Merge remote-tracking branch 'moduleh/module.h-split'
[linux-2.6/next.git] / fs / xfs / xfs_attr_leaf.c
blob8fad9602542bb4cf0fe5550ea1b970a1b7c0a3f9
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.
113 * Returns zero if not, else the di_forkoff fork offset to be used in the
114 * literal area for attribute data once the new bytes have been added.
116 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
117 * special case for dev/uuid inodes, they have fixed size data forks.
120 xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
122 int offset;
123 int minforkoff; /* lower limit on valid forkoff locations */
124 int maxforkoff; /* upper limit on valid forkoff locations */
125 int dsize;
126 xfs_mount_t *mp = dp->i_mount;
128 offset = (XFS_LITINO(mp) - bytes) >> 3; /* rounded down */
130 switch (dp->i_d.di_format) {
131 case XFS_DINODE_FMT_DEV:
132 minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
133 return (offset >= minforkoff) ? minforkoff : 0;
134 case XFS_DINODE_FMT_UUID:
135 minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
136 return (offset >= minforkoff) ? minforkoff : 0;
139 if (!(mp->m_flags & XFS_MOUNT_ATTR2)) {
140 if (bytes <= XFS_IFORK_ASIZE(dp))
141 return dp->i_d.di_forkoff;
142 return 0;
145 dsize = dp->i_df.if_bytes;
147 switch (dp->i_d.di_format) {
148 case XFS_DINODE_FMT_EXTENTS:
150 * If there is no attr fork and the data fork is extents,
151 * determine if creating the default attr fork will result
152 * in the extents form migrating to btree. If so, the
153 * minimum offset only needs to be the space required for
154 * the btree root.
156 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
157 xfs_default_attroffset(dp))
158 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
159 break;
161 case XFS_DINODE_FMT_BTREE:
163 * If have data btree then keep forkoff if we have one,
164 * otherwise we are adding a new attr, so then we set
165 * minforkoff to where the btree root can finish so we have
166 * plenty of room for attrs
168 if (dp->i_d.di_forkoff) {
169 if (offset < dp->i_d.di_forkoff)
170 return 0;
171 else
172 return dp->i_d.di_forkoff;
173 } else
174 dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
175 break;
179 * A data fork btree root must have space for at least
180 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
182 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
183 minforkoff = roundup(minforkoff, 8) >> 3;
185 /* attr fork btree root can have at least this many key/ptr pairs */
186 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
187 maxforkoff = maxforkoff >> 3; /* rounded down */
189 if (offset >= minforkoff && offset < maxforkoff)
190 return offset;
191 if (offset >= maxforkoff)
192 return maxforkoff;
193 return 0;
197 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
199 STATIC void
200 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
202 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
203 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
204 spin_lock(&mp->m_sb_lock);
205 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
206 xfs_sb_version_addattr2(&mp->m_sb);
207 spin_unlock(&mp->m_sb_lock);
208 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
209 } else
210 spin_unlock(&mp->m_sb_lock);
215 * Create the initial contents of a shortform attribute list.
217 void
218 xfs_attr_shortform_create(xfs_da_args_t *args)
220 xfs_attr_sf_hdr_t *hdr;
221 xfs_inode_t *dp;
222 xfs_ifork_t *ifp;
224 dp = args->dp;
225 ASSERT(dp != NULL);
226 ifp = dp->i_afp;
227 ASSERT(ifp != NULL);
228 ASSERT(ifp->if_bytes == 0);
229 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
230 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
231 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
232 ifp->if_flags |= XFS_IFINLINE;
233 } else {
234 ASSERT(ifp->if_flags & XFS_IFINLINE);
236 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
237 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
238 hdr->count = 0;
239 hdr->totsize = cpu_to_be16(sizeof(*hdr));
240 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
244 * Add a name/value pair to the shortform attribute list.
245 * Overflow from the inode has already been checked for.
247 void
248 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
250 xfs_attr_shortform_t *sf;
251 xfs_attr_sf_entry_t *sfe;
252 int i, offset, size;
253 xfs_mount_t *mp;
254 xfs_inode_t *dp;
255 xfs_ifork_t *ifp;
257 dp = args->dp;
258 mp = dp->i_mount;
259 dp->i_d.di_forkoff = forkoff;
260 dp->i_df.if_ext_max =
261 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
262 dp->i_afp->if_ext_max =
263 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
265 ifp = dp->i_afp;
266 ASSERT(ifp->if_flags & XFS_IFINLINE);
267 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
268 sfe = &sf->list[0];
269 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
270 #ifdef DEBUG
271 if (sfe->namelen != args->namelen)
272 continue;
273 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
274 continue;
275 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
276 continue;
277 ASSERT(0);
278 #endif
281 offset = (char *)sfe - (char *)sf;
282 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
283 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
284 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
285 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
287 sfe->namelen = args->namelen;
288 sfe->valuelen = args->valuelen;
289 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
290 memcpy(sfe->nameval, args->name, args->namelen);
291 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
292 sf->hdr.count++;
293 be16_add_cpu(&sf->hdr.totsize, size);
294 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
296 xfs_sbversion_add_attr2(mp, args->trans);
300 * After the last attribute is removed revert to original inode format,
301 * making all literal area available to the data fork once more.
303 STATIC void
304 xfs_attr_fork_reset(
305 struct xfs_inode *ip,
306 struct xfs_trans *tp)
308 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
309 ip->i_d.di_forkoff = 0;
310 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
312 ASSERT(ip->i_d.di_anextents == 0);
313 ASSERT(ip->i_afp == NULL);
315 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
316 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
320 * Remove an attribute from the shortform attribute list structure.
323 xfs_attr_shortform_remove(xfs_da_args_t *args)
325 xfs_attr_shortform_t *sf;
326 xfs_attr_sf_entry_t *sfe;
327 int base, size=0, end, totsize, i;
328 xfs_mount_t *mp;
329 xfs_inode_t *dp;
331 dp = args->dp;
332 mp = dp->i_mount;
333 base = sizeof(xfs_attr_sf_hdr_t);
334 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
335 sfe = &sf->list[0];
336 end = sf->hdr.count;
337 for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
338 base += size, i++) {
339 size = XFS_ATTR_SF_ENTSIZE(sfe);
340 if (sfe->namelen != args->namelen)
341 continue;
342 if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
343 continue;
344 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
345 continue;
346 break;
348 if (i == end)
349 return(XFS_ERROR(ENOATTR));
352 * Fix up the attribute fork data, covering the hole
354 end = base + size;
355 totsize = be16_to_cpu(sf->hdr.totsize);
356 if (end != totsize)
357 memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
358 sf->hdr.count--;
359 be16_add_cpu(&sf->hdr.totsize, -size);
362 * Fix up the start offset of the attribute fork
364 totsize -= size;
365 if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
366 (mp->m_flags & XFS_MOUNT_ATTR2) &&
367 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
368 !(args->op_flags & XFS_DA_OP_ADDNAME)) {
369 xfs_attr_fork_reset(dp, args->trans);
370 } else {
371 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
372 dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
373 ASSERT(dp->i_d.di_forkoff);
374 ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
375 (args->op_flags & XFS_DA_OP_ADDNAME) ||
376 !(mp->m_flags & XFS_MOUNT_ATTR2) ||
377 dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
378 dp->i_afp->if_ext_max =
379 XFS_IFORK_ASIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
380 dp->i_df.if_ext_max =
381 XFS_IFORK_DSIZE(dp) / (uint)sizeof(xfs_bmbt_rec_t);
382 xfs_trans_log_inode(args->trans, dp,
383 XFS_ILOG_CORE | XFS_ILOG_ADATA);
386 xfs_sbversion_add_attr2(mp, args->trans);
388 return(0);
392 * Look up a name in a shortform attribute list structure.
394 /*ARGSUSED*/
396 xfs_attr_shortform_lookup(xfs_da_args_t *args)
398 xfs_attr_shortform_t *sf;
399 xfs_attr_sf_entry_t *sfe;
400 int i;
401 xfs_ifork_t *ifp;
403 ifp = args->dp->i_afp;
404 ASSERT(ifp->if_flags & XFS_IFINLINE);
405 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
406 sfe = &sf->list[0];
407 for (i = 0; i < sf->hdr.count;
408 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
409 if (sfe->namelen != args->namelen)
410 continue;
411 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
412 continue;
413 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
414 continue;
415 return(XFS_ERROR(EEXIST));
417 return(XFS_ERROR(ENOATTR));
421 * Look up a name in a shortform attribute list structure.
423 /*ARGSUSED*/
425 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
427 xfs_attr_shortform_t *sf;
428 xfs_attr_sf_entry_t *sfe;
429 int i;
431 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
432 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
433 sfe = &sf->list[0];
434 for (i = 0; i < sf->hdr.count;
435 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
436 if (sfe->namelen != args->namelen)
437 continue;
438 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
439 continue;
440 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
441 continue;
442 if (args->flags & ATTR_KERNOVAL) {
443 args->valuelen = sfe->valuelen;
444 return(XFS_ERROR(EEXIST));
446 if (args->valuelen < sfe->valuelen) {
447 args->valuelen = sfe->valuelen;
448 return(XFS_ERROR(ERANGE));
450 args->valuelen = sfe->valuelen;
451 memcpy(args->value, &sfe->nameval[args->namelen],
452 args->valuelen);
453 return(XFS_ERROR(EEXIST));
455 return(XFS_ERROR(ENOATTR));
459 * Convert from using the shortform to the leaf.
462 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
464 xfs_inode_t *dp;
465 xfs_attr_shortform_t *sf;
466 xfs_attr_sf_entry_t *sfe;
467 xfs_da_args_t nargs;
468 char *tmpbuffer;
469 int error, i, size;
470 xfs_dablk_t blkno;
471 xfs_dabuf_t *bp;
472 xfs_ifork_t *ifp;
474 dp = args->dp;
475 ifp = dp->i_afp;
476 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
477 size = be16_to_cpu(sf->hdr.totsize);
478 tmpbuffer = kmem_alloc(size, KM_SLEEP);
479 ASSERT(tmpbuffer != NULL);
480 memcpy(tmpbuffer, ifp->if_u1.if_data, size);
481 sf = (xfs_attr_shortform_t *)tmpbuffer;
483 xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
484 bp = NULL;
485 error = xfs_da_grow_inode(args, &blkno);
486 if (error) {
488 * If we hit an IO error middle of the transaction inside
489 * grow_inode(), we may have inconsistent data. Bail out.
