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
44unsigned 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 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
75 ASSERT(dp->i_df.if_u1.if_data != NULL);
76
77 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
78
79 /*
80 * If the block number in the offset is out of range, we're done.
81 */
82 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
83 return 0;
84
85 /*
86 * Precalculate offsets for . and .. as we will always need them.
87 *
88 * XXX(hch): the second argument is sometimes 0 and sometimes
89 * geo->datablk
90 */
91 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
92 dp->d_ops->data_dot_offset);
93 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
94 dp->d_ops->data_dotdot_offset);
95
96 /*
97 * Put . entry unless we're starting past it.
98 */
99 if (ctx->pos <= dot_offset) {
100 ctx->pos = dot_offset & 0x7fffffff;
101 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
102 return 0;
103 }
104
105 /*
106 * Put .. entry unless we're starting past it.
107 */
108 if (ctx->pos <= dotdot_offset) {
109 ino = dp->d_ops->sf_get_parent_ino(sfp);
110 ctx->pos = dotdot_offset & 0x7fffffff;
111 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
112 return 0;
113 }
114
115 /*
116 * Loop while there are more entries and put'ing works.
117 */
118 sfep = xfs_dir2_sf_firstentry(sfp);
119 for (i = 0; i < sfp->count; i++) {
120 uint8_t filetype;
121
122 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
123 xfs_dir2_sf_get_offset(sfep));
124
125 if (ctx->pos > off) {
126 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
127 continue;
128 }
129
130 ino = dp->d_ops->sf_get_ino(sfp, sfep);
131 filetype = dp->d_ops->sf_get_ftype(sfep);
132 ctx->pos = off & 0x7fffffff;
133 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
134 xfs_dir3_get_dtype(dp->i_mount, filetype)))
135 return 0;
136 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
137 }
138
139 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
140 0x7fffffff;
141 return 0;
142}
143
144/*
145 * Readdir for block directories.
146 */
147STATIC int
148xfs_dir2_block_getdents(
149 struct xfs_da_args *args,
150 struct dir_context *ctx)
151{
152 struct xfs_inode *dp = args->dp; /* incore directory inode */
153 xfs_dir2_data_hdr_t *hdr; /* block header */
154 struct xfs_buf *bp; /* buffer for block */
155 xfs_dir2_data_entry_t *dep; /* block data entry */
156 xfs_dir2_data_unused_t *dup; /* block unused entry */
157 char *endptr; /* end of the data entries */
158 int error; /* error return value */
159 char *ptr; /* current data entry */
160 int wantoff; /* starting block offset */
161 xfs_off_t cook;
162 struct xfs_da_geometry *geo = args->geo;
163 int lock_mode;
164
165 /*
166 * If the block number in the offset is out of range, we're done.
167 */
168 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
169 return 0;
170
171 lock_mode = xfs_ilock_data_map_shared(dp);
172 error = xfs_dir3_block_read(args->trans, dp, &bp);
173 xfs_iunlock(dp, lock_mode);
174 if (error)
175 return error;
176
177 /*
178 * Extract the byte offset we start at from the seek pointer.
179 * We'll skip entries before this.
180 */
181 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
182 hdr = bp->b_addr;
183 xfs_dir3_data_check(dp, bp);
184 /*
185 * Set up values for the loop.
186 */
187 ptr = (char *)dp->d_ops->data_entry_p(hdr);
188 endptr = xfs_dir3_data_endp(geo, hdr);
189
190 /*
191 * Loop over the data portion of the block.
192 * Each object is a real entry (dep) or an unused one (dup).
193 */
194 while (ptr < endptr) {
195 uint8_t filetype;
196
197 dup = (xfs_dir2_data_unused_t *)ptr;
198 /*
199 * Unused, skip it.
200 */
201 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
202 ptr += be16_to_cpu(dup->length);
203 continue;
204 }
205
206 dep = (xfs_dir2_data_entry_t *)ptr;
207
208 /*
209 * Bump pointer for the next iteration.
210 */
211 ptr += dp->d_ops->data_entsize(dep->namelen);
212 /*
213 * The entry is before the desired starting point, skip it.
214 */
215 if ((char *)dep - (char *)hdr < wantoff)
216 continue;
217
218 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
219 (char *)dep - (char *)hdr);
220
221 ctx->pos = cook & 0x7fffffff;
222 filetype = dp->d_ops->data_get_ftype(dep);
223 /*
224 * If it didn't fit, set the final offset to here & return.
