Loading...
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_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_bit.h"
25#include "xfs_mount.h"
26#include "xfs_da_format.h"
27#include "xfs_da_btree.h"
28#include "xfs_inode.h"
29#include "xfs_dir2.h"
30#include "xfs_dir2_priv.h"
31#include "xfs_error.h"
32#include "xfs_trace.h"
33#include "xfs_bmap.h"
34#include "xfs_trans.h"
35
36/*
37 * Directory file type support functions
38 */
39static unsigned char xfs_dir3_filetype_table[] = {
40 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
41 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
42};
43
44static unsigned char
45xfs_dir3_get_dtype(
46 struct xfs_mount *mp,
47 __uint8_t filetype)
48{
49 if (!xfs_sb_version_hasftype(&mp->m_sb))
50 return DT_UNKNOWN;
51
52 if (filetype >= XFS_DIR3_FT_MAX)
53 return DT_UNKNOWN;
54
55 return xfs_dir3_filetype_table[filetype];
56}
57
58STATIC int
59xfs_dir2_sf_getdents(
60 struct xfs_da_args *args,
61 struct dir_context *ctx)
62{
63 int i; /* shortform entry number */
64 struct xfs_inode *dp = args->dp; /* incore directory inode */
65 xfs_dir2_dataptr_t off; /* current entry's offset */
66 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
67 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
68 xfs_dir2_dataptr_t dot_offset;
69 xfs_dir2_dataptr_t dotdot_offset;
70 xfs_ino_t ino;
71 struct xfs_da_geometry *geo = args->geo;
72
73 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
74 /*
75 * Give up if the directory is way too short.
76 */
77 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
78 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
79 return -EIO;
80 }
81
82 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
83 ASSERT(dp->i_df.if_u1.if_data != NULL);
84
85 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
86
87 if (dp->i_d.di_size < xfs_dir2_sf_hdr_size(sfp->i8count))
88 return -EFSCORRUPTED;
89
90 /*
91 * If the block number in the offset is out of range, we're done.
92 */
93 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
94 return 0;
95
96 /*
97 * Precalculate offsets for . and .. as we will always need them.
98 *
99 * XXX(hch): the second argument is sometimes 0 and sometimes
100 * geo->datablk
101 */
102 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
103 dp->d_ops->data_dot_offset);
104 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
105 dp->d_ops->data_dotdot_offset);
106
107 /*
108 * Put . entry unless we're starting past it.
109 */
110 if (ctx->pos <= dot_offset) {
111 ctx->pos = dot_offset & 0x7fffffff;
112 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
113 return 0;
114 }
115
116 /*
117 * Put .. entry unless we're starting past it.
118 */
119 if (ctx->pos <= dotdot_offset) {
120 ino = dp->d_ops->sf_get_parent_ino(sfp);
121 ctx->pos = dotdot_offset & 0x7fffffff;
122 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
123 return 0;
124 }
125
126 /*
127 * Loop while there are more entries and put'ing works.
128 */
129 sfep = xfs_dir2_sf_firstentry(sfp);
130 for (i = 0; i < sfp->count; i++) {
131 __uint8_t filetype;
132
133 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
134 xfs_dir2_sf_get_offset(sfep));
135
136 if (ctx->pos > off) {
137 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
138 continue;
139 }
140
141 ino = dp->d_ops->sf_get_ino(sfp, sfep);
142 filetype = dp->d_ops->sf_get_ftype(sfep);
143 ctx->pos = off & 0x7fffffff;
144 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
145 xfs_dir3_get_dtype(dp->i_mount, filetype)))
146 return 0;
147 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
148 }
149
150 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
151 0x7fffffff;
152 return 0;
153}
154
155/*
156 * Readdir for block directories.
157 */
158STATIC int
159xfs_dir2_block_getdents(
160 struct xfs_da_args *args,
161 struct dir_context *ctx)
162{
163 struct xfs_inode *dp = args->dp; /* incore directory inode */
164 xfs_dir2_data_hdr_t *hdr; /* block header */
165 struct xfs_buf *bp; /* buffer for block */
166 xfs_dir2_block_tail_t *btp; /* block tail */
167 xfs_dir2_data_entry_t *dep; /* block data entry */
168 xfs_dir2_data_unused_t *dup; /* block unused entry */
169 char *endptr; /* end of the data entries */
170 int error; /* error return value */
171 char *ptr; /* current data entry */
172 int wantoff; /* starting block offset */
173 xfs_off_t cook;
174 struct xfs_da_geometry *geo = args->geo;
175 int lock_mode;
176
177 /*
178 * If the block number in the offset is out of range, we're done.
179 */
180 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
181 return 0;
182
183 lock_mode = xfs_ilock_data_map_shared(dp);
184 error = xfs_dir3_block_read(NULL, dp, &bp);
185 xfs_iunlock(dp, lock_mode);
186 if (error)
187 return error;
188
189 /*
190 * Extract the byte offset we start at from the seek pointer.
191 * We'll skip entries before this.