491 if (error == EIO)
492 goto out;
493 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
494 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
495 goto out;
498 ASSERT(blkno == 0);
499 error = xfs_attr_leaf_create(args, blkno, &bp);
500 if (error) {
501 error = xfs_da_shrink_inode(args, 0, bp);
502 bp = NULL;
503 if (error)
504 goto out;
505 xfs_idata_realloc(dp, size, XFS_ATTR_FORK); /* try to put */
506 memcpy(ifp->if_u1.if_data, tmpbuffer, size); /* it back */
507 goto out;
510 memset((char *)&nargs, 0, sizeof(nargs));
511 nargs.dp = dp;
512 nargs.firstblock = args->firstblock;
513 nargs.flist = args->flist;
514 nargs.total = args->total;
515 nargs.whichfork = XFS_ATTR_FORK;
516 nargs.trans = args->trans;
517 nargs.op_flags = XFS_DA_OP_OKNOENT;
519 sfe = &sf->list[0];
520 for (i = 0; i < sf->hdr.count; i++) {
521 nargs.name = sfe->nameval;
522 nargs.namelen = sfe->namelen;
523 nargs.value = &sfe->nameval[nargs.namelen];
524 nargs.valuelen = sfe->valuelen;
525 nargs.hashval = xfs_da_hashname(sfe->nameval,
526 sfe->namelen);
527 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
528 error = xfs_attr_leaf_lookup_int(bp, &nargs); /* set a->index */
529 ASSERT(error == ENOATTR);
530 error = xfs_attr_leaf_add(bp, &nargs);
531 ASSERT(error != ENOSPC);
532 if (error)
533 goto out;
534 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
536 error = 0;
538 out:
539 if(bp)
540 xfs_da_buf_done(bp);
541 kmem_free(tmpbuffer);
542 return(error);
545 STATIC int
546 xfs_attr_shortform_compare(const void *a, const void *b)
548 xfs_attr_sf_sort_t *sa, *sb;
550 sa = (xfs_attr_sf_sort_t *)a;
551 sb = (xfs_attr_sf_sort_t *)b;
552 if (sa->hash < sb->hash) {
553 return(-1);
554 } else if (sa->hash > sb->hash) {
555 return(1);
556 } else {
557 return(sa->entno - sb->entno);
562 #define XFS_ISRESET_CURSOR(cursor) \
563 (!((cursor)->initted) && !((cursor)->hashval) && \
564 !((cursor)->blkno) && !((cursor)->offset))
566 * Copy out entries of shortform attribute lists for attr_list().
567 * Shortform attribute lists are not stored in hashval sorted order.
568 * If the output buffer is not large enough to hold them all, then we
569 * we have to calculate each entries' hashvalue and sort them before
570 * we can begin returning them to the user.
572 /*ARGSUSED*/
574 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
576 attrlist_cursor_kern_t *cursor;
577 xfs_attr_sf_sort_t *sbuf, *sbp;
578 xfs_attr_shortform_t *sf;
579 xfs_attr_sf_entry_t *sfe;
580 xfs_inode_t *dp;
581 int sbsize, nsbuf, count, i;
582 int error;
584 ASSERT(context != NULL);
585 dp = context->dp;
586 ASSERT(dp != NULL);
587 ASSERT(dp->i_afp != NULL);
588 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
589 ASSERT(sf != NULL);
590 if (!sf->hdr.count)
591 return(0);
592 cursor = context->cursor;
593 ASSERT(cursor != NULL);
595 trace_xfs_attr_list_sf(context);
598 * If the buffer is large enough and the cursor is at the start,
599 * do not bother with sorting since we will return everything in
600 * one buffer and another call using the cursor won't need to be
601 * made.
602 * Note the generous fudge factor of 16 overhead bytes per entry.
603 * If bufsize is zero then put_listent must be a search function
604 * and can just scan through what we have.
606 if (context->bufsize == 0 ||
607 (XFS_ISRESET_CURSOR(cursor) &&
608 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
609 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
610 error = context->put_listent(context,
611 sfe->flags,
612 sfe->nameval,
613 (int)sfe->namelen,
614 (int)sfe->valuelen,
615 &sfe->nameval[sfe->namelen]);
618 * Either search callback finished early or
619 * didn't fit it all in the buffer after all.
621 if (context->seen_enough)
622 break;
624 if (error)
625 return error;
626 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
628 trace_xfs_attr_list_sf_all(context);
629 return(0);
632 /* do no more for a search callback */
633 if (context->bufsize == 0)
634 return 0;
637 * It didn't all fit, so we have to sort everything on hashval.
639 sbsize = sf->hdr.count * sizeof(*sbuf);
640 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
643 * Scan the attribute list for the rest of the entries, storing
644 * the relevant info from only those that match into a buffer.
646 nsbuf = 0;
647 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
648 if (unlikely(
649 ((char *)sfe < (char *)sf) ||
650 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
651 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
652 XFS_ERRLEVEL_LOW,
653 context->dp->i_mount, sfe);
654 kmem_free(sbuf);
655 return XFS_ERROR(EFSCORRUPTED);
658 sbp->entno = i;
659 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
660 sbp->name = sfe->nameval;
661 sbp->namelen = sfe->namelen;
662 /* These are bytes, and both on-disk, don't endian-flip */
663 sbp->valuelen = sfe->valuelen;
664 sbp->flags = sfe->flags;
665 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
666 sbp++;
667 nsbuf++;
671 * Sort the entries on hash then entno.
673 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
676 * Re-find our place IN THE SORTED LIST.
678 count = 0;
679 cursor->initted = 1;
680 cursor->blkno = 0;
681 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
682 if (sbp->hash == cursor->hashval) {
683 if (cursor->offset == count) {
684 break;
686 count++;
687 } else if (sbp->hash > cursor->hashval) {
688 break;
691 if (i == nsbuf) {
692 kmem_free(sbuf);
693 return(0);
697 * Loop putting entries into the user buffer.
699 for ( ; i < nsbuf; i++, sbp++) {
700 if (cursor->hashval != sbp->hash) {
701 cursor->hashval = sbp->hash;
702 cursor->offset = 0;
704 error = context->put_listent(context,
705 sbp->flags,
706 sbp->name,
707 sbp->namelen,
708 sbp->valuelen,
709 &sbp->name[sbp->namelen]);
710 if (error)
711 return error;
712 if (context->seen_enough)
713 break;
714 cursor->offset++;
717 kmem_free(sbuf);
718 return(0);
722 * Check a leaf attribute block to see if all the entries would fit into
723 * a shortform attribute list.
726 xfs_attr_shortform_allfit(xfs_dabuf_t *bp, xfs_inode_t *dp)
728 xfs_attr_leafblock_t *leaf;
729 xfs_attr_leaf_entry_t *entry;
730 xfs_attr_leaf_name_local_t *name_loc;
731 int bytes, i;
733 leaf = bp->data;
734 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
736 entry = &leaf->entries[0];
737 bytes = sizeof(struct xfs_attr_sf_hdr);
738 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
739 if (entry->flags & XFS_ATTR_INCOMPLETE)
740 continue; /* don't copy partial entries */
741 if (!(entry->flags & XFS_ATTR_LOCAL))
742 return(0);
743 name_loc = xfs_attr_leaf_name_local(leaf, i);
744 if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
745 return(0);
746 if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
747 return(0);
748 bytes += sizeof(struct xfs_attr_sf_entry)-1
749 + name_loc->namelen
750 + be16_to_cpu(name_loc->valuelen);
752 if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
753 (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
754 (bytes == sizeof(struct xfs_attr_sf_hdr)))
755 return(-1);
756 return(xfs_attr_shortform_bytesfit(dp, bytes));
760 * Convert a leaf attribute list to shortform attribute list
763 xfs_attr_leaf_to_shortform(xfs_dabuf_t *bp, xfs_da_args_t *args, int forkoff)
765 xfs_attr_leafblock_t *leaf;
766 xfs_attr_leaf_entry_t *entry;
767 xfs_attr_leaf_name_local_t *name_loc;
768 xfs_da_args_t nargs;
769 xfs_inode_t *dp;
770 char *tmpbuffer;
771 int error, i;
773 dp = args->dp;
774 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
775 ASSERT(tmpbuffer != NULL);
777 ASSERT(bp != NULL);
778 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
779 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
780 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
781 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
784 * Clean out the prior contents of the attribute list.
786 error = xfs_da_shrink_inode(args, 0, bp);
787 if (error)
788 goto out;
790 if (forkoff == -1) {
791 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
792 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
793 xfs_attr_fork_reset(dp, args->trans);
794 goto out;
797 xfs_attr_shortform_create(args);
800 * Copy the attributes
802 memset((char *)&nargs, 0, sizeof(nargs));
803 nargs.dp = dp;
804 nargs.firstblock = args->firstblock;
805 nargs.flist = args->flist;
806 nargs.total = args->total;
807 nargs.whichfork = XFS_ATTR_FORK;
808 nargs.trans = args->trans;
809 nargs.op_flags = XFS_DA_OP_OKNOENT;
810 entry = &leaf->entries[0];
811 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
812 if (entry->flags & XFS_ATTR_INCOMPLETE)
813 continue; /* don't copy partial entries */
814 if (!entry->nameidx)
815 continue;
816 ASSERT(entry->flags & XFS_ATTR_LOCAL);
817 name_loc = xfs_attr_leaf_name_local(leaf, i);
818 nargs.name = name_loc->nameval;
819 nargs.namelen = name_loc->namelen;
820 nargs.value = &name_loc->nameval[nargs.namelen];
821 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
822 nargs.hashval = be32_to_cpu(entry->hashval);
823 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
824 xfs_attr_shortform_add(&nargs, forkoff);
826 error = 0;
828 out:
829 kmem_free(tmpbuffer);
830 return(error);
834 * Convert from using a single leaf to a root node and a leaf.