225 */
226 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
227 be64_to_cpu(dep->inumber),
228 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
229 xfs_trans_brelse(args->trans, bp);
230 return 0;
231 }
232 }
233
234 /*
235 * Reached the end of the block.
236 * Set the offset to a non-existent block 1 and return.
237 */
238 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
239 0x7fffffff;
240 xfs_trans_brelse(args->trans, bp);
241 return 0;
242}
243
244/*
245 * Read a directory block and initiate readahead for blocks beyond that.
246 * We maintain a sliding readahead window of the remaining space in the
247 * buffer rounded up to the nearest block.
248 */
249STATIC int
250xfs_dir2_leaf_readbuf(
251 struct xfs_da_args *args,
252 size_t bufsize,
253 xfs_dir2_off_t *cur_off,
254 xfs_dablk_t *ra_blk,
255 struct xfs_buf **bpp)
256{
257 struct xfs_inode *dp = args->dp;
258 struct xfs_buf *bp = NULL;
259 struct xfs_da_geometry *geo = args->geo;
260 struct xfs_ifork *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
261 struct xfs_bmbt_irec map;
262 struct blk_plug plug;
263 xfs_dir2_off_t new_off;
264 xfs_dablk_t next_ra;
265 xfs_dablk_t map_off;
266 xfs_dablk_t last_da;
267 struct xfs_iext_cursor icur;
268 int ra_want;
269 int error = 0;
270
271 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
272 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
273 if (error)
274 goto out;
275 }
276
277 /*
278 * Look for mapped directory blocks at or above the current offset.
279 * Truncate down to the nearest directory block to start the scanning
280 * operation.
281 */
282 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
283 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
284 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
285 goto out;
286 if (map.br_startoff >= last_da)
287 goto out;
288 xfs_trim_extent(&map, map_off, last_da - map_off);
289
290 /* Read the directory block of that first mapping. */
291 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
292 if (new_off > *cur_off)
293 *cur_off = new_off;
294 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, -1, &bp);
295 if (error)
296 goto out;
297
298 /*
299 * Start readahead for the next bufsize's worth of dir data blocks.
300 * We may have already issued readahead for some of that range;
301 * ra_blk tracks the last block we tried to read(ahead).
302 */
303 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
304 if (*ra_blk >= last_da)
305 goto out;
306 else if (*ra_blk == 0)
307 *ra_blk = map.br_startoff;
308 next_ra = map.br_startoff + geo->fsbcount;
309 if (next_ra >= last_da)
310 goto out_no_ra;
311 if (map.br_blockcount < geo->fsbcount &&
312 !xfs_iext_next_extent(ifp, &icur, &map))
313 goto out_no_ra;
314 if (map.br_startoff >= last_da)
315 goto out_no_ra;
316 xfs_trim_extent(&map, next_ra, last_da - next_ra);
317
318 /* Start ra for each dir (not fs) block that has a mapping. */
319 blk_start_plug(&plug);
320 while (ra_want > 0) {
321 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
322 while (ra_want > 0 &&
323 next_ra < map.br_startoff + map.br_blockcount) {
324 if (next_ra >= last_da) {
325 *ra_blk = last_da;
326 break;
327 }
328 if (next_ra > *ra_blk) {
329 xfs_dir3_data_readahead(dp, next_ra, -2);
330 *ra_blk = next_ra;
331 }
332 ra_want -= geo->fsbcount;
333 next_ra += geo->fsbcount;
334 }
335 if (!xfs_iext_next_extent(ifp, &icur, &map)) {
336 *ra_blk = last_da;
337 break;
338 }
339 }
340 blk_finish_plug(&plug);
341
342out:
343 *bpp = bp;
344 return error;
345out_no_ra:
346 *ra_blk = last_da;
347 goto out;
348}
349
350/*
351 * Getdents (readdir) for leaf and node directories.
352 * This reads the data blocks only, so is the same for both forms.
