Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_dir2.h"
16#include "xfs_dir2_priv.h"
17#include "xfs_trace.h"
18#include "xfs_bmap.h"
19#include "xfs_trans.h"
20#include "xfs_error.h"
21
22/*
23 * Directory file type support functions
24 */
25static unsigned char xfs_dir3_filetype_table[] = {
26 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
27 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
28};
29
30unsigned char
31xfs_dir3_get_dtype(
32 struct xfs_mount *mp,
33 uint8_t filetype)
34{
35 if (!xfs_has_ftype(mp))
36 return DT_UNKNOWN;
37
38 if (filetype >= XFS_DIR3_FT_MAX)
39 return DT_UNKNOWN;
40
41 return xfs_dir3_filetype_table[filetype];
42}
43
44STATIC int
45xfs_dir2_sf_getdents(
46 struct xfs_da_args *args,
47 struct dir_context *ctx)
48{
49 int i; /* shortform entry number */
50 struct xfs_inode *dp = args->dp; /* incore directory inode */
51 struct xfs_mount *mp = dp->i_mount;
52 xfs_dir2_dataptr_t off; /* current entry's offset */
53 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
54 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
55 xfs_dir2_dataptr_t dot_offset;
56 xfs_dir2_dataptr_t dotdot_offset;
57 xfs_ino_t ino;
58 struct xfs_da_geometry *geo = args->geo;
59
60 ASSERT(dp->i_df.if_format == XFS_DINODE_FMT_LOCAL);
61 ASSERT(dp->i_df.if_bytes == dp->i_disk_size);
62 ASSERT(dp->i_df.if_u1.if_data != NULL);
63
64 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
65
66 /*
67 * If the block number in the offset is out of range, we're done.
68 */
69 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
70 return 0;
71
72 /*
73 * Precalculate offsets for "." and ".." as we will always need them.
74 * This relies on the fact that directories always start with the
75 * entries for "." and "..".
76 */
77 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
78 geo->data_entry_offset);
79 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
80 geo->data_entry_offset +
81 xfs_dir2_data_entsize(mp, sizeof(".") - 1));
82
83 /*
84 * Put . entry unless we're starting past it.
85 */
86 if (ctx->pos <= dot_offset) {
87 ctx->pos = dot_offset & 0x7fffffff;
88 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
89 return 0;
90 }
91
92 /*
93 * Put .. entry unless we're starting past it.
94 */
95 if (ctx->pos <= dotdot_offset) {
96 ino = xfs_dir2_sf_get_parent_ino(sfp);
97 ctx->pos = dotdot_offset & 0x7fffffff;
98 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
99 return 0;
100 }
101
102 /*
103 * Loop while there are more entries and put'ing works.
104 */
105 sfep = xfs_dir2_sf_firstentry(sfp);
106 for (i = 0; i < sfp->count; i++) {
107 uint8_t filetype;
108
109 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
110 xfs_dir2_sf_get_offset(sfep));
111
112 if (ctx->pos > off) {
113 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
114 continue;
115 }
116
117 ino = xfs_dir2_sf_get_ino(mp, sfp, sfep);
118 filetype = xfs_dir2_sf_get_ftype(mp, sfep);
119 ctx->pos = off & 0x7fffffff;
120 if (XFS_IS_CORRUPT(dp->i_mount,
121 !xfs_dir2_namecheck(sfep->name,
122 sfep->namelen)))
123 return -EFSCORRUPTED;
124 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
125 xfs_dir3_get_dtype(mp, filetype)))
126 return 0;
127 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
128 }
129
130 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
131 0x7fffffff;
132 return 0;
133}
134
135/*
136 * Readdir for block directories.
137 */
138STATIC int
139xfs_dir2_block_getdents(
140 struct xfs_da_args *args,
141 struct dir_context *ctx,
142 unsigned int *lock_mode)
143{
144 struct xfs_inode *dp = args->dp; /* incore directory inode */
145 struct xfs_buf *bp; /* buffer for block */
146 int error; /* error return value */
147 int wantoff; /* starting block offset */
148 xfs_off_t cook;
149 struct xfs_da_geometry *geo = args->geo;
150 unsigned int offset, next_offset;
151 unsigned int end;
152
153 /*
154 * If the block number in the offset is out of range, we're done.
155 */
156 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
157 return 0;
158
159 error = xfs_dir3_block_read(args->trans, dp, &bp);
160 if (error)
161 return error;
162
163 xfs_iunlock(dp, *lock_mode);
164 *lock_mode = 0;
165
166 /*
167 * Extract the byte offset we start at from the seek pointer.
168 * We'll skip entries before this.
