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_sb_version_hasftype(&mp->m_sb))
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_flags & XFS_IFINLINE);
61 ASSERT(dp->i_df.if_bytes == dp->i_d.di_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)139 xfs_dir2_block_getdents(
140 struct xfs_da_args *args,
141 struct dir_context *ctx)
142 {
143 struct xfs_inode *dp = args->dp; /* incore directory inode */
144 struct xfs_buf *bp; /* buffer for block */
145 int error; /* error return value */
146 int wantoff; /* starting block offset */
147 xfs_off_t cook;
148 struct xfs_da_geometry *geo = args->geo;
149 int lock_mode;
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 lock_mode = xfs_ilock_data_map_shared(dp);
160 error = xfs_dir3_block_read(args->trans, dp, &bp);
161 xfs_iunlock(dp, lock_mode);
162 if (error)
163 return error;
164
165 /*
166 * Extract the byte offset we start at from the seek pointer.
167 * We'll skip entries before this.
168 */
169 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
170 xfs_dir3_data_check(dp, bp);
171
172 /*
173 * Loop over the data portion of the block.
174 * Each object is a real entry (dep) or an unused one (dup).
175 */
176 end = xfs_dir3_data_end_offset(geo, bp->b_addr);
177 for (offset = geo->data_entry_offset;
178 offset < end;
179 offset = next_offset) {
180 struct xfs_dir2_data_unused *dup = bp->b_addr + offset;
181 struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
182 uint8_t filetype;
183
184 /*
185 * Unused, skip it.
186 */
187 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
188 next_offset = offset + be16_to_cpu(dup->length);
189 continue;
190 }
191
192 /*
193 * Bump pointer for the next iteration.
194 */
195 next_offset = offset +
196 xfs_dir2_data_entsize(dp->i_mount, dep->namelen);
197
198 /*
199 * The entry is before the desired starting point, skip it.
200 */
201 if (offset < wantoff)
202 continue;
203
204 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, offset);
205
206 ctx->pos = cook & 0x7fffffff;
207 filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
208 /*
209 * If it didn't fit, set the final offset to here & return.
210 */
211 if (XFS_IS_CORRUPT(dp->i_mount,
212 !xfs_dir2_namecheck(dep->name,
213 dep->namelen))) {
214 error = -EFSCORRUPTED;
215 goto out_rele;
216 }
217 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
218 be64_to_cpu(dep->inumber),
219 xfs_dir3_get_dtype(dp->i_mount, filetype)))
220 goto out_rele;
221 }
222
223 /*
224 * Reached the end of the block.
225 * Set the offset to a non-existent block 1 and return.
226 */
227 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
228 0x7fffffff;
229 out_rele:
230 xfs_trans_brelse(args->trans, bp);
231 return error;
232 }
233
234 /*
235 * Read a directory block and initiate readahead for blocks beyond that.
236 * We maintain a sliding readahead window of the remaining space in the
237 * buffer rounded up to the nearest block.
238 */
239 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)240 xfs_dir2_leaf_readbuf(
241 struct xfs_da_args *args,
242 size_t bufsize,
243 xfs_dir2_off_t *cur_off,
244 xfs_dablk_t *ra_blk,
245 struct xfs_buf **bpp)
246 {
247 struct xfs_inode *dp = args->dp;
248 struct xfs_buf *bp = NULL;
249 struct xfs_da_geometry *geo = args->geo;
250 struct xfs_ifork *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
251 struct xfs_bmbt_irec map;
252 struct blk_plug plug;
253 xfs_dir2_off_t new_off;
254 xfs_dablk_t next_ra;
255 xfs_dablk_t map_off;
256 xfs_dablk_t last_da;
257 struct xfs_iext_cursor icur;
258 int ra_want;
259 int error = 0;
260
261 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
262 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
263 if (error)
264 goto out;
265 }
266
267 /*
268 * Look for mapped directory blocks at or above the current offset.
269 * Truncate down to the nearest directory block to start the scanning
270 * operation.
271 */
272 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
273 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
274 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
275 goto out;
276 if (map.br_startoff >= last_da)
277 goto out;
278 xfs_trim_extent(&map, map_off, last_da - map_off);
279
280 /* Read the directory block of that first mapping. */
281 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
282 if (new_off > *cur_off)
283 *cur_off = new_off;
284 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
285 if (error)
286 goto out;
287
288 /*
289 * Start readahead for the next bufsize's worth of dir data blocks.
290 * We may have already issued readahead for some of that range;
291 * ra_blk tracks the last block we tried to read(ahead).
292 */
293 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
294 if (*ra_blk >= last_da)
295 goto out;
296 else if (*ra_blk == 0)
297 *ra_blk = map.br_startoff;
298 next_ra = map.br_startoff + geo->fsbcount;
299 if (next_ra >= last_da)
300 goto out_no_ra;
301 if (map.br_blockcount < geo->fsbcount &&
302 !xfs_iext_next_extent(ifp, &icur, &map))
303 goto out_no_ra;
304 if (map.br_startoff >= last_da)
305 goto out_no_ra;
306 xfs_trim_extent(&map, next_ra, last_da - next_ra);
307
308 /* Start ra for each dir (not fs) block that has a mapping. */
309 blk_start_plug(&plug);
310 while (ra_want > 0) {
311 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
312 while (ra_want > 0 &&
313 next_ra < map.br_startoff + map.br_blockcount) {
314 if (next_ra >= last_da) {
315 *ra_blk = last_da;
316 break;
317 }
318 if (next_ra > *ra_blk) {
319 xfs_dir3_data_readahead(dp, next_ra,
320 XFS_DABUF_MAP_HOLE_OK);
321 *ra_blk = next_ra;
322 }
323 ra_want -= geo->fsbcount;
324 next_ra += geo->fsbcount;
325 }
326 if (!xfs_iext_next_extent(ifp, &icur, &map)) {
327 *ra_blk = last_da;
328 break;
329 }
330 }
331 blk_finish_plug(&plug);
332
333 out:
334 *bpp = bp;
335 return error;
336 out_no_ra:
337 *ra_blk = last_da;
338 goto out;
339 }
340
341 /*
342 * Getdents (readdir) for leaf and node directories.