353 */
354STATIC int
355xfs_dir2_leaf_getdents(
356 struct xfs_da_args *args,
357 struct dir_context *ctx,
358 size_t bufsize)
359{
360 struct xfs_inode *dp = args->dp;
361 struct xfs_buf *bp = NULL; /* data block buffer */
362 xfs_dir2_data_hdr_t *hdr; /* data block header */
363 xfs_dir2_data_entry_t *dep; /* data entry */
364 xfs_dir2_data_unused_t *dup; /* unused entry */
365 char *ptr = NULL; /* pointer to current data */
366 struct xfs_da_geometry *geo = args->geo;
367 xfs_dablk_t rablk = 0; /* current readahead block */
368 xfs_dir2_off_t curoff; /* current overall offset */
369 int length; /* temporary length value */
370 int byteoff; /* offset in current block */
371 int lock_mode;
372 int error = 0; /* error return value */
373
374 /*
375 * If the offset is at or past the largest allowed value,
376 * give up right away.
377 */
378 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
379 return 0;
380
381 /*
382 * Inside the loop we keep the main offset value as a byte offset
383 * in the directory file.
384 */
385 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
386
387 /*
388 * Loop over directory entries until we reach the end offset.
389 * Get more blocks and readahead as necessary.
390 */
391 while (curoff < XFS_DIR2_LEAF_OFFSET) {
392 uint8_t filetype;
393
394 /*
395 * If we have no buffer, or we're off the end of the
396 * current buffer, need to get another one.
397 */
398 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
399 if (bp) {
400 xfs_trans_brelse(args->trans, bp);
401 bp = NULL;
402 }
403
404 lock_mode = xfs_ilock_data_map_shared(dp);
405 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
406 &rablk, &bp);
407 xfs_iunlock(dp, lock_mode);
408 if (error || !bp)
409 break;
410
411 hdr = bp->b_addr;
412 xfs_dir3_data_check(dp, bp);
413 /*
414 * Find our position in the block.
415 */
416 ptr = (char *)dp->d_ops->data_entry_p(hdr);
417 byteoff = xfs_dir2_byte_to_off(geo, curoff);
418 /*
419 * Skip past the header.
420 */
421 if (byteoff == 0)
422 curoff += dp->d_ops->data_entry_offset;
423 /*
424 * Skip past entries until we reach our offset.
425 */
426 else {
427 while ((char *)ptr - (char *)hdr < byteoff) {
428 dup = (xfs_dir2_data_unused_t *)ptr;
429
430 if (be16_to_cpu(dup->freetag)
431 == XFS_DIR2_DATA_FREE_TAG) {
432
433 length = be16_to_cpu(dup->length);
434 ptr += length;
435 continue;
436 }
437 dep = (xfs_dir2_data_entry_t *)ptr;
438 length =
439 dp->d_ops->data_entsize(dep->namelen);
440 ptr += length;
441 }
442 /*
443 * Now set our real offset.
444 */
445 curoff =
446 xfs_dir2_db_off_to_byte(geo,
447 xfs_dir2_byte_to_db(geo, curoff),
448 (char *)ptr - (char *)hdr);
449 if (ptr >= (char *)hdr + geo->blksize) {
450 continue;
451 }
452 }
453 }
454 /*
455 * We have a pointer to an entry.
456 * Is it a live one?
457 */
458 dup = (xfs_dir2_data_unused_t *)ptr;
459 /*
460 * No, it's unused, skip over it.
461 */
462 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
463 length = be16_to_cpu(dup->length);
464 ptr += length;
465 curoff += length;
466 continue;
467 }
468
469 dep = (xfs_dir2_data_entry_t *)ptr;
470 length = dp->d_ops->data_entsize(dep->namelen);
471 filetype = dp->d_ops->data_get_ftype(dep);
472
473 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
474 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
475 be64_to_cpu(dep->inumber),
476 xfs_dir3_get_dtype(dp->i_mount, filetype)))
477 break;
478
479 /*
480 * Advance to next entry in the block.
481 */
482 ptr += length;
483 curoff += length;
484 /* bufsize may have just been a guess; don't go negative */
485 bufsize = bufsize > length ? bufsize - length : 0;
486 }
487
488 /*
489 * All done. Set output offset value to current offset.
490 */
491 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
492 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
493 else
494 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
495 if (bp)
496 xfs_trans_brelse(args->trans, bp);
497 return error;
498}
499
500/*
501 * Read a directory.
502 *
503 * If supplied, the transaction collects locked dir buffers to avoid
504 * nested buffer deadlocks. This function does not dirty the
505 * transaction. The caller should ensure that the inode is locked
506 * before calling this function.