169 */
170 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
171 xfs_dir3_data_check(dp, bp);
172
173 /*
174 * Loop over the data portion of the block.
175 * Each object is a real entry (dep) or an unused one (dup).
176 */
177 end = xfs_dir3_data_end_offset(geo, bp->b_addr);
178 for (offset = geo->data_entry_offset;
179 offset < end;
180 offset = next_offset) {
181 struct xfs_dir2_data_unused *dup = bp->b_addr + offset;
182 struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
183 uint8_t filetype;
184
185 /*
186 * Unused, skip it.
187 */
188 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
189 next_offset = offset + be16_to_cpu(dup->length);
190 continue;
191 }
192
193 /*
194 * Bump pointer for the next iteration.
195 */
196 next_offset = offset +
197 xfs_dir2_data_entsize(dp->i_mount, dep->namelen);
198
199 /*
200 * The entry is before the desired starting point, skip it.
201 */
202 if (offset < wantoff)
203 continue;
204
205 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, offset);
206
207 ctx->pos = cook & 0x7fffffff;
208 filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
209 /*
210 * If it didn't fit, set the final offset to here & return.
211 */
212 if (XFS_IS_CORRUPT(dp->i_mount,
213 !xfs_dir2_namecheck(dep->name,
214 dep->namelen))) {
215 error = -EFSCORRUPTED;
216 goto out_rele;
217 }
218 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
219 be64_to_cpu(dep->inumber),
220 xfs_dir3_get_dtype(dp->i_mount, filetype)))
221 goto out_rele;
222 }
223
224 /*
225 * Reached the end of the block.
226 * Set the offset to a non-existent block 1 and return.
227 */
228 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
229 0x7fffffff;
230out_rele:
231 xfs_trans_brelse(args->trans, bp);
232 return error;
233}
234
235/*
236 * Read a directory block and initiate readahead for blocks beyond that.
237 * We maintain a sliding readahead window of the remaining space in the
238 * buffer rounded up to the nearest block.
239 */
240STATIC int
241xfs_dir2_leaf_readbuf(
242 struct xfs_da_args *args,
243 size_t bufsize,
244 xfs_dir2_off_t *cur_off,
245 xfs_dablk_t *ra_blk,
246 struct xfs_buf **bpp)
247{
248 struct xfs_inode *dp = args->dp;
249 struct xfs_buf *bp = NULL;
250 struct xfs_da_geometry *geo = args->geo;
251 struct xfs_ifork *ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);
252 struct xfs_bmbt_irec map;
253 struct blk_plug plug;
254 xfs_dir2_off_t new_off;
255 xfs_dablk_t next_ra;
256 xfs_dablk_t map_off;
257 xfs_dablk_t last_da;
258 struct xfs_iext_cursor icur;
259 int ra_want;
260 int error = 0;
261
262 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
263 if (error)
264 goto out;
265
266 /*
267 * Look for mapped directory blocks at or above the current offset.
268 * Truncate down to the nearest directory block to start the scanning
269 * operation.
270 */
271 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
272 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
273 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
274 goto out;
275 if (map.br_startoff >= last_da)
276 goto out;
277 xfs_trim_extent(&map, map_off, last_da - map_off);
278
279 /* Read the directory block of that first mapping. */
280 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
281 if (new_off > *cur_off)
282 *cur_off = new_off;
283 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
284 if (error)
285 goto out;
286
287 /*
288 * Start readahead for the next bufsize's worth of dir data blocks.
289 * We may have already issued readahead for some of that range;
290 * ra_blk tracks the last block we tried to read(ahead).
291 */
292 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
293 if (*ra_blk >= last_da)
294 goto out;
295 else if (*ra_blk == 0)
296 *ra_blk = map.br_startoff;
297 next_ra = map.br_startoff + geo->fsbcount;
298 if (next_ra >= last_da)
299 goto out_no_ra;
300 if (map.br_blockcount < geo->fsbcount &&
301 !xfs_iext_next_extent(ifp, &icur, &map))
302 goto out_no_ra;
303 if (map.br_startoff >= last_da)
304 goto out_no_ra;
305 xfs_trim_extent(&map, next_ra, last_da - next_ra);
306
307 /* Start ra for each dir (not fs) block that has a mapping. */
308 blk_start_plug(&plug);
309 while (ra_want > 0) {
310 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
311 while (ra_want > 0 &&
312 next_ra < map.br_startoff + map.br_blockcount) {
313 if (next_ra >= last_da) {
314 *ra_blk = last_da;
315 break;
316 }
317 if (next_ra > *ra_blk) {
318 xfs_dir3_data_readahead(dp, next_ra,
319 XFS_DABUF_MAP_HOLE_OK);
320 *ra_blk = next_ra;
321 }
322 ra_want -= geo->fsbcount;
323 next_ra += geo->fsbcount;
324 }
325 if (!xfs_iext_next_extent(ifp, &icur, &map)) {
326 *ra_blk = last_da;
327 break;
328 }
329 }
330 blk_finish_plug(&plug);
331
332out:
333 *bpp = bp;
334 return error;
335out_no_ra:
336 *ra_blk = last_da;
337 goto out;
338}
339
340/*
341 * Getdents (readdir) for leaf and node directories.
