target: refactor core_update_device_list_for_node()
[linux-2.6.git] / fs / xfs / xfs_attr_leaf.c
blob7d89d800f5173f2d5d1c34a36278cab703f92d33
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_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_alloc.h"
32 #include "xfs_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_bmap.h"
38 #include "xfs_attr.h"
39 #include "xfs_attr_leaf.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
44 * xfs_attr_leaf.c
46 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
49 /*========================================================================
50 * Function prototypes for the kernel.
51 *========================================================================*/
54 * Routines used for growing the Btree.
56 STATIC int xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t which_block,
57 xfs_dabuf_t **bpp);
58 STATIC int xfs_attr_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
59 int freemap_index);
60 STATIC void xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer);
61 STATIC void xfs_attr_leaf_rebalance(xfs_da_state_t *state,
62 xfs_da_state_blk_t *blk1,
63 xfs_da_state_blk_t *blk2);
64 STATIC int xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
65 xfs_da_state_blk_t *leaf_blk_1,
66 xfs_da_state_blk_t *leaf_blk_2,
67 int *number_entries_in_blk1,
68 int *number_usedbytes_in_blk1);
71 * Routines used for shrinking the Btree.
73 STATIC int xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
74 xfs_dabuf_t *bp, int level);
75 STATIC int xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp,
76 xfs_dabuf_t *bp);
77 STATIC int xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
78 xfs_dablk_t blkno, int blkcnt);
81 * Utility routines.
83 STATIC void xfs_attr_leaf_moveents(xfs_attr_leafblock_t *src_leaf,
84 int src_start,
85 xfs_attr_leafblock_t *dst_leaf,
86 int dst_start, int move_count,
87 xfs_mount_t *mp);
88 STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
90 /*========================================================================
91 * Namespace helper routines
92 *========================================================================*/
95 * If namespace bits don't match return 0.
96 * If all match then return 1.
98 STATIC int
99 xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
101 return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
105 /*========================================================================
106 * External routines when attribute fork size < XFS_LITINO(mp).
107 *========================================================================*/
110 * Query whether the requested number of additional bytes of extended
111 * 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;
140 * If the requested numbers of bytes is smaller or equal to the
141 * current attribute fork size we can always proceed.
143 * Note that if_bytes in the data fork might actually be larger than
144 * the current data fork size is due to delalloc extents. In that
145 * case either the extent count will go down when they are converted
146 * to real extents, or the delalloc conversion will take care of the
147 * literal area rebalancing.
149 if (bytes <= XFS_IFORK_ASIZE(dp))
150 return dp->i_d.di_forkoff;
153 * For attr2 we can try to move the forkoff if there is space in the
154 * literal area, but for the old format we are done if there is no
155 * space in the fixed attribute fork.
157 if (!(mp->m_flags & XFS_MOUNT_ATTR2))
158 return 0;
160 dsize = dp->i_df.if_bytes;
162 switch (dp->i_d.di_format) {
163 case XFS_DINODE_FMT_EXTENTS:
165 * If there is no attr fork and the data fork is extents,
166 * determine if creating the default attr fork will result
167 * in the extents form migrating to btree. If so, the
168 * minimum offset only needs to be the space required for
169 * the btree root.
171 if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
172 xfs_default_attroffset(dp))
173 dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
174 break;
175 case XFS_DINODE_FMT_BTREE:
177 * If we have a data btree then keep forkoff if we have one,
178 * otherwise we are adding a new attr, so then we set
179 * minforkoff to where the btree root can finish so we have
180 * plenty of room for attrs
182 if (dp->i_d.di_forkoff) {
183 if (offset < dp->i_d.di_forkoff)
184 return 0;
185 return dp->i_d.di_forkoff;
187 dsize = XFS_BMAP_BROOT_SPACE(dp->i_df.if_broot);
188 break;
192 * A data fork btree root must have space for at least
193 * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
195 minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
196 minforkoff = roundup(minforkoff, 8) >> 3;
198 /* attr fork btree root can have at least this many key/ptr pairs */
199 maxforkoff = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
200 maxforkoff = maxforkoff >> 3; /* rounded down */
202 if (offset >= maxforkoff)
203 return maxforkoff;
204 if (offset >= minforkoff)
205 return offset;
206 return 0;
210 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
212 STATIC void
213 xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
215 if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
216 !(xfs_sb_version_hasattr2(&mp->m_sb))) {
217 spin_lock(&mp->m_sb_lock);
218 if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
219 xfs_sb_version_addattr2(&mp->m_sb);
220 spin_unlock(&mp->m_sb_lock);
221 xfs_mod_sb(tp, XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
222 } else
223 spin_unlock(&mp->m_sb_lock);
228 * Create the initial contents of a shortform attribute list.
230 void
231 xfs_attr_shortform_create(xfs_da_args_t *args)
233 xfs_attr_sf_hdr_t *hdr;
234 xfs_inode_t *dp;
235 xfs_ifork_t *ifp;
237 trace_xfs_attr_sf_create(args);
239 dp = args->dp;
240 ASSERT(dp != NULL);
241 ifp = dp->i_afp;
242 ASSERT(ifp != NULL);
243 ASSERT(ifp->if_bytes == 0);
244 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
245 ifp->if_flags &= ~XFS_IFEXTENTS; /* just in case */
246 dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
247 ifp->if_flags |= XFS_IFINLINE;
248 } else {
249 ASSERT(ifp->if_flags & XFS_IFINLINE);
251 xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
252 hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
253 hdr->count = 0;
254 hdr->totsize = cpu_to_be16(sizeof(*hdr));
255 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
259 * Add a name/value pair to the shortform attribute list.
260 * Overflow from the inode has already been checked for.
262 void
263 xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
265 xfs_attr_shortform_t *sf;
266 xfs_attr_sf_entry_t *sfe;
267 int i, offset, size;
268 xfs_mount_t *mp;
269 xfs_inode_t *dp;
270 xfs_ifork_t *ifp;
272 trace_xfs_attr_sf_add(args);
274 dp = args->dp;
275 mp = dp->i_mount;
276 dp->i_d.di_forkoff = forkoff;
278 ifp = dp->i_afp;
279 ASSERT(ifp->if_flags & XFS_IFINLINE);
280 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
281 sfe = &sf->list[0];
282 for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
283 #ifdef DEBUG
284 if (sfe->namelen != args->namelen)
285 continue;
286 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
287 continue;
288 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
289 continue;
290 ASSERT(0);
291 #endif
294 offset = (char *)sfe - (char *)sf;
295 size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
296 xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
297 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
298 sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
300 sfe->namelen = args->namelen;
301 sfe->valuelen = args->valuelen;
302 sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
303 memcpy(sfe->nameval, args->name, args->namelen);
304 memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
305 sf->hdr.count++;
306 be16_add_cpu(&sf->hdr.totsize, size);
307 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
309 xfs_sbversion_add_attr2(mp, args->trans);
313 * After the last attribute is removed revert to original inode format,
314 * making all literal area available to the data fork once more.
316 STATIC void
317 xfs_attr_fork_reset(
318 struct xfs_inode *ip,
319 struct xfs_trans *tp)
321 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
322 ip->i_d.di_forkoff = 0;
323 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
325 ASSERT(ip->i_d.di_anextents == 0);
326 ASSERT(ip->i_afp == NULL);
328 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
332 * Remove an attribute from the shortform attribute list structure.
335 xfs_attr_shortform_remove(xfs_da_args_t *args)
337 xfs_attr_shortform_t *sf;
338 xfs_attr_sf_entry_t *sfe;
339 int base, size=0, end, totsize, i;
340 xfs_mount_t *mp;
341 xfs_inode_t *dp;
343 trace_xfs_attr_sf_remove(args);
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 xfs_trans_log_inode(args->trans, dp,
393 XFS_ILOG_CORE | XFS_ILOG_ADATA);
396 xfs_sbversion_add_attr2(mp, args->trans);
398 return(0);
402 * Look up a name in a shortform attribute list structure.
404 /*ARGSUSED*/
406 xfs_attr_shortform_lookup(xfs_da_args_t *args)
408 xfs_attr_shortform_t *sf;
409 xfs_attr_sf_entry_t *sfe;
410 int i;
411 xfs_ifork_t *ifp;
413 trace_xfs_attr_sf_lookup(args);
415 ifp = args->dp->i_afp;
416 ASSERT(ifp->if_flags & XFS_IFINLINE);
417 sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
418 sfe = &sf->list[0];
419 for (i = 0; i < sf->hdr.count;
420 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
421 if (sfe->namelen != args->namelen)
422 continue;
423 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
424 continue;
425 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
426 continue;
427 return(XFS_ERROR(EEXIST));
429 return(XFS_ERROR(ENOATTR));
433 * Look up a name in a shortform attribute list structure.
435 /*ARGSUSED*/
437 xfs_attr_shortform_getvalue(xfs_da_args_t *args)
439 xfs_attr_shortform_t *sf;
440 xfs_attr_sf_entry_t *sfe;
441 int i;
443 ASSERT(args->dp->i_d.di_aformat == XFS_IFINLINE);
444 sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
445 sfe = &sf->list[0];
446 for (i = 0; i < sf->hdr.count;
447 sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
448 if (sfe->namelen != args->namelen)
449 continue;
450 if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
451 continue;
452 if (!xfs_attr_namesp_match(args->flags, sfe->flags))
453 continue;
454 if (args->flags & ATTR_KERNOVAL) {
455 args->valuelen = sfe->valuelen;
456 return(XFS_ERROR(EEXIST));
458 if (args->valuelen < sfe->valuelen) {
459 args->valuelen = sfe->valuelen;
460 return(XFS_ERROR(ERANGE));
462 args->valuelen = sfe->valuelen;
463 memcpy(args->value, &sfe->nameval[args->namelen],
464 args->valuelen);
465 return(XFS_ERROR(EEXIST));
467 return(XFS_ERROR(ENOATTR));
471 * Convert from using the shortform to the leaf.
474 xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
476 xfs_inode_t *dp;
477 xfs_attr_shortform_t *sf;
478 xfs_attr_sf_entry_t *sfe;
479 xfs_da_args_t nargs;
480 char *tmpbuffer;
481 int error, i, size;
482 xfs_dablk_t blkno;
483 xfs_dabuf_t *bp;
484 xfs_ifork_t *ifp;
486 trace_xfs_attr_sf_to_leaf(args);
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(leaf->hdr.info.magic == cpu_to_be16(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 trace_xfs_attr_leaf_to_sf(args);
789 dp = args->dp;
790 tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
791 ASSERT(tmpbuffer != NULL);
793 ASSERT(bp != NULL);
794 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
795 leaf = (xfs_attr_leafblock_t *)tmpbuffer;
796 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
797 memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
800 * Clean out the prior contents of the attribute list.
