2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_inode.h"
32 #include "xfs_alloc.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
36 #include "xfs_bmap_util.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_attr_remote.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_trace.h"
42 #include "xfs_cksum.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_error.h"
46 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
49 * Each contiguous block has a header, so it is not just a simple attribute
50 * length to FSB conversion.
57 if (xfs_sb_version_hascrc(&mp->m_sb)) {
58 int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
59 return (attrlen + buflen - 1) / buflen;
61 return XFS_B_TO_FSB(mp, attrlen);
65 * Checking of the remote attribute header is split into two parts. The verifier
66 * does CRC, location and bounds checking, the unpacking function checks the
67 * attribute parameters and owner.
78 struct xfs_attr3_rmt_hdr *rmt = ptr;
80 if (bno != be64_to_cpu(rmt->rm_blkno))
82 if (offset != be32_to_cpu(rmt->rm_offset))
84 if (size != be32_to_cpu(rmt->rm_bytes))
86 if (ino != be64_to_cpu(rmt->rm_owner))
100 struct xfs_attr3_rmt_hdr *rmt = ptr;
102 if (!xfs_sb_version_hascrc(&mp->m_sb))
104 if (rmt->rm_magic != cpu_to_be32(XFS_ATTR3_RMT_MAGIC))
106 if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_uuid))
108 if (be64_to_cpu(rmt->rm_blkno) != bno)
110 if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
112 if (be32_to_cpu(rmt->rm_offset) +
113 be32_to_cpu(rmt->rm_bytes) > XATTR_SIZE_MAX)
115 if (rmt->rm_owner == 0)
122 xfs_attr3_rmt_read_verify(
125 struct xfs_mount *mp = bp->b_target->bt_mount;
128 bool corrupt = false;
131 /* no verification of non-crc buffers */
132 if (!xfs_sb_version_hascrc(&mp->m_sb))
137 len = BBTOB(bp->b_length);
138 ASSERT(len >= XFS_LBSIZE(mp));
141 if (!xfs_verify_cksum(ptr, XFS_LBSIZE(mp),
142 XFS_ATTR3_RMT_CRC_OFF)) {
146 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
150 len -= XFS_LBSIZE(mp);
151 ptr += XFS_LBSIZE(mp);
156 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
157 xfs_buf_ioerror(bp, EFSCORRUPTED);
163 xfs_attr3_rmt_write_verify(
166 struct xfs_mount *mp = bp->b_target->bt_mount;
167 struct xfs_buf_log_item *bip = bp->b_fspriv;
172 /* no verification of non-crc buffers */
173 if (!xfs_sb_version_hascrc(&mp->m_sb))
178 len = BBTOB(bp->b_length);
179 ASSERT(len >= XFS_LBSIZE(mp));
182 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
183 XFS_CORRUPTION_ERROR(__func__,
184 XFS_ERRLEVEL_LOW, mp, bp->b_addr);
185 xfs_buf_ioerror(bp, EFSCORRUPTED);
189 struct xfs_attr3_rmt_hdr *rmt;
191 rmt = (struct xfs_attr3_rmt_hdr *)ptr;
192 rmt->rm_lsn = cpu_to_be64(bip->bli_item.li_lsn);
194 xfs_update_cksum(ptr, XFS_LBSIZE(mp), XFS_ATTR3_RMT_CRC_OFF);
196 len -= XFS_LBSIZE(mp);
197 ptr += XFS_LBSIZE(mp);
203 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
204 .verify_read = xfs_attr3_rmt_read_verify,
205 .verify_write = xfs_attr3_rmt_write_verify,
209 xfs_attr3_rmt_hdr_set(
210 struct xfs_mount *mp,
217 struct xfs_attr3_rmt_hdr *rmt = ptr;
219 if (!xfs_sb_version_hascrc(&mp->m_sb))
222 rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
223 rmt->rm_offset = cpu_to_be32(offset);
224 rmt->rm_bytes = cpu_to_be32(size);
225 uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_uuid);
226 rmt->rm_owner = cpu_to_be64(ino);
227 rmt->rm_blkno = cpu_to_be64(bno);
229 return sizeof(struct xfs_attr3_rmt_hdr);
233 * Helper functions to copy attribute data in and out of the one disk extents
236 xfs_attr_rmtval_copyout(
237 struct xfs_mount *mp,
244 char *src = bp->b_addr;
245 xfs_daddr_t bno = bp->b_bn;
246 int len = BBTOB(bp->b_length);
248 ASSERT(len >= XFS_LBSIZE(mp));
250 while (len > 0 && *valuelen > 0) {
252 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
254 byte_cnt = min(*valuelen, byte_cnt);
256 if (xfs_sb_version_hascrc(&mp->m_sb)) {
257 if (!xfs_attr3_rmt_hdr_ok(mp, src, ino, *offset,
260 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
261 bno, *offset, byte_cnt, ino);
264 hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
267 memcpy(*dst, src + hdr_size, byte_cnt);
269 /* roll buffer forwards */
270 len -= XFS_LBSIZE(mp);
271 src += XFS_LBSIZE(mp);
274 /* roll attribute data forwards */
275 *valuelen -= byte_cnt;
283 xfs_attr_rmtval_copyin(
284 struct xfs_mount *mp,
291 char *dst = bp->b_addr;
292 xfs_daddr_t bno = bp->b_bn;
293 int len = BBTOB(bp->b_length);
295 ASSERT(len >= XFS_LBSIZE(mp));
297 while (len > 0 && *valuelen > 0) {
299 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
301 byte_cnt = min(*valuelen, byte_cnt);
302 hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
305 memcpy(dst + hdr_size, *src, byte_cnt);
308 * If this is the last block, zero the remainder of it.