342 * This reads the data blocks only, so is the same for both forms.
343 */
344STATIC int
345xfs_dir2_leaf_getdents(
346 struct xfs_da_args *args,
347 struct dir_context *ctx,
348 size_t bufsize,
349 unsigned int *lock_mode)
350{
351 struct xfs_inode *dp = args->dp;
352 struct xfs_mount *mp = dp->i_mount;
353 struct xfs_buf *bp = NULL; /* data block buffer */
354 xfs_dir2_data_entry_t *dep; /* data entry */
355 xfs_dir2_data_unused_t *dup; /* unused entry */
356 struct xfs_da_geometry *geo = args->geo;
357 xfs_dablk_t rablk = 0; /* current readahead block */
358 xfs_dir2_off_t curoff; /* current overall offset */
359 int length; /* temporary length value */
360 int byteoff; /* offset in current block */
361 unsigned int offset = 0;
362 int error = 0; /* error return value */
363
364 /*
365 * If the offset is at or past the largest allowed value,
366 * give up right away.
367 */
368 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
369 return 0;
370
371 /*
372 * Inside the loop we keep the main offset value as a byte offset
373 * in the directory file.
374 */
375 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
376
377 /*
378 * Loop over directory entries until we reach the end offset.
379 * Get more blocks and readahead as necessary.
380 */
381 while (curoff < XFS_DIR2_LEAF_OFFSET) {
382 uint8_t filetype;
383
384 /*
385 * If we have no buffer, or we're off the end of the
386 * current buffer, need to get another one.
387 */
388 if (!bp || offset >= geo->blksize) {
389 if (bp) {
390 xfs_trans_brelse(args->trans, bp);
391 bp = NULL;
392 }
393
394 if (*lock_mode == 0)
395 *lock_mode = xfs_ilock_data_map_shared(dp);
396 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
397 &rablk, &bp);
398 if (error || !bp)
399 break;
400
401 xfs_iunlock(dp, *lock_mode);
402 *lock_mode = 0;
403
404 xfs_dir3_data_check(dp, bp);
405 /*
406 * Find our position in the block.
407 */
408 offset = geo->data_entry_offset;
409 byteoff = xfs_dir2_byte_to_off(geo, curoff);
410 /*
411 * Skip past the header.
412 */
413 if (byteoff == 0)
414 curoff += geo->data_entry_offset;
415 /*
416 * Skip past entries until we reach our offset.
417 */
418 else {
419 while (offset < byteoff) {
420 dup = bp->b_addr + offset;
421
422 if (be16_to_cpu(dup->freetag)
423 == XFS_DIR2_DATA_FREE_TAG) {
424
425 length = be16_to_cpu(dup->length);
426 offset += length;
427 continue;
428 }
429 dep = bp->b_addr + offset;
430 length = xfs_dir2_data_entsize(mp,
431 dep->namelen);
432 offset += length;
433 }
434 /*
435 * Now set our real offset.
436 */
437 curoff =
438 xfs_dir2_db_off_to_byte(geo,
439 xfs_dir2_byte_to_db(geo, curoff),
440 offset);
441 if (offset >= geo->blksize)
442 continue;
443 }
444 }
445
446 /*
447 * We have a pointer to an entry. Is it a live one?
448 */
449 dup = bp->b_addr + offset;
450
451 /*
452 * No, it's unused, skip over it.
453 */
454 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
455 length = be16_to_cpu(dup->length);
456 offset += length;
457 curoff += length;
458 continue;
459 }
460
461 dep = bp->b_addr + offset;
462 length = xfs_dir2_data_entsize(mp, dep->namelen);
463 filetype = xfs_dir2_data_get_ftype(mp, dep);
464
465 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
466 if (XFS_IS_CORRUPT(dp->i_mount,
467 !xfs_dir2_namecheck(dep->name,
468 dep->namelen))) {
469 error = -EFSCORRUPTED;
470 break;
471 }
472 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
473 be64_to_cpu(dep->inumber),
474 xfs_dir3_get_dtype(dp->i_mount, filetype)))
475 break;
476
477 /*
478 * Advance to next entry in the block.