802 error = xfs_da_shrink_inode(args, 0, bp);
803 if (error)
804 goto out;
806 if (forkoff == -1) {
807 ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
808 ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
809 xfs_attr_fork_reset(dp, args->trans);
810 goto out;
813 xfs_attr_shortform_create(args);
816 * Copy the attributes
818 memset((char *)&nargs, 0, sizeof(nargs));
819 nargs.dp = dp;
820 nargs.firstblock = args->firstblock;
821 nargs.flist = args->flist;
822 nargs.total = args->total;
823 nargs.whichfork = XFS_ATTR_FORK;
824 nargs.trans = args->trans;
825 nargs.op_flags = XFS_DA_OP_OKNOENT;
826 entry = &leaf->entries[0];
827 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
828 if (entry->flags & XFS_ATTR_INCOMPLETE)
829 continue; /* don't copy partial entries */
830 if (!entry->nameidx)
831 continue;
832 ASSERT(entry->flags & XFS_ATTR_LOCAL);
833 name_loc = xfs_attr_leaf_name_local(leaf, i);
834 nargs.name = name_loc->nameval;
835 nargs.namelen = name_loc->namelen;
836 nargs.value = &name_loc->nameval[nargs.namelen];
837 nargs.valuelen = be16_to_cpu(name_loc->valuelen);
838 nargs.hashval = be32_to_cpu(entry->hashval);
839 nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
840 xfs_attr_shortform_add(&nargs, forkoff);
842 error = 0;
844 out:
845 kmem_free(tmpbuffer);
846 return(error);
850 * Convert from using a single leaf to a root node and a leaf.
853 xfs_attr_leaf_to_node(xfs_da_args_t *args)
855 xfs_attr_leafblock_t *leaf;
856 xfs_da_intnode_t *node;
857 xfs_inode_t *dp;
858 xfs_dabuf_t *bp1, *bp2;
859 xfs_dablk_t blkno;
860 int error;
862 trace_xfs_attr_leaf_to_node(args);
864 dp = args->dp;
865 bp1 = bp2 = NULL;
866 error = xfs_da_grow_inode(args, &blkno);
867 if (error)
868 goto out;
869 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
870 XFS_ATTR_FORK);
871 if (error)
872 goto out;
873 ASSERT(bp1 != NULL);
874 bp2 = NULL;
875 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp2,
876 XFS_ATTR_FORK);
877 if (error)
878 goto out;
879 ASSERT(bp2 != NULL);
880 memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
881 xfs_da_buf_done(bp1);
882 bp1 = NULL;
883 xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
886 * Set up the new root node.
888 error = xfs_da_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
889 if (error)
890 goto out;
891 node = bp1->data;
892 leaf = bp2->data;
893 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
894 /* both on-disk, don't endian-flip twice */
895 node->btree[0].hashval =
896 leaf->entries[be16_to_cpu(leaf->hdr.count)-1 ].hashval;
897 node->btree[0].before = cpu_to_be32(blkno);
898 node->hdr.count = cpu_to_be16(1);
899 xfs_da_log_buf(args->trans, bp1, 0, XFS_LBSIZE(dp->i_mount) - 1);
900 error = 0;
901 out:
902 if (bp1)
903 xfs_da_buf_done(bp1);
904 if (bp2)
905 xfs_da_buf_done(bp2);
906 return(error);
910 /*========================================================================
911 * Routines used for growing the Btree.
912 *========================================================================*/
915 * Create the initial contents of a leaf attribute list
916 * or a leaf in a node attribute list.
918 STATIC int
919 xfs_attr_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
921 xfs_attr_leafblock_t *leaf;
922 xfs_attr_leaf_hdr_t *hdr;
923 xfs_inode_t *dp;
924 xfs_dabuf_t *bp;
925 int error;
927 trace_xfs_attr_leaf_create(args);
929 dp = args->dp;
930 ASSERT(dp != NULL);
931 error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
932 XFS_ATTR_FORK);
933 if (error)
934 return(error);
935 ASSERT(bp != NULL);
936 leaf = bp->data;
937 memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
938 hdr = &leaf->hdr;
939 hdr->info.magic = cpu_to_be16(XFS_ATTR_LEAF_MAGIC);
940 hdr->firstused = cpu_to_be16(XFS_LBSIZE(dp->i_mount));
941 if (!hdr->firstused) {
942 hdr->firstused = cpu_to_be16(
943 XFS_LBSIZE(dp->i_mount) - XFS_ATTR_LEAF_NAME_ALIGN);
946 hdr->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
947 hdr->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr->firstused) -
948 sizeof(xfs_attr_leaf_hdr_t));
950 xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
952 *bpp = bp;
953 return(0);
957 * Split the leaf node, rebalance, then add the new entry.
960 xfs_attr_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
961 xfs_da_state_blk_t *newblk)
963 xfs_dablk_t blkno;
964 int error;
966 trace_xfs_attr_leaf_split(state->args);
969 * Allocate space for a new leaf node.
971 ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
972 error = xfs_da_grow_inode(state->args, &blkno);
973 if (error)
974 return(error);
975 error = xfs_attr_leaf_create(state->args, blkno, &newblk->bp);
976 if (error)
977 return(error);
978 newblk->blkno = blkno;
979 newblk->magic = XFS_ATTR_LEAF_MAGIC;
982 * Rebalance the entries across the two leaves.
983 * NOTE: rebalance() currently depends on the 2nd block being empty.
985 xfs_attr_leaf_rebalance(state, oldblk, newblk);
986 error = xfs_da_blk_link(state, oldblk, newblk);
987 if (error)
988 return(error);
991 * Save info on "old" attribute for "atomic rename" ops, leaf_add()
992 * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
993 * "new" attrs info. Will need the "old" info to remove it later.
995 * Insert the "new" entry in the correct block.
997 if (state->inleaf) {
998 trace_xfs_attr_leaf_add_old(state->args);
999 error = xfs_attr_leaf_add(oldblk->bp, state->args);
1000 } else {
1001 trace_xfs_attr_leaf_add_new(state->args);
1002 error = xfs_attr_leaf_add(newblk->bp, state->args);
1006 * Update last hashval in each block since we added the name.
1008 oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1009 newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1010 return(error);
1014 * Add a name to the leaf attribute list structure.
1017 xfs_attr_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args)
1019 xfs_attr_leafblock_t *leaf;
1020 xfs_attr_leaf_hdr_t *hdr;
1021 xfs_attr_leaf_map_t *map;
1022 int tablesize, entsize, sum, tmp, i;
1024 trace_xfs_attr_leaf_add(args);
1026 leaf = bp->data;
1027 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1028 ASSERT((args->index >= 0)
1029 && (args->index <= be16_to_cpu(leaf->hdr.count)));
1030 hdr = &leaf->hdr;
1031 entsize = xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1032 args->trans->t_mountp->m_sb.sb_blocksize, NULL);
1035 * Search through freemap for first-fit on new name length.
1036 * (may need to figure in size of entry struct too)
1038 tablesize = (be16_to_cpu(hdr->count) + 1)
1039 * sizeof(xfs_attr_leaf_entry_t)
1040 + sizeof(xfs_attr_leaf_hdr_t);
1041 map = &hdr->freemap[XFS_ATTR_LEAF_MAPSIZE-1];
1042 for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
1043 if (tablesize > be16_to_cpu(hdr->firstused)) {
1044 sum += be16_to_cpu(map->size);
1045 continue;
1047 if (!map->size)
1048 continue; /* no space in this map */
1049 tmp = entsize;
1050 if (be16_to_cpu(map->base) < be16_to_cpu(hdr->firstused))
1051 tmp += sizeof(xfs_attr_leaf_entry_t);
1052 if (be16_to_cpu(map->size) >= tmp) {
1053 tmp = xfs_attr_leaf_add_work(bp, args, i);
1054 return(tmp);
1056 sum += be16_to_cpu(map->size);
1060 * If there are no holes in the address space of the block,
1061 * and we don't have enough freespace, then compaction will do us
1062 * no good and we should just give up.
1064 if (!hdr->holes && (sum < entsize))
1065 return(XFS_ERROR(ENOSPC));
1068 * Compact the entries to coalesce free space.
1069 * This may change the hdr->count via dropping INCOMPLETE entries.
1071 xfs_attr_leaf_compact(args->trans, bp);
1074 * After compaction, the block is guaranteed to have only one
1075 * free region, in freemap[0]. If it is not big enough, give up.
1077 if (be16_to_cpu(hdr->freemap[0].size)
1078 < (entsize + sizeof(xfs_attr_leaf_entry_t)))
1079 return(XFS_ERROR(ENOSPC));
1081 return(xfs_attr_leaf_add_work(bp, args, 0));
1085 * Add a name to a leaf attribute list structure.
1087 STATIC int
1088 xfs_attr_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int mapindex)
1090 xfs_attr_leafblock_t *leaf;
1091 xfs_attr_leaf_hdr_t *hdr;
1092 xfs_attr_leaf_entry_t *entry;
1093 xfs_attr_leaf_name_local_t *name_loc;
1094 xfs_attr_leaf_name_remote_t *name_rmt;
1095 xfs_attr_leaf_map_t *map;
1096 xfs_mount_t *mp;
1097 int tmp, i;
1099 leaf = bp->data;
1100 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1101 hdr = &leaf->hdr;
1102 ASSERT((mapindex >= 0) && (mapindex < XFS_ATTR_LEAF_MAPSIZE));
1103 ASSERT((args->index >= 0) && (args->index <= be16_to_cpu(hdr->count)));
1106 * Force open some space in the entry array and fill it in.
1108 entry = &leaf->entries[args->index];
1109 if (args->index < be16_to_cpu(hdr->count)) {
1110 tmp = be16_to_cpu(hdr->count) - args->index;
1111 tmp *= sizeof(xfs_attr_leaf_entry_t);
1112 memmove((char *)(entry+1), (char *)entry, tmp);
1113 xfs_da_log_buf(args->trans, bp,
1114 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1116 be16_add_cpu(&hdr->count, 1);
1119 * Allocate space for the new string (at the end of the run).
