->hdr.count);
1610 bytes = state->blocksize - (state->blocksize>>2);
1611 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1612 leaf = bp->data;
1613 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1614 count += be16_to_cpu(leaf->hdr.count);
1615 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1616 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1617 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1618 xfs_da_brelse(state->args->trans, bp);
1619 if (bytes >= 0)
1620 break; /* fits with at least 25% to spare */
1622 if (i >= 2) {
1623 *action = 0;
1624 return(0);
1628 * Make altpath point to the block we want to keep (the lower
1629 * numbered block) and path point to the block we want to drop.
1631 memcpy(&state->altpath, &state->path, sizeof(state->path));
1632 if (blkno < blk->blkno) {
1633 error = xfs_da_path_shift(state, &state->altpath, forward,
1634 0, &retval);
1635 } else {
1636 error = xfs_da_path_shift(state, &state->path, forward,
1637 0, &retval);
1639 if (error)
1640 return(error);
1641 if (retval) {
1642 *action = 0;
1643 } else {
1644 *action = 1;
1646 return(0);
1650 * Remove a name from the leaf attribute list structure.
1652 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1653 * If two leaves are 37% full, when combined they will leave 25% free.
1656 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1658 xfs_attr_leafblock_t *leaf;
1659 xfs_attr_leaf_hdr_t *hdr;
1660 xfs_attr_leaf_map_t *map;
1661 xfs_attr_leaf_entry_t *entry;
1662 int before, after, smallest, entsize;
1663 int tablesize, tmp, i;
1664 xfs_mount_t *mp;
1666 leaf = bp->data;
1667 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1668 hdr = &leaf->hdr;
1669 mp = args->trans->t_mountp;
1670 ASSERT((be16_to_cpu(hdr->count) > 0)
1671 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1672 ASSERT((args->index >= 0)
1673 && (args->index < be16_to_cpu(hdr->count)));
1674 ASSERT(be16_to_cpu(hdr->firstused) >=
1675 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1676 entry = &leaf->entries[args->index];
1677 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1678 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1681 * Scan through free region table:
1682 * check for adjacency of free'd entry with an existing one,
1683 * find smallest free region in case we need to replace it,
1684 * adjust any map that borders the entry table,
1686 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1687 + sizeof(xfs_attr_leaf_hdr_t);
1688 map = &hdr->freemap[0];
1689 tmp = be16_to_cpu(map->size);
1690 before = after = -1;
1691 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1692 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1693 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1694 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1695 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1696 if (be16_to_cpu(map->base) == tablesize) {
1697 be16_add_cpu(&map->base,
1698 -((int)sizeof(xfs_attr_leaf_entry_t)));
1699 be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1702 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1703 == be16_to_cpu(entry->nameidx)) {
1704 before = i;
1705 } else if (be16_to_cpu(map->base)
1706 == (be16_to_cpu(entry->nameidx) + entsize)) {
1707 after = i;
1708 } else if (be16_to_cpu(map->size) < tmp) {
1709 tmp = be16_to_cpu(map->size);
1710 smallest = i;
1715 * Coalesce adjacent freemap regions,
1716 * or replace the smallest region.
1718 if ((before >= 0) || (after >= 0)) {
1719 if ((before >= 0) && (after >= 0)) {
1720 map = &hdr->freemap[before];
1721 be16_add_cpu(&map->size, entsize);
1722 be16_add_cpu(&map->size,
1723 be16_to_cpu(hdr->freemap[after].size));
1724 hdr->freemap[after].base = 0;
1725 hdr->freemap[after].size = 0;
1726 } else if (before >= 0) {
1727 map = &hdr->freemap[before];
1728 be16_add_cpu(&map->size, entsize);
1729 } else {
1730 map = &hdr->freemap[after];
1731 /* both on-disk, don't endian flip twice */
1732 map->base = entry->nameidx;
1733 be16_add_cpu(&map->size, entsize);
1735 } else {
1737 * Replace smallest region (if it is smaller than free'd entry)
1739 map = &hdr->freemap[smallest];
1740 if (be16_to_cpu(map->size) < entsize) {
1741 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1742 map->size = cpu_to_be16(entsize);
1747 * Did we remove the first entry?
1749 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1750 smallest = 1;
1751 else
1752 smallest = 0;
1755 * Compress the remaining entries and zero out the removed stuff.
1757 memset(xfs_attr_leaf_name(leaf, args->index), 0, entsize);
1758 be16_add_cpu(&hdr->usedbytes, -entsize);
1759 xfs_da_log_buf(args->trans, bp,
1760 XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1761 entsize));
1763 tmp = (be16_to_cpu(hdr->count) - args->index)
1764 * sizeof(xfs_attr_leaf_entry_t);
1765 memmove((char *)entry, (char *)(entry+1), tmp);
1766 be16_add_cpu(&hdr->count, -1);
1767 xfs_da_log_buf(args->trans, bp,
1768 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1769 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1770 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1773 * If we removed the first entry, re-find the first used byte
1774 * in the name area. Note that if the entry was the "firstused",
1775 * then we don't have a "hole" in our block resulting from
1776 * removing the name.