192 */
193 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
194 hdr = bp->b_addr;
195 xfs_dir3_data_check(dp, bp);
196 /*
197 * Set up values for the loop.
198 */
199 btp = xfs_dir2_block_tail_p(geo, hdr);
200 ptr = (char *)dp->d_ops->data_entry_p(hdr);
201 endptr = (char *)xfs_dir2_block_leaf_p(btp);
202
203 /*
204 * Loop over the data portion of the block.
205 * Each object is a real entry (dep) or an unused one (dup).
206 */
207 while (ptr < endptr) {
208 __uint8_t filetype;
209
210 dup = (xfs_dir2_data_unused_t *)ptr;
211 /*
212 * Unused, skip it.
213 */
214 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
215 ptr += be16_to_cpu(dup->length);
216 continue;
217 }
218
219 dep = (xfs_dir2_data_entry_t *)ptr;
220
221 /*
222 * Bump pointer for the next iteration.
223 */
224 ptr += dp->d_ops->data_entsize(dep->namelen);
225 /*
226 * The entry is before the desired starting point, skip it.
227 */
228 if ((char *)dep - (char *)hdr < wantoff)
229 continue;
230
231 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
232 (char *)dep - (char *)hdr);
233
234 ctx->pos = cook & 0x7fffffff;
235 filetype = dp->d_ops->data_get_ftype(dep);
236 /*
237 * If it didn't fit, set the final offset to here & return.
238 */
239 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
240 be64_to_cpu(dep->inumber),
241 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
242 xfs_trans_brelse(NULL, bp);
243 return 0;
244 }
245 }
246
247 /*
248 * Reached the end of the block.
249 * Set the offset to a non-existent block 1 and return.
250 */
251 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
252 0x7fffffff;
253 xfs_trans_brelse(NULL, bp);
254 return 0;
255}
256
257struct xfs_dir2_leaf_map_info {
258 xfs_extlen_t map_blocks; /* number of fsbs in map */
259 xfs_dablk_t map_off; /* last mapped file offset */
260 int map_size; /* total entries in *map */
261 int map_valid; /* valid entries in *map */
262 int nmap; /* mappings to ask xfs_bmapi */
263 xfs_dir2_db_t curdb; /* db for current block */
264 int ra_current; /* number of read-ahead blks */
265 int ra_index; /* *map index for read-ahead */
266 int ra_offset; /* map entry offset for ra */
267 int ra_want; /* readahead count wanted */
268 struct xfs_bmbt_irec map[]; /* map vector for blocks */
269};
270
271STATIC int
272xfs_dir2_leaf_readbuf(
273 struct xfs_da_args *args,
274 size_t bufsize,
275 struct xfs_dir2_leaf_map_info *mip,
276 xfs_dir2_off_t *curoff,
277 struct xfs_buf **bpp,
278 bool trim_map)
279{
280 struct xfs_inode *dp = args->dp;
281 struct xfs_buf *bp = NULL;
282 struct xfs_bmbt_irec *map = mip->map;
283 struct blk_plug plug;
284 int error = 0;
285 int length;
286 int i;
287 int j;
288 struct xfs_da_geometry *geo = args->geo;
289
290 /*
291 * If the caller just finished processing a buffer, it will tell us
292 * we need to trim that block out of the mapping now it is done.
293 */
294 if (trim_map) {
295 mip->map_blocks -= geo->fsbcount;
296 /*
297 * Loop to get rid of the extents for the
298 * directory block.
299 */
300 for (i = geo->fsbcount; i > 0; ) {
301 j = min_t(int, map->br_blockcount, i);
302 map->br_blockcount -= j;
303 map->br_startblock += j;
304 map->br_startoff += j;
305 /*
306 * If mapping is done, pitch it from
307 * the table.
308 */
309 if (!map->br_blockcount && --mip->map_valid)
310 memmove(&map[0], &map[1],
311 sizeof(map[0]) * mip->map_valid);
312 i -= j;
313 }
314 }
315
316 /*
317 * Recalculate the readahead blocks wanted.
318 */
319 mip->ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog)) - 1;
320 ASSERT(mip->ra_want >= 0);
321
322 /*
323 * If we don't have as many as we want, and we haven't
324 * run out of data blocks, get some more mappings.
325 */
326 if (1 + mip->ra_want > mip->map_blocks &&
327 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
328 /*
329 * Get more bmaps, fill in after the ones
330 * we already have in the table.
331 */
332 mip->nmap = mip->map_size - mip->map_valid;
333 error = xfs_bmapi_read(dp, mip->map_off,
334 xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET) -
335 mip->map_off,
336 &map[mip->map_valid], &mip->nmap, 0);
337
338 /*
339 * Don't know if we should ignore this or try to return an
340 * error. The trouble with returning errors is that readdir
341 * will just stop without actually passing the error through.
342 */
343 if (error)
344 goto out; /* XXX */
345
346 /*
347 * If we got all the mappings we asked for, set the final map
348 * offset based on the last bmap value received. Otherwise,
349 * we've reached the end.