1121 map = &hdr->freemap[mapindex];
1122 mp = args->trans->t_mountp;
1123 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1124 ASSERT((be16_to_cpu(map->base) & 0x3) == 0);
1125 ASSERT(be16_to_cpu(map->size) >=
1126 xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1127 mp->m_sb.sb_blocksize, NULL));
1128 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1129 ASSERT((be16_to_cpu(map->size) & 0x3) == 0);
1130 be16_add_cpu(&map->size,
1131 -xfs_attr_leaf_newentsize(args->namelen, args->valuelen,
1132 mp->m_sb.sb_blocksize, &tmp));
1133 entry->nameidx = cpu_to_be16(be16_to_cpu(map->base) +
1134 be16_to_cpu(map->size));
1135 entry->hashval = cpu_to_be32(args->hashval);
1136 entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1137 entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1138 if (args->op_flags & XFS_DA_OP_RENAME) {
1139 entry->flags |= XFS_ATTR_INCOMPLETE;
1140 if ((args->blkno2 == args->blkno) &&
1141 (args->index2 <= args->index)) {
1142 args->index2++;
1145 xfs_da_log_buf(args->trans, bp,
1146 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1147 ASSERT((args->index == 0) ||
1148 (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1149 ASSERT((args->index == be16_to_cpu(hdr->count)-1) ||
1150 (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1153 * For "remote" attribute values, simply note that we need to
1154 * allocate space for the "remote" value. We can't actually
1155 * allocate the extents in this transaction, and we can't decide
1156 * which blocks they should be as we might allocate more blocks
1157 * as part of this transaction (a split operation for example).
1159 if (entry->flags & XFS_ATTR_LOCAL) {
1160 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
1161 name_loc->namelen = args->namelen;
1162 name_loc->valuelen = cpu_to_be16(args->valuelen);
1163 memcpy((char *)name_loc->nameval, args->name, args->namelen);
1164 memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1165 be16_to_cpu(name_loc->valuelen));
1166 } else {
1167 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
1168 name_rmt->namelen = args->namelen;
1169 memcpy((char *)name_rmt->name, args->name, args->namelen);
1170 entry->flags |= XFS_ATTR_INCOMPLETE;
1171 /* just in case */
1172 name_rmt->valuelen = 0;
1173 name_rmt->valueblk = 0;
1174 args->rmtblkno = 1;
1175 args->rmtblkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1177 xfs_da_log_buf(args->trans, bp,
1178 XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1179 xfs_attr_leaf_entsize(leaf, args->index)));
1182 * Update the control info for this leaf node
1184 if (be16_to_cpu(entry->nameidx) < be16_to_cpu(hdr->firstused)) {
1185 /* both on-disk, don't endian-flip twice */
1186 hdr->firstused = entry->nameidx;
1188 ASSERT(be16_to_cpu(hdr->firstused) >=
1189 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1190 tmp = (be16_to_cpu(hdr->count)-1) * sizeof(xfs_attr_leaf_entry_t)
1191 + sizeof(xfs_attr_leaf_hdr_t);
1192 map = &hdr->freemap[0];
1193 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1194 if (be16_to_cpu(map->base) == tmp) {
1195 be16_add_cpu(&map->base, sizeof(xfs_attr_leaf_entry_t));
1196 be16_add_cpu(&map->size,
1197 -((int)sizeof(xfs_attr_leaf_entry_t)));
1200 be16_add_cpu(&hdr->usedbytes, xfs_attr_leaf_entsize(leaf, args->index));
1201 xfs_da_log_buf(args->trans, bp,
1202 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1203 return(0);
1207 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1209 STATIC void
1210 xfs_attr_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp)
1212 xfs_attr_leafblock_t *leaf_s, *leaf_d;
1213 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
1214 xfs_mount_t *mp;
1215 char *tmpbuffer;
1217 mp = trans->t_mountp;
1218 tmpbuffer = kmem_alloc(XFS_LBSIZE(mp), KM_SLEEP);
1219 ASSERT(tmpbuffer != NULL);
1220 memcpy(tmpbuffer, bp->data, XFS_LBSIZE(mp));
1221 memset(bp->data, 0, XFS_LBSIZE(mp));
1224 * Copy basic information
1226 leaf_s = (xfs_attr_leafblock_t *)tmpbuffer;
1227 leaf_d = bp->data;
1228 hdr_s = &leaf_s->hdr;
1229 hdr_d = &leaf_d->hdr;
1230 hdr_d->info = hdr_s->info; /* struct copy */
1231 hdr_d->firstused = cpu_to_be16(XFS_LBSIZE(mp));
1232 /* handle truncation gracefully */
1233 if (!hdr_d->firstused) {
1234 hdr_d->firstused = cpu_to_be16(
1235 XFS_LBSIZE(mp) - XFS_ATTR_LEAF_NAME_ALIGN);
1237 hdr_d->usedbytes = 0;
1238 hdr_d->count = 0;
1239 hdr_d->holes = 0;
1240 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
1241 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused) -
1242 sizeof(xfs_attr_leaf_hdr_t));
1245 * Copy all entry's in the same (sorted) order,
1246 * but allocate name/value pairs packed and in sequence.
1248 xfs_attr_leaf_moveents(leaf_s, 0, leaf_d, 0,
1249 be16_to_cpu(hdr_s->count), mp);
1250 xfs_da_log_buf(trans, bp, 0, XFS_LBSIZE(mp) - 1);
1252 kmem_free(tmpbuffer);
1256 * Redistribute the attribute list entries between two leaf nodes,
1257 * taking into account the size of the new entry.
1259 * NOTE: if new block is empty, then it will get the upper half of the
1260 * old block. At present, all (one) callers pass in an empty second block.
1262 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1263 * to match what it is doing in splitting the attribute leaf block. Those
1264 * values are used in "atomic rename" operations on attributes. Note that
1265 * the "new" and "old" values can end up in different blocks.
1267 STATIC void
1268 xfs_attr_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1269 xfs_da_state_blk_t *blk2)
1271 xfs_da_args_t *args;
1272 xfs_da_state_blk_t *tmp_blk;
1273 xfs_attr_leafblock_t *leaf1, *leaf2;
1274 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1275 int count, totallen, max, space, swap;
1278 * Set up environment.
1280 ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1281 ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1282 leaf1 = blk1->bp->data;
1283 leaf2 = blk2->bp->data;
1284 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1285 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1286 args = state->args;
1288 trace_xfs_attr_leaf_rebalance(args);
1291 * Check ordering of blocks, reverse if it makes things simpler.
1293 * NOTE: Given that all (current) callers pass in an empty
1294 * second block, this code should never set "swap".
1296 swap = 0;
1297 if (xfs_attr_leaf_order(blk1->bp, blk2->bp)) {
1298 tmp_blk = blk1;
1299 blk1 = blk2;
1300 blk2 = tmp_blk;
1301 leaf1 = blk1->bp->data;
1302 leaf2 = blk2->bp->data;
1303 swap = 1;
1305 hdr1 = &leaf1->hdr;
1306 hdr2 = &leaf2->hdr;
1309 * Examine entries until we reduce the absolute difference in
1310 * byte usage between the two blocks to a minimum. Then get
1311 * the direction to copy and the number of elements to move.
1313 * "inleaf" is true if the new entry should be inserted into blk1.
1314 * If "swap" is also true, then reverse the sense of "inleaf".
1316 state->inleaf = xfs_attr_leaf_figure_balance(state, blk1, blk2,
1317 &count, &totallen);
1318 if (swap)
1319 state->inleaf = !state->inleaf;
1322 * Move any entries required from leaf to leaf:
1324 if (count < be16_to_cpu(hdr1->count)) {
1326 * Figure the total bytes to be added to the destination leaf.
1328 /* number entries being moved */
1329 count = be16_to_cpu(hdr1->count) - count;
1330 space = be16_to_cpu(hdr1->usedbytes) - totallen;
1331 space += count * sizeof(xfs_attr_leaf_entry_t);
1334 * leaf2 is the destination, compact it if it looks tight.
1336 max = be16_to_cpu(hdr2->firstused)
1337 - sizeof(xfs_attr_leaf_hdr_t);
1338 max -= be16_to_cpu(hdr2->count) * sizeof(xfs_attr_leaf_entry_t);
1339 if (space > max) {
1340 xfs_attr_leaf_compact(args->trans, blk2->bp);
1344 * Move high entries from leaf1 to low end of leaf2.
1346 xfs_attr_leaf_moveents(leaf1, be16_to_cpu(hdr1->count) - count,
1347 leaf2, 0, count, state->mp);
1349 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1350 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1351 } else if (count > be16_to_cpu(hdr1->count)) {
1353 * I assert that since all callers pass in an empty
1354 * second buffer, this code should never execute.
1358 * Figure the total bytes to be added to the destination leaf.
1360 /* number entries being moved */
1361 count -= be16_to_cpu(hdr1->count);
1362 space = totallen - be16_to_cpu(hdr1->usedbytes);
1363 space += count * sizeof(xfs_attr_leaf_entry_t);
1366 * leaf1 is the destination, compact it if it looks tight.
1368 max = be16_to_cpu(hdr1->firstused)
1369 - sizeof(xfs_attr_leaf_hdr_t);
1370 max -= be16_to_cpu(hdr1->count) * sizeof(xfs_attr_leaf_entry_t);
1371 if (space > max) {
1372 xfs_attr_leaf_compact(args->trans, blk1->bp);
1376 * Move low entries from leaf2 to high end of leaf1.
1378 xfs_attr_leaf_moveents(leaf2, 0, leaf1,
1379 be16_to_cpu(hdr1->count), count, state->mp);
1381 xfs_da_log_buf(args->trans, blk1->bp, 0, state->blocksize-1);
1382 xfs_da_log_buf(args->trans, blk2->bp, 0, state->blocksize-1);
1386 * Copy out last hashval in each block for B-tree code.
1388 blk1->hashval = be32_to_cpu(
1389 leaf1->entries[be16_to_cpu(leaf1->hdr.count)-1].hashval);
1390 blk2->hashval = be32_to_cpu(
1391 leaf2->entries[be16_to_cpu(leaf2->hdr.count)-1].hashval);
1394 * Adjust the expected index for insertion.
1395 * NOTE: this code depends on the (current) situation that the
1396 * second block was originally empty.
1398 * If the insertion point moved to the 2nd block, we must adjust
1399 * the index. We must also track the entry just following the
1400 * new entry for use in an "atomic rename" operation, that entry
1401 * is always the "old" entry and the "new" entry is what we are
1402 * inserting. The index/blkno fields refer to the "old" entry,
1403 * while the index2/blkno2 fields refer to the "new" entry.
1405 if (blk1->index > be16_to_cpu(leaf1->hdr.count)) {
1406 ASSERT(state->inleaf == 0);
1407 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
1408 args->index = args->index2 = blk2->index;
1409 args->blkno = args->blkno2 = blk2->blkno;
1410 } else if (blk1->index == be16_to_cpu(leaf1->hdr.count)) {
1411 if (state->inleaf) {
1412 args->index = blk1->index;
1413 args->blkno = blk1->blkno;
1414 args->index2 = 0;
1415 args->blkno2 = blk2->blkno;
1416 } else {
1417 blk2->index = blk1->index
1418 - be16_to_cpu(leaf1->hdr.count);
1419 args->index = args->index2 = blk2->index;
1420 args->blkno = args->blkno2 = blk2->blkno;
1422 } else {
1423 ASSERT(state->inleaf == 1);
1424 args->index = args->index2 = blk1->index;
1425 args->blkno = args->blkno2 = blk1->blkno;
1430 * Examine entries until we reduce the absolute difference in
1431 * byte usage between the two blocks to a minimum.