1778 if (smallest) {
1779 tmp = XFS_LBSIZE(mp);
1780 entry = &leaf->entries[0];
1781 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1782 ASSERT(be16_to_cpu(entry->nameidx) >=
1783 be16_to_cpu(hdr->firstused));
1784 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1786 if (be16_to_cpu(entry->nameidx) < tmp)
1787 tmp = be16_to_cpu(entry->nameidx);
1789 hdr->firstused = cpu_to_be16(tmp);
1790 if (!hdr->firstused) {
1791 hdr->firstused = cpu_to_be16(
1792 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1794 } else {
1795 hdr->holes = 1; /* mark as needing compaction */
1797 xfs_da_log_buf(args->trans, bp,
1798 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1801 * Check if leaf is less than 50% full, caller may want to
1802 * "join" the leaf with a sibling if so.
1804 tmp = sizeof(xfs_attr_leaf_hdr_t);
1805 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1806 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1807 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1811 * Move all the attribute list entries from drop_leaf into save_leaf.
1813 void
1814 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1815 xfs_da_state_blk_t *save_blk)
1817 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1818 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1819 xfs_mount_t *mp;
1820 char *tmpbuffer;
1823 * Set up environment.
1825 mp = state->mp;
1826 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1827 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1828 drop_leaf = drop_blk->bp->data;
1829 save_leaf = save_blk->bp->data;
1830 ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1831 ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1832 drop_hdr = &drop_leaf->hdr;
1833 save_hdr = &save_leaf->hdr;
1836 * Save last hashval from dying block for later Btree fixup.
1838 drop_blk->hashval = be32_to_cpu(
1839 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1842 * Check if we need a temp buffer, or can we do it in place.
1843 * Note that we don't check "leaf" for holes because we will
1844 * always be dropping it, toosmall() decided that for us already.
1846 if (save_hdr->holes == 0) {
1848 * dest leaf has no holes, so we add there. May need
1849 * to make some room in the entry array.
1851 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1852 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1853 be16_to_cpu(drop_hdr->count), mp);
1854 } else {
1855 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1856 be16_to_cpu(save_hdr->count),
1857 be16_to_cpu(drop_hdr->count), mp);
1859 } else {
1861 * Destination has holes, so we make a temporary copy
1862 * of the leaf and add them both to that.
1864 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1865 ASSERT(tmpbuffer != NULL);
1866 memset(tmpbuffer, 0, state->blocksize);
1867 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1868 tmp_hdr = &tmp_leaf->hdr;
1869 tmp_hdr->info = save_hdr->info; /* struct copy */
1870 tmp_hdr->count = 0;
1871 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1872 if (!tmp_hdr->firstused) {
1873 tmp_hdr->firstused = cpu_to_be16(
1874 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1876 tmp_hdr->usedbytes = 0;
1877 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1878 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1879 be16_to_cpu(drop_hdr->count), mp);
1880 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1881 be16_to_cpu(tmp_leaf->hdr.count),
1882 be16_to_cpu(save_hdr->count), mp);
1883 } else {
1884 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1885 be16_to_cpu(save_hdr->count), mp);
1886 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1887 be16_to_cpu(tmp_leaf->hdr.count),
1888 be16_to_cpu(drop_hdr->count), mp);
1890 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1891 kmem_free(tmpbuffer);
1894 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1895 state->blocksize - 1);
1898 * Copy out last hashval in each block for B-tree code.
1900 save_blk->hashval = be32_to_cpu(
1901 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1904 /*========================================================================
1905 * Routines used for finding things in the Btree.
1906 *========================================================================*/
1909 * Look up a name in a leaf attribute list structure.
1910 * This is the internal routine, it uses the caller's buffer.
1912 * Note that duplicate keys are allowed, but only check within the
1913 * current leaf node. The Btree code must check in adjacent leaf nodes.
1915 * Return in args->index the index into the entry[] array of either
1916 * the found entry, or where the entry should have been (insert before
1917 * that entry).
1919 * Don't change the args->value unless we find the attribute.