350 */
351 if (mip->nmap == mip->map_size - mip->map_valid) {
352 i = mip->map_valid + mip->nmap - 1;
353 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
354 } else
355 mip->map_off = xfs_dir2_byte_to_da(geo,
356 XFS_DIR2_LEAF_OFFSET);
357
358 /*
359 * Look for holes in the mapping, and eliminate them. Count up
360 * the valid blocks.
361 */
362 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
363 if (map[i].br_startblock == HOLESTARTBLOCK) {
364 mip->nmap--;
365 length = mip->map_valid + mip->nmap - i;
366 if (length)
367 memmove(&map[i], &map[i + 1],
368 sizeof(map[i]) * length);
369 } else {
370 mip->map_blocks += map[i].br_blockcount;
371 i++;
372 }
373 }
374 mip->map_valid += mip->nmap;
375 }
376
377 /*
378 * No valid mappings, so no more data blocks.
379 */
380 if (!mip->map_valid) {
381 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
382 goto out;
383 }
384
385 /*
386 * Read the directory block starting at the first mapping.
387 */
388 mip->curdb = xfs_dir2_da_to_db(geo, map->br_startoff);
389 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
390 map->br_blockcount >= geo->fsbcount ?
391 XFS_FSB_TO_DADDR(dp->i_mount, map->br_startblock) :
392 -1, &bp);
393 /*
394 * Should just skip over the data block instead of giving up.
395 */
396 if (error)
397 goto out; /* XXX */
398
399 /*
400 * Adjust the current amount of read-ahead: we just read a block that
401 * was previously ra.
402 */
403 if (mip->ra_current)
404 mip->ra_current -= geo->fsbcount;
405
406 /*
407 * Do we need more readahead?
408 */
409 blk_start_plug(&plug);
410 for (mip->ra_index = mip->ra_offset = i = 0;
411 mip->ra_want > mip->ra_current && i < mip->map_blocks;
412 i += geo->fsbcount) {
413 ASSERT(mip->ra_index < mip->map_valid);
414 /*
415 * Read-ahead a contiguous directory block.
416 */
417 if (i > mip->ra_current &&
418 map[mip->ra_index].br_blockcount >= geo->fsbcount) {
419 xfs_dir3_data_readahead(dp,
420 map[mip->ra_index].br_startoff + mip->ra_offset,
421 XFS_FSB_TO_DADDR(dp->i_mount,
422 map[mip->ra_index].br_startblock +
423 mip->ra_offset));
424 mip->ra_current = i;
425 }
426
427 /*
428 * Read-ahead a non-contiguous directory block. This doesn't
429 * use our mapping, but this is a very rare case.
430 */
431 else if (i > mip->ra_current) {
432 xfs_dir3_data_readahead(dp,
433 map[mip->ra_index].br_startoff +
434 mip->ra_offset, -1);
435 mip->ra_current = i;
436 }
437
438 /*
439 * Advance offset through the mapping table.
440 */
441 for (j = 0; j < geo->fsbcount; j += length ) {
442 /*
443 * The rest of this extent but not more than a dir
444 * block.
445 */
446 length = min_t(int, geo->fsbcount,
447 map[mip->ra_index].br_blockcount -
448 mip->ra_offset);
449 mip->ra_offset += length;
450
451 /*
452 * Advance to the next mapping if this one is used up.
453 */
454 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
455 mip->ra_offset = 0;
456 mip->ra_index++;
457 }
458 }
459 }
460 blk_finish_plug(&plug);
461
462out:
463 *bpp = bp;
464 return error;
465}
466
467/*
468 * Getdents (readdir) for leaf and node directories.
469 * This reads the data blocks only, so is the same for both forms.
470 */
471STATIC int
472xfs_dir2_leaf_getdents(
473 struct xfs_da_args *args,
474 struct dir_context *ctx,
475 size_t bufsize)
476{
477 struct xfs_inode *dp = args->dp;
478 struct xfs_buf *bp = NULL; /* data block buffer */
479 xfs_dir2_data_hdr_t *hdr; /* data block header */
480 xfs_dir2_data_entry_t *dep; /* data entry */
481 xfs_dir2_data_unused_t *dup; /* unused entry */
482 int error = 0; /* error return value */
483 int length; /* temporary length value */
484 int byteoff; /* offset in current block */
485 xfs_dir2_off_t curoff; /* current overall offset */
486 xfs_dir2_off_t newoff; /* new curoff after new blk */
487 char *ptr = NULL; /* pointer to current data */
488 struct xfs_dir2_leaf_map_info *map_info;
489 struct xfs_da_geometry *geo = args->geo;
490
491 /*
492 * If the offset is at or past the largest allowed value,
493 * give up right away.
494 */
495 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
496 return 0;
497
498 /*
499 * Set up to bmap a number of blocks based on the caller's
500 * buffer size, the directory block size, and the filesystem
501 * block size.