479 */
480 offset += length;
481 curoff += length;
482 /* bufsize may have just been a guess; don't go negative */
483 bufsize = bufsize > length ? bufsize - length : 0;
484 }
485
486 /*
487 * All done. Set output offset value to current offset.
488 */
489 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
490 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
491 else
492 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
493 if (bp)
494 xfs_trans_brelse(args->trans, bp);
495 return error;
496}
497
498/*
499 * Read a directory.
500 *
501 * If supplied, the transaction collects locked dir buffers to avoid
502 * nested buffer deadlocks. This function does not dirty the
503 * transaction. The caller must hold the IOLOCK (shared or exclusive)
504 * before calling this function.
505 */
506int
507xfs_readdir(
508 struct xfs_trans *tp,
509 struct xfs_inode *dp,
510 struct dir_context *ctx,
511 size_t bufsize)
512{
513 struct xfs_da_args args = { NULL };
514 unsigned int lock_mode;
515 bool isblock;
516 int error;
517
518 trace_xfs_readdir(dp);
519
520 if (xfs_is_shutdown(dp->i_mount))
521 return -EIO;
522
523 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
524 ASSERT(xfs_isilocked(dp, XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
525 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
526
527 args.dp = dp;
528 args.geo = dp->i_mount->m_dir_geo;
529 args.trans = tp;
530
531 if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
532 return xfs_dir2_sf_getdents(&args, ctx);
533
534 lock_mode = xfs_ilock_data_map_shared(dp);
535 error = xfs_dir2_isblock(&args, &isblock);
536 if (error)
537 goto out_unlock;
538
539 if (isblock) {
540 error = xfs_dir2_block_getdents(&args, ctx, &lock_mode);
541 goto out_unlock;
542 }
543
544 error = xfs_dir2_leaf_getdents(&args, ctx, bufsize, &lock_mode);
545
546out_unlock:
547 if (lock_mode)
548 xfs_iunlock(dp, lock_mode);
549 return error;
550}
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_sb.h"
26#include "xfs_ag.h"
27#include "xfs_mount.h"
28#include "xfs_da_format.h"
29#include "xfs_da_btree.h"
30#include "xfs_inode.h"
31#include "xfs_dir2.h"
32#include "xfs_dir2_priv.h"
33#include "xfs_error.h"
34#include "xfs_trace.h"
35#include "xfs_bmap.h"
36#include "xfs_trans.h"
37#include "xfs_dinode.h"
38
39/*
40 * Directory file type support functions
41 */
42static unsigned char xfs_dir3_filetype_table[] = {
43 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
44 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
45};
46
47unsigned char
48xfs_dir3_get_dtype(
49 struct xfs_mount *mp,
50 __uint8_t filetype)
51{
52 if (!xfs_sb_version_hasftype(&mp->m_sb))
53 return DT_UNKNOWN;
54
55 if (filetype >= XFS_DIR3_FT_MAX)
56 return DT_UNKNOWN;
57
58 return xfs_dir3_filetype_table[filetype];
59}
60/*
61 * @mode, if set, indicates that the type field needs to be set up.
62 * This uses the transformation from file mode to DT_* as defined in linux/fs.h
63 * for file type specification. This will be propagated into the directory
64 * structure if appropriate for the given operation and filesystem config.
65 */
66const unsigned char xfs_mode_to_ftype[S_IFMT >> S_SHIFT] = {
67 [0] = XFS_DIR3_FT_UNKNOWN,
68 [S_IFREG >> S_SHIFT] = XFS_DIR3_FT_REG_FILE,
69 [S_IFDIR >> S_SHIFT] = XFS_DIR3_FT_DIR,
70 [S_IFCHR >> S_SHIFT] = XFS_DIR3_FT_CHRDEV,
71 [S_IFBLK >> S_SHIFT] = XFS_DIR3_FT_BLKDEV,
72 [S_IFIFO >> S_SHIFT] = XFS_DIR3_FT_FIFO,
73 [S_IFSOCK >> S_SHIFT] = XFS_DIR3_FT_SOCK,
74 [S_IFLNK >> S_SHIFT] = XFS_DIR3_FT_SYMLINK,
75};
76
77STATIC int
78xfs_dir2_sf_getdents(
79 xfs_inode_t *dp, /* incore directory inode */
80 struct dir_context *ctx)
81{
82 int i; /* shortform entry number */
83 xfs_mount_t *mp; /* filesystem mount point */
84 xfs_dir2_dataptr_t off; /* current entry's offset */
85 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
86 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
87 xfs_dir2_dataptr_t dot_offset;
88 xfs_dir2_dataptr_t dotdot_offset;
89 xfs_ino_t ino;
90
91 mp = dp->i_mount;
92
93 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
94 /*
95 * Give up if the directory is way too short.