1922 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1924 xfs_attr_leafblock_t *leaf;
1925 xfs_attr_leaf_entry_t *entry;
1926 xfs_attr_leaf_name_local_t *name_loc;
1927 xfs_attr_leaf_name_remote_t *name_rmt;
1928 int probe, span;
1929 xfs_dahash_t hashval;
1931 leaf = bp->data;
1932 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
1933 ASSERT(be16_to_cpu(leaf->hdr.count)
1934 < (XFS_LBSIZE(args->dp->i_mount)/8));
1937 * Binary search. (note: small blocks will skip this loop)
1939 hashval = args->hashval;
1940 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1941 for (entry = &leaf->entries[probe]; span > 4;
1942 entry = &leaf->entries[probe]) {
1943 span /= 2;
1944 if (be32_to_cpu(entry->hashval) < hashval)
1945 probe += span;
1946 else if (be32_to_cpu(entry->hashval) > hashval)
1947 probe -= span;
1948 else
1949 break;
1951 ASSERT((probe >= 0) &&
1952 (!leaf->hdr.count
1953 || (probe < be16_to_cpu(leaf->hdr.count))));
1954 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1957 * Since we may have duplicate hashval's, find the first matching
1958 * hashval in the leaf.
1960 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1961 entry--;
1962 probe--;
1964 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1965 (be32_to_cpu(entry->hashval) < hashval)) {
1966 entry++;
1967 probe++;
1969 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1970 (be32_to_cpu(entry->hashval) != hashval)) {
1971 args->index = probe;
1972 return(XFS_ERROR(ENOATTR));
1976 * Duplicate keys may be present, so search all of them for a match.
1978 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1979 (be32_to_cpu(entry->hashval) == hashval);
1980 entry++, probe++) {
1982 * GROT: Add code to remove incomplete entries.
1985 * If we are looking for INCOMPLETE entries, show only those.
1986 * If we are looking for complete entries, show only those.
1988 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
1989 (entry->flags & XFS_ATTR_INCOMPLETE)) {
1990 continue;
1992 if (entry->flags & XFS_ATTR_LOCAL) {
1993 name_loc = xfs_attr_leaf_name_local(leaf, probe);
1994 if (name_loc->namelen != args->namelen)
1995 continue;
1996 if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
1997 continue;
1998 if (!xfs_attr_namesp_match(args->flags, entry->flags))
1999 continue;
2000 args->index = probe;
2001 return(XFS_ERROR(EEXIST));
2002 } else {
2003 name_rmt = xfs_attr_leaf_name_remote(leaf, probe);
2004 if (name_rmt->namelen != args->namelen)
2005 continue;
2006 if (memcmp(args->name, (char *)name_rmt->name,
2007 args->namelen) != 0)
2008 continue;
2009 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2010 continue;
2011 args->index = probe;
2012 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2013 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2014 be32_to_cpu(name_rmt->valuelen));
2015 return(XFS_ERROR(EEXIST));
2018 args->index = probe;
2019 return(XFS_ERROR(ENOATTR));
2023 * Get the value associated with an attribute name from a leaf attribute
2024 * list structure.
2027 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2029 int valuelen;
2030 xfs_attr_leafblock_t *leaf;
2031 xfs_attr_leaf_entry_t *entry;
2032 xfs_attr_leaf_name_local_t *name_loc;
2033 xfs_attr_leaf_name_remote_t *name_rmt;
2035 leaf = bp->data;
2036 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2037 ASSERT(be16_to_cpu(leaf->hdr.count)
2038 < (XFS_LBSIZE(args->dp->i_mount)/8));
2039 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2041 entry = &leaf->entries[args->index];
2042 if (entry->flags & XFS_ATTR_LOCAL) {
2043 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2044 ASSERT(name_loc->namelen == args->namelen);
2045 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2046 valuelen = be16_to_cpu(name_loc->valuelen);
2047 if (args->flags & ATTR_KERNOVAL) {
2048 args->valuelen = valuelen;
2049 return(0);
2051 if (args->valuelen < valuelen) {
2052 args->valuelen = valuelen;
2053 return(XFS_ERROR(ERANGE));
2055 args->valuelen = valuelen;
2056 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2057 } else {
2058 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2059 ASSERT(name_rmt->namelen == args->namelen);
2060 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2061 valuelen = be32_to_cpu(name_rmt->valuelen);
2062 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2063 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2064 if (args->flags & ATTR_KERNOVAL) {
2065 args->valuelen = valuelen;
2066 return(0);
2068 if (args->valuelen < valuelen) {
2069 args->valuelen = valuelen;
2070 return(XFS_ERROR(ERANGE));
2072 args->valuelen = valuelen;
2074 return(0);
2077 /*========================================================================
2078 * Utility routines.
2079 *========================================================================*/
2082 * Move the indicated entries from one leaf to another.
2083 * NOTE: this routine modifies both source and destination leaves.
2085 /*ARGSUSED*/
2086 STATIC void
2087 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2088 xfs_attr_leafblock_t *leaf_d, int start_d,
2089 int count, xfs_mount_t *mp)
2091 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2092 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2093 int desti, tmp, i;
2096 * Check for nothing to do.
2098 if (count == 0)
2099 return;
2102 * Set up environment.