502 */
503 length = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
504 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
505 (length * sizeof(struct xfs_bmbt_irec)),
506 KM_SLEEP | KM_NOFS);
507 map_info->map_size = length;
508
509 /*
510 * Inside the loop we keep the main offset value as a byte offset
511 * in the directory file.
512 */
513 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
514
515 /*
516 * Force this conversion through db so we truncate the offset
517 * down to get the start of the data block.
518 */
519 map_info->map_off = xfs_dir2_db_to_da(geo,
520 xfs_dir2_byte_to_db(geo, curoff));
521
522 /*
523 * Loop over directory entries until we reach the end offset.
524 * Get more blocks and readahead as necessary.
525 */
526 while (curoff < XFS_DIR2_LEAF_OFFSET) {
527 __uint8_t filetype;
528
529 /*
530 * If we have no buffer, or we're off the end of the
531 * current buffer, need to get another one.
532 */
533 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
534 int lock_mode;
535 bool trim_map = false;
536
537 if (bp) {
538 xfs_trans_brelse(NULL, bp);
539 bp = NULL;
540 trim_map = true;
541 }
542
543 lock_mode = xfs_ilock_data_map_shared(dp);
544 error = xfs_dir2_leaf_readbuf(args, bufsize, map_info,
545 &curoff, &bp, trim_map);
546 xfs_iunlock(dp, lock_mode);
547 if (error || !map_info->map_valid)
548 break;
549
550 /*
551 * Having done a read, we need to set a new offset.
552 */
553 newoff = xfs_dir2_db_off_to_byte(geo,
554 map_info->curdb, 0);
555 /*
556 * Start of the current block.
557 */
558 if (curoff < newoff)
559 curoff = newoff;
560 /*
561 * Make sure we're in the right block.
562 */
563 else if (curoff > newoff)
564 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
565 map_info->curdb);
566 hdr = bp->b_addr;
567 xfs_dir3_data_check(dp, bp);
568 /*
569 * Find our position in the block.
570 */
571 ptr = (char *)dp->d_ops->data_entry_p(hdr);
572 byteoff = xfs_dir2_byte_to_off(geo, curoff);
573 /*
574 * Skip past the header.
575 */
576 if (byteoff == 0)
577 curoff += dp->d_ops->data_entry_offset;
578 /*
579 * Skip past entries until we reach our offset.
580 */
581 else {
582 while ((char *)ptr - (char *)hdr < byteoff) {
583 dup = (xfs_dir2_data_unused_t *)ptr;
584
585 if (be16_to_cpu(dup->freetag)
586 == XFS_DIR2_DATA_FREE_TAG) {
587
588 length = be16_to_cpu(dup->length);
589 ptr += length;
590 continue;
591 }
592 dep = (xfs_dir2_data_entry_t *)ptr;
593 length =
594 dp->d_ops->data_entsize(dep->namelen);
595 ptr += length;
596 }
597 /*
598 * Now set our real offset.
599 */
600 curoff =
601 xfs_dir2_db_off_to_byte(geo,
602 xfs_dir2_byte_to_db(geo, curoff),
603 (char *)ptr - (char *)hdr);
604 if (ptr >= (char *)hdr + geo->blksize) {
605 continue;
606 }
607 }
608 }
609 /*
610 * We have a pointer to an entry.
611 * Is it a live one?
612 */
613 dup = (xfs_dir2_data_unused_t *)ptr;
614 /*
615 * No, it's unused, skip over it.
616 */
617 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
618 length = be16_to_cpu(dup->length);
619 ptr += length;
620 curoff += length;
621 continue;
622 }
623
624 dep = (xfs_dir2_data_entry_t *)ptr;
625 length = dp->d_ops->data_entsize(dep->namelen);
626 filetype = dp->d_ops->data_get_ftype(dep);
627
628 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
629 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
630 be64_to_cpu(dep->inumber),
631 xfs_dir3_get_dtype(dp->i_mount, filetype)))
632 break;
633
634 /*
635 * Advance to next entry in the block.
636 */
637 ptr += length;
638 curoff += length;
639 /* bufsize may have just been a guess; don't go negative */
640 bufsize = bufsize > length ? bufsize - length : 0;
641 }
642
643 /*
644 * All done. Set output offset value to current offset.
645 */
646 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
647 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
648 else
649 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
650 kmem_free(map_info);
651 if (bp)
652 xfs_trans_brelse(NULL, bp);
653 return error;
654}
655
656/*
657 * Read a directory.
