1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
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.
9 *
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.
14 *
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
18 */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_types.h"
22 #include "xfs_bit.h"
23 #include "xfs_log.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_alloc.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_bmap.h"
36 #include "xfs_attr.h"
37 #include "xfs_attr_leaf.h"
38 #include "xfs_attr_remote.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_trace.h"
41 #include "xfs_cksum.h"
42 #include "xfs_buf_item.h"
43
44 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
45
46 /*
47 * Each contiguous block has a header, so it is not just a simple attribute
48 * length to FSB conversion.
49 */
50 int
xfs_attr3_rmt_blocks(struct xfs_mount * mp,int attrlen)51 xfs_attr3_rmt_blocks(
52 struct xfs_mount *mp,
53 int attrlen)
54 {
55 if (xfs_sb_version_hascrc(&mp->m_sb)) {
56 int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
57 return (attrlen + buflen - 1) / buflen;
58 }
59 return XFS_B_TO_FSB(mp, attrlen);
60 }
61
62 /*
63 * Checking of the remote attribute header is split into two parts. The verifier
64 * does CRC, location and bounds checking, the unpacking function checks the
65 * attribute parameters and owner.
66 */
67 static bool
xfs_attr3_rmt_hdr_ok(struct xfs_mount * mp,void * ptr,xfs_ino_t ino,uint32_t offset,uint32_t size,xfs_daddr_t bno)68 xfs_attr3_rmt_hdr_ok(
69 struct xfs_mount *mp,
70 void *ptr,
71 xfs_ino_t ino,
72 uint32_t offset,
73 uint32_t size,
74 xfs_daddr_t bno)
75 {
76 struct xfs_attr3_rmt_hdr *rmt = ptr;
77
78 if (bno != be64_to_cpu(rmt->rm_blkno))
79 return false;
80 if (offset != be32_to_cpu(rmt->rm_offset))
81 return false;
82 if (size != be32_to_cpu(rmt->rm_bytes))
83 return false;
84 if (ino != be64_to_cpu(rmt->rm_owner))
85 return false;
86
87 /* ok */
88 return true;
89 }
90
91 static bool
xfs_attr3_rmt_verify(struct xfs_mount * mp,void * ptr,int fsbsize,xfs_daddr_t bno)92 xfs_attr3_rmt_verify(
93 struct xfs_mount *mp,
94 void *ptr,
95 int fsbsize,
96 xfs_daddr_t bno)
97 {
98 struct xfs_attr3_rmt_hdr *rmt = ptr;
99
100 if (!xfs_sb_version_hascrc(&mp->m_sb))
101 return false;
102 if (rmt->rm_magic != cpu_to_be32(XFS_ATTR3_RMT_MAGIC))
103 return false;
104 if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_uuid))
105 return false;
106 if (be64_to_cpu(rmt->rm_blkno) != bno)
107 return false;
108 if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
109 return false;
110 if (be32_to_cpu(rmt->rm_offset) +
111 be32_to_cpu(rmt->rm_bytes) >= XATTR_SIZE_MAX)
112 return false;
113 if (rmt->rm_owner == 0)
114 return false;
115
116 return true;
117 }
118
119 static void