96 */
97 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
98 ASSERT(XFS_FORCED_SHUTDOWN(mp));
99 return XFS_ERROR(EIO);
100 }
101
102 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
103 ASSERT(dp->i_df.if_u1.if_data != NULL);
104
105 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
106
107 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
108
109 /*
110 * If the block number in the offset is out of range, we're done.
111 */
112 if (xfs_dir2_dataptr_to_db(mp, ctx->pos) > mp->m_dirdatablk)
113 return 0;
114
115 /*
116 * Precalculate offsets for . and .. as we will always need them.
117 *
118 * XXX(hch): the second argument is sometimes 0 and sometimes
119 * mp->m_dirdatablk.
120 */
121 dot_offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
122 dp->d_ops->data_dot_offset);
123 dotdot_offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
124 dp->d_ops->data_dotdot_offset);
125
126 /*
127 * Put . entry unless we're starting past it.
128 */
129 if (ctx->pos <= dot_offset) {
130 ctx->pos = dot_offset & 0x7fffffff;
131 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
132 return 0;
133 }
134
135 /*
136 * Put .. entry unless we're starting past it.
137 */
138 if (ctx->pos <= dotdot_offset) {
139 ino = dp->d_ops->sf_get_parent_ino(sfp);
140 ctx->pos = dotdot_offset & 0x7fffffff;
141 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
142 return 0;
143 }
144
145 /*
146 * Loop while there are more entries and put'ing works.
147 */
148 sfep = xfs_dir2_sf_firstentry(sfp);
149 for (i = 0; i < sfp->count; i++) {
150 __uint8_t filetype;
151
152 off = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
153 xfs_dir2_sf_get_offset(sfep));
154
155 if (ctx->pos > off) {
156 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
157 continue;
158 }
159
160 ino = dp->d_ops->sf_get_ino(sfp, sfep);
161 filetype = dp->d_ops->sf_get_ftype(sfep);
162 ctx->pos = off & 0x7fffffff;
163 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
164 xfs_dir3_get_dtype(mp, filetype)))
165 return 0;
166 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
167 }
168
169 ctx->pos = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
170 0x7fffffff;
171 return 0;
172}
173
174/*
175 * Readdir for block directories.
176 */
177STATIC int
178xfs_dir2_block_getdents(
179 xfs_inode_t *dp, /* incore inode */
180 struct dir_context *ctx)
181{
182 xfs_dir2_data_hdr_t *hdr; /* block header */
183 struct xfs_buf *bp; /* buffer for block */
184 xfs_dir2_block_tail_t *btp; /* block tail */
185 xfs_dir2_data_entry_t *dep; /* block data entry */
186 xfs_dir2_data_unused_t *dup; /* block unused entry */
187 char *endptr; /* end of the data entries */
188 int error; /* error return value */
189 xfs_mount_t *mp; /* filesystem mount point */
190 char *ptr; /* current data entry */
191 int wantoff; /* starting block offset */
192 xfs_off_t cook;
193
194 mp = dp->i_mount;
195 /*
196 * If the block number in the offset is out of range, we're done.
197 */
198 if (xfs_dir2_dataptr_to_db(mp, ctx->pos) > mp->m_dirdatablk)
199 return 0;
200
201 error = xfs_dir3_block_read(NULL, dp, &bp);
202 if (error)
203 return error;
204
205 /*
206 * Extract the byte offset we start at from the seek pointer.
207 * We'll skip entries before this.
208 */
209 wantoff = xfs_dir2_dataptr_to_off(mp, ctx->pos);
210 hdr = bp->b_addr;
211 xfs_dir3_data_check(dp, bp);
212 /*
213 * Set up values for the loop.
214 */
215 btp = xfs_dir2_block_tail_p(mp, hdr);
216 ptr = (char *)dp->d_ops->data_entry_p(hdr);
217 endptr = (char *)xfs_dir2_block_leaf_p(btp);
218
219 /*
220 * Loop over the data portion of the block.
221 * Each object is a real entry (dep) or an unused one (dup).
222 */
223 while (ptr < endptr) {
224 __uint8_t filetype;
225
226 dup = (xfs_dir2_data_unused_t *)ptr;
227 /*
228 * Unused, skip it.
229 */
230 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
231 ptr += be16_to_cpu(dup->length);
232 continue;
233 }
234
235 dep = (xfs_dir2_data_entry_t *)ptr;
236
237 /*
238 * Bump pointer for the next iteration.
239 */
240 ptr += dp->d_ops->data_entsize(dep->namelen);
241 /*
242 * The entry is before the desired starting point, skip it.