837 xfs_attr_leaf_to_node(xfs_da_args_t *args)
839 xfs_attr_leafblock_t *leaf;
840 xfs_da_intnode_t *node;
841 xfs_inode_t *dp;
842 xfs_dabuf_t *bp1, *bp2;
843 xfs_dablk_t blkno;
844 int error;
846 dp = args->dp;
847 bp1 = bp2 = NULL;
848 error = xfs_da_grow_inode(args, &blkno);
849 if (error)
850 goto out;
851 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
852 XFS_ATTR_FORK);
853 if (error)
854 goto out;
855 ASSERT(bp1 != NULL);
856 bp2 = NULL;
857 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
858 XFS_ATTR_FORK);
859 if (error)
860 goto out;
861 ASSERT(bp2 != NULL);
862 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
863 xfs_da_buf_done(bp1);
864 bp1 = NULL;
865 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
868 * Set up the new root node.
870 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
871 if (error)
872 goto out;
873 node = bp1->data;
874 leaf = bp2->data;
875 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
876 /* both on-disk, don't endian-flip twice */
877 node->btree[0].hashval =
878 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
879 node->btree[0].before = cpu_to_be32(blkno);
880 node->hdr.count = cpu_to_be16(1);
881 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
882 error = 0;
883 out:
884 if (bp1)
885 xfs_da_buf_done(bp1);
886 if (bp2)
887 xfs_da_buf_done(bp2);
888 return(error);
892 /*========================================================================
893 * Routines used for growing the Btree.
894 *========================================================================*/
897 * Create the initial contents of a leaf attribute list
898 * or a leaf in a node attribute list.
900 STATIC int
901 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
903 xfs_attr_leafblock_t *leaf;
904 xfs_attr_leaf_hdr_t *hdr;
905 xfs_inode_t *dp;
906 xfs_dabuf_t *bp;
907 int error;
909 dp = args->dp;
910 ASSERT(dp != NULL);
911 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
912 XFS_ATTR_FORK);
913 if (error)
914 return(error);
915 ASSERT(bp != NULL);
916 leaf = bp->data;
917 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
918 hdr = &leaf->hdr;
919 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
920 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
921 if (!hdr->firstused) {
922 hdr->firstused = cpu_to_be16(
923 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
926 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
927 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
928 sizeof(xfs_attr_leaf_hdr_t));
930 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
932 *bpp = bp;
933 return(0);
937 * Split the leaf node, rebalance, then add the new entry.
940 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
941 xfs_da_state_blk_t *newblk)
943 xfs_dablk_t blkno;
944 int error;
947 * Allocate space for a new leaf node.
949 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
950 error = xfs_da_grow_inode(state->args, &blkno);
951 if (error)
952 return(error);
953 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
954 if (error)
955 return(error);
956 newblk->blkno = blkno;
957 newblk->magic = XFS_ATTR_LEAF_MAGIC;
960 * Rebalance the entries across the two leaves.
961 * NOTE: rebalance() currently depends on the 2nd block being empty.
963 xfs_attr_leaf_rebalance(state, oldblk, newblk);
964 error = xfs_da_blk_link(state, oldblk, newblk);
965 if (error)
966 return(error);
969 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
970 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
971 * "new" attrs info. Will need the "old" info to remove it later.
973 * Insert the "new" entry in the correct block.
975 if (state->inleaf)
976 error = xfs_attr_leaf_add(oldblk->bp, state->args);
977 else
978 error = xfs_attr_leaf_add(newblk->bp, state->args);
981 * Update last hashval in each block since we added the name.
983 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
984 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
985 return(error);
989 * Add a name to the leaf attribute list structure.
992 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
994 xfs_attr_leafblock_t *leaf;
995 xfs_attr_leaf_hdr_t *hdr;
996 xfs_attr_leaf_map_t *map;
997 int tablesize, entsize, sum, tmp, i;
999 leaf = bp->data;
1000 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1001 ASSERT((args->index >= 0)
1002 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1003 hdr = &leaf->hdr;
1004 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1005 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1008 * Search through freemap for first-fit on new name length.
1009 * (may need to figure in size of entry struct too)
1011 tablesize = (be16_to_cpu(hdr->count) + 1)
1012 * sizeof(xfs_attr_leaf_entry_t)
1013 + sizeof(xfs_attr_leaf_hdr_t);
1014 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1015 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1016 if (tablesize > be16_to_cpu(hdr->firstused)) {
1017 sum += be16_to_cpu(map->size);
1018 continue;
1020 if (!map->size)
1021 continue; /* no space in this map */
1022 tmp = entsize;
1023 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1024 tmp += sizeof(xfs_attr_leaf_entry_t);
1025 if (be16_to_cpu(map->size) >= tmp) {
1026 tmp = xfs_attr_leaf_add_work(bp, args, i);
1027 return(tmp);
1029 sum += be16_to_cpu(map->size);
1033 * If there are no holes in the address space of the block,
1034 * and we don't have enough freespace, then compaction will do us
1035 * no good and we should just give up.
1037 if (!hdr->holes && (sum < entsize))
1038 return(XFS_ERROR(ENOSPC));
1041 * Compact the entries to coalesce free space.
1042 * This may change the hdr->count via dropping INCOMPLETE entries.
1044 xfs_attr_leaf_compact(args->trans, bp);
1047 * After compaction, the block is guaranteed to have only one
1048 * free region, in freemap[0]. If it is not big enough, give up.
1050 if (be16_to_cpu(hdr->freemap[0].size)
1051 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1052 return(XFS_ERROR(ENOSPC));
1054 return(xfs_attr_leaf_add_work(bp, args, 0));
1058 * Add a name to a leaf attribute list structure.
1060 STATIC int
1061 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1063 xfs_attr_leafblock_t *leaf;
1064 xfs_attr_leaf_hdr_t *hdr;
1065 xfs_attr_leaf_entry_t *entry;
1066 xfs_attr_leaf_name_local_t *name_loc;
1067 xfs_attr_leaf_name_remote_t *name_rmt;
1068 xfs_attr_leaf_map_t *map;
1069 xfs_mount_t *mp;
1070 int tmp, i;
1072 leaf = bp->data;
1073 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1074 hdr = &leaf->hdr;
1075 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1076 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1079 * Force open some space in the entry array and fill it in.
1081 entry = &leaf->entries[args->index];
1082 if (args->index < be16_to_cpu(hdr->count)) {
1083 tmp = be16_to_cpu(hdr->count) - args->index;
1084 tmp *= sizeof(xfs_attr_leaf_entry_t);
1085 memmove((char *)(entry+1), (char *)entry, tmp);
1086 xfs_da_log_buf(args->trans, bp,
1087 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1089 be16_add_cpu(&hdr->count, 1);
1092 * Allocate space for the new string (at the end of the run).
1094 map = &hdr->freemap[mapindex];
1095 mp = args->trans->t_mountp;
1096 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1097 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1098 ASSERT(be16_to_cpu(map->size) >=
1099 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1100 mp->m_sb.sb_blocksize, NULL));
1101 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1102 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1103 be16_add_cpu(&map->size,
1104 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1105 mp->m_sb.sb_blocksize, &tmp));
1106 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1107 be16_to_cpu(map->size));
1108 entry->hashval = cpu_to_be32(args->hashval);
1109 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1110 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1111 if (args->op_flags & XFS_DA_OP_RENAME) {
1112 entry->flags |= XFS_ATTR_INCOMPLETE;
1113 if ((args->blkno2 == args->blkno) &&
1114 (args->index2 <= args->index)) {
1115 args->index2++;
1118 xfs_da_log_buf(args->trans, bp,
1119 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1120 ASSERT((args->index == 0) ||
1121 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1122 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1123 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1126 * Copy the attribute name and value into the new space.
1128 * For "remote" attribute values, simply note that we need to
1129 * allocate space for the "remote" value. We can't actually
1130 * allocate the extents in this transaction, and we can't decide
1131 * which blocks they should be as we might allocate more blocks
1132 * as part of this transaction (a split operation for example).
1134 if (entry->flags & XFS_ATTR_LOCAL) {
1135 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
1136 name_loc->namelen = args->namelen;
1137 name_loc->valuelen = cpu_to_be16(args->valuelen);
1138 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1139 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1140 be16_to_cpu(name_loc->valuelen));
1141 } else {
1142 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
1143 name_rmt->namelen = args->namelen;
1144 memcpy((char *)name_rmt->name, args->name, args->namelen);
1145 entry->flags |= XFS_ATTR_INCOMPLETE;
1146 /* just in case */
1147 name_rmt->valuelen = 0;
1148 name_rmt->valueblk = 0;
1149 args->rmtblkno = 1;
1150 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1152 xfs_da_log_buf(args->trans, bp,
1153 XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1154 xfs_attr_leaf_entsize(leaf, args->index)));
1157 * Update the control info for this leaf node
1159 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1160 /* both on-disk, don't endian-flip twice */
1161 hdr->firstused = entry->nameidx;
1163 ASSERT(be16_to_cpu(hdr->firstused) >=
1164 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1165 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1166 + sizeof(xfs_attr_leaf_hdr_t);
1167 map = &hdr->freemap[0];
1168 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1169 if (be16_to_cpu(map->base) == tmp) {
1170 be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1171 be16_add_cpu(&map->size,
1172 -((int)sizeof(xfs_attr_leaf_entry_t)));
1175 be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1176 xfs_da_log_buf(args->trans, bp,
1177 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1178 return(0);
1182 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1184 STATIC void
1185 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1187 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1188 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1189 xfs_mount_t *mp;
1190 char *tmpbuffer;
1192 mp = trans->t_mountp;
1193 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1194 ASSERT(tmpbuffer != NULL);
1195 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1196 memset(bp->data, 0, XFS_LBSIZE(mp));
1199 * Copy basic information
1201 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1202 leaf_d = bp->data;
1203 hdr_s = &leaf_s->hdr;
1204 hdr_d = &leaf_d->hdr;
1205 hdr_d->info = hdr_s->info; /* struct copy */
1206 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1207 /* handle truncation gracefully */
1208 if (!hdr_d->firstused) {
1209 hdr_d->firstused = cpu_to_be16(
1210 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1212 hdr_d->usedbytes = 0;
1213 hdr_d->count = 0;
1214 hdr_d->holes = 0;
1215 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1216 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1217 sizeof(xfs_attr_leaf_hdr_t));
1220 * Copy all entry's in the same (sorted) order,
1221 * but allocate name/value pairs packed and in sequence.