2104 ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2105 ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2106 hdr_s = &leaf_s->hdr;
2107 hdr_d = &leaf_d->hdr;
2108 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2109 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2110 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2111 ((be16_to_cpu(hdr_s->count)
2112 * sizeof(*entry_s))+sizeof(*hdr_s)));
2113 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2114 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2115 ((be16_to_cpu(hdr_d->count)
2116 * sizeof(*entry_d))+sizeof(*hdr_d)));
2118 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2119 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2120 ASSERT(count <= be16_to_cpu(hdr_s->count));
2123 * Move the entries in the destination leaf up to make a hole?
2125 if (start_d < be16_to_cpu(hdr_d->count)) {
2126 tmp = be16_to_cpu(hdr_d->count) - start_d;
2127 tmp *= sizeof(xfs_attr_leaf_entry_t);
2128 entry_s = &leaf_d->entries[start_d];
2129 entry_d = &leaf_d->entries[start_d + count];
2130 memmove((char *)entry_d, (char *)entry_s, tmp);
2134 * Copy all entry's in the same (sorted) order,
2135 * but allocate attribute info packed and in sequence.
2137 entry_s = &leaf_s->entries[start_s];
2138 entry_d = &leaf_d->entries[start_d];
2139 desti = start_d;
2140 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2141 ASSERT(be16_to_cpu(entry_s->nameidx)
2142 >= be16_to_cpu(hdr_s->firstused));
2143 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2144 #ifdef GROT
2146 * Code to drop INCOMPLETE entries. Difficult to use as we
2147 * may also need to change the insertion index. Code turned
2148 * off for 6.2, should be revisited later.
2150 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2151 memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2152 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2153 be16_add_cpu(&hdr_s->count, -1);
2154 entry_d--; /* to compensate for ++ in loop hdr */
2155 desti--;
2156 if ((start_s + i) < offset)
2157 result++; /* insertion index adjustment */
2158 } else {
2159 #endif /* GROT */
2160 be16_add_cpu(&hdr_d->firstused, -tmp);
2161 /* both on-disk, don't endian flip twice */
2162 entry_d->hashval = entry_s->hashval;
2163 /* both on-disk, don't endian flip twice */
2164 entry_d->nameidx = hdr_d->firstused;
2165 entry_d->flags = entry_s->flags;
2166 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2167 <= XFS_LBSIZE(mp));
2168 memmove(xfs_attr_leaf_name(leaf_d, desti),
2169 xfs_attr_leaf_name(leaf_s, start_s + i), tmp);
2170 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2171 <= XFS_LBSIZE(mp));
2172 memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2173 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2174 be16_add_cpu(&hdr_d->usedbytes, tmp);
2175 be16_add_cpu(&hdr_s->count, -1);
2176 be16_add_cpu(&hdr_d->count, 1);
2177 tmp = be16_to_cpu(hdr_d->count)
2178 * sizeof(xfs_attr_leaf_entry_t)
2179 + sizeof(xfs_attr_leaf_hdr_t);
2180 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2181 #ifdef GROT
2183 #endif /* GROT */
2187 * Zero out the entries we just copied.
2189 if (start_s == be16_to_cpu(hdr_s->count)) {
2190 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2191 entry_s = &leaf_s->entries[start_s];
2192 ASSERT(((char *)entry_s + tmp) <=
2193 ((char *)leaf_s + XFS_LBSIZE(mp)));
2194 memset((char *)entry_s, 0, tmp);
2195 } else {
2197 * Move the remaining entries down to fill the hole,
2198 * then zero the entries at the top.
2200 tmp = be16_to_cpu(hdr_s->count) - count;
2201 tmp *= sizeof(xfs_attr_leaf_entry_t);
2202 entry_s = &leaf_s->entries[start_s + count];
2203 entry_d = &leaf_s->entries[start_s];
2204 memmove((char *)entry_d, (char *)entry_s, tmp);
2206 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2207 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2208 ASSERT(((char *)entry_s + tmp) <=
2209 ((char *)leaf_s + XFS_LBSIZE(mp)));
2210 memset((char *)entry_s, 0, tmp);
2214 * Fill in the freemap information
2216 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2217 be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2218 sizeof(xfs_attr_leaf_entry_t));
2219 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2220 - be16_to_cpu(hdr_d->freemap[0].base));
2221 hdr_d->freemap[1].base = 0;
2222 hdr_d->freemap[2].base = 0;
2223 hdr_d->freemap[1].size = 0;
2224 hdr_d->freemap[2].size = 0;
2225 hdr_s->holes = 1; /* leaf may not be compact */
2229 * Compare two leaf blocks "order".
2230 * Return 0 unless leaf2 should go before leaf1.