309 * Check that we are actually the last block, too.
311 if (byte_cnt + hdr_size < XFS_LBSIZE(mp)) {
312 ASSERT(*valuelen - byte_cnt == 0);
313 ASSERT(len == XFS_LBSIZE(mp));
314 memset(dst + hdr_size + byte_cnt, 0,
315 XFS_LBSIZE(mp) - hdr_size - byte_cnt);
318 /* roll buffer forwards */
319 len -= XFS_LBSIZE(mp);
320 dst += XFS_LBSIZE(mp);
323 /* roll attribute data forwards */
324 *valuelen -= byte_cnt;
331 * Read the value associated with an attribute from the out-of-line buffer
332 * that we stored it in.
336 struct xfs_da_args *args)
338 struct xfs_bmbt_irec map[ATTR_RMTVALUE_MAPSIZE];
339 struct xfs_mount *mp = args->dp->i_mount;
341 xfs_dablk_t lblkno = args->rmtblkno;
342 __uint8_t *dst = args->value;
343 int valuelen = args->valuelen;
346 int blkcnt = args->rmtblkcnt;
350 trace_xfs_attr_rmtval_get(args);
352 ASSERT(!(args->flags & ATTR_KERNOVAL));
354 while (valuelen > 0) {
355 nmap = ATTR_RMTVALUE_MAPSIZE;
356 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
363 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
367 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
368 (map[i].br_startblock != HOLESTARTBLOCK));
369 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
370 dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
371 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
372 dblkno, dblkcnt, 0, &bp,
373 &xfs_attr3_rmt_buf_ops);
377 error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
384 /* roll attribute extent map forwards */
385 lblkno += map[i].br_blockcount;
386 blkcnt -= map[i].br_blockcount;
389 ASSERT(valuelen == 0);
394 * Write the value associated with an attribute into the out-of-line buffer
395 * that we have defined for it.
399 struct xfs_da_args *args)
401 struct xfs_inode *dp = args->dp;
402 struct xfs_mount *mp = dp->i_mount;
403 struct xfs_bmbt_irec map;
405 xfs_fileoff_t lfileoff = 0;
406 __uint8_t *src = args->value;
413 trace_xfs_attr_rmtval_set(args);
416 * Find a "hole" in the attribute address space large enough for
417 * us to drop the new attribute's value into. Because CRC enable
418 * attributes have headers, we can't just do a straight byte to FSB
419 * conversion and have to take the header space into account.
421 blkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
422 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
427 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
428 args->rmtblkcnt = blkcnt;
431 * Roll through the "value", allocating blocks on disk as required.
437 * Allocate a single extent, up to the size of the value.
439 xfs_bmap_init(args->flist, args->firstblock);
441 error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
443 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
444 args->firstblock, args->total, &map, &nmap,
447 error = xfs_bmap_finish(&args->trans, args->flist,
453 xfs_bmap_cancel(args->flist);
458 * bmap_finish() may have committed the last trans and started
459 * a new one. We need the inode to be in all transactions.
462 xfs_trans_ijoin(args->trans, dp, 0);
465 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
466 (map.br_startblock != HOLESTARTBLOCK));
467 lblkno += map.br_blockcount;
468 blkcnt -= map.br_blockcount;
471 * Start the next trans in the chain.
473 error = xfs_trans_roll(&args->trans, dp);
479 * Roll through the "value", copying the attribute value to the
480 * already-allocated blocks. Blocks are written synchronously
481 * so that we can know they are all on disk before we turn off
482 * the INCOMPLETE flag.
484 lblkno = args->rmtblkno;
485 blkcnt = args->rmtblkcnt;
486 valuelen = args->valuelen;
487 while (valuelen > 0) {
494 xfs_bmap_init(args->flist, args->firstblock);
496 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
502 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
503 (map.br_startblock != HOLESTARTBLOCK));
505 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
506 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
508 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
511 bp->b_ops = &xfs_attr3_rmt_buf_ops;
513 xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
516 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
522 /* roll attribute extent map forwards */
523 lblkno += map.br_blockcount;
524 blkcnt -= map.br_blockcount;
526 ASSERT(valuelen == 0);
531 * Remove the value associated with an attribute by deleting the
532 * out-of-line buffer that it is stored on.
535 xfs_attr_rmtval_remove(
536 struct xfs_da_args *args)
538 struct xfs_mount *mp = args->dp->i_mount;
544 trace_xfs_attr_rmtval_remove(args);
547 * Roll through the "value", invalidating the attribute value's blocks.
549 lblkno = args->rmtblkno;
550 blkcnt = args->rmtblkcnt;
552 struct xfs_bmbt_irec map;
559 * Try to remember where we decided to put the value.
562 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
563 blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
567 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
568 (map.br_startblock != HOLESTARTBLOCK));
570 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
571 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
574 * If the "remote" value is in the cache, remove it.
576 bp = xfs_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
583 lblkno += map.br_blockcount;
584 blkcnt -= map.br_blockcount;
588 * Keep de-allocating extents until the remote-value region is gone.
590 lblkno = args->rmtblkno;
591 blkcnt = args->rmtblkcnt;
596 xfs_bmap_init(args->flist, args->firstblock);
597 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
598 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
599 1, args->firstblock, args->flist,
602 error = xfs_bmap_finish(&args->trans, args->flist,
608 xfs_bmap_cancel(args->flist);
613 * bmap_finish() may have committed the last trans and started
614 * a new one. We need the inode to be in all transactions.
617 xfs_trans_ijoin(args->trans, args->dp, 0);
620 * Close out trans and start the next one in the chain.
622 error = xfs_trans_roll(&args->trans, args->dp);