658 */
659int
660xfs_readdir(
661 struct xfs_inode *dp,
662 struct dir_context *ctx,
663 size_t bufsize)
664{
665 struct xfs_da_args args = { NULL };
666 int rval;
667 int v;
668
669 trace_xfs_readdir(dp);
670
671 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
672 return -EIO;
673
674 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
675 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
676
677 args.dp = dp;
678 args.geo = dp->i_mount->m_dir_geo;
679
680 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
681 rval = xfs_dir2_sf_getdents(&args, ctx);
682 else if ((rval = xfs_dir2_isblock(&args, &v)))
683 ;
684 else if (v)
685 rval = xfs_dir2_block_getdents(&args, ctx);
686 else
687 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
688
689 return rval;
690}
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_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_bit.h"
25#include "xfs_mount.h"
26#include "xfs_da_format.h"
27#include "xfs_da_btree.h"
28#include "xfs_inode.h"
29#include "xfs_dir2.h"
30#include "xfs_dir2_priv.h"
31#include "xfs_error.h"
32#include "xfs_trace.h"
33#include "xfs_bmap.h"
34#include "xfs_trans.h"
35
36/*
37 * Directory file type support functions
38 */
39static unsigned char xfs_dir3_filetype_table[] = {
40 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
41 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
42};
43
44static unsigned char
45xfs_dir3_get_dtype(
46 struct xfs_mount *mp,
47 __uint8_t filetype)
48{
49 if (!xfs_sb_version_hasftype(&mp->m_sb))
50 return DT_UNKNOWN;
51
52 if (filetype >= XFS_DIR3_FT_MAX)
53 return DT_UNKNOWN;
54
55 return xfs_dir3_filetype_table[filetype];
56}
57
58STATIC int
59xfs_dir2_sf_getdents(
60 struct xfs_da_args *args,
61 struct dir_context *ctx)
62{
63 int i; /* shortform entry number */
64 struct xfs_inode *dp = args->dp; /* incore directory inode */
65 xfs_dir2_dataptr_t off; /* current entry's offset */
66 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
67 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
68 xfs_dir2_dataptr_t dot_offset;
69 xfs_dir2_dataptr_t dotdot_offset;
70 xfs_ino_t ino;
71 struct xfs_da_geometry *geo = args->geo;
72
73 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
74 /*
75 * Give up if the directory is way too short.
76 */
77 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
78 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
79 return -EIO;
80 }
81
82 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
83 ASSERT(dp->i_df.if_u1.if_data != NULL);
84
85 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
86
87 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
88
89 /*
90 * If the block number in the offset is out of range, we're done.
91 */
92 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
93 return 0;
94
95 /*
96 * Precalculate offsets for . and .. as we will always need them.
97 *
98 * XXX(hch): the second argument is sometimes 0 and sometimes
99 * geo->datablk
100 */
101 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
102 dp->d_ops->data_dot_offset);
103 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
104 dp->d_ops->data_dotdot_offset);
105
106 /*
107 * Put . entry unless we're starting past it.
108 */
109 if (ctx->pos <= dot_offset) {
110 ctx->pos = dot_offset & 0x7fffffff;
111 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
112 return 0;
113 }
114
115 /*
116 * Put .. entry unless we're starting past it.
117 */
118 if (ctx->pos <= dotdot_offset) {
119 ino = dp->d_ops->sf_get_parent_ino(sfp);
120 ctx->pos = dotdot_offset & 0x7fffffff;
121 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
122 return 0;
123 }
124
125 /*
126 * Loop while there are more entries and put'ing works.
127 */
128 sfep = xfs_dir2_sf_firstentry(sfp);
129 for (i = 0; i < sfp->count; i++) {
130 __uint8_t filetype;
131
132 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
133 xfs_dir2_sf_get_offset(sfep));
134
135 if (ctx->pos > off) {
136 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
137 continue;
138 }
139
140 ino = dp->d_ops->sf_get_ino(sfp, sfep);
141 filetype = dp->d_ops->sf_get_ftype(sfep);
142 ctx->pos = off & 0x7fffffff;
143 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
144 xfs_dir3_get_dtype(dp->i_mount, filetype)))
145 return 0;
146 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
147 }
148
149 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
150 0x7fffffff;
151 return 0;
152}
153
154/*
155 * Readdir for block directories.
156 */
157STATIC int
158xfs_dir2_block_getdents(
159 struct xfs_da_args *args,
160 struct dir_context *ctx)
161{
162 struct xfs_inode *dp = args->dp; /* incore directory inode */
163 xfs_dir2_data_hdr_t *hdr; /* block header */
164 struct xfs_buf *bp; /* buffer for block */
165 xfs_dir2_block_tail_t *btp; /* block tail */
166 xfs_dir2_data_entry_t *dep; /* block data entry */
167 xfs_dir2_data_unused_t *dup; /* block unused entry */
168 char *endptr; /* end of the data entries */
169 int error; /* error return value */
170 char *ptr; /* current data entry */
171 int wantoff; /* starting block offset */
172 xfs_off_t cook;
173 struct xfs_da_geometry *geo = args->geo;
174 int lock_mode;
175
176 /*
177 * If the block number in the offset is out of range, we're done.
178 */
179 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
180 return 0;
181
182 lock_mode = xfs_ilock_data_map_shared(dp);
183 error = xfs_dir3_block_read(NULL, dp, &bp);
184 xfs_iunlock(dp, lock_mode);
185 if (error)
186 return error;
187
188 /*
189 * Extract the byte offset we start at from the seek pointer.
190 * We'll skip entries before this.