2233 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2235 xfs_attr_leafblock_t *leaf1, *leaf2;
2237 leaf1 = leaf1_bp->data;
2238 leaf2 = leaf2_bp->data;
2239 ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC) &&
2240 (be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC));
2241 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2242 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2243 ((be32_to_cpu(leaf2->entries[0].hashval) <
2244 be32_to_cpu(leaf1->entries[0].hashval)) ||
2245 (be32_to_cpu(leaf2->entries[
2246 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2247 be32_to_cpu(leaf1->entries[
2248 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2249 return(1);
2251 return(0);
2255 * Pick up the last hashvalue from a leaf block.
2257 xfs_dahash_t
2258 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2260 xfs_attr_leafblock_t *leaf;
2262 leaf = bp->data;
2263 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2264 if (count)
2265 *count = be16_to_cpu(leaf->hdr.count);
2266 if (!leaf->hdr.count)
2267 return(0);
2268 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2272 * Calculate the number of bytes used to store the indicated attribute
2273 * (whether local or remote only calculate bytes in this block).
2275 STATIC int
2276 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2278 xfs_attr_leaf_name_local_t *name_loc;
2279 xfs_attr_leaf_name_remote_t *name_rmt;
2280 int size;
2282 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2283 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2284 name_loc = xfs_attr_leaf_name_local(leaf, index);
2285 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2286 be16_to_cpu(name_loc->valuelen));
2287 } else {
2288 name_rmt = xfs_attr_leaf_name_remote(leaf, index);
2289 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2291 return(size);
2295 * Calculate the number of bytes that would be required to store the new
2296 * attribute (whether local or remote only calculate bytes in this block).
2297 * This routine decides as a side effect whether the attribute will be
2298 * a "local" or a "remote" attribute.
2301 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2303 int size;
2305 size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2306 if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2307 if (local) {
2308 *local = 1;
2310 } else {
2311 size = xfs_attr_leaf_entsize_remote(namelen);
2312 if (local) {
2313 *local = 0;
2316 return(size);
2320 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2323 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2325 attrlist_cursor_kern_t *cursor;
2326 xfs_attr_leafblock_t *leaf;
2327 xfs_attr_leaf_entry_t *entry;
2328 int retval, i;
2330 ASSERT(bp != NULL);
2331 leaf = bp->data;
2332 cursor = context->cursor;
2333 cursor->initted = 1;
2335 trace_xfs_attr_list_leaf(context);
2338 * Re-find our place in the leaf block if this is a new syscall.
2340 if (context->resynch) {
2341 entry = &leaf->entries[0];
2342 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2343 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2344 if (cursor->offset == context->dupcnt) {
2345 context->dupcnt = 0;
2346 break;
2348 context->dupcnt++;
2349 } else if (be32_to_cpu(entry->hashval) >
2350 cursor->hashval) {
2351 context->dupcnt = 0;
2352 break;
2355 if (i == be16_to_cpu(leaf->hdr.count)) {
2356 trace_xfs_attr_list_notfound(context);
2357 return(0);
2359 } else {
2360 entry = &leaf->entries[0];
2361 i = 0;
2363 context->resynch = 0;
2366 * We have found our place, start copying out the new attributes.
2368 retval = 0;
2369 for ( ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2370 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2371 cursor->hashval = be32_to_cpu(entry->hashval);
2372 cursor->offset = 0;
2375 if (entry->flags & XFS_ATTR_INCOMPLETE)
2376 continue; /* skip incomplete entries */
2378 if (entry->flags & XFS_ATTR_LOCAL) {
2379 xfs_attr_leaf_name_local_t *name_loc =
2380 xfs_attr_leaf_name_local(leaf, i);
2382 retval = context->put_listent(context,
2383 entry->flags,
2384 name_loc->nameval,
2385 (int)name_loc->namelen,
2386 be16_to_cpu(name_loc->valuelen),
2387 &name_loc->nameval[name_loc->namelen]);
2388 if (retval)
2389 return retval;
2390 } else {
2391 xfs_attr_leaf_name_remote_t *name_rmt =
2392 xfs_attr_leaf_name_remote(leaf, i);
2394 int valuelen = be32_to_cpu(name_rmt->valuelen);
2396 if (context->put_value) {
2397 xfs_da_args_t args;
2399 memset((char *)&args, 0, sizeof(args));
2400 args.dp = context->dp;
2401 args.whichfork = XFS_ATTR_FORK;
2402 args.valuelen = valuelen;
2403 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2404 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2405 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2406 retval = xfs_attr_rmtval_get(&args);
2407 if (retval)
2408 return retval;
2409 retval = context->put_listent(context,
2410 entry->flags,
2411 name_rmt->name,
2412 (int)name_rmt->namelen,
2413 valuelen,
2414 args.value);
2415 kmem_free(args.value);
2416 } else {
2417 retval = context->put_listent(context,
2418 entry->flags,
2419 name_rmt->name,
2420 (int)name_rmt->namelen,
2421 valuelen,
2422 NULL);
2424 if (retval)
2425 return retval;
2427 if (context->seen_enough)
2428 break;
2429 cursor->offset++;
2431 trace_xfs_attr_list_leaf_end(context);
2432 return(retval);
2436 /*========================================================================
2437 * Manage the INCOMPLETE flag in a leaf entry
2438 *========================================================================*/
2441 * Clear the INCOMPLETE flag on an entry in a leaf block.