243 */
244 if ((char *)dep - (char *)hdr < wantoff)
245 continue;
246
247 cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
248 (char *)dep - (char *)hdr);
249
250 ctx->pos = cook & 0x7fffffff;
251 filetype = dp->d_ops->data_get_ftype(dep);
252 /*
253 * If it didn't fit, set the final offset to here & return.
254 */
255 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
256 be64_to_cpu(dep->inumber),
257 xfs_dir3_get_dtype(mp, filetype))) {
258 xfs_trans_brelse(NULL, bp);
259 return 0;
260 }
261 }
262
263 /*
264 * Reached the end of the block.
265 * Set the offset to a non-existent block 1 and return.
266 */
267 ctx->pos = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
268 0x7fffffff;
269 xfs_trans_brelse(NULL, bp);
270 return 0;
271}
272
273struct xfs_dir2_leaf_map_info {
274 xfs_extlen_t map_blocks; /* number of fsbs in map */
275 xfs_dablk_t map_off; /* last mapped file offset */
276 int map_size; /* total entries in *map */
277 int map_valid; /* valid entries in *map */
278 int nmap; /* mappings to ask xfs_bmapi */
279 xfs_dir2_db_t curdb; /* db for current block */
280 int ra_current; /* number of read-ahead blks */
281 int ra_index; /* *map index for read-ahead */
282 int ra_offset; /* map entry offset for ra */
283 int ra_want; /* readahead count wanted */
284 struct xfs_bmbt_irec map[]; /* map vector for blocks */
285};
286
287STATIC int
288xfs_dir2_leaf_readbuf(
289 struct xfs_inode *dp,
290 size_t bufsize,
291 struct xfs_dir2_leaf_map_info *mip,
292 xfs_dir2_off_t *curoff,
293 struct xfs_buf **bpp)
294{
295 struct xfs_mount *mp = dp->i_mount;
296 struct xfs_buf *bp = *bpp;
297 struct xfs_bmbt_irec *map = mip->map;
298 struct blk_plug plug;
299 int error = 0;
300 int length;
301 int i;
302 int j;
303
304 /*
305 * If we have a buffer, we need to release it and
306 * take it out of the mapping.
307 */
308
309 if (bp) {
310 xfs_trans_brelse(NULL, bp);
311 bp = NULL;
312 mip->map_blocks -= mp->m_dirblkfsbs;
313 /*
314 * Loop to get rid of the extents for the
315 * directory block.
316 */
317 for (i = mp->m_dirblkfsbs; i > 0; ) {
318 j = min_t(int, map->br_blockcount, i);
319 map->br_blockcount -= j;
320 map->br_startblock += j;
321 map->br_startoff += j;
322 /*
323 * If mapping is done, pitch it from
324 * the table.
325 */
326 if (!map->br_blockcount && --mip->map_valid)
327 memmove(&map[0], &map[1],
328 sizeof(map[0]) * mip->map_valid);
329 i -= j;
330 }
331 }
332
333 /*
334 * Recalculate the readahead blocks wanted.
335 */
336 mip->ra_want = howmany(bufsize + mp->m_dirblksize,
337 mp->m_sb.sb_blocksize) - 1;
338 ASSERT(mip->ra_want >= 0);
339
340 /*
341 * If we don't have as many as we want, and we haven't
342 * run out of data blocks, get some more mappings.
343 */
344 if (1 + mip->ra_want > mip->map_blocks &&
345 mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
346 /*
347 * Get more bmaps, fill in after the ones
348 * we already have in the table.
349 */
350 mip->nmap = mip->map_size - mip->map_valid;
351 error = xfs_bmapi_read(dp, mip->map_off,
352 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
353 mip->map_off,
354 &map[mip->map_valid], &mip->nmap, 0);
355
356 /*
357 * Don't know if we should ignore this or try to return an
358 * error. The trouble with returning errors is that readdir
359 * will just stop without actually passing the error through.
360 */
361 if (error)
362 goto out; /* XXX */
363
364 /*
365 * If we got all the mappings we asked for, set the final map
366 * offset based on the last bmap value received. Otherwise,
367 * we've reached the end.
368 */
369 if (mip->nmap == mip->map_size - mip->map_valid) {
370 i = mip->map_valid + mip->nmap - 1;
371 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
372 } else
373 mip->map_off = xfs_dir2_byte_to_da(mp,
374 XFS_DIR2_LEAF_OFFSET);
375
376 /*
377 * Look for holes in the mapping, and eliminate them. Count up
378 * the valid blocks.