1432 * GROT: Is this really necessary? With other than a 512 byte blocksize,
1433 * GROT: there will always be enough room in either block for a new entry.
1434 * GROT: Do a double-split for this case?
1436 STATIC int
1437 xfs_attr_leaf_figure_balance(xfs_da_state_t *state,
1438 xfs_da_state_blk_t *blk1,
1439 xfs_da_state_blk_t *blk2,
1440 int *countarg, int *usedbytesarg)
1442 xfs_attr_leafblock_t *leaf1, *leaf2;
1443 xfs_attr_leaf_hdr_t *hdr1, *hdr2;
1444 xfs_attr_leaf_entry_t *entry;
1445 int count, max, index, totallen, half;
1446 int lastdelta, foundit, tmp;
1449 * Set up environment.
1451 leaf1 = blk1->bp->data;
1452 leaf2 = blk2->bp->data;
1453 hdr1 = &leaf1->hdr;
1454 hdr2 = &leaf2->hdr;
1455 foundit = 0;
1456 totallen = 0;
1459 * Examine entries until we reduce the absolute difference in
1460 * byte usage between the two blocks to a minimum.
1462 max = be16_to_cpu(hdr1->count) + be16_to_cpu(hdr2->count);
1463 half = (max+1) * sizeof(*entry);
1464 half += be16_to_cpu(hdr1->usedbytes) +
1465 be16_to_cpu(hdr2->usedbytes) +
1466 xfs_attr_leaf_newentsize(
1467 state->args->namelen,
1468 state->args->valuelen,
1469 state->blocksize, NULL);
1470 half /= 2;
1471 lastdelta = state->blocksize;
1472 entry = &leaf1->entries[0];
1473 for (count = index = 0; count < max; entry++, index++, count++) {
1475 #define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1477 * The new entry is in the first block, account for it.
1479 if (count == blk1->index) {
1480 tmp = totallen + sizeof(*entry) +
1481 xfs_attr_leaf_newentsize(
1482 state->args->namelen,
1483 state->args->valuelen,
1484 state->blocksize, NULL);
1485 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1486 break;
1487 lastdelta = XFS_ATTR_ABS(half - tmp);
1488 totallen = tmp;
1489 foundit = 1;
1493 * Wrap around into the second block if necessary.
1495 if (count == be16_to_cpu(hdr1->count)) {
1496 leaf1 = leaf2;
1497 entry = &leaf1->entries[0];
1498 index = 0;
1502 * Figure out if next leaf entry would be too much.
1504 tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1505 index);
1506 if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1507 break;
1508 lastdelta = XFS_ATTR_ABS(half - tmp);
1509 totallen = tmp;
1510 #undef XFS_ATTR_ABS
1514 * Calculate the number of usedbytes that will end up in lower block.
1515 * If new entry not in lower block, fix up the count.
1517 totallen -= count * sizeof(*entry);
1518 if (foundit) {
1519 totallen -= sizeof(*entry) +
1520 xfs_attr_leaf_newentsize(
1521 state->args->namelen,
1522 state->args->valuelen,
1523 state->blocksize, NULL);
1526 *countarg = count;
1527 *usedbytesarg = totallen;
1528 return(foundit);
1531 /*========================================================================
1532 * Routines used for shrinking the Btree.
1533 *========================================================================*/
1536 * Check a leaf block and its neighbors to see if the block should be
1537 * collapsed into one or the other neighbor. Always keep the block
1538 * with the smaller block number.
1539 * If the current block is over 50% full, don't try to join it, return 0.
1540 * If the block is empty, fill in the state structure and return 2.
1541 * If it can be collapsed, fill in the state structure and return 1.
1542 * If nothing can be done, return 0.
1544 * GROT: allow for INCOMPLETE entries in calculation.
1547 xfs_attr_leaf_toosmall(xfs_da_state_t *state, int *action)
1549 xfs_attr_leafblock_t *leaf;
1550 xfs_da_state_blk_t *blk;
1551 xfs_da_blkinfo_t *info;
1552 int count, bytes, forward, error, retval, i;
1553 xfs_dablk_t blkno;
1554 xfs_dabuf_t *bp;
1557 * Check for the degenerate case of the block being over 50% full.
1558 * If so, it's not worth even looking to see if we might be able
1559 * to coalesce with a sibling.
1561 blk = &state->path.blk[ state->path.active-1 ];
1562 info = blk->bp->data;
1563 ASSERT(info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1564 leaf = (xfs_attr_leafblock_t *)info;
1565 count = be16_to_cpu(leaf->hdr.count);
1566 bytes = sizeof(xfs_attr_leaf_hdr_t) +
1567 count * sizeof(xfs_attr_leaf_entry_t) +
1568 be16_to_cpu(leaf->hdr.usedbytes);
1569 if (bytes > (state->blocksize >> 1)) {
1570 *action = 0; /* blk over 50%, don't try to join */
1571 return(0);
1575 * Check for the degenerate case of the block being empty.
1576 * If the block is empty, we'll simply delete it, no need to
1577 * coalesce it with a sibling block. We choose (arbitrarily)
1578 * to merge with the forward block unless it is NULL.
1580 if (count == 0) {
1582 * Make altpath point to the block we want to keep and
1583 * path point to the block we want to drop (this one).
1585 forward = (info->forw != 0);
1586 memcpy(&state->altpath, &state->path, sizeof(state->path));
1587 error = xfs_da_path_shift(state, &state->altpath, forward,
1588 0, &retval);
1589 if (error)
1590 return(error);
1591 if (retval) {
1592 *action = 0;
1593 } else {
1594 *action = 2;
1596 return(0);
1600 * Examine each sibling block to see if we can coalesce with
1601 * at least 25% free space to spare. We need to figure out
1602 * whether to merge with the forward or the backward block.
1603 * We prefer coalescing with the lower numbered sibling so as
1604 * to shrink an attribute list over time.
1606 /* start with smaller blk num */
1607 forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));
1608 for (i = 0; i < 2; forward = !forward, i++) {
1609 if (forward)
1610 blkno = be32_to_cpu(info->forw);
1611 else
1612 blkno = be32_to_cpu(info->back);
1613 if (blkno == 0)
1614 continue;
1615 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1616 blkno, -1, &bp, XFS_ATTR_FORK);
1617 if (error)
1618 return(error);
1619 ASSERT(bp != NULL);
1621 leaf = (xfs_attr_leafblock_t *)info;
1622 count = be16_to_cpu(leaf->hdr.count);
1623 bytes = state->blocksize - (state->blocksize>>2);
1624 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1625 leaf = bp->data;
1626 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1627 count += be16_to_cpu(leaf->hdr.count);
1628 bytes -= be16_to_cpu(leaf->hdr.usedbytes);
1629 bytes -= count * sizeof(xfs_attr_leaf_entry_t);
1630 bytes -= sizeof(xfs_attr_leaf_hdr_t);
1631 xfs_da_brelse(state->args->trans, bp);
1632 if (bytes >= 0)
1633 break; /* fits with at least 25% to spare */
1635 if (i >= 2) {
1636 *action = 0;
1637 return(0);
1641 * Make altpath point to the block we want to keep (the lower
1642 * numbered block) and path point to the block we want to drop.
1644 memcpy(&state->altpath, &state->path, sizeof(state->path));
1645 if (blkno < blk->blkno) {
1646 error = xfs_da_path_shift(state, &state->altpath, forward,
1647 0, &retval);
1648 } else {
1649 error = xfs_da_path_shift(state, &state->path, forward,
1650 0, &retval);
1652 if (error)
1653 return(error);
1654 if (retval) {
1655 *action = 0;
1656 } else {
1657 *action = 1;
1659 return(0);
1663 * Remove a name from the leaf attribute list structure.
1665 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1666 * If two leaves are 37% full, when combined they will leave 25% free.
1669 xfs_attr_leaf_remove(xfs_dabuf_t *bp, xfs_da_args_t *args)
1671 xfs_attr_leafblock_t *leaf;
1672 xfs_attr_leaf_hdr_t *hdr;
1673 xfs_attr_leaf_map_t *map;
1674 xfs_attr_leaf_entry_t *entry;
1675 int before, after, smallest, entsize;
1676 int tablesize, tmp, i;
1677 xfs_mount_t *mp;
1679 leaf = bp->data;
1680 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1681 hdr = &leaf->hdr;
1682 mp = args->trans->t_mountp;
1683 ASSERT((be16_to_cpu(hdr->count) > 0)
1684 && (be16_to_cpu(hdr->count) < (XFS_LBSIZE(mp)/8)));
1685 ASSERT((args->index >= 0)
1686 && (args->index < be16_to_cpu(hdr->count)));
1687 ASSERT(be16_to_cpu(hdr->firstused) >=
1688 ((be16_to_cpu(hdr->count) * sizeof(*entry)) + sizeof(*hdr)));
1689 entry = &leaf->entries[args->index];
1690 ASSERT(be16_to_cpu(entry->nameidx) >= be16_to_cpu(hdr->firstused));
1691 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1694 * Scan through free region table:
1695 * check for adjacency of free'd entry with an existing one,
1696 * find smallest free region in case we need to replace it,
1697 * adjust any map that borders the entry table,
1699 tablesize = be16_to_cpu(hdr->count) * sizeof(xfs_attr_leaf_entry_t)
1700 + sizeof(xfs_attr_leaf_hdr_t);
1701 map = &hdr->freemap[0];
1702 tmp = be16_to_cpu(map->size);
1703 before = after = -1;
1704 smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1705 entsize = xfs_attr_leaf_entsize(leaf, args->index);
1706 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; map++, i++) {
1707 ASSERT(be16_to_cpu(map->base) < XFS_LBSIZE(mp));
1708 ASSERT(be16_to_cpu(map->size) < XFS_LBSIZE(mp));
1709 if (be16_to_cpu(map->base) == tablesize) {
1710 be16_add_cpu(&map->base,
1711 -((int)sizeof(xfs_attr_leaf_entry_t)));
1712 be16_add_cpu(&map->size, sizeof(xfs_attr_leaf_entry_t));
1715 if ((be16_to_cpu(map->base) + be16_to_cpu(map->size))
1716 == be16_to_cpu(entry->nameidx)) {
1717 before = i;
1718 } else if (be16_to_cpu(map->base)
1719 == (be16_to_cpu(entry->nameidx) + entsize)) {
1720 after = i;
1721 } else if (be16_to_cpu(map->size) < tmp) {
1722 tmp = be16_to_cpu(map->size);
1723 smallest = i;
1728 * Coalesce adjacent freemap regions,
1729 * or replace the smallest region.