2444 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2446 xfs_attr_leafblock_t *leaf;
2447 xfs_attr_leaf_entry_t *entry;
2448 xfs_attr_leaf_name_remote_t *name_rmt;
2449 xfs_dabuf_t *bp;
2450 int error;
2451 #ifdef DEBUG
2452 xfs_attr_leaf_name_local_t *name_loc;
2453 int namelen;
2454 char *name;
2455 #endif /* DEBUG */
2458 * Set up the operation.
2460 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2461 XFS_ATTR_FORK);
2462 if (error) {
2463 return(error);
2465 ASSERT(bp != NULL);
2467 leaf = bp->data;
2468 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2469 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2470 ASSERT(args->index >= 0);
2471 entry = &leaf->entries[ args->index ];
2472 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2474 #ifdef DEBUG
2475 if (entry->flags & XFS_ATTR_LOCAL) {
2476 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2477 namelen = name_loc->namelen;
2478 name = (char *)name_loc->nameval;
2479 } else {
2480 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2481 namelen = name_rmt->namelen;
2482 name = (char *)name_rmt->name;
2484 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2485 ASSERT(namelen == args->namelen);
2486 ASSERT(memcmp(name, args->name, namelen) == 0);
2487 #endif /* DEBUG */
2489 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2490 xfs_da_log_buf(args->trans, bp,
2491 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2493 if (args->rmtblkno) {
2494 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2495 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2496 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2497 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2498 xfs_da_log_buf(args->trans, bp,
2499 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2501 xfs_da_buf_done(bp);
2504 * Commit the flag value change and start the next trans in series.
2506 return xfs_trans_roll(&args->trans, args->dp);
2510 * Set the INCOMPLETE flag on an entry in a leaf block.
2513 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2515 xfs_attr_leafblock_t *leaf;
2516 xfs_attr_leaf_entry_t *entry;
2517 xfs_attr_leaf_name_remote_t *name_rmt;
2518 xfs_dabuf_t *bp;
2519 int error;
2522 * Set up the operation.
2524 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2525 XFS_ATTR_FORK);
2526 if (error) {
2527 return(error);
2529 ASSERT(bp != NULL);
2531 leaf = bp->data;
2532 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2533 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2534 ASSERT(args->index >= 0);
2535 entry = &leaf->entries[ args->index ];
2537 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2538 entry->flags |= XFS_ATTR_INCOMPLETE;
2539 xfs_da_log_buf(args->trans, bp,
2540 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2541 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2542 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2543 name_rmt->valueblk = 0;
2544 name_rmt->valuelen = 0;
2545 xfs_da_log_buf(args->trans, bp,
2546 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2548 xfs_da_buf_done(bp);
2551 * Commit the flag value change and start the next trans in series.
2553 return xfs_trans_roll(&args->trans, args->dp);
2557 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2558 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2559 * entry given by args->blkno2/index2.
2561 * Note that they could be in different blocks, or in the same block.
2564 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2566 xfs_attr_leafblock_t *leaf1, *leaf2;
2567 xfs_attr_leaf_entry_t *entry1, *entry2;
2568 xfs_attr_leaf_name_remote_t *name_rmt;
2569 xfs_dabuf_t *bp1, *bp2;
2570 int error;
2571 #ifdef DEBUG
2572 xfs_attr_leaf_name_local_t *name_loc;
2573 int namelen1, namelen2;
2574 char *name1, *name2;
2575 #endif /* DEBUG */
2578 * Read the block containing the "old" attr
2580 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2581 XFS_ATTR_FORK);
2582 if (error) {
2583 return(error);
2585 ASSERT(bp1 != NULL);
2588 * Read the block containing the "new" attr, if it is different
2590 if (args->blkno2 != args->blkno) {
2591 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2592 -1, &bp2, XFS_ATTR_FORK);
2593 if (error) {
2594 return(error);
2596 ASSERT(bp2 != NULL);
2597 } else {
2598 bp2 = bp1;
2601 leaf1 = bp1->data;
2602 ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2603 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2604 ASSERT(args->index >= 0);
2605 entry1 = &leaf1->entries[ args->index ];
2607 leaf2 = bp2->data;
2608 ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2609 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2610 ASSERT(args->index2 >= 0);
2611 entry2 = &leaf2->entries[ args->index2 ];
2613 #ifdef DEBUG
2614 if (entry1->flags & XFS_ATTR_LOCAL) {
2615 name_loc = xfs_attr_leaf_name_local(leaf1, args->index);
2616 namelen1 = name_loc->namelen;
2617 name1 = (char *)name_loc->nameval;
2618 } else {
2619 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2620 namelen1 = name_rmt->namelen;
2621 name1 = (char *)name_rmt->name;