379 */
380 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
381 if (map[i].br_startblock == HOLESTARTBLOCK) {
382 mip->nmap--;
383 length = mip->map_valid + mip->nmap - i;
384 if (length)
385 memmove(&map[i], &map[i + 1],
386 sizeof(map[i]) * length);
387 } else {
388 mip->map_blocks += map[i].br_blockcount;
389 i++;
390 }
391 }
392 mip->map_valid += mip->nmap;
393 }
394
395 /*
396 * No valid mappings, so no more data blocks.
397 */
398 if (!mip->map_valid) {
399 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
400 goto out;
401 }
402
403 /*
404 * Read the directory block starting at the first mapping.
405 */
406 mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
407 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
408 map->br_blockcount >= mp->m_dirblkfsbs ?
409 XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
410
411 /*
412 * Should just skip over the data block instead of giving up.
413 */
414 if (error)
415 goto out; /* XXX */
416
417 /*
418 * Adjust the current amount of read-ahead: we just read a block that
419 * was previously ra.
420 */
421 if (mip->ra_current)
422 mip->ra_current -= mp->m_dirblkfsbs;
423
424 /*
425 * Do we need more readahead?
426 */
427 blk_start_plug(&plug);
428 for (mip->ra_index = mip->ra_offset = i = 0;
429 mip->ra_want > mip->ra_current && i < mip->map_blocks;
430 i += mp->m_dirblkfsbs) {
431 ASSERT(mip->ra_index < mip->map_valid);
432 /*
433 * Read-ahead a contiguous directory block.
434 */
435 if (i > mip->ra_current &&
436 map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
437 xfs_dir3_data_readahead(NULL, dp,
438 map[mip->ra_index].br_startoff + mip->ra_offset,
439 XFS_FSB_TO_DADDR(mp,
440 map[mip->ra_index].br_startblock +
441 mip->ra_offset));
442 mip->ra_current = i;
443 }
444
445 /*
446 * Read-ahead a non-contiguous directory block. This doesn't
447 * use our mapping, but this is a very rare case.
448 */
449 else if (i > mip->ra_current) {
450 xfs_dir3_data_readahead(NULL, dp,
451 map[mip->ra_index].br_startoff +
452 mip->ra_offset, -1);
453 mip->ra_current = i;
454 }
455
456 /*
457 * Advance offset through the mapping table.
458 */
459 for (j = 0; j < mp->m_dirblkfsbs; j++) {
460 /*
461 * The rest of this extent but not more than a dir
462 * block.
463 */
464 length = min_t(int, mp->m_dirblkfsbs,
465 map[mip->ra_index].br_blockcount -
466 mip->ra_offset);
467 j += length;
468 mip->ra_offset += length;
469
470 /*
471 * Advance to the next mapping if this one is used up.
472 */
473 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
474 mip->ra_offset = 0;
475 mip->ra_index++;
476 }
477 }
478 }
479 blk_finish_plug(&plug);
480
481out:
482 *bpp = bp;
483 return error;
484}
485
486/*
487 * Getdents (readdir) for leaf and node directories.
488 * This reads the data blocks only, so is the same for both forms.
489 */
490STATIC int
491xfs_dir2_leaf_getdents(
492 xfs_inode_t *dp, /* incore directory inode */
493 struct dir_context *ctx,
494 size_t bufsize)
495{
496 struct xfs_buf *bp = NULL; /* data block buffer */
497 xfs_dir2_data_hdr_t *hdr; /* data block header */
498 xfs_dir2_data_entry_t *dep; /* data entry */
499 xfs_dir2_data_unused_t *dup; /* unused entry */
500 int error = 0; /* error return value */
501 int length; /* temporary length value */
502 xfs_mount_t *mp; /* filesystem mount point */
503 int byteoff; /* offset in current block */
504 xfs_dir2_off_t curoff; /* current overall offset */
505 xfs_dir2_off_t newoff; /* new curoff after new blk */
506 char *ptr = NULL; /* pointer to current data */
507 struct xfs_dir2_leaf_map_info *map_info;
508
509 /*
510 * If the offset is at or past the largest allowed value,
511 * give up right away.
512 */
513 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
514 return 0;
515
516 mp = dp->i_mount;
517
518 /*
519 * Set up to bmap a number of blocks based on the caller's
520 * buffer size, the directory block size, and the filesystem
521 * block size.
522 */
523 length = howmany(bufsize + mp->m_dirblksize,
524 mp->m_sb.sb_blocksize);
525 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
526 (length * sizeof(struct xfs_bmbt_irec)),
527 KM_SLEEP | KM_NOFS);
528 map_info->map_size = length;
529
530 /*
531 * Inside the loop we keep the main offset value as a byte offset
532 * in the directory file.
533 */
534 curoff = xfs_dir2_dataptr_to_byte(mp, ctx->pos);
535
536 /*
537 * Force this conversion through db so we truncate the offset
538 * down to get the start of the data block.