1223 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1224 be16_to_cpu(hdr_s->count), mp);
1225 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1227 kmem_free(tmpbuffer);
1231 * Redistribute the attribute list entries between two leaf nodes,
1232 * taking into account the size of the new entry.
1234 * NOTE: if new block is empty, then it will get the upper half of the
1235 * old block. At present, all (one) callers pass in an empty second block.
1237 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1238 * to match what it is doing in splitting the attribute leaf block. Those
1239 * values are used in "atomic rename" operations on attributes. Note that
1240 * the "new" and "old" values can end up in different blocks.
1242 STATIC void
1243 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1244 xfs_da_state_blk_t *blk2)
1246 xfs_da_args_t *args;
1247 xfs_da_state_blk_t *tmp_blk;
1248 xfs_attr_leafblock_t *leaf1, *leaf2;
1249 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1250 int count, totallen, max, space, swap;
1253 * Set up environment.
1255 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1256 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1257 leaf1 = blk1->bp->data;
1258 leaf2 = blk2->bp->data;
1259 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1260 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1261 args = state->args;
1264 * Check ordering of blocks, reverse if it makes things simpler.
1266 * NOTE: Given that all (current) callers pass in an empty
1267 * second block, this code should never set "swap".
1269 swap = 0;
1270 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1271 tmp_blk = blk1;
1272 blk1 = blk2;
1273 blk2 = tmp_blk;
1274 leaf1 = blk1->bp->data;
1275 leaf2 = blk2->bp->data;
1276 swap = 1;
1278 hdr1 = &leaf1->hdr;
1279 hdr2 = &leaf2->hdr;
1282 * Examine entries until we reduce the absolute difference in
1283 * byte usage between the two blocks to a minimum. Then get
1284 * the direction to copy and the number of elements to move.
1286 * "inleaf" is true if the new entry should be inserted into blk1.
1287 * If "swap" is also true, then reverse the sense of "inleaf".
1289 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1290 &count, &totallen);
1291 if (swap)
1292 state->inleaf = !state->inleaf;
1295 * Move any entries required from leaf to leaf:
1297 if (count < be16_to_cpu(hdr1->count)) {
1299 * Figure the total bytes to be added to the destination leaf.
1301 /* number entries being moved */
1302 count = be16_to_cpu(hdr1->count) - count;
1303 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1304 space += count * sizeof(xfs_attr_leaf_entry_t);
1307 * leaf2 is the destination, compact it if it looks tight.
1309 max = be16_to_cpu(hdr2->firstused)
1310 - sizeof(xfs_attr_leaf_hdr_t);
1311 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1312 if (space > max) {
1313 xfs_attr_leaf_compact(args->trans, blk2->bp);
1317 * Move high entries from leaf1 to low end of leaf2.
1319 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1320 leaf2, 0, count, state->mp);
1322 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1323 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1324 } else if (count > be16_to_cpu(hdr1->count)) {
1326 * I assert that since all callers pass in an empty
1327 * second buffer, this code should never execute.
1331 * Figure the total bytes to be added to the destination leaf.
1333 /* number entries being moved */
1334 count -= be16_to_cpu(hdr1->count);
1335 space = totallen - be16_to_cpu(hdr1->usedbytes);
1336 space += count * sizeof(xfs_attr_leaf_entry_t);
1339 * leaf1 is the destination, compact it if it looks tight.
1341 max = be16_to_cpu(hdr1->firstused)
1342 - sizeof(xfs_attr_leaf_hdr_t);
1343 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1344 if (space > max) {
1345 xfs_attr_leaf_compact(args->trans, blk1->bp);
1349 * Move low entries from leaf2 to high end of leaf1.
1351 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1352 be16_to_cpu(hdr1->count), count, state->mp);
1354 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1355 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1359 * Copy out last hashval in each block for B-tree code.
1361 blk1->hashval = be32_to_cpu(
1362 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1363 blk2->hashval = be32_to_cpu(
1364 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1367 * Adjust the expected index for insertion.
1368 * NOTE: this code depends on the (current) situation that the
1369 * second block was originally empty.
1371 * If the insertion point moved to the 2nd block, we must adjust
1372 * the index. We must also track the entry just following the
1373 * new entry for use in an "atomic rename" operation, that entry
1374 * is always the "old" entry and the "new" entry is what we are
1375 * inserting. The index/blkno fields refer to the "old" entry,
1376 * while the index2/blkno2 fields refer to the "new" entry.
1378 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1379 ASSERT(state->inleaf == 0);
1380 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1381 args->index = args->index2 = blk2->index;
1382 args->blkno = args->blkno2 = blk2->blkno;
1383 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1384 if (state->inleaf) {
1385 args->index = blk1->index;
1386 args->blkno = blk1->blkno;
1387 args->index2 = 0;
1388 args->blkno2 = blk2->blkno;
1389 } else {
1390 blk2->index = blk1->index
1391 - be16_to_cpu(leaf1->hdr.count);
1392 args->index = args->index2 = blk2->index;
1393 args->blkno = args->blkno2 = blk2->blkno;
1395 } else {
1396 ASSERT(state->inleaf == 1);
1397 args->index = args->index2 = blk1->index;
1398 args->blkno = args->blkno2 = blk1->blkno;
1403 * Examine entries until we reduce the absolute difference in
1404 * byte usage between the two blocks to a minimum.
1405 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1406 * GROT: there will always be enough room in either block for a new entry.
1407 * GROT: Do a double-split for this case?
1409 STATIC int
1410 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1411 xfs_da_state_blk_t *blk1,
1412 xfs_da_state_blk_t *blk2,
1413 int *countarg, int *usedbytesarg)
1415 xfs_attr_leafblock_t *leaf1, *leaf2;
1416 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1417 xfs_attr_leaf_entry_t *entry;
1418 int count, max, index, totallen, half;
1419 int lastdelta, foundit, tmp;
1422 * Set up environment.
1424 leaf1 = blk1->bp->data;
1425 leaf2 = blk2->bp->data;
1426 hdr1 = &leaf1->hdr;
1427 hdr2 = &leaf2->hdr;
1428 foundit = 0;
1429 totallen = 0;
1432 * Examine entries until we reduce the absolute difference in
1433 * byte usage between the two blocks to a minimum.
1435 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1436 half = (max+1) * sizeof(*entry);
1437 half += be16_to_cpu(hdr1->usedbytes) +
1438 be16_to_cpu(hdr2->usedbytes) +
1439 xfs_attr_leaf_newentsize(
1440 state->args->namelen,
1441 state->args->valuelen,
1442 state->blocksize, NULL);
1443 half /= 2;
1444 lastdelta = state->blocksize;
1445 entry = &leaf1->entries[0];
1446 for (count = index = 0; count < max; entry++, index++, count++) {
1448 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1450 * The new entry is in the first block, account for it.
1452 if (count == blk1->index) {
1453 tmp = totallen + sizeof(*entry) +
1454 xfs_attr_leaf_newentsize(
1455 state->args->namelen,
1456 state->args->valuelen,
1457 state->blocksize, NULL);
1458 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1459 break;
1460 lastdelta = XFS_ATTR_ABS(half - tmp);
1461 totallen = tmp;
1462 foundit = 1;
1466 * Wrap around into the second block if necessary.
1468 if (count == be16_to_cpu(hdr1->count)) {
1469 leaf1 = leaf2;
1470 entry = &leaf1->entries[0];
1471 index = 0;
1475 * Figure out if next leaf entry would be too much.
1477 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1478 index);
1479 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1480 break;
1481 lastdelta = XFS_ATTR_ABS(half - tmp);
1482 totallen = tmp;
1483 #undef XFS_ATTR_ABS
1487 * Calculate the number of usedbytes that will end up in lower block.
1488 * If new entry not in lower block, fix up the count.
1490 totallen -= count * sizeof(*entry);
1491 if (foundit) {
1492 totallen -= sizeof(*entry) +
1493 xfs_attr_leaf_newentsize(
1494 state->args->namelen,
1495 state->args->valuelen,
1496 state->blocksize, NULL);
1499 *countarg = count;
1500 *usedbytesarg = totallen;
1501 return(foundit);
1504 /*========================================================================
1505 * Routines used for shrinking the Btree.
1506 *========================================================================*/
1509 * Check a leaf block and its neighbors to see if the block should be
1510 * collapsed into one or the other neighbor. Always keep the block
1511 * with the smaller block number.
1512 * If the current block is over 50% full, don't try to join it, return 0.
1513 * If the block is empty, fill in the state structure and return 2.
1514 * If it can be collapsed, fill in the state structure and return 1.
1515 * If nothing can be done, return 0.
1517 * GROT: allow for INCOMPLETE entries in calculation.
1520 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1522 xfs_attr_leafblock_t *leaf;
1523 xfs_da_state_blk_t *blk;
1524 xfs_da_blkinfo_t *info;
1525 int count, bytes, forward, error, retval, i;
1526 xfs_dablk_t blkno;
1527 xfs_dabuf_t *bp;
1530 * Check for the degenerate case of the block being over 50% full.
1531 * If so, it's not worth even looking to see if we might be able
1532 * to coalesce with a sibling.