xfs_attr3_rmt_read_verify(struct xfs_buf * bp)120 xfs_attr3_rmt_read_verify(
121 struct xfs_buf *bp)
122 {
123 struct xfs_mount *mp = bp->b_target->bt_mount;
124 char *ptr;
125 int len;
126 bool corrupt = false;
127 xfs_daddr_t bno;
128
129 /* no verification of non-crc buffers */
130 if (!xfs_sb_version_hascrc(&mp->m_sb))
131 return;
132
133 ptr = bp->b_addr;
134 bno = bp->b_bn;
135 len = BBTOB(bp->b_length);
136 ASSERT(len >= XFS_LBSIZE(mp));
137
138 while (len > 0) {
139 if (!xfs_verify_cksum(ptr, XFS_LBSIZE(mp),
140 XFS_ATTR3_RMT_CRC_OFF)) {
141 corrupt = true;
142 break;
143 }
144 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
145 corrupt = true;
146 break;
147 }
148 len -= XFS_LBSIZE(mp);
149 ptr += XFS_LBSIZE(mp);
150 bno += mp->m_bsize;
151 }
152
153 if (corrupt) {
154 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
155 xfs_buf_ioerror(bp, EFSCORRUPTED);
156 } else
157 ASSERT(len == 0);
158 }
159
160 static void
xfs_attr3_rmt_write_verify(struct xfs_buf * bp)161 xfs_attr3_rmt_write_verify(
162 struct xfs_buf *bp)
163 {
164 struct xfs_mount *mp = bp->b_target->bt_mount;
165 struct xfs_buf_log_item *bip = bp->b_fspriv;
166 char *ptr;
167 int len;
168 xfs_daddr_t bno;
169
170 /* no verification of non-crc buffers */
171 if (!xfs_sb_version_hascrc(&mp->m_sb))
172 return;
173
174 ptr = bp->b_addr;
175 bno = bp->b_bn;
176 len = BBTOB(bp->b_length);
177 ASSERT(len >= XFS_LBSIZE(mp));
178
179 while (len > 0) {
180 if (!xfs_attr3_rmt_verify(mp, ptr, XFS_LBSIZE(mp), bno)) {
181 XFS_CORRUPTION_ERROR(__func__,
182 XFS_ERRLEVEL_LOW, mp, bp->b_addr);
183 xfs_buf_ioerror(bp, EFSCORRUPTED);
184 return;
185 }
186 if (bip) {
187 struct xfs_attr3_rmt_hdr *rmt;
188
189 rmt = (struct xfs_attr3_rmt_hdr *)ptr;
190 rmt->rm_lsn = cpu_to_be64(bip->bli_item.li_lsn);
191 }
192 xfs_update_cksum(ptr, XFS_LBSIZE(mp), XFS_ATTR3_RMT_CRC_OFF);
193
194 len -= XFS_LBSIZE(mp);
195 ptr += XFS_LBSIZE(mp);
196 bno += mp->m_bsize;
197 }
198 ASSERT(len == 0);
199 }
200
201 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
202 .verify_read = xfs_attr3_rmt_read_verify,
203 .verify_write = xfs_attr3_rmt_write_verify,
204 };
205
206 STATIC int
xfs_attr3_rmt_hdr_set(struct xfs_mount * mp,void * ptr,xfs_ino_t ino,uint32_t offset,uint32_t size,xfs_daddr_t bno)207 xfs_attr3_rmt_hdr_set(
208 struct xfs_mount *mp,
209 void *ptr,
210 xfs_ino_t ino,
211 uint32_t offset,
212 uint32_t size,
213 xfs_daddr_t bno)
214 {
215 struct xfs_attr3_rmt_hdr *rmt = ptr;
216
217 if (!xfs_sb_version_hascrc(&mp->m_sb))
218 return 0;
219
220 rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
221 rmt->rm_offset = cpu_to_be32(offset);
222 rmt->rm_bytes = cpu_to_be32(size);
223 uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_uuid);
224 rmt->rm_owner = cpu_to_be64(ino);
225 rmt->rm_blkno = cpu_to_be64(bno);
226