191 */
192 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
193 hdr = bp->b_addr;
194 xfs_dir3_data_check(dp, bp);
195 /*
196 * Set up values for the loop.
197 */
198 btp = xfs_dir2_block_tail_p(geo, hdr);
199 ptr = (char *)dp->d_ops->data_entry_p(hdr);
200 endptr = (char *)xfs_dir2_block_leaf_p(btp);
201
202 /*
203 * Loop over the data portion of the block.
204 * Each object is a real entry (dep) or an unused one (dup).
205 */
206 while (ptr < endptr) {
207 __uint8_t filetype;
208
209 dup = (xfs_dir2_data_unused_t *)ptr;
210 /*
211 * Unused, skip it.
212 */
213 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
214 ptr += be16_to_cpu(dup->length);
215 continue;
216 }
217
218 dep = (xfs_dir2_data_entry_t *)ptr;
219
220 /*
221 * Bump pointer for the next iteration.
222 */
223 ptr += dp->d_ops->data_entsize(dep->namelen);
224 /*
225 * The entry is before the desired starting point, skip it.
226 */
227 if ((char *)dep - (char *)hdr < wantoff)
228 continue;
229
230 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
231 (char *)dep - (char *)hdr);
232
233 ctx->pos = cook & 0x7fffffff;
234 filetype = dp->d_ops->data_get_ftype(dep);
235 /*
236 * If it didn't fit, set the final offset to here & return.
237 */
238 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
239 be64_to_cpu(dep->inumber),
240 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
241 xfs_trans_brelse(NULL, bp);
242 return 0;
243 }
244 }
245
246 /*
247 * Reached the end of the block.
248 * Set the offset to a non-existent block 1 and return.
249 */
250 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
251 0x7fffffff;
252 xfs_trans_brelse(NULL, bp);
253 return 0;
254}
255
256struct xfs_dir2_leaf_map_info {
257 xfs_extlen_t map_blocks; /* number of fsbs in map */
258 xfs_dablk_t map_off; /* last mapped file offset */
259 int map_size; /* total entries in *map */
260 int map_valid; /* valid entries in *map */
261 int nmap; /* mappings to ask xfs_bmapi */
262 xfs_dir2_db_t curdb; /* db for current block */
263 int ra_current; /* number of read-ahead blks */
264 int ra_index; /* *map index for read-ahead */
265 int ra_offset; /* map entry offset for ra */
266 int ra_want; /* readahead count wanted */
267 struct xfs_bmbt_irec map[]; /* map vector for blocks */
268};
269
270STATIC int
271xfs_dir2_leaf_readbuf(
272 struct xfs_da_args *args,
273 size_t bufsize,
274 struct xfs_dir2_leaf_map_info *mip,
275 xfs_dir2_off_t *curoff,
276 struct xfs_buf **bpp)
277{
278 struct xfs_inode *dp = args->dp;
279 struct xfs_buf *bp = *bpp;
280 struct xfs_bmbt_irec *map = mip->map;
281 struct blk_plug plug;
282 int error = 0;
283 int length;
284 int i;
285 int j;
286 struct xfs_da_geometry *geo = args->geo;
287
288 /*
289 * If we have a buffer, we need to release it and
290 * take it out of the mapping.
291 */
292
293 if (bp) {
294 xfs_trans_brelse(NULL, bp);
295 bp = NULL;
296 mip->map_blocks -= geo->fsbcount;
297 /*
298 * Loop to get rid of the extents for the
299 * directory block.
300 */
301 for (i = geo->fsbcount; i > 0; ) {
302 j = min_t(int, map->br_blockcount, i);
303 map->br_blockcount -= j;
304 map->br_startblock += j;
305 map->br_startoff += j;
306 /*
307 * If mapping is done, pitch it from
308 * the table.
309 */
310 if (!map->br_blockcount && --mip->map_valid)
311 memmove(&map[0], &map[1],
312 sizeof(map[0]) * mip->map_valid);
313 i -= j;
314 }
315 }
316
317 /*
318 * Recalculate the readahead blocks wanted.
319 */
320 mip->ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog)) - 1;
321 ASSERT(mip->ra_want >= 0);
322
323 /*
324 * If we don't have as many as we want, and we haven't
325 * run out of data blocks, get some more mappings.
326 */
327 if (1 + mip->ra_want > mip->map_blocks &&
328 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
329 /*
330 * Get more bmaps, fill in after the ones
331 * we already have in the table.
332 */
333 mip->nmap = mip->map_size - mip->map_valid;
334 error = xfs_bmapi_read(dp, mip->map_off,
335 xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET) -
336 mip->map_off,
337 &map[mip->map_valid], &mip->nmap, 0);
338
339 /*
340 * Don't know if we should ignore this or try to return an
341 * error. The trouble with returning errors is that readdir
342 * will just stop without actually passing the error through.
343 */
344 if (error)
345 goto out; /* XXX */
346
347 /*
348 * If we got all the mappings we asked for, set the final map
349 * offset based on the last bmap value received. Otherwise,
350 * we've reached the end.