1534 blk = &state->path.blk[ state->path.active-1 ];
1535 info = blk->bp->data;
1536 ASSERT(info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1537 leaf = (xfs_attr_leafblock_t *)info;
1538 count = be16_to_cpu(leaf->hdr.count);
1539 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1540 count * sizeof(xfs_attr_leaf_entry_t) +
1541 be16_to_cpu(leaf->hdr.usedbytes);
1542 if (bytes > (state->blocksize >> 1)) {
1543 *action = 0; /* blk over 50%, don't try to join */
1544 return(0);
1548 * Check for the degenerate case of the block being empty.
1549 * If the block is empty, we'll simply delete it, no need to
1550 * coalesce it with a sibling block. We choose (arbitrarily)
1551 * to merge with the forward block unless it is NULL.
1553 if (count == 0) {
1555 * Make altpath point to the block we want to keep and
1556 * path point to the block we want to drop (this one).
1558 forward = (info->forw != 0);
1559 memcpy(&state->altpath, &state->path, sizeof(state->path));
1560 error = xfs_da_path_shift(state, &state->altpath, forward,
1561 0, &retval);
1562 if (error)
1563 return(error);
1564 if (retval) {
1565 *action = 0;
1566 } else {
1567 *action = 2;
1569 return(0);
1573 * Examine each sibling block to see if we can coalesce with
1574 * at least 25% free space to spare. We need to figure out
1575 * whether to merge with the forward or the backward block.
1576 * We prefer coalescing with the lower numbered sibling so as
1577 * to shrink an attribute list over time.
1579 /* start with smaller blk num */
1580 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1581 for (i = 0; i < 2; forward = !forward, i++) {
1582 if (forward)
1583 blkno = be32_to_cpu(info->forw);
1584 else
1585 blkno = be32_to_cpu(info->back);
1586 if (blkno == 0)
1587 continue;
1588 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1589 blkno, -1, &bp, XFS_ATTR_FORK);
1590 if (error)
1591 return(error);
1592 ASSERT(bp != NULL);
1594 leaf = (xfs_attr_leafblock_t *)info;
1595 count = be16_to_cpu(leaf->hdr.count);
1596 bytes = state->blocksize - (state->blocksize>>2);
1597 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1598 leaf = bp->data;
1599 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1600 count += be16_to_cpu(leaf->hdr.count);
1601 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1602 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1603 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1604 xfs_da_brelse(state->args->trans, bp);
1605 if (bytes >= 0)
1606 break; /* fits with at least 25% to spare */
1608 if (i >= 2) {
1609 *action = 0;
1610 return(0);
1614 * Make altpath point to the block we want to keep (the lower
1615 * numbered block) and path point to the block we want to drop.
1617 memcpy(&state->altpath, &state->path, sizeof(state->path));
1618 if (blkno < blk->blkno) {
1619 error = xfs_da_path_shift(state, &state->altpath, forward,
1620 0, &retval);
1621 } else {
1622 error = xfs_da_path_shift(state, &state->path, forward,
1623 0, &retval);
1625 if (error)
1626 return(error);
1627 if (retval) {
1628 *action = 0;
1629 } else {
1630 *action = 1;
1632 return(0);
1636 * Remove a name from the leaf attribute list structure.
1638 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1639 * If two leaves are 37% full, when combined they will leave 25% free.
1642 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1644 xfs_attr_leafblock_t *leaf;
1645 xfs_attr_leaf_hdr_t *hdr;
1646 xfs_attr_leaf_map_t *map;
1647 xfs_attr_leaf_entry_t *entry;
1648 int before, after, smallest, entsize;
1649 int tablesize, tmp, i;
1650 xfs_mount_t *mp;
1652 leaf = bp->data;
1653 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1654 hdr = &leaf->hdr;
1655 mp = args->trans->t_mountp;
1656 ASSERT((be16_to_cpu(hdr->count) > 0)
1657 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1658 ASSERT((args->index >= 0)
1659 && (args->index < be16_to_cpu(hdr->count)));
1660 ASSERT(be16_to_cpu(hdr->firstused) >=
1661 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1662 entry = &leaf->entries[args->index];
1663 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1664 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1667 * Scan through free region table:
1668 * check for adjacency of free'd entry with an existing one,
1669 * find smallest free region in case we need to replace it,
1670 * adjust any map that borders the entry table,
1672 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1673 + sizeof(xfs_attr_leaf_hdr_t);
1674 map = &hdr->freemap[0];
1675 tmp = be16_to_cpu(map->size);
1676 before = after = -1;
1677 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1678 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1679 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1680 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1681 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1682 if (be16_to_cpu(map->base) == tablesize) {
1683 be16_add_cpu(&map->base,
1684 -((int)sizeof(xfs_attr_leaf_entry_t)));
1685 be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1688 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1689 == be16_to_cpu(entry->nameidx)) {
1690 before = i;
1691 } else if (be16_to_cpu(map->base)
1692 == (be16_to_cpu(entry->nameidx) + entsize)) {
1693 after = i;
1694 } else if (be16_to_cpu(map->size) < tmp) {
1695 tmp = be16_to_cpu(map->size);
1696 smallest = i;
1701 * Coalesce adjacent freemap regions,
1702 * or replace the smallest region.
1704 if ((before >= 0) || (after >= 0)) {
1705 if ((before >= 0) && (after >= 0)) {
1706 map = &hdr->freemap[before];
1707 be16_add_cpu(&map->size, entsize);
1708 be16_add_cpu(&map->size,
1709 be16_to_cpu(hdr->freemap[after].size));
1710 hdr->freemap[after].base = 0;
1711 hdr->freemap[after].size = 0;
1712 } else if (before >= 0) {
1713 map = &hdr->freemap[before];
1714 be16_add_cpu(&map->size, entsize);
1715 } else {
1716 map = &hdr->freemap[after];
1717 /* both on-disk, don't endian flip twice */
1718 map->base = entry->nameidx;
1719 be16_add_cpu(&map->size, entsize);
1721 } else {
1723 * Replace smallest region (if it is smaller than free'd entry)
1725 map = &hdr->freemap[smallest];
1726 if (be16_to_cpu(map->size) < entsize) {
1727 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1728 map->size = cpu_to_be16(entsize);
1733 * Did we remove the first entry?
1735 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1736 smallest = 1;
1737 else
1738 smallest = 0;
1741 * Compress the remaining entries and zero out the removed stuff.
1743 memset(xfs_attr_leaf_name(leaf, args->index), 0, entsize);
1744 be16_add_cpu(&hdr->usedbytes, -entsize);
1745 xfs_da_log_buf(args->trans, bp,
1746 XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1747 entsize));
1749 tmp = (be16_to_cpu(hdr->count) - args->index)
1750 * sizeof(xfs_attr_leaf_entry_t);
1751 memmove((char *)entry, (char *)(entry+1), tmp);
1752 be16_add_cpu(&hdr->count, -1);
1753 xfs_da_log_buf(args->trans, bp,
1754 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1755 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1756 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1759 * If we removed the first entry, re-find the first used byte
1760 * in the name area. Note that if the entry was the "firstused",
1761 * then we don't have a "hole" in our block resulting from
1762 * removing the name.
1764 if (smallest) {
1765 tmp = XFS_LBSIZE(mp);
1766 entry = &leaf->entries[0];
1767 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1768 ASSERT(be16_to_cpu(entry->nameidx) >=
1769 be16_to_cpu(hdr->firstused));
1770 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1772 if (be16_to_cpu(entry->nameidx) < tmp)
1773 tmp = be16_to_cpu(entry->nameidx);
1775 hdr->firstused = cpu_to_be16(tmp);
1776 if (!hdr->firstused) {
1777 hdr->firstused = cpu_to_be16(
1778 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1780 } else {
1781 hdr->holes = 1; /* mark as needing compaction */
1783 xfs_da_log_buf(args->trans, bp,
1784 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1787 * Check if leaf is less than 50% full, caller may want to
1788 * "join" the leaf with a sibling if so.
1790 tmp = sizeof(xfs_attr_leaf_hdr_t);
1791 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1792 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1793 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1797 * Move all the attribute list entries from drop_leaf into save_leaf.
1799 void
1800 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1801 xfs_da_state_blk_t *save_blk)
1803 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1804 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1805 xfs_mount_t *mp;
1806 char *tmpbuffer;
1809 * Set up environment.
1811 mp = state->mp;
1812 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1813 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1814 drop_leaf = drop_blk->bp->data;
1815 save_leaf = save_blk->bp->data;
1816 ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1817 ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1818 drop_hdr = &drop_leaf->hdr;
1819 save_hdr = &save_leaf->hdr;
1822 * Save last hashval from dying block for later Btree fixup.
1824 drop_blk->hashval = be32_to_cpu(
1825 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1828 * Check if we need a temp buffer, or can we do it in place.
1829 * Note that we don't check "leaf" for holes because we will
1830 * always be dropping it, toosmall() decided that for us already.
1832 if (save_hdr->holes == 0) {
1834 * dest leaf has no holes, so we add there. May need
1835 * to make some room in the entry array.
1837 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1838 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1839 be16_to_cpu(drop_hdr->count), mp);
1840 } else {
1841 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1842 be16_to_cpu(save_hdr->count),
1843 be16_to_cpu(drop_hdr->count), mp);
1845 } else {
1847 * Destination has holes, so we make a temporary copy
1848 * of the leaf and add them both to that.
1850 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1851 ASSERT(tmpbuffer != NULL);
1852 memset(tmpbuffer, 0, state->blocksize);
1853 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1854 tmp_hdr = &tmp_leaf->hdr;
1855 tmp_hdr->info = save_hdr->info; /* struct copy */
1856 tmp_hdr->count = 0;
1857 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1858 if (!tmp_hdr->firstused) {
1859 tmp_hdr->firstused = cpu_to_be16(
1860 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1862 tmp_hdr->usedbytes = 0;
1863 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1864 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1865 be16_to_cpu(drop_hdr->count), mp);
1866 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1867 be16_to_cpu(tmp_leaf->hdr.count),
1868 be16_to_cpu(save_hdr->count), mp);
1869 } else {
1870 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1871 be16_to_cpu(save_hdr->count), mp);
1872 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1873 be16_to_cpu(tmp_leaf->hdr.count),
1874 be16_to_cpu(drop_hdr->count), mp);
1876 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1877 kmem_free(tmpbuffer);
1880 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1881 state->blocksize - 1);
1884 * Copy out last hashval in each block for B-tree code.