227 return sizeof(struct xfs_attr3_rmt_hdr);
228 }
229
230 /*
231 * Helper functions to copy attribute data in and out of the one disk extents
232 */
233 STATIC int
xfs_attr_rmtval_copyout(struct xfs_mount * mp,struct xfs_buf * bp,xfs_ino_t ino,int * offset,int * valuelen,char ** dst)234 xfs_attr_rmtval_copyout(
235 struct xfs_mount *mp,
236 struct xfs_buf *bp,
237 xfs_ino_t ino,
238 int *offset,
239 int *valuelen,
240 char **dst)
241 {
242 char *src = bp->b_addr;
243 xfs_daddr_t bno = bp->b_bn;
244 int len = BBTOB(bp->b_length);
245
246 ASSERT(len >= XFS_LBSIZE(mp));
247
248 while (len > 0 && *valuelen > 0) {
249 int hdr_size = 0;
250 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
251
252 byte_cnt = min_t(int, *valuelen, byte_cnt);
253
254 if (xfs_sb_version_hascrc(&mp->m_sb)) {
255 if (!xfs_attr3_rmt_hdr_ok(mp, src, ino, *offset,
256 byte_cnt, bno)) {
257 xfs_alert(mp,
258 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
259 bno, *offset, byte_cnt, ino);
260 return EFSCORRUPTED;
261 }
262 hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
263 }
264
265 memcpy(*dst, src + hdr_size, byte_cnt);
266
267 /* roll buffer forwards */
268 len -= XFS_LBSIZE(mp);
269 src += XFS_LBSIZE(mp);
270 bno += mp->m_bsize;
271
272 /* roll attribute data forwards */
273 *valuelen -= byte_cnt;
274 *dst += byte_cnt;
275 *offset += byte_cnt;
276 }
277 return 0;
278 }
279
280 STATIC void
xfs_attr_rmtval_copyin(struct xfs_mount * mp,struct xfs_buf * bp,xfs_ino_t ino,int * offset,int * valuelen,char ** src)281 xfs_attr_rmtval_copyin(
282 struct xfs_mount *mp,
283 struct xfs_buf *bp,
284 xfs_ino_t ino,
285 int *offset,
286 int *valuelen,
287 char **src)
288 {
289 char *dst = bp->b_addr;
290 xfs_daddr_t bno = bp->b_bn;
291 int len = BBTOB(bp->b_length);
292
293 ASSERT(len >= XFS_LBSIZE(mp));
294
295 while (len > 0 && *valuelen > 0) {
296 int hdr_size;
297 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, XFS_LBSIZE(mp));
298
299 byte_cnt = min(*valuelen, byte_cnt);
300 hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
301 byte_cnt, bno);
302
303 memcpy(dst + hdr_size, *src, byte_cnt);
304
305 /*
306 * If this is the last block, zero the remainder of it.
307 * Check that we are actually the last block, too.
308 */
309 if (byte_cnt + hdr_size < XFS_LBSIZE(mp)) {
310 ASSERT(*valuelen - byte_cnt == 0);
311 ASSERT(len == XFS_LBSIZE(mp));
312 memset(dst + hdr_size + byte_cnt, 0,
313 XFS_LBSIZE(mp) - hdr_size - byte_cnt);
314 }
315
316 /* roll buffer forwards */
317 len -= XFS_LBSIZE(mp);
318 dst += XFS_LBSIZE(mp);
319 bno += mp->m_bsize;
320
321 /* roll attribute data forwards */
322 *valuelen -= byte_cnt;
323 *src += byte_cnt;
324 *offset += byte_cnt;
325 }
326 }
327
328 /*
329 * Read the value associated with an attribute from the out-of-line buffer
330 * that we stored it in.