2623 if (entry2->flags & XFS_ATTR_LOCAL) {
2624 name_loc = xfs_attr_leaf_name_local(leaf2, args->index2);
2625 namelen2 = name_loc->namelen;
2626 name2 = (char *)name_loc->nameval;
2627 } else {
2628 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2629 namelen2 = name_rmt->namelen;
2630 name2 = (char *)name_rmt->name;
2632 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2633 ASSERT(namelen1 == namelen2);
2634 ASSERT(memcmp(name1, name2, namelen1) == 0);
2635 #endif /* DEBUG */
2637 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2638 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2640 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2641 xfs_da_log_buf(args->trans, bp1,
2642 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2643 if (args->rmtblkno) {
2644 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2645 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2646 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2647 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2648 xfs_da_log_buf(args->trans, bp1,
2649 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2652 entry2->flags |= XFS_ATTR_INCOMPLETE;
2653 xfs_da_log_buf(args->trans, bp2,
2654 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2655 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2656 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2657 name_rmt->valueblk = 0;
2658 name_rmt->valuelen = 0;
2659 xfs_da_log_buf(args->trans, bp2,
2660 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2662 xfs_da_buf_done(bp1);
2663 if (bp1 != bp2)
2664 xfs_da_buf_done(bp2);
2667 * Commit the flag value change and start the next trans in series.
2669 error = xfs_trans_roll(&args->trans, args->dp);
2671 return(error);
2674 /*========================================================================
2675 * Indiscriminately delete the entire attribute fork
2676 *========================================================================*/
2679 * Recurse (gasp!) through the attribute nodes until we find leaves.
2680 * We're doing a depth-first traversal in order to invalidate everything.
2683 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2685 xfs_da_blkinfo_t *info;
2686 xfs_daddr_t blkno;
2687 xfs_dabuf_t *bp;
2688 int error;
2691 * Read block 0 to see what we have to work with.
2692 * We only get here if we have extents, since we remove
2693 * the extents in reverse order the extent containing
2694 * block 0 must still be there.
2696 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2697 if (error)
2698 return(error);
2699 blkno = xfs_da_blkno(bp);
2702 * Invalidate the tree, even if the "tree" is only a single leaf block.
2703 * This is a depth-first traversal!
2705 info = bp->data;
2706 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2707 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2708 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2709 error = xfs_attr_leaf_inactive(trans, dp, bp);
2710 } else {
2711 error = XFS_ERROR(EIO);
2712 xfs_da_brelse(*trans, bp);
2714 if (error)
2715 return(error);
2718 * Invalidate the incore copy of the root block.
2720 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2721 if (error)
2722 return(error);
2723 xfs_da_binval(*trans, bp); /* remove from cache */
2725 * Commit the invalidate and start the next transaction.
2727 error = xfs_trans_roll(trans, dp);
2729 return (error);
2733 * Recurse (gasp!) through the attribute nodes until we find leaves.
2734 * We're doing a depth-first traversal in order to invalidate everything.
2736 STATIC int
2737 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2738 int level)
2740 xfs_da_blkinfo_t *info;
2741 xfs_da_intnode_t *node;
2742 xfs_dablk_t child_fsb;
2743 xfs_daddr_t parent_blkno, child_blkno;
2744 int error, count, i;
2745 xfs_dabuf_t *child_bp;
2748 * Since this code is recursive (gasp!) we must protect ourselves.
2750 if (level > XFS_DA_NODE_MAXDEPTH) {
2751 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2752 return(XFS_ERROR(EIO));
2755 node = bp->data;
2756 ASSERT(be16_to_cpu(node->hdr.info.magic) == XFS_DA_NODE_MAGIC);
2757 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2758 count = be16_to_cpu(node->hdr.count);
2759 if (!count) {
2760 xfs_da_brelse(*trans, bp);
2761 return(0);
2763 child_fsb = be32_to_cpu(node->btree[0].before);
2764 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2767 * If this is the node level just above the leaves, simply loop
2768 * over the leaves removing all of them. If this is higher up
2769 * in the tree, recurse downward.
2771 for (i = 0; i < count; i++) {
2773 * Read the subsidiary block to see what we have to work with.
2774 * Don't do this in a transaction. This is a depth-first
2775 * traversal of the tree so we may deal with many blocks
2776 * before we come back to this one.
2778 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2779 XFS_ATTR_FORK);
2780 if (error)
2781 return(error);
2782 if (child_bp) {
2783 /* save for re-read later */
2784 child_blkno = xfs_da_blkno(child_bp);
2787 * Invalidate the subtree, however we have to.