1886 save_blk->hashval = be32_to_cpu(
1887 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1890 /*========================================================================
1891 * Routines used for finding things in the Btree.
1892 *========================================================================*/
1895 * Look up a name in a leaf attribute list structure.
1896 * This is the internal routine, it uses the caller's buffer.
1898 * Note that duplicate keys are allowed, but only check within the
1899 * current leaf node. The Btree code must check in adjacent leaf nodes.
1901 * Return in args->index the index into the entry[] array of either
1902 * the found entry, or where the entry should have been (insert before
1903 * that entry).
1905 * Don't change the args->value unless we find the attribute.
1908 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1910 xfs_attr_leafblock_t *leaf;
1911 xfs_attr_leaf_entry_t *entry;
1912 xfs_attr_leaf_name_local_t *name_loc;
1913 xfs_attr_leaf_name_remote_t *name_rmt;
1914 int probe, span;
1915 xfs_dahash_t hashval;
1917 leaf = bp->data;
1918 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1919 ASSERT(be16_to_cpu(leaf->hdr.count)
1920 < (XFS_LBSIZE(args->dp->i_mount)/8));
1923 * Binary search. (note: small blocks will skip this loop)
1925 hashval = args->hashval;
1926 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1927 for (entry = &leaf->entries[probe]; span > 4;
1928 entry = &leaf->entries[probe]) {
1929 span /= 2;
1930 if (be32_to_cpu(entry->hashval) < hashval)
1931 probe += span;
1932 else if (be32_to_cpu(entry->hashval) > hashval)
1933 probe -= span;
1934 else
1935 break;
1937 ASSERT((probe >= 0) &&
1938 (!leaf->hdr.count
1939 || (probe < be16_to_cpu(leaf->hdr.count))));
1940 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1943 * Since we may have duplicate hashval's, find the first matching
1944 * hashval in the leaf.
1946 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1947 entry--;
1948 probe--;
1950 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1951 (be32_to_cpu(entry->hashval) < hashval)) {
1952 entry++;
1953 probe++;
1955 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1956 (be32_to_cpu(entry->hashval) != hashval)) {
1957 args->index = probe;
1958 return(XFS_ERROR(ENOATTR));
1962 * Duplicate keys may be present, so search all of them for a match.
1964 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1965 (be32_to_cpu(entry->hashval) == hashval);
1966 entry++, probe++) {
1968 * GROT: Add code to remove incomplete entries.
1971 * If we are looking for INCOMPLETE entries, show only those.
1972 * If we are looking for complete entries, show only those.
1974 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
1975 (entry->flags & XFS_ATTR_INCOMPLETE)) {
1976 continue;
1978 if (entry->flags & XFS_ATTR_LOCAL) {
1979 name_loc = xfs_attr_leaf_name_local(leaf, probe);
1980 if (name_loc->namelen != args->namelen)
1981 continue;
1982 if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
1983 continue;
1984 if (!xfs_attr_namesp_match(args->flags, entry->flags))
1985 continue;
1986 args->index = probe;
1987 return(XFS_ERROR(EEXIST));
1988 } else {
1989 name_rmt = xfs_attr_leaf_name_remote(leaf, probe);
1990 if (name_rmt->namelen != args->namelen)
1991 continue;
1992 if (memcmp(args->name, (char *)name_rmt->name,
1993 args->namelen) != 0)
1994 continue;
1995 if (!xfs_attr_namesp_match(args->flags, entry->flags))
1996 continue;
1997 args->index = probe;
1998 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
1999 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2000 be32_to_cpu(name_rmt->valuelen));
2001 return(XFS_ERROR(EEXIST));
2004 args->index = probe;
2005 return(XFS_ERROR(ENOATTR));
2009 * Get the value associated with an attribute name from a leaf attribute
2010 * list structure.
2013 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2015 int valuelen;
2016 xfs_attr_leafblock_t *leaf;
2017 xfs_attr_leaf_entry_t *entry;
2018 xfs_attr_leaf_name_local_t *name_loc;
2019 xfs_attr_leaf_name_remote_t *name_rmt;
2021 leaf = bp->data;
2022 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2023 ASSERT(be16_to_cpu(leaf->hdr.count)
2024 < (XFS_LBSIZE(args->dp->i_mount)/8));
2025 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2027 entry = &leaf->entries[args->index];
2028 if (entry->flags & XFS_ATTR_LOCAL) {
2029 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2030 ASSERT(name_loc->namelen == args->namelen);
2031 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2032 valuelen = be16_to_cpu(name_loc->valuelen);
2033 if (args->flags & ATTR_KERNOVAL) {
2034 args->valuelen = valuelen;
2035 return(0);
2037 if (args->valuelen < valuelen) {
2038 args->valuelen = valuelen;
2039 return(XFS_ERROR(ERANGE));
2041 args->valuelen = valuelen;
2042 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2043 } else {
2044 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2045 ASSERT(name_rmt->namelen == args->namelen);
2046 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2047 valuelen = be32_to_cpu(name_rmt->valuelen);
2048 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2049 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2050 if (args->flags & ATTR_KERNOVAL) {
2051 args->valuelen = valuelen;
2052 return(0);
2054 if (args->valuelen < valuelen) {
2055 args->valuelen = valuelen;
2056 return(XFS_ERROR(ERANGE));
2058 args->valuelen = valuelen;
2060 return(0);
2063 /*========================================================================
2064 * Utility routines.
2065 *========================================================================*/
2068 * Move the indicated entries from one leaf to another.
2069 * NOTE: this routine modifies both source and destination leaves.
2071 /*ARGSUSED*/
2072 STATIC void
2073 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2074 xfs_attr_leafblock_t *leaf_d, int start_d,
2075 int count, xfs_mount_t *mp)
2077 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2078 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2079 int desti, tmp, i;
2082 * Check for nothing to do.
2084 if (count == 0)
2085 return;
2088 * Set up environment.
2090 ASSERT(leaf_s->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2091 ASSERT(leaf_d->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2092 hdr_s = &leaf_s->hdr;
2093 hdr_d = &leaf_d->hdr;
2094 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2095 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2096 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2097 ((be16_to_cpu(hdr_s->count)
2098 * sizeof(*entry_s))+sizeof(*hdr_s)));
2099 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2100 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2101 ((be16_to_cpu(hdr_d->count)
2102 * sizeof(*entry_d))+sizeof(*hdr_d)));
2104 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2105 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2106 ASSERT(count <= be16_to_cpu(hdr_s->count));
2109 * Move the entries in the destination leaf up to make a hole?
2111 if (start_d < be16_to_cpu(hdr_d->count)) {
2112 tmp = be16_to_cpu(hdr_d->count) - start_d;
2113 tmp *= sizeof(xfs_attr_leaf_entry_t);
2114 entry_s = &leaf_d->entries[start_d];
2115 entry_d = &leaf_d->entries[start_d + count];
2116 memmove((char *)entry_d, (char *)entry_s, tmp);
2120 * Copy all entry's in the same (sorted) order,
2121 * but allocate attribute info packed and in sequence.
2123 entry_s = &leaf_s->entries[start_s];
2124 entry_d = &leaf_d->entries[start_d];
2125 desti = start_d;
2126 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2127 ASSERT(be16_to_cpu(entry_s->nameidx)
2128 >= be16_to_cpu(hdr_s->firstused));
2129 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2130 #ifdef GROT
2132 * Code to drop INCOMPLETE entries. Difficult to use as we
2133 * may also need to change the insertion index. Code turned
2134 * off for 6.2, should be revisited later.
2136 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2137 memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2138 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2139 be16_add_cpu(&hdr_s->count, -1);
2140 entry_d--; /* to compensate for ++ in loop hdr */
2141 desti--;
2142 if ((start_s + i) < offset)
2143 result++; /* insertion index adjustment */
2144 } else {
2145 #endif /* GROT */
2146 be16_add_cpu(&hdr_d->firstused, -tmp);
2147 /* both on-disk, don't endian flip twice */
2148 entry_d->hashval = entry_s->hashval;
2149 /* both on-disk, don't endian flip twice */
2150 entry_d->nameidx = hdr_d->firstused;
2151 entry_d->flags = entry_s->flags;
2152 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2153 <= XFS_LBSIZE(mp));
2154 memmove(xfs_attr_leaf_name(leaf_d, desti),
2155 xfs_attr_leaf_name(leaf_s, start_s + i), tmp);
2156 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2157 <= XFS_LBSIZE(mp));
2158 memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2159 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2160 be16_add_cpu(&hdr_d->usedbytes, tmp);
2161 be16_add_cpu(&hdr_s->count, -1);
2162 be16_add_cpu(&hdr_d->count, 1);
2163 tmp = be16_to_cpu(hdr_d->count)
2164 * sizeof(xfs_attr_leaf_entry_t)
2165 + sizeof(xfs_attr_leaf_hdr_t);
2166 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2167 #ifdef GROT
2169 #endif /* GROT */
2173 * Zero out the entries we just copied.
2175 if (start_s == be16_to_cpu(hdr_s->count)) {
2176 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2177 entry_s = &leaf_s->entries[start_s];
2178 ASSERT(((char *)entry_s + tmp) <=
2179 ((char *)leaf_s + XFS_LBSIZE(mp)));
2180 memset((char *)entry_s, 0, tmp);
2181 } else {
2183 * Move the remaining entries down to fill the hole,
2184 * then zero the entries at the top.