351 */
352 if (mip->nmap == mip->map_size - mip->map_valid) {
353 i = mip->map_valid + mip->nmap - 1;
354 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
355 } else
356 mip->map_off = xfs_dir2_byte_to_da(geo,
357 XFS_DIR2_LEAF_OFFSET);
358
359 /*
360 * Look for holes in the mapping, and eliminate them. Count up
361 * the valid blocks.
362 */
363 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
364 if (map[i].br_startblock == HOLESTARTBLOCK) {
365 mip->nmap--;
366 length = mip->map_valid + mip->nmap - i;
367 if (length)
368 memmove(&map[i], &map[i + 1],
369 sizeof(map[i]) * length);
370 } else {
371 mip->map_blocks += map[i].br_blockcount;
372 i++;
373 }
374 }
375 mip->map_valid += mip->nmap;
376 }
377
378 /*
379 * No valid mappings, so no more data blocks.
380 */
381 if (!mip->map_valid) {
382 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
383 goto out;
384 }
385
386 /*
387 * Read the directory block starting at the first mapping.
388 */
389 mip->curdb = xfs_dir2_da_to_db(geo, map->br_startoff);
390 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
391 map->br_blockcount >= geo->fsbcount ?
392 XFS_FSB_TO_DADDR(dp->i_mount, map->br_startblock) :
393 -1, &bp);
394 /*
395 * Should just skip over the data block instead of giving up.
396 */
397 if (error)
398 goto out; /* XXX */
399
400 /*
401 * Adjust the current amount of read-ahead: we just read a block that
402 * was previously ra.
403 */
404 if (mip->ra_current)
405 mip->ra_current -= geo->fsbcount;
406
407 /*
408 * Do we need more readahead?
409 */
410 blk_start_plug(&plug);
411 for (mip->ra_index = mip->ra_offset = i = 0;
412 mip->ra_want > mip->ra_current && i < mip->map_blocks;
413 i += geo->fsbcount) {
414 ASSERT(mip->ra_index < mip->map_valid);
415 /*
416 * Read-ahead a contiguous directory block.
417 */
418 if (i > mip->ra_current &&
419 map[mip->ra_index].br_blockcount >= geo->fsbcount) {
420 xfs_dir3_data_readahead(dp,
421 map[mip->ra_index].br_startoff + mip->ra_offset,
422 XFS_FSB_TO_DADDR(dp->i_mount,
423 map[mip->ra_index].br_startblock +
424 mip->ra_offset));
425 mip->ra_current = i;
426 }
427
428 /*
429 * Read-ahead a non-contiguous directory block. This doesn't
430 * use our mapping, but this is a very rare case.
431 */
432 else if (i > mip->ra_current) {
433 xfs_dir3_data_readahead(dp,
434 map[mip->ra_index].br_startoff +
435 mip->ra_offset, -1);
436 mip->ra_current = i;
437 }
438
439 /*
440 * Advance offset through the mapping table.
441 */
442 for (j = 0; j < geo->fsbcount; j += length ) {
443 /*
444 * The rest of this extent but not more than a dir
445 * block.
446 */
447 length = min_t(int, geo->fsbcount,
448 map[mip->ra_index].br_blockcount -
449 mip->ra_offset);
450 mip->ra_offset += length;
451
452 /*
453 * Advance to the next mapping if this one is used up.
454 */
455 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
456 mip->ra_offset = 0;
457 mip->ra_index++;
458 }
459 }
460 }
461 blk_finish_plug(&plug);
462
463out:
464 *bpp = bp;
465 return error;
466}
467
468/*
469 * Getdents (readdir) for leaf and node directories.
470 * This reads the data blocks only, so is the same for both forms.
471 */
472STATIC int
473xfs_dir2_leaf_getdents(
474 struct xfs_da_args *args,
475 struct dir_context *ctx,
476 size_t bufsize)
477{
478 struct xfs_inode *dp = args->dp;
479 struct xfs_buf *bp = NULL; /* data block buffer */
480 xfs_dir2_data_hdr_t *hdr; /* data block header */
481 xfs_dir2_data_entry_t *dep; /* data entry */
482 xfs_dir2_data_unused_t *dup; /* unused entry */
483 int error = 0; /* error return value */
484 int length; /* temporary length value */
485 int byteoff; /* offset in current block */
486 xfs_dir2_off_t curoff; /* current overall offset */
487 xfs_dir2_off_t newoff; /* new curoff after new blk */
488 char *ptr = NULL; /* pointer to current data */
489 struct xfs_dir2_leaf_map_info *map_info;
490 struct xfs_da_geometry *geo = args->geo;
491
492 /*
493 * If the offset is at or past the largest allowed value,
494 * give up right away.
495 */
496 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
497 return 0;
498
499 /*
500 * Set up to bmap a number of blocks based on the caller's
501 * buffer size, the directory block size, and the filesystem
502 * block size.