1731 if ((before >= 0) || (after >= 0)) {
1732 if ((before >= 0) && (after >= 0)) {
1733 map = &hdr->freemap[before];
1734 be16_add_cpu(&map->size, entsize);
1735 be16_add_cpu(&map->size,
1736 be16_to_cpu(hdr->freemap[after].size));
1737 hdr->freemap[after].base = 0;
1738 hdr->freemap[after].size = 0;
1739 } else if (before >= 0) {
1740 map = &hdr->freemap[before];
1741 be16_add_cpu(&map->size, entsize);
1742 } else {
1743 map = &hdr->freemap[after];
1744 /* both on-disk, don't endian flip twice */
1745 map->base = entry->nameidx;
1746 be16_add_cpu(&map->size, entsize);
1748 } else {
1750 * Replace smallest region (if it is smaller than free'd entry)
1752 map = &hdr->freemap[smallest];
1753 if (be16_to_cpu(map->size) < entsize) {
1754 map->base = cpu_to_be16(be16_to_cpu(entry->nameidx));
1755 map->size = cpu_to_be16(entsize);
1760 * Did we remove the first entry?
1762 if (be16_to_cpu(entry->nameidx) == be16_to_cpu(hdr->firstused))
1763 smallest = 1;
1764 else
1765 smallest = 0;
1768 * Compress the remaining entries and zero out the removed stuff.
1770 memset(xfs_attr_leaf_name(leaf, args->index), 0, entsize);
1771 be16_add_cpu(&hdr->usedbytes, -entsize);
1772 xfs_da_log_buf(args->trans, bp,
1773 XFS_DA_LOGRANGE(leaf, xfs_attr_leaf_name(leaf, args->index),
1774 entsize));
1776 tmp = (be16_to_cpu(hdr->count) - args->index)
1777 * sizeof(xfs_attr_leaf_entry_t);
1778 memmove((char *)entry, (char *)(entry+1), tmp);
1779 be16_add_cpu(&hdr->count, -1);
1780 xfs_da_log_buf(args->trans, bp,
1781 XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1782 entry = &leaf->entries[be16_to_cpu(hdr->count)];
1783 memset((char *)entry, 0, sizeof(xfs_attr_leaf_entry_t));
1786 * If we removed the first entry, re-find the first used byte
1787 * in the name area. Note that if the entry was the "firstused",
1788 * then we don't have a "hole" in our block resulting from
1789 * removing the name.
1791 if (smallest) {
1792 tmp = XFS_LBSIZE(mp);
1793 entry = &leaf->entries[0];
1794 for (i = be16_to_cpu(hdr->count)-1; i >= 0; entry++, i--) {
1795 ASSERT(be16_to_cpu(entry->nameidx) >=
1796 be16_to_cpu(hdr->firstused));
1797 ASSERT(be16_to_cpu(entry->nameidx) < XFS_LBSIZE(mp));
1799 if (be16_to_cpu(entry->nameidx) < tmp)
1800 tmp = be16_to_cpu(entry->nameidx);
1802 hdr->firstused = cpu_to_be16(tmp);
1803 if (!hdr->firstused) {
1804 hdr->firstused = cpu_to_be16(
1805 tmp - XFS_ATTR_LEAF_NAME_ALIGN);
1807 } else {
1808 hdr->holes = 1; /* mark as needing compaction */
1810 xfs_da_log_buf(args->trans, bp,
1811 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1814 * Check if leaf is less than 50% full, caller may want to
1815 * "join" the leaf with a sibling if so.
1817 tmp = sizeof(xfs_attr_leaf_hdr_t);
1818 tmp += be16_to_cpu(leaf->hdr.count) * sizeof(xfs_attr_leaf_entry_t);
1819 tmp += be16_to_cpu(leaf->hdr.usedbytes);
1820 return(tmp < mp->m_attr_magicpct); /* leaf is < 37% full */
1824 * Move all the attribute list entries from drop_leaf into save_leaf.
1826 void
1827 xfs_attr_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1828 xfs_da_state_blk_t *save_blk)
1830 xfs_attr_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1831 xfs_attr_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1832 xfs_mount_t *mp;
1833 char *tmpbuffer;
1835 trace_xfs_attr_leaf_unbalance(state->args);
1838 * Set up environment.
1840 mp = state->mp;
1841 ASSERT(drop_blk->magic == XFS_ATTR_LEAF_MAGIC);
1842 ASSERT(save_blk->magic == XFS_ATTR_LEAF_MAGIC);
1843 drop_leaf = drop_blk->bp->data;
1844 save_leaf = save_blk->bp->data;
1845 ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1846 ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1847 drop_hdr = &drop_leaf->hdr;
1848 save_hdr = &save_leaf->hdr;
1851 * Save last hashval from dying block for later Btree fixup.
1853 drop_blk->hashval = be32_to_cpu(
1854 drop_leaf->entries[be16_to_cpu(drop_leaf->hdr.count)-1].hashval);
1857 * Check if we need a temp buffer, or can we do it in place.
1858 * Note that we don't check "leaf" for holes because we will
1859 * always be dropping it, toosmall() decided that for us already.
1861 if (save_hdr->holes == 0) {
1863 * dest leaf has no holes, so we add there. May need
1864 * to make some room in the entry array.
1866 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1867 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1868 be16_to_cpu(drop_hdr->count), mp);
1869 } else {
1870 xfs_attr_leaf_moveents(drop_leaf, 0, save_leaf,
1871 be16_to_cpu(save_hdr->count),
1872 be16_to_cpu(drop_hdr->count), mp);
1874 } else {
1876 * Destination has holes, so we make a temporary copy
1877 * of the leaf and add them both to that.
1879 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1880 ASSERT(tmpbuffer != NULL);
1881 memset(tmpbuffer, 0, state->blocksize);
1882 tmp_leaf = (xfs_attr_leafblock_t *)tmpbuffer;
1883 tmp_hdr = &tmp_leaf->hdr;
1884 tmp_hdr->info = save_hdr->info; /* struct copy */
1885 tmp_hdr->count = 0;
1886 tmp_hdr->firstused = cpu_to_be16(state->blocksize);
1887 if (!tmp_hdr->firstused) {
1888 tmp_hdr->firstused = cpu_to_be16(
1889 state->blocksize - XFS_ATTR_LEAF_NAME_ALIGN);
1891 tmp_hdr->usedbytes = 0;
1892 if (xfs_attr_leaf_order(save_blk->bp, drop_blk->bp)) {
1893 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1894 be16_to_cpu(drop_hdr->count), mp);
1895 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf,
1896 be16_to_cpu(tmp_leaf->hdr.count),
1897 be16_to_cpu(save_hdr->count), mp);
1898 } else {
1899 xfs_attr_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1900 be16_to_cpu(save_hdr->count), mp);
1901 xfs_attr_leaf_moveents(drop_leaf, 0, tmp_leaf,
1902 be16_to_cpu(tmp_leaf->hdr.count),
1903 be16_to_cpu(drop_hdr->count), mp);
1905 memcpy((char *)save_leaf, (char *)tmp_leaf, state->blocksize);
1906 kmem_free(tmpbuffer);
1909 xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1910 state->blocksize - 1);
1913 * Copy out last hashval in each block for B-tree code.
1915 save_blk->hashval = be32_to_cpu(
1916 save_leaf->entries[be16_to_cpu(save_leaf->hdr.count)-1].hashval);
1919 /*========================================================================
1920 * Routines used for finding things in the Btree.
1921 *========================================================================*/
1924 * Look up a name in a leaf attribute list structure.
1925 * This is the internal routine, it uses the caller's buffer.
1927 * Note that duplicate keys are allowed, but only check within the
1928 * current leaf node. The Btree code must check in adjacent leaf nodes.
1930 * Return in args->index the index into the entry[] array of either
1931 * the found entry, or where the entry should have been (insert before
1932 * that entry).
1934 * Don't change the args->value unless we find the attribute.
1937 xfs_attr_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args)
1939 xfs_attr_leafblock_t *leaf;
1940 xfs_attr_leaf_entry_t *entry;
1941 xfs_attr_leaf_name_local_t *name_loc;
1942 xfs_attr_leaf_name_remote_t *name_rmt;
1943 int probe, span;
1944 xfs_dahash_t hashval;
1946 trace_xfs_attr_leaf_lookup(args);
1948 leaf = bp->data;
1949 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
1950 ASSERT(be16_to_cpu(leaf->hdr.count)
1951 < (XFS_LBSIZE(args->dp->i_mount)/8));
1954 * Binary search. (note: small blocks will skip this loop)
1956 hashval = args->hashval;
1957 probe = span = be16_to_cpu(leaf->hdr.count) / 2;
1958 for (entry = &leaf->entries[probe]; span > 4;
1959 entry = &leaf->entries[probe]) {
1960 span /= 2;
1961 if (be32_to_cpu(entry->hashval) < hashval)
1962 probe += span;
1963 else if (be32_to_cpu(entry->hashval) > hashval)
1964 probe -= span;
1965 else
1966 break;
1968 ASSERT((probe >= 0) &&
1969 (!leaf->hdr.count
1970 || (probe < be16_to_cpu(leaf->hdr.count))));
1971 ASSERT((span <= 4) || (be32_to_cpu(entry->hashval) == hashval));
1974 * Since we may have duplicate hashval's, find the first matching
1975 * hashval in the leaf.
1977 while ((probe > 0) && (be32_to_cpu(entry->hashval) >= hashval)) {
1978 entry--;
1979 probe--;
1981 while ((probe < be16_to_cpu(leaf->hdr.count)) &&
1982 (be32_to_cpu(entry->hashval) < hashval)) {
1983 entry++;
1984 probe++;
1986 if ((probe == be16_to_cpu(leaf->hdr.count)) ||
1987 (be32_to_cpu(entry->hashval) != hashval)) {
1988 args->index = probe;
1989 return(XFS_ERROR(ENOATTR));
1993 * Duplicate keys may be present, so search all of them for a match.
1995 for ( ; (probe < be16_to_cpu(leaf->hdr.count)) &&
1996 (be32_to_cpu(entry->hashval) == hashval);
1997 entry++, probe++) {
1999 * GROT: Add code to remove incomplete entries.
2002 * If we are looking for INCOMPLETE entries, show only those.
2003 * If we are looking for complete entries, show only those.