2186 tmp = be16_to_cpu(hdr_s->count) - count;
2187 tmp *= sizeof(xfs_attr_leaf_entry_t);
2188 entry_s = &leaf_s->entries[start_s + count];
2189 entry_d = &leaf_s->entries[start_s];
2190 memmove((char *)entry_d, (char *)entry_s, tmp);
2192 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2193 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2194 ASSERT(((char *)entry_s + tmp) <=
2195 ((char *)leaf_s + XFS_LBSIZE(mp)));
2196 memset((char *)entry_s, 0, tmp);
2200 * Fill in the freemap information
2202 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2203 be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2204 sizeof(xfs_attr_leaf_entry_t));
2205 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2206 - be16_to_cpu(hdr_d->freemap[0].base));
2207 hdr_d->freemap[1].base = 0;
2208 hdr_d->freemap[2].base = 0;
2209 hdr_d->freemap[1].size = 0;
2210 hdr_d->freemap[2].size = 0;
2211 hdr_s->holes = 1; /* leaf may not be compact */
2215 * Compare two leaf blocks "order".
2216 * Return 0 unless leaf2 should go before leaf1.
2219 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2221 xfs_attr_leafblock_t *leaf1, *leaf2;
2223 leaf1 = leaf1_bp->data;
2224 leaf2 = leaf2_bp->data;
2225 ASSERT((leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) &&
2226 (leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)));
2227 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2228 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2229 ((be32_to_cpu(leaf2->entries[0].hashval) <
2230 be32_to_cpu(leaf1->entries[0].hashval)) ||
2231 (be32_to_cpu(leaf2->entries[
2232 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2233 be32_to_cpu(leaf1->entries[
2234 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2235 return(1);
2237 return(0);
2241 * Pick up the last hashvalue from a leaf block.
2243 xfs_dahash_t
2244 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2246 xfs_attr_leafblock_t *leaf;
2248 leaf = bp->data;
2249 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2250 if (count)
2251 *count = be16_to_cpu(leaf->hdr.count);
2252 if (!leaf->hdr.count)
2253 return(0);
2254 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2258 * Calculate the number of bytes used to store the indicated attribute
2259 * (whether local or remote only calculate bytes in this block).
2261 STATIC int
2262 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2264 xfs_attr_leaf_name_local_t *name_loc;
2265 xfs_attr_leaf_name_remote_t *name_rmt;
2266 int size;
2268 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2269 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2270 name_loc = xfs_attr_leaf_name_local(leaf, index);
2271 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2272 be16_to_cpu(name_loc->valuelen));
2273 } else {
2274 name_rmt = xfs_attr_leaf_name_remote(leaf, index);
2275 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2277 return(size);
2281 * Calculate the number of bytes that would be required to store the new
2282 * attribute (whether local or remote only calculate bytes in this block).
2283 * This routine decides as a side effect whether the attribute will be
2284 * a "local" or a "remote" attribute.
2287 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2289 int size;
2291 size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2292 if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2293 if (local) {
2294 *local = 1;
2296 } else {
2297 size = xfs_attr_leaf_entsize_remote(namelen);
2298 if (local) {
2299 *local = 0;
2302 return(size);
2306 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2309 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2311 attrlist_cursor_kern_t *cursor;
2312 xfs_attr_leafblock_t *leaf;
2313 xfs_attr_leaf_entry_t *entry;
2314 int retval, i;
2316 ASSERT(bp != NULL);
2317 leaf = bp->data;
2318 cursor = context->cursor;
2319 cursor->initted = 1;
2321 trace_xfs_attr_list_leaf(context);
2324 * Re-find our place in the leaf block if this is a new syscall.
2326 if (context->resynch) {
2327 entry = &leaf->entries[0];
2328 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2329 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2330 if (cursor->offset == context->dupcnt) {
2331 context->dupcnt = 0;
2332 break;
2334 context->dupcnt++;
2335 } else if (be32_to_cpu(entry->hashval) >
2336 cursor->hashval) {
2337 context->dupcnt = 0;
2338 break;
2341 if (i == be16_to_cpu(leaf->hdr.count)) {
2342 trace_xfs_attr_list_notfound(context);
2343 return(0);
2345 } else {
2346 entry = &leaf->entries[0];
2347 i = 0;
2349 context->resynch = 0;
2352 * We have found our place, start copying out the new attributes.
2354 retval = 0;
2355 for ( ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2356 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2357 cursor->hashval = be32_to_cpu(entry->hashval);
2358 cursor->offset = 0;
2361 if (entry->flags & XFS_ATTR_INCOMPLETE)
2362 continue; /* skip incomplete entries */
2364 if (entry->flags & XFS_ATTR_LOCAL) {
2365 xfs_attr_leaf_name_local_t *name_loc =
2366 xfs_attr_leaf_name_local(leaf, i);
2368 retval = context->put_listent(context,
2369 entry->flags,
2370 name_loc->nameval,
2371 (int)name_loc->namelen,
2372 be16_to_cpu(name_loc->valuelen),
2373 &name_loc->nameval[name_loc->namelen]);
2374 if (retval)
2375 return retval;
2376 } else {
2377 xfs_attr_leaf_name_remote_t *name_rmt =
2378 xfs_attr_leaf_name_remote(leaf, i);
2380 int valuelen = be32_to_cpu(name_rmt->valuelen);
2382 if (context->put_value) {
2383 xfs_da_args_t args;
2385 memset((char *)&args, 0, sizeof(args));
2386 args.dp = context->dp;
2387 args.whichfork = XFS_ATTR_FORK;
2388 args.valuelen = valuelen;
2389 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2390 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2391 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2392 retval = xfs_attr_rmtval_get(&args);
2393 if (retval)
2394 return retval;
2395 retval = context->put_listent(context,
2396 entry->flags,
2397 name_rmt->name,
2398 (int)name_rmt->namelen,
2399 valuelen,
2400 args.value);
2401 kmem_free(args.value);
2402 } else {
2403 retval = context->put_listent(context,
2404 entry->flags,
2405 name_rmt->name,
2406 (int)name_rmt->namelen,
2407 valuelen,
2408 NULL);
2410 if (retval)
2411 return retval;
2413 if (context->seen_enough)
2414 break;
2415 cursor->offset++;
2417 trace_xfs_attr_list_leaf_end(context);
2418 return(retval);
2422 /*========================================================================
2423 * Manage the INCOMPLETE flag in a leaf entry
2424 *========================================================================*/
2427 * Clear the INCOMPLETE flag on an entry in a leaf block.
2430 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2432 xfs_attr_leafblock_t *leaf;
2433 xfs_attr_leaf_entry_t *entry;
2434 xfs_attr_leaf_name_remote_t *name_rmt;
2435 xfs_dabuf_t *bp;
2436 int error;
2437 #ifdef DEBUG
2438 xfs_attr_leaf_name_local_t *name_loc;
2439 int namelen;
2440 char *name;
2441 #endif /* DEBUG */
2444 * Set up the operation.
2446 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2447 XFS_ATTR_FORK);
2448 if (error) {
2449 return(error);
2451 ASSERT(bp != NULL);
2453 leaf = bp->data;
2454 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2455 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2456 ASSERT(args->index >= 0);
2457 entry = &leaf->entries[ args->index ];
2458 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2460 #ifdef DEBUG
2461 if (entry->flags & XFS_ATTR_LOCAL) {
2462 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2463 namelen = name_loc->namelen;
2464 name = (char *)name_loc->nameval;
2465 } else {
2466 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2467 namelen = name_rmt->namelen;
2468 name = (char *)name_rmt->name;
2470 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2471 ASSERT(namelen == args->namelen);
2472 ASSERT(memcmp(name, args->name, namelen) == 0);
2473 #endif /* DEBUG */
2475 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2476 xfs_da_log_buf(args->trans, bp,
2477 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2479 if (args->rmtblkno) {
2480 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2481 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2482 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2483 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2484 xfs_da_log_buf(args->trans, bp,
2485 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2487 xfs_da_buf_done(bp);
2490 * Commit the flag value change and start the next trans in series.
2492 return xfs_trans_roll(&args->trans, args->dp);
2496 * Set the INCOMPLETE flag on an entry in a leaf block.
2499 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2501 xfs_attr_leafblock_t *leaf;
2502 xfs_attr_leaf_entry_t *entry;
2503 xfs_attr_leaf_name_remote_t *name_rmt;
2504 xfs_dabuf_t *bp;
2505 int error;
2508 * Set up the operation.
2510 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2511 XFS_ATTR_FORK);
2512 if (error) {
2513 return(error);
2515 ASSERT(bp != NULL);
2517 leaf = bp->data;
2518 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2519 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2520 ASSERT(args->index >= 0);
2521 entry = &leaf->entries[ args->index ];
2523 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2524 entry->flags |= XFS_ATTR_INCOMPLETE;
2525 xfs_da_log_buf(args->trans, bp,
2526 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2527 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2528 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2529 name_rmt->valueblk = 0;
2530 name_rmt->valuelen = 0;
2531 xfs_da_log_buf(args->trans, bp,
2532 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2534 xfs_da_buf_done(bp);
2537 * Commit the flag value change and start the next trans in series.
2539 return xfs_trans_roll(&args->trans, args->dp);
2543 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2544 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2545 * entry given by args->blkno2/index2.
2547 * Note that they could be in different blocks, or in the same block.