331 */
332 int
xfs_attr_rmtval_get(struct xfs_da_args * args)333 xfs_attr_rmtval_get(
334 struct xfs_da_args *args)
335 {
336 struct xfs_bmbt_irec map[ATTR_RMTVALUE_MAPSIZE];
337 struct xfs_mount *mp = args->dp->i_mount;
338 struct xfs_buf *bp;
339 xfs_dablk_t lblkno = args->rmtblkno;
340 char *dst = args->value;
341 int valuelen = args->valuelen;
342 int nmap;
343 int error;
344 int blkcnt = args->rmtblkcnt;
345 int i;
346 int offset = 0;
347
348 trace_xfs_attr_rmtval_get(args);
349
350 ASSERT(!(args->flags & ATTR_KERNOVAL));
351
352 while (valuelen > 0) {
353 nmap = ATTR_RMTVALUE_MAPSIZE;
354 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
355 blkcnt, map, &nmap,
356 XFS_BMAPI_ATTRFORK);
357 if (error)
358 return error;
359 ASSERT(nmap >= 1);
360
361 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
362 xfs_daddr_t dblkno;
363 int dblkcnt;
364
365 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
366 (map[i].br_startblock != HOLESTARTBLOCK));
367 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
368 dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
369 error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
370 dblkno, dblkcnt, 0, &bp,
371 &xfs_attr3_rmt_buf_ops);
372 if (error)
373 return error;
374
375 error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
376 &offset, &valuelen,
377 &dst);
378 xfs_buf_relse(bp);
379 if (error)
380 return error;
381
382 /* roll attribute extent map forwards */
383 lblkno += map[i].br_blockcount;
384 blkcnt -= map[i].br_blockcount;
385 }
386 }
387 ASSERT(valuelen == 0);
388 return 0;
389 }
390
391 /*
392 * Write the value associated with an attribute into the out-of-line buffer
393 * that we have defined for it.
394 */
395 int
xfs_attr_rmtval_set(struct xfs_da_args * args)396 xfs_attr_rmtval_set(
397 struct xfs_da_args *args)
398 {
399 struct xfs_inode *dp = args->dp;
400 struct xfs_mount *mp = dp->i_mount;
401 struct xfs_bmbt_irec map;
402 xfs_dablk_t lblkno;
403 xfs_fileoff_t lfileoff = 0;
404 char *src = args->value;
405 int blkcnt;
406 int valuelen;
407 int nmap;
408 int error;
409 int offset = 0;
410
411 trace_xfs_attr_rmtval_set(args);
412
413 /*
414 * Find a "hole" in the attribute address space large enough for
415 * us to drop the new attribute's value into. Because CRC enable
416 * attributes have headers, we can't just do a straight byte to FSB
417 * conversion and have to take the header space into account.
418 */
419 blkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
420 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
421 XFS_ATTR_FORK);
422 if (error)
423 return error;
424
425 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
426 args->rmtblkcnt = blkcnt;
427
428 /*
429 * Roll through the "value", allocating blocks on disk as required.
430 */
431 while (blkcnt > 0) {
432 int committed;
433
434 /*
435 * Allocate a single extent, up to the size of the value.
436 */
437 xfs_bmap_init(args->flist, args->firstblock);
438 nmap = 1;
439 error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
440 blkcnt,
441 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
442 args->firstblock, args->total, &map, &nmap,
443 args->flist);
444 if (!error) {
445 error = xfs_bmap_finish(&args->trans, args->flist,
446 &committed);
447 }
448 if (error) {
449 ASSERT(committed);
450 args->trans = NULL;
451 xfs_bmap_cancel(args->flist);
452 return(error);
453 }
454
455 /*
456 * bmap_finish() may have committed the last trans and started
457 * a new one. We need the inode to be in all transactions.
458 */
459 if (committed)
460 xfs_trans_ijoin(args->trans, dp, 0);
461
462 ASSERT(nmap == 1);
463 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
464 (map.br_startblock != HOLESTARTBLOCK));
465 lblkno += map.br_blockcount;
466 blkcnt -= map.br_blockcount;
467
468 /*
469 * Start the next trans in the chain.
470 */
471 error = xfs_trans_roll(&args->trans, dp);
472 if (error)
473 return (error);
474 }
475
476 /*
477 * Roll through the "value", copying the attribute value to the
478 * already-allocated blocks. Blocks are written synchronously
479 * so that we can know they are all on disk before we turn off
480 * the INCOMPLETE flag.