2005 if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2006 (entry->flags & XFS_ATTR_INCOMPLETE)) {
2007 continue;
2009 if (entry->flags & XFS_ATTR_LOCAL) {
2010 name_loc = xfs_attr_leaf_name_local(leaf, probe);
2011 if (name_loc->namelen != args->namelen)
2012 continue;
2013 if (memcmp(args->name, (char *)name_loc->nameval, args->namelen) != 0)
2014 continue;
2015 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2016 continue;
2017 args->index = probe;
2018 return(XFS_ERROR(EEXIST));
2019 } else {
2020 name_rmt = xfs_attr_leaf_name_remote(leaf, probe);
2021 if (name_rmt->namelen != args->namelen)
2022 continue;
2023 if (memcmp(args->name, (char *)name_rmt->name,
2024 args->namelen) != 0)
2025 continue;
2026 if (!xfs_attr_namesp_match(args->flags, entry->flags))
2027 continue;
2028 args->index = probe;
2029 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2030 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount,
2031 be32_to_cpu(name_rmt->valuelen));
2032 return(XFS_ERROR(EEXIST));
2035 args->index = probe;
2036 return(XFS_ERROR(ENOATTR));
2040 * Get the value associated with an attribute name from a leaf attribute
2041 * list structure.
2044 xfs_attr_leaf_getvalue(xfs_dabuf_t *bp, xfs_da_args_t *args)
2046 int valuelen;
2047 xfs_attr_leafblock_t *leaf;
2048 xfs_attr_leaf_entry_t *entry;
2049 xfs_attr_leaf_name_local_t *name_loc;
2050 xfs_attr_leaf_name_remote_t *name_rmt;
2052 leaf = bp->data;
2053 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2054 ASSERT(be16_to_cpu(leaf->hdr.count)
2055 < (XFS_LBSIZE(args->dp->i_mount)/8));
2056 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2058 entry = &leaf->entries[args->index];
2059 if (entry->flags & XFS_ATTR_LOCAL) {
2060 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2061 ASSERT(name_loc->namelen == args->namelen);
2062 ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2063 valuelen = be16_to_cpu(name_loc->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;
2073 memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2074 } else {
2075 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2076 ASSERT(name_rmt->namelen == args->namelen);
2077 ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2078 valuelen = be32_to_cpu(name_rmt->valuelen);
2079 args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2080 args->rmtblkcnt = XFS_B_TO_FSB(args->dp->i_mount, valuelen);
2081 if (args->flags & ATTR_KERNOVAL) {
2082 args->valuelen = valuelen;
2083 return(0);
2085 if (args->valuelen < valuelen) {
2086 args->valuelen = valuelen;
2087 return(XFS_ERROR(ERANGE));
2089 args->valuelen = valuelen;
2091 return(0);
2094 /*========================================================================
2095 * Utility routines.
2096 *========================================================================*/
2099 * Move the indicated entries from one leaf to another.
2100 * NOTE: this routine modifies both source and destination leaves.
2102 /*ARGSUSED*/
2103 STATIC void
2104 xfs_attr_leaf_moveents(xfs_attr_leafblock_t *leaf_s, int start_s,
2105 xfs_attr_leafblock_t *leaf_d, int start_d,
2106 int count, xfs_mount_t *mp)
2108 xfs_attr_leaf_hdr_t *hdr_s, *hdr_d;
2109 xfs_attr_leaf_entry_t *entry_s, *entry_d;
2110 int desti, tmp, i;
2113 * Check for nothing to do.
2115 if (count == 0)
2116 return;
2119 * Set up environment.
2121 ASSERT(leaf_s->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2122 ASSERT(leaf_d->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2123 hdr_s = &leaf_s->hdr;
2124 hdr_d = &leaf_d->hdr;
2125 ASSERT((be16_to_cpu(hdr_s->count) > 0) &&
2126 (be16_to_cpu(hdr_s->count) < (XFS_LBSIZE(mp)/8)));
2127 ASSERT(be16_to_cpu(hdr_s->firstused) >=
2128 ((be16_to_cpu(hdr_s->count)
2129 * sizeof(*entry_s))+sizeof(*hdr_s)));
2130 ASSERT(be16_to_cpu(hdr_d->count) < (XFS_LBSIZE(mp)/8));
2131 ASSERT(be16_to_cpu(hdr_d->firstused) >=
2132 ((be16_to_cpu(hdr_d->count)
2133 * sizeof(*entry_d))+sizeof(*hdr_d)));
2135 ASSERT(start_s < be16_to_cpu(hdr_s->count));
2136 ASSERT(start_d <= be16_to_cpu(hdr_d->count));
2137 ASSERT(count <= be16_to_cpu(hdr_s->count));
2140 * Move the entries in the destination leaf up to make a hole?
2142 if (start_d < be16_to_cpu(hdr_d->count)) {
2143 tmp = be16_to_cpu(hdr_d->count) - start_d;
2144 tmp *= sizeof(xfs_attr_leaf_entry_t);
2145 entry_s = &leaf_d->entries[start_d];
2146 entry_d = &leaf_d->entries[start_d + count];
2147 memmove((char *)entry_d, (char *)entry_s, tmp);
2151 * Copy all entry's in the same (sorted) order,
2152 * but allocate attribute info packed and in sequence.
2154 entry_s = &leaf_s->entries[start_s];
2155 entry_d = &leaf_d->entries[start_d];
2156 desti = start_d;
2157 for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2158 ASSERT(be16_to_cpu(entry_s->nameidx)
2159 >= be16_to_cpu(hdr_s->firstused));
2160 tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2161 #ifdef GROT
2163 * Code to drop INCOMPLETE entries. Difficult to use as we
2164 * may also need to change the insertion index. Code turned
2165 * off for 6.2, should be revisited later.
2167 if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2168 memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2169 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2170 be16_add_cpu(&hdr_s->count, -1);
2171 entry_d--; /* to compensate for ++ in loop hdr */
2172 desti--;
2173 if ((start_s + i) < offset)
2174 result++; /* insertion index adjustment */
2175 } else {
2176 #endif /* GROT */
2177 be16_add_cpu(&hdr_d->firstused, -tmp);
2178 /* both on-disk, don't endian flip twice */
2179 entry_d->hashval = entry_s->hashval;
2180 /* both on-disk, don't endian flip twice */
2181 entry_d->nameidx = hdr_d->firstused;
2182 entry_d->flags = entry_s->flags;
2183 ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2184 <= XFS_LBSIZE(mp));
2185 memmove(xfs_attr_leaf_name(leaf_d, desti),
2186 xfs_attr_leaf_name(leaf_s, start_s + i), tmp);
2187 ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2188 <= XFS_LBSIZE(mp));
2189 memset(xfs_attr_leaf_name(leaf_s, start_s + i), 0, tmp);
2190 be16_add_cpu(&hdr_s->usedbytes, -tmp);
2191 be16_add_cpu(&hdr_d->usedbytes, tmp);
2192 be16_add_cpu(&hdr_s->count, -1);
2193 be16_add_cpu(&hdr_d->count, 1);
2194 tmp = be16_to_cpu(hdr_d->count)
2195 * sizeof(xfs_attr_leaf_entry_t)
2196 + sizeof(xfs_attr_leaf_hdr_t);
2197 ASSERT(be16_to_cpu(hdr_d->firstused) >= tmp);
2198 #ifdef GROT
2200 #endif /* GROT */
2204 * Zero out the entries we just copied.
2206 if (start_s == be16_to_cpu(hdr_s->count)) {
2207 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2208 entry_s = &leaf_s->entries[start_s];
2209 ASSERT(((char *)entry_s + tmp) <=
2210 ((char *)leaf_s + XFS_LBSIZE(mp)));
2211 memset((char *)entry_s, 0, tmp);
2212 } else {
2214 * Move the remaining entries down to fill the hole,
2215 * then zero the entries at the top.
2217 tmp = be16_to_cpu(hdr_s->count) - count;
2218 tmp *= sizeof(xfs_attr_leaf_entry_t);
2219 entry_s = &leaf_s->entries[start_s + count];
2220 entry_d = &leaf_s->entries[start_s];
2221 memmove((char *)entry_d, (char *)entry_s, tmp);
2223 tmp = count * sizeof(xfs_attr_leaf_entry_t);
2224 entry_s = &leaf_s->entries[be16_to_cpu(hdr_s->count)];
2225 ASSERT(((char *)entry_s + tmp) <=
2226 ((char *)leaf_s + XFS_LBSIZE(mp)));
2227 memset((char *)entry_s, 0, tmp);
2231 * Fill in the freemap information
2233 hdr_d->freemap[0].base = cpu_to_be16(sizeof(xfs_attr_leaf_hdr_t));
2234 be16_add_cpu(&hdr_d->freemap[0].base, be16_to_cpu(hdr_d->count) *
2235 sizeof(xfs_attr_leaf_entry_t));
2236 hdr_d->freemap[0].size = cpu_to_be16(be16_to_cpu(hdr_d->firstused)
2237 - be16_to_cpu(hdr_d->freemap[0].base));
2238 hdr_d->freemap[1].base = 0;
2239 hdr_d->freemap[2].base = 0;
2240 hdr_d->freemap[1].size = 0;
2241 hdr_d->freemap[2].size = 0;
2242 hdr_s->holes = 1; /* leaf may not be compact */
2246 * Compare two leaf blocks "order".
2247 * Return 0 unless leaf2 should go before leaf1.
2250 xfs_attr_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
2252 xfs_attr_leafblock_t *leaf1, *leaf2;
2254 leaf1 = leaf1_bp->data;
2255 leaf2 = leaf2_bp->data;
2256 ASSERT((leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) &&
2257 (leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)));
2258 if ((be16_to_cpu(leaf1->hdr.count) > 0) &&
2259 (be16_to_cpu(leaf2->hdr.count) > 0) &&
2260 ((be32_to_cpu(leaf2->entries[0].hashval) <
2261 be32_to_cpu(leaf1->entries[0].hashval)) ||
2262 (be32_to_cpu(leaf2->entries[
2263 be16_to_cpu(leaf2->hdr.count)-1].hashval) <
2264 be32_to_cpu(leaf1->entries[
2265 be16_to_cpu(leaf1->hdr.count)-1].hashval)))) {
2266 return(1);
2268 return(0);
2272 * Pick up the last hashvalue from a leaf block.
2274 xfs_dahash_t
2275 xfs_attr_leaf_lasthash(xfs_dabuf_t *bp, int *count)
2277 xfs_attr_leafblock_t *leaf;
2279 leaf = bp->data;
2280 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2281 if (count)
2282 *count = be16_to_cpu(leaf->hdr.count);
2283 if (!leaf->hdr.count)
2284 return(0);
2285 return be32_to_cpu(leaf->entries[be16_to_cpu(leaf->hdr.count)-1].hashval);
2289 * Calculate the number of bytes used to store the indicated attribute
2290 * (whether local or remote only calculate bytes in this block).