503 */
504 length = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
505 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
506 (length * sizeof(struct xfs_bmbt_irec)),
507 KM_SLEEP | KM_NOFS);
508 map_info->map_size = length;
509
510 /*
511 * Inside the loop we keep the main offset value as a byte offset
512 * in the directory file.
513 */
514 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
515
516 /*
517 * Force this conversion through db so we truncate the offset
518 * down to get the start of the data block.
519 */
520 map_info->map_off = xfs_dir2_db_to_da(geo,
521 xfs_dir2_byte_to_db(geo, curoff));
522
523 /*
524 * Loop over directory entries until we reach the end offset.
525 * Get more blocks and readahead as necessary.
526 */
527 while (curoff < XFS_DIR2_LEAF_OFFSET) {
528 __uint8_t filetype;
529
530 /*
531 * If we have no buffer, or we're off the end of the
532 * current buffer, need to get another one.
533 */
534 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
535 int lock_mode;
536
537 lock_mode = xfs_ilock_data_map_shared(dp);
538 error = xfs_dir2_leaf_readbuf(args, bufsize, map_info,
539 &curoff, &bp);
540 xfs_iunlock(dp, lock_mode);
541 if (error || !map_info->map_valid)
542 break;
543
544 /*
545 * Having done a read, we need to set a new offset.
546 */
547 newoff = xfs_dir2_db_off_to_byte(geo,
548 map_info->curdb, 0);
549 /*
550 * Start of the current block.
551 */
552 if (curoff < newoff)
553 curoff = newoff;
554 /*
555 * Make sure we're in the right block.
556 */
557 else if (curoff > newoff)
558 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
559 map_info->curdb);
560 hdr = bp->b_addr;
561 xfs_dir3_data_check(dp, bp);
562 /*
563 * Find our position in the block.
564 */
565 ptr = (char *)dp->d_ops->data_entry_p(hdr);
566 byteoff = xfs_dir2_byte_to_off(geo, curoff);
567 /*
568 * Skip past the header.
569 */
570 if (byteoff == 0)
571 curoff += dp->d_ops->data_entry_offset;
572 /*
573 * Skip past entries until we reach our offset.
574 */
575 else {
576 while ((char *)ptr - (char *)hdr < byteoff) {
577 dup = (xfs_dir2_data_unused_t *)ptr;
578
579 if (be16_to_cpu(dup->freetag)
580 == XFS_DIR2_DATA_FREE_TAG) {
581
582 length = be16_to_cpu(dup->length);
583 ptr += length;
584 continue;
585 }
586 dep = (xfs_dir2_data_entry_t *)ptr;
587 length =
588 dp->d_ops->data_entsize(dep->namelen);
589 ptr += length;
590 }
591 /*
592 * Now set our real offset.
593 */
594 curoff =
595 xfs_dir2_db_off_to_byte(geo,
596 xfs_dir2_byte_to_db(geo, curoff),
597 (char *)ptr - (char *)hdr);
598 if (ptr >= (char *)hdr + geo->blksize) {
599 continue;
600 }
601 }
602 }
603 /*
604 * We have a pointer to an entry.
605 * Is it a live one?
606 */
607 dup = (xfs_dir2_data_unused_t *)ptr;
608 /*
609 * No, it's unused, skip over it.
610 */
611 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
612 length = be16_to_cpu(dup->length);
613 ptr += length;
614 curoff += length;
615 continue;
616 }
617
618 dep = (xfs_dir2_data_entry_t *)ptr;
619 length = dp->d_ops->data_entsize(dep->namelen);
620 filetype = dp->d_ops->data_get_ftype(dep);
621
622 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
623 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
624 be64_to_cpu(dep->inumber),
625 xfs_dir3_get_dtype(dp->i_mount, filetype)))
626 break;
627
628 /*
629 * Advance to next entry in the block.
630 */
631 ptr += length;
632 curoff += length;
633 /* bufsize may have just been a guess; don't go negative */
634 bufsize = bufsize > length ? bufsize - length : 0;
635 }
636
637 /*
638 * All done. Set output offset value to current offset.
639 */
640 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
641 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
642 else
643 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
644 kmem_free(map_info);
645 if (bp)
646 xfs_trans_brelse(NULL, bp);
647 return error;
648}
649
650/*
651 * Read a directory.
652 */
653int
654xfs_readdir(
655 struct xfs_inode *dp,
656 struct dir_context *ctx,
657 size_t bufsize)
658{
659 struct xfs_da_args args = { NULL };
660 int rval;
661 int v;
662
663 trace_xfs_readdir(dp);
664
665 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
666 return -EIO;
667
668 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
669 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
670
671 args.dp = dp;
672 args.geo = dp->i_mount->m_dir_geo;
673
674 xfs_ilock(dp, XFS_IOLOCK_SHARED);
675 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
676 rval = xfs_dir2_sf_getdents(&args, ctx);
677 else if ((rval = xfs_dir2_isblock(&args, &v)))
678 ;
679 else if (v)
680 rval = xfs_dir2_block_getdents(&args, ctx);
681 else
682 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
683 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
684
685 return rval;
686}