539 */
540 map_info->map_off = xfs_dir2_db_to_da(mp,
541 xfs_dir2_byte_to_db(mp, curoff));
542
543 /*
544 * Loop over directory entries until we reach the end offset.
545 * Get more blocks and readahead as necessary.
546 */
547 while (curoff < XFS_DIR2_LEAF_OFFSET) {
548 __uint8_t filetype;
549
550 /*
551 * If we have no buffer, or we're off the end of the
552 * current buffer, need to get another one.
553 */
554 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
555
556 error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
557 &curoff, &bp);
558 if (error || !map_info->map_valid)
559 break;
560
561 /*
562 * Having done a read, we need to set a new offset.
563 */
564 newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
565 /*
566 * Start of the current block.
567 */
568 if (curoff < newoff)
569 curoff = newoff;
570 /*
571 * Make sure we're in the right block.
572 */
573 else if (curoff > newoff)
574 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
575 map_info->curdb);
576 hdr = bp->b_addr;
577 xfs_dir3_data_check(dp, bp);
578 /*
579 * Find our position in the block.
580 */
581 ptr = (char *)dp->d_ops->data_entry_p(hdr);
582 byteoff = xfs_dir2_byte_to_off(mp, curoff);
583 /*
584 * Skip past the header.
585 */
586 if (byteoff == 0)
587 curoff += dp->d_ops->data_entry_offset;
588 /*
589 * Skip past entries until we reach our offset.
590 */
591 else {
592 while ((char *)ptr - (char *)hdr < byteoff) {
593 dup = (xfs_dir2_data_unused_t *)ptr;
594
595 if (be16_to_cpu(dup->freetag)
596 == XFS_DIR2_DATA_FREE_TAG) {
597
598 length = be16_to_cpu(dup->length);
599 ptr += length;
600 continue;
601 }
602 dep = (xfs_dir2_data_entry_t *)ptr;
603 length =
604 dp->d_ops->data_entsize(dep->namelen);
605 ptr += length;
606 }
607 /*
608 * Now set our real offset.
609 */
610 curoff =
611 xfs_dir2_db_off_to_byte(mp,
612 xfs_dir2_byte_to_db(mp, curoff),
613 (char *)ptr - (char *)hdr);
614 if (ptr >= (char *)hdr + mp->m_dirblksize) {
615 continue;
616 }
617 }
618 }
619 /*
620 * We have a pointer to an entry.
621 * Is it a live one?
622 */
623 dup = (xfs_dir2_data_unused_t *)ptr;
624 /*
625 * No, it's unused, skip over it.
626 */
627 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
628 length = be16_to_cpu(dup->length);
629 ptr += length;
630 curoff += length;
631 continue;
632 }
633
634 dep = (xfs_dir2_data_entry_t *)ptr;
635 length = dp->d_ops->data_entsize(dep->namelen);
636 filetype = dp->d_ops->data_get_ftype(dep);
637
638 ctx->pos = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
639 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
640 be64_to_cpu(dep->inumber),
641 xfs_dir3_get_dtype(mp, filetype)))
642 break;
643
644 /*
645 * Advance to next entry in the block.
646 */
647 ptr += length;
648 curoff += length;
649 /* bufsize may have just been a guess; don't go negative */
650 bufsize = bufsize > length ? bufsize - length : 0;
651 }
652
653 /*
654 * All done. Set output offset value to current offset.
655 */
656 if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
657 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
658 else
659 ctx->pos = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
660 kmem_free(map_info);
661 if (bp)
662 xfs_trans_brelse(NULL, bp);
663 return error;
664}
665
666/*
667 * Read a directory.
668 */
669int
670xfs_readdir(
671 xfs_inode_t *dp,
672 struct dir_context *ctx,
673 size_t bufsize)
674{
675 int rval; /* return value */
676 int v; /* type-checking value */
677 uint lock_mode;
678
679 trace_xfs_readdir(dp);
680
681 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
682 return XFS_ERROR(EIO);
683
684 ASSERT(S_ISDIR(dp->i_d.di_mode));
685 XFS_STATS_INC(xs_dir_getdents);
686
687 lock_mode = xfs_ilock_data_map_shared(dp);
688 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
689 rval = xfs_dir2_sf_getdents(dp, ctx);
690 else if ((rval = xfs_dir2_isblock(NULL, dp, &v)))
691 ;
692 else if (v)
693 rval = xfs_dir2_block_getdents(dp, ctx);
694 else
695 rval = xfs_dir2_leaf_getdents(dp, ctx, bufsize);
696 xfs_iunlock(dp, lock_mode);
697
698 return rval;
699}