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 */
25 static 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
30 unsigned char
xfs_dir3_get_dtype(struct xfs_mount * mp,uint8_t filetype)31 xfs_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
44 STATIC int
xfs_dir2_sf_getdents(struct xfs_da_args * args,struct dir_context * ctx)45 xfs_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 */
138 STATIC int
xfs_dir2_block_getdents(struct xfs_da_args * args,struct dir_context * ctx,unsigned int * lock_mode)139 xfs_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;
230 out_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 */
240 STATIC int
xfs_dir2_leaf_readbuf(struct xfs_da_args * args,size_t bufsize,xfs_dir2_off_t * cur_off,xfs_dablk_t * ra_blk,struct xfs_buf ** bpp)241 xfs_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
332 out:
333 *bpp = bp;
334 return error;
335 out_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 */
344 STATIC int
xfs_dir2_leaf_getdents(struct xfs_da_args * args,struct dir_context * ctx,size_t bufsize,unsigned int * lock_mode)345 xfs_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 */
506 int
xfs_readdir(struct xfs_trans * tp,struct xfs_inode * dp,struct dir_context * ctx,size_t bufsize)507 xfs_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
546 out_unlock:
547 if (lock_mode)
548 xfs_iunlock(dp, lock_mode);
549 return error;
550 }
551