481 */
482 lblkno = args->rmtblkno;
483 blkcnt = args->rmtblkcnt;
484 valuelen = args->valuelen;
485 while (valuelen > 0) {
486 struct xfs_buf *bp;
487 xfs_daddr_t dblkno;
488 int dblkcnt;
489
490 ASSERT(blkcnt > 0);
491
492 xfs_bmap_init(args->flist, args->firstblock);
493 nmap = 1;
494 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
495 blkcnt, &map, &nmap,
496 XFS_BMAPI_ATTRFORK);
497 if (error)
498 return(error);
499 ASSERT(nmap == 1);
500 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
501 (map.br_startblock != HOLESTARTBLOCK));
502
503 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
504 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
505
506 bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
507 if (!bp)
508 return ENOMEM;
509 bp->b_ops = &xfs_attr3_rmt_buf_ops;
510
511 xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
512 &valuelen, &src);
513
514 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
515 xfs_buf_relse(bp);
516 if (error)
517 return error;
518
519
520 /* roll attribute extent map forwards */
521 lblkno += map.br_blockcount;
522 blkcnt -= map.br_blockcount;
523 }
524 ASSERT(valuelen == 0);
525 return 0;
526 }
527
528 /*
529 * Remove the value associated with an attribute by deleting the
530 * out-of-line buffer that it is stored on.
531 */
532 int
xfs_attr_rmtval_remove(struct xfs_da_args * args)533 xfs_attr_rmtval_remove(
534 struct xfs_da_args *args)
535 {
536 struct xfs_mount *mp = args->dp->i_mount;
537 xfs_dablk_t lblkno;
538 int blkcnt;
539 int error;
540 int done;
541
542 trace_xfs_attr_rmtval_remove(args);
543
544 /*
545 * Roll through the "value", invalidating the attribute value's blocks.
546 * Note that args->rmtblkcnt is the minimum number of data blocks we'll
547 * see for a CRC enabled remote attribute. Each extent will have a
548 * header, and so we may have more blocks than we realise here. If we
549 * fail to map the blocks correctly, we'll have problems with the buffer
550 * lookups.
551 */
552 lblkno = args->rmtblkno;
553 blkcnt = args->rmtblkcnt;
554 while (blkcnt > 0) {
555 struct xfs_bmbt_irec map;
556 struct xfs_buf *bp;
557 xfs_daddr_t dblkno;
558 int dblkcnt;
559 int nmap;
560
561 /*
562 * Try to remember where we decided to put the value.
563 */
564 nmap = 1;
565 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
566 blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
567 if (error)
568 return(error);
569 ASSERT(nmap == 1);
570 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
571 (map.br_startblock != HOLESTARTBLOCK));
572
573 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
574 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
575
576 /*
577 * If the "remote" value is in the cache, remove it.
578 */
579 bp = xfs_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
580 if (bp) {
581 xfs_buf_stale(bp);
582 xfs_buf_relse(bp);
583 bp = NULL;
584 }
585
586 lblkno += map.br_blockcount;
587 blkcnt -= map.br_blockcount;
588 }
589
590 /*
591 * Keep de-allocating extents until the remote-value region is gone.
592 */
593 lblkno = args->rmtblkno;
594 blkcnt = args->rmtblkcnt;
595 done = 0;
596 while (!done) {
597 int committed;
598
599 xfs_bmap_init(args->flist, args->firstblock);
600 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
601 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
602 1, args->firstblock, args->flist,
603 &done);
604 if (!error) {
605 error = xfs_bmap_finish(&args->trans, args->flist,
606 &committed);
607 }
608 if (error) {
609 ASSERT(committed);
610 args->trans = NULL;
611 xfs_bmap_cancel(args->flist);
612 return error;
613 }
614
615 /*
616 * bmap_finish() may have committed the last trans and started
617 * a new one. We need the inode to be in all transactions.
618 */
619 if (committed)
620 xfs_trans_ijoin(args->trans, args->dp, 0);
621
622 /*
623 * Close out trans and start the next one in the chain.
624 */
625 error = xfs_trans_roll(&args->trans, args->dp);
626 if (error)
627 return (error);
628 }
629 return(0);
630 }
631
632