2550 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2552 xfs_attr_leafblock_t *leaf1, *leaf2;
2553 xfs_attr_leaf_entry_t *entry1, *entry2;
2554 xfs_attr_leaf_name_remote_t *name_rmt;
2555 xfs_dabuf_t *bp1, *bp2;
2556 int error;
2557 #ifdef DEBUG
2558 xfs_attr_leaf_name_local_t *name_loc;
2559 int namelen1, namelen2;
2560 char *name1, *name2;
2561 #endif /* DEBUG */
2564 * Read the block containing the "old" attr
2566 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2567 XFS_ATTR_FORK);
2568 if (error) {
2569 return(error);
2571 ASSERT(bp1 != NULL);
2574 * Read the block containing the "new" attr, if it is different
2576 if (args->blkno2 != args->blkno) {
2577 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2578 -1, &bp2, XFS_ATTR_FORK);
2579 if (error) {
2580 return(error);
2582 ASSERT(bp2 != NULL);
2583 } else {
2584 bp2 = bp1;
2587 leaf1 = bp1->data;
2588 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2589 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2590 ASSERT(args->index >= 0);
2591 entry1 = &leaf1->entries[ args->index ];
2593 leaf2 = bp2->data;
2594 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2595 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2596 ASSERT(args->index2 >= 0);
2597 entry2 = &leaf2->entries[ args->index2 ];
2599 #ifdef DEBUG
2600 if (entry1->flags & XFS_ATTR_LOCAL) {
2601 name_loc = xfs_attr_leaf_name_local(leaf1, args->index);
2602 namelen1 = name_loc->namelen;
2603 name1 = (char *)name_loc->nameval;
2604 } else {
2605 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2606 namelen1 = name_rmt->namelen;
2607 name1 = (char *)name_rmt->name;
2609 if (entry2->flags & XFS_ATTR_LOCAL) {
2610 name_loc = xfs_attr_leaf_name_local(leaf2, args->index2);
2611 namelen2 = name_loc->namelen;
2612 name2 = (char *)name_loc->nameval;
2613 } else {
2614 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2615 namelen2 = name_rmt->namelen;
2616 name2 = (char *)name_rmt->name;
2618 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2619 ASSERT(namelen1 == namelen2);
2620 ASSERT(memcmp(name1, name2, namelen1) == 0);
2621 #endif /* DEBUG */
2623 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2624 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2626 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2627 xfs_da_log_buf(args->trans, bp1,
2628 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2629 if (args->rmtblkno) {
2630 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2631 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2632 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2633 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2634 xfs_da_log_buf(args->trans, bp1,
2635 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2638 entry2->flags |= XFS_ATTR_INCOMPLETE;
2639 xfs_da_log_buf(args->trans, bp2,
2640 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2641 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2642 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2643 name_rmt->valueblk = 0;
2644 name_rmt->valuelen = 0;
2645 xfs_da_log_buf(args->trans, bp2,
2646 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2648 xfs_da_buf_done(bp1);
2649 if (bp1 != bp2)
2650 xfs_da_buf_done(bp2);
2653 * Commit the flag value change and start the next trans in series.
2655 error = xfs_trans_roll(&args->trans, args->dp);
2657 return(error);
2660 /*========================================================================
2661 * Indiscriminately delete the entire attribute fork
2662 *========================================================================*/
2665 * Recurse (gasp!) through the attribute nodes until we find leaves.
2666 * We're doing a depth-first traversal in order to invalidate everything.
2669 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2671 xfs_da_blkinfo_t *info;
2672 xfs_daddr_t blkno;
2673 xfs_dabuf_t *bp;
2674 int error;
2677 * Read block 0 to see what we have to work with.
2678 * We only get here if we have extents, since we remove
2679 * the extents in reverse order the extent containing
2680 * block 0 must still be there.
2682 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2683 if (error)
2684 return(error);
2685 blkno = xfs_da_blkno(bp);
2688 * Invalidate the tree, even if the "tree" is only a single leaf block.
2689 * This is a depth-first traversal!
2691 info = bp->data;
2692 if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2693 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2694 } else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2695 error = xfs_attr_leaf_inactive(trans, dp, bp);
2696 } else {
2697 error = XFS_ERROR(EIO);
2698 xfs_da_brelse(*trans, bp);
2700 if (error)
2701 return(error);
2704 * Invalidate the incore copy of the root block.
2706 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2707 if (error)
2708 return(error);
2709 xfs_da_binval(*trans, bp); /* remove from cache */
2711 * Commit the invalidate and start the next transaction.
2713 error = xfs_trans_roll(trans, dp);
2715 return (error);
2719 * Recurse (gasp!) through the attribute nodes until we find leaves.
2720 * We're doing a depth-first traversal in order to invalidate everything.
2722 STATIC int
2723 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2724 int level)
2726 xfs_da_blkinfo_t *info;
2727 xfs_da_intnode_t *node;
2728 xfs_dablk_t child_fsb;
2729 xfs_daddr_t parent_blkno, child_blkno;
2730 int error, count, i;
2731 xfs_dabuf_t *child_bp;
2734 * Since this code is recursive (gasp!) we must protect ourselves.
2736 if (level > XFS_DA_NODE_MAXDEPTH) {
2737 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2738 return(XFS_ERROR(EIO));
2741 node = bp->data;
2742 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
2743 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2744 count = be16_to_cpu(node->hdr.count);
2745 if (!count) {
2746 xfs_da_brelse(*trans, bp);
2747 return(0);
2749 child_fsb = be32_to_cpu(node->btree[0].before);
2750 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2753 * If this is the node level just above the leaves, simply loop
2754 * over the leaves removing all of them. If this is higher up
2755 * in the tree, recurse downward.
2757 for (i = 0; i < count; i++) {
2759 * Read the subsidiary block to see what we have to work with.
2760 * Don't do this in a transaction. This is a depth-first
2761 * traversal of the tree so we may deal with many blocks
2762 * before we come back to this one.
2764 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2765 XFS_ATTR_FORK);
2766 if (error)
2767 return(error);
2768 if (child_bp) {
2769 /* save for re-read later */
2770 child_blkno = xfs_da_blkno(child_bp);
2773 * Invalidate the subtree, however we have to.
2775 info = child_bp->data;
2776 if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2777 error = xfs_attr_node_inactive(trans, dp,
2778 child_bp, level+1);
2779 } else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2780 error = xfs_attr_leaf_inactive(trans, dp,
2781 child_bp);
2782 } else {
2783 error = XFS_ERROR(EIO);
2784 xfs_da_brelse(*trans, child_bp);
2786 if (error)
2787 return(error);
2790 * Remove the subsidiary block from the cache
2791 * and from the log.
2793 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2794 &child_bp, XFS_ATTR_FORK);
2795 if (error)
2796 return(error);
2797 xfs_da_binval(*trans, child_bp);
2801 * If we're not done, re-read the parent to get the next
2802 * child block number.
2804 if ((i+1) < count) {
2805 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2806 &bp, XFS_ATTR_FORK);
2807 if (error)
2808 return(error);
2809 child_fsb = be32_to_cpu(node->btree[i+1].before);
2810 xfs_da_brelse(*trans, bp);
2813 * Atomically commit the whole invalidate stuff.
2815 error = xfs_trans_roll(trans, dp);
2816 if (error)
2817 return (error);
2820 return(0);
2824 * Invalidate all of the "remote" value regions pointed to by a particular
2825 * leaf block.
2826 * Note that we must release the lock on the buffer so that we are not
2827 * caught holding something that the logging code wants to flush to disk.
2829 STATIC int
2830 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2832 xfs_attr_leafblock_t *leaf;
2833 xfs_attr_leaf_entry_t *entry;
2834 xfs_attr_leaf_name_remote_t *name_rmt;
2835 xfs_attr_inactive_list_t *list, *lp;
2836 int error, count, size, tmp, i;
2838 leaf = bp->data;
2839 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2842 * Count the number of "remote" value extents.
2844 count = 0;
2845 entry = &leaf->entries[0];
2846 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2847 if (be16_to_cpu(entry->nameidx) &&
2848 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2849 name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2850 if (name_rmt->valueblk)
2851 count++;
2856 * If there are no "remote" values, we're done.
2858 if (count == 0) {
2859 xfs_da_brelse(*trans, bp);
2860 return(0);
2864 * Allocate storage for a list of all the "remote" value extents.
2866 size = count * sizeof(xfs_attr_inactive_list_t);
2867 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2870 * Identify each of the "remote" value extents.
2872 lp = list;
2873 entry = &leaf->entries[0];
2874 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2875 if (be16_to_cpu(entry->nameidx) &&
2876 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2877 name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2878 if (name_rmt->valueblk) {
2879 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2880 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2881 be32_to_cpu(name_rmt->valuelen));
2882 lp++;
2886 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2889 * Invalidate each of the "remote" value extents.
2891 error = 0;
2892 for (lp = list, i = 0; i < count; i++, lp++) {
2893 tmp = xfs_attr_leaf_freextent(trans, dp,
2894 lp->valueblk, lp->valuelen);
2896 if (error == 0)
2897 error = tmp; /* save only the 1st errno */
2900 kmem_free((xfs_caddr_t)list);
2901 return(error);
2905 * Look at all the extents for this logical region,
2906 * invalidate any buffers that are incore/in transactions.
2908 STATIC int
2909 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2910 xfs_dablk_t blkno, int blkcnt)
2912 xfs_bmbt_irec_t map;
2913 xfs_dablk_t tblkno;
2914 int tblkcnt, dblkcnt, nmap, error;
2915 xfs_daddr_t dblkno;
2916 xfs_buf_t *bp;
2919 * Roll through the "value", invalidating the attribute value's
2920 * blocks.
2922 tblkno = blkno;
2923 tblkcnt = blkcnt;
2924 while (tblkcnt > 0) {
2926 * Try to remember where we decided to put the value.
2928 nmap = 1;
2929 error = xfs_bmapi(*trans, dp, (xfs_fileoff_t)tblkno, tblkcnt,
2930 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2931 NULL, 0, &map, &nmap, NULL);
2932 if (error) {
2933 return(error);
2935 ASSERT(nmap == 1);
2936 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2939 * If it's a hole, these are already unmapped
2940 * so there's nothing to invalidate.
2942 if (map.br_startblock != HOLESTARTBLOCK) {
2944 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
2945 map.br_startblock);
2946 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
2947 map.br_blockcount);
2948 bp = xfs_trans_get_buf(*trans,
2949 dp->i_mount->m_ddev_targp,
2950 dblkno, dblkcnt, XBF_LOCK);
2951 xfs_trans_binval(*trans, bp);
2953 * Roll to next transaction.
2955 error = xfs_trans_roll(trans, dp);
2956 if (error)
2957 return (error);
2960 tblkno += map.br_blockcount;
2961 tblkcnt -= map.br_blockcount;
2964 return(0);