2292 STATIC int
2293 xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2295 xfs_attr_leaf_name_local_t *name_loc;
2296 xfs_attr_leaf_name_remote_t *name_rmt;
2297 int size;
2299 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2300 if (leaf->entries[index].flags & XFS_ATTR_LOCAL) {
2301 name_loc = xfs_attr_leaf_name_local(leaf, index);
2302 size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2303 be16_to_cpu(name_loc->valuelen));
2304 } else {
2305 name_rmt = xfs_attr_leaf_name_remote(leaf, index);
2306 size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2308 return(size);
2312 * Calculate the number of bytes that would be required to store the new
2313 * attribute (whether local or remote only calculate bytes in this block).
2314 * This routine decides as a side effect whether the attribute will be
2315 * a "local" or a "remote" attribute.
2318 xfs_attr_leaf_newentsize(int namelen, int valuelen, int blocksize, int *local)
2320 int size;
2322 size = xfs_attr_leaf_entsize_local(namelen, valuelen);
2323 if (size < xfs_attr_leaf_entsize_local_max(blocksize)) {
2324 if (local) {
2325 *local = 1;
2327 } else {
2328 size = xfs_attr_leaf_entsize_remote(namelen);
2329 if (local) {
2330 *local = 0;
2333 return(size);
2337 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
2340 xfs_attr_leaf_list_int(xfs_dabuf_t *bp, xfs_attr_list_context_t *context)
2342 attrlist_cursor_kern_t *cursor;
2343 xfs_attr_leafblock_t *leaf;
2344 xfs_attr_leaf_entry_t *entry;
2345 int retval, i;
2347 ASSERT(bp != NULL);
2348 leaf = bp->data;
2349 cursor = context->cursor;
2350 cursor->initted = 1;
2352 trace_xfs_attr_list_leaf(context);
2355 * Re-find our place in the leaf block if this is a new syscall.
2357 if (context->resynch) {
2358 entry = &leaf->entries[0];
2359 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2360 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
2361 if (cursor->offset == context->dupcnt) {
2362 context->dupcnt = 0;
2363 break;
2365 context->dupcnt++;
2366 } else if (be32_to_cpu(entry->hashval) >
2367 cursor->hashval) {
2368 context->dupcnt = 0;
2369 break;
2372 if (i == be16_to_cpu(leaf->hdr.count)) {
2373 trace_xfs_attr_list_notfound(context);
2374 return(0);
2376 } else {
2377 entry = &leaf->entries[0];
2378 i = 0;
2380 context->resynch = 0;
2383 * We have found our place, start copying out the new attributes.
2385 retval = 0;
2386 for ( ; (i < be16_to_cpu(leaf->hdr.count)); entry++, i++) {
2387 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
2388 cursor->hashval = be32_to_cpu(entry->hashval);
2389 cursor->offset = 0;
2392 if (entry->flags & XFS_ATTR_INCOMPLETE)
2393 continue; /* skip incomplete entries */
2395 if (entry->flags & XFS_ATTR_LOCAL) {
2396 xfs_attr_leaf_name_local_t *name_loc =
2397 xfs_attr_leaf_name_local(leaf, i);
2399 retval = context->put_listent(context,
2400 entry->flags,
2401 name_loc->nameval,
2402 (int)name_loc->namelen,
2403 be16_to_cpu(name_loc->valuelen),
2404 &name_loc->nameval[name_loc->namelen]);
2405 if (retval)
2406 return retval;
2407 } else {
2408 xfs_attr_leaf_name_remote_t *name_rmt =
2409 xfs_attr_leaf_name_remote(leaf, i);
2411 int valuelen = be32_to_cpu(name_rmt->valuelen);
2413 if (context->put_value) {
2414 xfs_da_args_t args;
2416 memset((char *)&args, 0, sizeof(args));
2417 args.dp = context->dp;
2418 args.whichfork = XFS_ATTR_FORK;
2419 args.valuelen = valuelen;
2420 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
2421 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
2422 args.rmtblkcnt = XFS_B_TO_FSB(args.dp->i_mount, valuelen);
2423 retval = xfs_attr_rmtval_get(&args);
2424 if (retval)
2425 return retval;
2426 retval = context->put_listent(context,
2427 entry->flags,
2428 name_rmt->name,
2429 (int)name_rmt->namelen,
2430 valuelen,
2431 args.value);
2432 kmem_free(args.value);
2433 } else {
2434 retval = context->put_listent(context,
2435 entry->flags,
2436 name_rmt->name,
2437 (int)name_rmt->namelen,
2438 valuelen,
2439 NULL);
2441 if (retval)
2442 return retval;
2444 if (context->seen_enough)
2445 break;
2446 cursor->offset++;
2448 trace_xfs_attr_list_leaf_end(context);
2449 return(retval);
2453 /*========================================================================
2454 * Manage the INCOMPLETE flag in a leaf entry
2455 *========================================================================*/
2458 * Clear the INCOMPLETE flag on an entry in a leaf block.
2461 xfs_attr_leaf_clearflag(xfs_da_args_t *args)
2463 xfs_attr_leafblock_t *leaf;
2464 xfs_attr_leaf_entry_t *entry;
2465 xfs_attr_leaf_name_remote_t *name_rmt;
2466 xfs_dabuf_t *bp;
2467 int error;
2468 #ifdef DEBUG
2469 xfs_attr_leaf_name_local_t *name_loc;
2470 int namelen;
2471 char *name;
2472 #endif /* DEBUG */
2474 trace_xfs_attr_leaf_clearflag(args);
2476 * Set up the operation.
2478 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2479 XFS_ATTR_FORK);
2480 if (error) {
2481 return(error);
2483 ASSERT(bp != NULL);
2485 leaf = bp->data;
2486 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2487 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2488 ASSERT(args->index >= 0);
2489 entry = &leaf->entries[ args->index ];
2490 ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2492 #ifdef DEBUG
2493 if (entry->flags & XFS_ATTR_LOCAL) {
2494 name_loc = xfs_attr_leaf_name_local(leaf, args->index);
2495 namelen = name_loc->namelen;
2496 name = (char *)name_loc->nameval;
2497 } else {
2498 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2499 namelen = name_rmt->namelen;
2500 name = (char *)name_rmt->name;
2502 ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2503 ASSERT(namelen == args->namelen);
2504 ASSERT(memcmp(name, args->name, namelen) == 0);
2505 #endif /* DEBUG */
2507 entry->flags &= ~XFS_ATTR_INCOMPLETE;
2508 xfs_da_log_buf(args->trans, bp,
2509 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2511 if (args->rmtblkno) {
2512 ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2513 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2514 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2515 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2516 xfs_da_log_buf(args->trans, bp,
2517 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2519 xfs_da_buf_done(bp);
2522 * Commit the flag value change and start the next trans in series.
2524 return xfs_trans_roll(&args->trans, args->dp);
2528 * Set the INCOMPLETE flag on an entry in a leaf block.
2531 xfs_attr_leaf_setflag(xfs_da_args_t *args)
2533 xfs_attr_leafblock_t *leaf;
2534 xfs_attr_leaf_entry_t *entry;
2535 xfs_attr_leaf_name_remote_t *name_rmt;
2536 xfs_dabuf_t *bp;
2537 int error;
2539 trace_xfs_attr_leaf_setflag(args);
2542 * Set up the operation.
2544 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
2545 XFS_ATTR_FORK);
2546 if (error) {
2547 return(error);
2549 ASSERT(bp != NULL);
2551 leaf = bp->data;
2552 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2553 ASSERT(args->index < be16_to_cpu(leaf->hdr.count));
2554 ASSERT(args->index >= 0);
2555 entry = &leaf->entries[ args->index ];
2557 ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2558 entry->flags |= XFS_ATTR_INCOMPLETE;
2559 xfs_da_log_buf(args->trans, bp,
2560 XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2561 if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2562 name_rmt = xfs_attr_leaf_name_remote(leaf, args->index);
2563 name_rmt->valueblk = 0;
2564 name_rmt->valuelen = 0;
2565 xfs_da_log_buf(args->trans, bp,
2566 XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2568 xfs_da_buf_done(bp);
2571 * Commit the flag value change and start the next trans in series.
2573 return xfs_trans_roll(&args->trans, args->dp);
2577 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2578 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2579 * entry given by args->blkno2/index2.
2581 * Note that they could be in different blocks, or in the same block.
2584 xfs_attr_leaf_flipflags(xfs_da_args_t *args)
2586 xfs_attr_leafblock_t *leaf1, *leaf2;
2587 xfs_attr_leaf_entry_t *entry1, *entry2;
2588 xfs_attr_leaf_name_remote_t *name_rmt;
2589 xfs_dabuf_t *bp1, *bp2;
2590 int error;
2591 #ifdef DEBUG
2592 xfs_attr_leaf_name_local_t *name_loc;
2593 int namelen1, namelen2;
2594 char *name1, *name2;
2595 #endif /* DEBUG */
2597 trace_xfs_attr_leaf_flipflags(args);
2600 * Read the block containing the "old" attr
2602 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp1,
2603 XFS_ATTR_FORK);
2604 if (error) {
2605 return(error);
2607 ASSERT(bp1 != NULL);
2610 * Read the block containing the "new" attr, if it is different
2612 if (args->blkno2 != args->blkno) {
2613 error = xfs_da_read_buf(args->trans, args->dp, args->blkno2,
2614 -1, &bp2, XFS_ATTR_FORK);
2615 if (error) {
2616 return(error);
2618 ASSERT(bp2 != NULL);
2619 } else {
2620 bp2 = bp1;
2623 leaf1 = bp1->data;
2624 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2625 ASSERT(args->index < be16_to_cpu(leaf1->hdr.count));
2626 ASSERT(args->index >= 0);
2627 entry1 = &leaf1->entries[ args->index ];
2629 leaf2 = bp2->data;
2630 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2631 ASSERT(args->index2 < be16_to_cpu(leaf2->hdr.count));
2632 ASSERT(args->index2 >= 0);
2633 entry2 = &leaf2->entries[ args->index2 ];
2635 #ifdef DEBUG
2636 if (entry1->flags & XFS_ATTR_LOCAL) {
2637 name_loc = xfs_attr_leaf_name_local(leaf1, args->index);
2638 namelen1 = name_loc->namelen;
2639 name1 = (char *)name_loc->nameval;
2640 } else {
2641 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2642 namelen1 = name_rmt->namelen;
2643 name1 = (char *)name_rmt->name;
2645 if (entry2->flags & XFS_ATTR_LOCAL) {
2646 name_loc = xfs_attr_leaf_name_local(leaf2, args->index2);
2647 namelen2 = name_loc->namelen;
2648 name2 = (char *)name_loc->nameval;
2649 } else {
2650 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2651 namelen2 = name_rmt->namelen;
2652 name2 = (char *)name_rmt->name;
2654 ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2655 ASSERT(namelen1 == namelen2);
2656 ASSERT(memcmp(name1, name2, namelen1) == 0);
2657 #endif /* DEBUG */
2659 ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2660 ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2662 entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2663 xfs_da_log_buf(args->trans, bp1,
2664 XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2665 if (args->rmtblkno) {
2666 ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2667 name_rmt = xfs_attr_leaf_name_remote(leaf1, args->index);
2668 name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2669 name_rmt->valuelen = cpu_to_be32(args->valuelen);
2670 xfs_da_log_buf(args->trans, bp1,
2671 XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2674 entry2->flags |= XFS_ATTR_INCOMPLETE;
2675 xfs_da_log_buf(args->trans, bp2,
2676 XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2677 if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2678 name_rmt = xfs_attr_leaf_name_remote(leaf2, args->index2);
2679 name_rmt->valueblk = 0;
2680 name_rmt->valuelen = 0;
2681 xfs_da_log_buf(args->trans, bp2,
2682 XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2684 xfs_da_buf_done(bp1);
2685 if (bp1 != bp2)
2686 xfs_da_buf_done(bp2);
2689 * Commit the flag value change and start the next trans in series.