2789 info = child_bp->data;
2790 if (be16_to_cpu(info->magic) == XFS_DA_NODE_MAGIC) {
2791 error = xfs_attr_node_inactive(trans, dp,
2792 child_bp, level+1);
2793 } else if (be16_to_cpu(info->magic) == XFS_ATTR_LEAF_MAGIC) {
2794 error = xfs_attr_leaf_inactive(trans, dp,
2795 child_bp);
2796 } else {
2797 error = XFS_ERROR(EIO);
2798 xfs_da_brelse(*trans, child_bp);
2800 if (error)
2801 return(error);
2804 * Remove the subsidiary block from the cache
2805 * and from the log.
2807 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2808 &child_bp, XFS_ATTR_FORK);
2809 if (error)
2810 return(error);
2811 xfs_da_binval(*trans, child_bp);
2815 * If we're not done, re-read the parent to get the next
2816 * child block number.
2818 if ((i+1) < count) {
2819 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2820 &bp, XFS_ATTR_FORK);
2821 if (error)
2822 return(error);
2823 child_fsb = be32_to_cpu(node->btree[i+1].before);
2824 xfs_da_brelse(*trans, bp);
2827 * Atomically commit the whole invalidate stuff.
2829 error = xfs_trans_roll(trans, dp);
2830 if (error)
2831 return (error);
2834 return(0);
2838 * Invalidate all of the "remote" value regions pointed to by a particular
2839 * leaf block.
2840 * Note that we must release the lock on the buffer so that we are not
2841 * caught holding something that the logging code wants to flush to disk.
2843 STATIC int
2844 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2846 xfs_attr_leafblock_t *leaf;
2847 xfs_attr_leaf_entry_t *entry;
2848 xfs_attr_leaf_name_remote_t *name_rmt;
2849 xfs_attr_inactive_list_t *list, *lp;
2850 int error, count, size, tmp, i;
2852 leaf = bp->data;
2853 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_ATTR_LEAF_MAGIC);
2856 * Count the number of "remote" value extents.
2858 count = 0;
2859 entry = &leaf->entries[0];
2860 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2861 if (be16_to_cpu(entry->nameidx) &&
2862 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2863 name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2864 if (name_rmt->valueblk)
2865 count++;
2870 * If there are no "remote" values, we're done.
2872 if (count == 0) {
2873 xfs_da_brelse(*trans, bp);
2874 return(0);
2878 * Allocate storage for a list of all the "remote" value extents.
2880 size = count * sizeof(xfs_attr_inactive_list_t);
2881 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2884 * Identify each of the "remote" value extents.
2886 lp = list;
2887 entry = &leaf->entries[0];
2888 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2889 if (be16_to_cpu(entry->nameidx) &&
2890 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2891 name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2892 if (name_rmt->valueblk) {
2893 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2894 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2895 be32_to_cpu(name_rmt->valuelen));
2896 lp++;
2900 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2903 * Invalidate each of the "remote" value extents.
2905 error = 0;
2906 for (lp = list, i = 0; i < count; i++, lp++) {
2907 tmp = xfs_attr_leaf_freextent(trans, dp,
2908 lp->valueblk, lp->valuelen);
2910 if (error == 0)
2911 error = tmp; /* save only the 1st errno */
2914 kmem_free((xfs_caddr_t)list);
2915 return(error);
2919 * Look at all the extents for this logical region,
2920 * invalidate any buffers that are incore/in transactions.
2922 STATIC int
2923 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2924 xfs_dablk_t blkno, int blkcnt)
2926 xfs_bmbt_irec_t map;
2927 xfs_dablk_t tblkno;
2928 int tblkcnt, dblkcnt, nmap, error;
2929 xfs_daddr_t dblkno;
2930 xfs_buf_t *bp;
2933 * Roll through the "value", invalidating the attribute value's
2934 * blocks.
2936 tblkno = blkno;
2937 tblkcnt = blkcnt;
2938 while (tblkcnt > 0) {
2940 * Try to remember where we decided to put the value.
2942 nmap = 1;
2943 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
2944 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2945 NULL, 0, &map, &nmap, NULL);
2946 if (error) {
2947 return(error);
2949 ASSERT(nmap == 1);
2950 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2953 * If it's a hole, these are already unmapped
2954 * so there's nothing to invalidate.
2956 if (map.br_startblock != HOLESTARTBLOCK) {
2958 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
2959 map.br_startblock);
2960 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
2961 map.br_blockcount);
2962 bp = xfs_trans_get_buf(*trans,
2963 dp->i_mount->m_ddev_targp,
2964 dblkno, dblkcnt, XBF_LOCK);
2965 xfs_trans_binval(*trans, bp);
2967 * Roll to next transaction.
2969 error = xfs_trans_roll(trans, dp);
2970 if (error)
2971 return (error);
2974 tblkno += map.br_blockcount;
2975 tblkcnt -= map.br_blockcount;
2978 return(0);