343 * This reads the data blocks only, so is the same for both forms.
344 */
345 STATIC int
xfs_dir2_leaf_getdents(struct xfs_da_args * args,struct dir_context * ctx,size_t bufsize)346 xfs_dir2_leaf_getdents(
347 struct xfs_da_args *args,
348 struct dir_context *ctx,
349 size_t bufsize)
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 int lock_mode;
362 unsigned int offset = 0;
363 int error = 0; /* error return value */
364
365 /*
366 * If the offset is at or past the largest allowed value,
367 * give up right away.
368 */
369 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
370 return 0;
371
372 /*
373 * Inside the loop we keep the main offset value as a byte offset
374 * in the directory file.
375 */
376 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
377
378 /*
379 * Loop over directory entries until we reach the end offset.
380 * Get more blocks and readahead as necessary.
381 */
382 while (curoff < XFS_DIR2_LEAF_OFFSET) {
383 uint8_t filetype;
384
385 /*
386 * If we have no buffer, or we're off the end of the
387 * current buffer, need to get another one.
388 */
389 if (!bp || offset >= geo->blksize) {
390 if (bp) {
391 xfs_trans_brelse(args->trans, bp);
392 bp = NULL;
393 }
394
395 lock_mode = xfs_ilock_data_map_shared(dp);
396 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
397 &rablk, &bp);
398 xfs_iunlock(dp, lock_mode);
399 if (error || !bp)
400 break;
401
402 xfs_dir3_data_check(dp, bp);
403 /*
404 * Find our position in the block.
405 */
406 offset = geo->data_entry_offset;
407 byteoff = xfs_dir2_byte_to_off(geo, curoff);
408 /*
409 * Skip past the header.
410 */
411 if (byteoff == 0)
412 curoff += geo->data_entry_offset;
413 /*
414 * Skip past entries until we reach our offset.
415 */
416 else {
417 while (offset < byteoff) {
418 dup = bp->b_addr + offset;
419
420 if (be16_to_cpu(dup->freetag)
421 == XFS_DIR2_DATA_FREE_TAG) {
422
423 length = be16_to_cpu(dup->length);
424 offset += length;
425 continue;
426 }
427 dep = bp->b_addr + offset;
428 length = xfs_dir2_data_entsize(mp,
429 dep->namelen);
430 offset += length;
431 }
432 /*
433 * Now set our real offset.
434 */
435 curoff =
436 xfs_dir2_db_off_to_byte(geo,
437 xfs_dir2_byte_to_db(geo, curoff),
438 offset);
439 if (offset >= geo->blksize)
440 continue;
441 }
442 }
443
444 /*
445 * We have a pointer to an entry. Is it a live one?
446 */
447 dup = bp->b_addr + offset;
448
449 /*
450 * No, it's unused, skip over it.
451 */
452 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
453 length = be16_to_cpu(dup->length);
454 offset += length;
455 curoff += length;
456 continue;
457 }
458
459 dep = bp->b_addr + offset;
460 length = xfs_dir2_data_entsize(mp, dep->namelen);
461 filetype = xfs_dir2_data_get_ftype(mp, dep);
462
463 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
464 if (XFS_IS_CORRUPT(dp->i_mount,
465 !xfs_dir2_namecheck(dep->name,
466 dep->namelen))) {
467 error = -EFSCORRUPTED;
468 break;
469 }
470 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
471 be64_to_cpu(dep->inumber),
472 xfs_dir3_get_dtype(dp->i_mount, filetype)))
473 break;
474
475 /*
476 * Advance to next entry in the block.
477 */
478 offset += length;
479 curoff += length;
480 /* bufsize may have just been a guess; don't go negative */
481 bufsize = bufsize > length ? bufsize - length : 0;
482 }
483
484 /*
485 * All done. Set output offset value to current offset.
486 */
487 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
488 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
489 else
490 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
491 if (bp)
492 xfs_trans_brelse(args->trans, bp);
493 return error;
494 }
495
496 /*
497 * Read a directory.
498 *
499 * If supplied, the transaction collects locked dir buffers to avoid
500 * nested buffer deadlocks. This function does not dirty the
501 * transaction. The caller should ensure that the inode is locked
502 * before calling this function.
503 */
504 int
xfs_readdir(struct xfs_trans * tp,struct xfs_inode * dp,struct dir_context * ctx,size_t bufsize)505 xfs_readdir(
506 struct xfs_trans *tp,
507 struct xfs_inode *dp,
508 struct dir_context *ctx,
509 size_t bufsize)
510 {
511 struct xfs_da_args args = { NULL };
512 int rval;
513 int v;
514
515 trace_xfs_readdir(dp);
516
517 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
518 return -EIO;
519
520 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
521 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
522
523 args.dp = dp;
524 args.geo = dp->i_mount->m_dir_geo;
525 args.trans = tp;
526
527 if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
528 rval = xfs_dir2_sf_getdents(&args, ctx);
529 else if ((rval = xfs_dir2_isblock(&args, &v)))
530 ;
531 else if (v)
532 rval = xfs_dir2_block_getdents(&args, ctx);
533 else
534 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
535
536 return rval;
537 }
538