2691 error = xfs_trans_roll(&args->trans, args->dp);
2693 return(error);
2696 /*========================================================================
2697 * Indiscriminately delete the entire attribute fork
2698 *========================================================================*/
2701 * Recurse (gasp!) through the attribute nodes until we find leaves.
2702 * We're doing a depth-first traversal in order to invalidate everything.
2705 xfs_attr_root_inactive(xfs_trans_t **trans, xfs_inode_t *dp)
2707 xfs_da_blkinfo_t *info;
2708 xfs_daddr_t blkno;
2709 xfs_dabuf_t *bp;
2710 int error;
2713 * Read block 0 to see what we have to work with.
2714 * We only get here if we have extents, since we remove
2715 * the extents in reverse order the extent containing
2716 * block 0 must still be there.
2718 error = xfs_da_read_buf(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
2719 if (error)
2720 return(error);
2721 blkno = xfs_da_blkno(bp);
2724 * Invalidate the tree, even if the "tree" is only a single leaf block.
2725 * This is a depth-first traversal!
2727 info = bp->data;
2728 if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2729 error = xfs_attr_node_inactive(trans, dp, bp, 1);
2730 } else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2731 error = xfs_attr_leaf_inactive(trans, dp, bp);
2732 } else {
2733 error = XFS_ERROR(EIO);
2734 xfs_da_brelse(*trans, bp);
2736 if (error)
2737 return(error);
2740 * Invalidate the incore copy of the root block.
2742 error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
2743 if (error)
2744 return(error);
2745 xfs_da_binval(*trans, bp); /* remove from cache */
2747 * Commit the invalidate and start the next transaction.
2749 error = xfs_trans_roll(trans, dp);
2751 return (error);
2755 * Recurse (gasp!) through the attribute nodes until we find leaves.
2756 * We're doing a depth-first traversal in order to invalidate everything.
2758 STATIC int
2759 xfs_attr_node_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp,
2760 int level)
2762 xfs_da_blkinfo_t *info;
2763 xfs_da_intnode_t *node;
2764 xfs_dablk_t child_fsb;
2765 xfs_daddr_t parent_blkno, child_blkno;
2766 int error, count, i;
2767 xfs_dabuf_t *child_bp;
2770 * Since this code is recursive (gasp!) we must protect ourselves.
2772 if (level > XFS_DA_NODE_MAXDEPTH) {
2773 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2774 return(XFS_ERROR(EIO));
2777 node = bp->data;
2778 ASSERT(node->hdr.info.magic == cpu_to_be16(XFS_DA_NODE_MAGIC));
2779 parent_blkno = xfs_da_blkno(bp); /* save for re-read later */
2780 count = be16_to_cpu(node->hdr.count);
2781 if (!count) {
2782 xfs_da_brelse(*trans, bp);
2783 return(0);
2785 child_fsb = be32_to_cpu(node->btree[0].before);
2786 xfs_da_brelse(*trans, bp); /* no locks for later trans */
2789 * If this is the node level just above the leaves, simply loop
2790 * over the leaves removing all of them. If this is higher up
2791 * in the tree, recurse downward.
2793 for (i = 0; i < count; i++) {
2795 * Read the subsidiary block to see what we have to work with.
2796 * Don't do this in a transaction. This is a depth-first
2797 * traversal of the tree so we may deal with many blocks
2798 * before we come back to this one.
2800 error = xfs_da_read_buf(*trans, dp, child_fsb, -2, &child_bp,
2801 XFS_ATTR_FORK);
2802 if (error)
2803 return(error);
2804 if (child_bp) {
2805 /* save for re-read later */
2806 child_blkno = xfs_da_blkno(child_bp);
2809 * Invalidate the subtree, however we have to.
2811 info = child_bp->data;
2812 if (info->magic == cpu_to_be16(XFS_DA_NODE_MAGIC)) {
2813 error = xfs_attr_node_inactive(trans, dp,
2814 child_bp, level+1);
2815 } else if (info->magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC)) {
2816 error = xfs_attr_leaf_inactive(trans, dp,
2817 child_bp);
2818 } else {
2819 error = XFS_ERROR(EIO);
2820 xfs_da_brelse(*trans, child_bp);
2822 if (error)
2823 return(error);
2826 * Remove the subsidiary block from the cache
2827 * and from the log.
2829 error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
2830 &child_bp, XFS_ATTR_FORK);
2831 if (error)
2832 return(error);
2833 xfs_da_binval(*trans, child_bp);
2837 * If we're not done, re-read the parent to get the next
2838 * child block number.
2840 if ((i+1) < count) {
2841 error = xfs_da_read_buf(*trans, dp, 0, parent_blkno,
2842 &bp, XFS_ATTR_FORK);
2843 if (error)
2844 return(error);
2845 child_fsb = be32_to_cpu(node->btree[i+1].before);
2846 xfs_da_brelse(*trans, bp);
2849 * Atomically commit the whole invalidate stuff.
2851 error = xfs_trans_roll(trans, dp);
2852 if (error)
2853 return (error);
2856 return(0);
2860 * Invalidate all of the "remote" value regions pointed to by a particular
2861 * leaf block.
2862 * Note that we must release the lock on the buffer so that we are not
2863 * caught holding something that the logging code wants to flush to disk.
2865 STATIC int
2866 xfs_attr_leaf_inactive(xfs_trans_t **trans, xfs_inode_t *dp, xfs_dabuf_t *bp)
2868 xfs_attr_leafblock_t *leaf;
2869 xfs_attr_leaf_entry_t *entry;
2870 xfs_attr_leaf_name_remote_t *name_rmt;
2871 xfs_attr_inactive_list_t *list, *lp;
2872 int error, count, size, tmp, i;
2874 leaf = bp->data;
2875 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC));
2878 * Count the number of "remote" value extents.
2880 count = 0;
2881 entry = &leaf->entries[0];
2882 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2883 if (be16_to_cpu(entry->nameidx) &&
2884 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2885 name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2886 if (name_rmt->valueblk)
2887 count++;
2892 * If there are no "remote" values, we're done.
2894 if (count == 0) {
2895 xfs_da_brelse(*trans, bp);
2896 return(0);
2900 * Allocate storage for a list of all the "remote" value extents.
2902 size = count * sizeof(xfs_attr_inactive_list_t);
2903 list = (xfs_attr_inactive_list_t *)kmem_alloc(size, KM_SLEEP);
2906 * Identify each of the "remote" value extents.
2908 lp = list;
2909 entry = &leaf->entries[0];
2910 for (i = 0; i < be16_to_cpu(leaf->hdr.count); entry++, i++) {
2911 if (be16_to_cpu(entry->nameidx) &&
2912 ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
2913 name_rmt = xfs_attr_leaf_name_remote(leaf, i);
2914 if (name_rmt->valueblk) {
2915 lp->valueblk = be32_to_cpu(name_rmt->valueblk);
2916 lp->valuelen = XFS_B_TO_FSB(dp->i_mount,
2917 be32_to_cpu(name_rmt->valuelen));
2918 lp++;
2922 xfs_da_brelse(*trans, bp); /* unlock for trans. in freextent() */
2925 * Invalidate each of the "remote" value extents.
2927 error = 0;
2928 for (lp = list, i = 0; i < count; i++, lp++) {
2929 tmp = xfs_attr_leaf_freextent(trans, dp,
2930 lp->valueblk, lp->valuelen);
2932 if (error == 0)
2933 error = tmp; /* save only the 1st errno */
2936 kmem_free((xfs_caddr_t)list);
2937 return(error);
2941 * Look at all the extents for this logical region,
2942 * invalidate any buffers that are incore/in transactions.
2944 STATIC int
2945 xfs_attr_leaf_freextent(xfs_trans_t **trans, xfs_inode_t *dp,
2946 xfs_dablk_t blkno, int blkcnt)
2948 xfs_bmbt_irec_t map;
2949 xfs_dablk_t tblkno;
2950 int tblkcnt, dblkcnt, nmap, error;
2951 xfs_daddr_t dblkno;
2952 xfs_buf_t *bp;
2955 * Roll through the "value", invalidating the attribute value's
2956 * blocks.
2958 tblkno = blkno;
2959 tblkcnt = blkcnt;
2960 while (tblkcnt > 0) {
2962 * Try to remember where we decided to put the value.
2964 nmap = 1;
2965 error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
2966 &map, &nmap, XFS_BMAPI_ATTRFORK);
2967 if (error) {
2968 return(error);
2970 ASSERT(nmap == 1);
2971 ASSERT(map.br_startblock != DELAYSTARTBLOCK);
2974 * If it's a hole, these are already unmapped
2975 * so there's nothing to invalidate.
2977 if (map.br_startblock != HOLESTARTBLOCK) {
2979 dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
2980 map.br_startblock);
2981 dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
2982 map.br_blockcount);
2983 bp = xfs_trans_get_buf(*trans,
2984 dp->i_mount->m_ddev_targp,
2985 dblkno, dblkcnt, 0);
2986 if (!bp)
2987 return ENOMEM;
2988 xfs_trans_binval(*trans, bp);
2990 * Roll to next transaction.
2992 error = xfs_trans_roll(trans, dp);
2993 if (error)
2994 return (error);
2997 tblkno += map.br_blockcount;
2998 tblkcnt -= map.br_blockcount;
3001 return(0);