507 */
508int
509xfs_readdir(
510 struct xfs_trans *tp,
511 struct xfs_inode *dp,
512 struct dir_context *ctx,
513 size_t bufsize)
514{
515 struct xfs_da_args args = { NULL };
516 int rval;
517 int v;
518
519 trace_xfs_readdir(dp);
520
521 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
522 return -EIO;
523
524 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
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_d.di_format == XFS_DINODE_FMT_LOCAL)
532 rval = xfs_dir2_sf_getdents(&args, ctx);
533 else if ((rval = xfs_dir2_isblock(&args, &v)))
534 ;
535 else if (v)
536 rval = xfs_dir2_block_getdents(&args, ctx);
537 else
538 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
539
540 return rval;
541}
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#include "xfs_health.h"
22
23/*
24 * Directory file type support functions
25 */
26static unsigned char xfs_dir3_filetype_table[] = {
27 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
28 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
29};
30
31unsigned char
32xfs_dir3_get_dtype(
33 struct xfs_mount *mp,
34 uint8_t filetype)
35{
36 if (!xfs_has_ftype(mp))
37 return DT_UNKNOWN;
38
39 if (filetype >= XFS_DIR3_FT_MAX)
40 return DT_UNKNOWN;
41
42 return xfs_dir3_filetype_table[filetype];
43}
44
45STATIC int
46xfs_dir2_sf_getdents(
47 struct xfs_da_args *args,
48 struct dir_context *ctx)
49{
50 int i; /* shortform entry number */
51 struct xfs_inode *dp = args->dp; /* incore directory inode */
52 struct xfs_mount *mp = dp->i_mount;
53 xfs_dir2_dataptr_t off; /* current entry's offset */
54 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
55 struct xfs_dir2_sf_hdr *sfp = dp->i_df.if_data;
56 xfs_dir2_dataptr_t dot_offset;
57 xfs_dir2_dataptr_t dotdot_offset;
58 xfs_ino_t ino;
59 struct xfs_da_geometry *geo = args->geo;
60
61 ASSERT(dp->i_df.if_format == XFS_DINODE_FMT_LOCAL);
62 ASSERT(dp->i_df.if_bytes == dp->i_disk_size);
63 ASSERT(sfp != NULL);
64
65 /*
66 * If the block number in the offset is out of range, we're done.
67 */
68 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
69 return 0;
70
71 /*
72 * Precalculate offsets for "." and ".." as we will always need them.
73 * This relies on the fact that directories always start with the
74 * entries for "." and "..".
75 */
76 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
77 geo->data_entry_offset);
78 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
79 geo->data_entry_offset +
80 xfs_dir2_data_entsize(mp, sizeof(".") - 1));
81
82 /*
83 * Put . entry unless we're starting past it.
84 */
85 if (ctx->pos <= dot_offset) {
86 ctx->pos = dot_offset & 0x7fffffff;
87 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
88 return 0;
89 }
90
91 /*
92 * Put .. entry unless we're starting past it.
93 */
94 if (ctx->pos <= dotdot_offset) {
95 ino = xfs_dir2_sf_get_parent_ino(sfp);
96 ctx->pos = dotdot_offset & 0x7fffffff;
97 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
98 return 0;
99 }
100
101 /*
102 * Loop while there are more entries and put'ing works.
103 */
104 sfep = xfs_dir2_sf_firstentry(sfp);
105 for (i = 0; i < sfp->count; i++) {
106 uint8_t filetype;
107
108 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
109 xfs_dir2_sf_get_offset(sfep));
110
111 if (ctx->pos > off) {
112 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
113 continue;
114 }
115
116 ino = xfs_dir2_sf_get_ino(mp, sfp, sfep);
117 filetype = xfs_dir2_sf_get_ftype(mp, sfep);
118 ctx->pos = off & 0x7fffffff;
119 if (XFS_IS_CORRUPT(dp->i_mount,
120 !xfs_dir2_namecheck(sfep->name,
121 sfep->namelen)))
122 return -EFSCORRUPTED;
123 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
124 xfs_dir3_get_dtype(mp, filetype)))
125 return 0;
126 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
127 }
128
129 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
130 0x7fffffff;
131 return 0;
132}
133
134/*
135 * Readdir for block directories.
136 */
137STATIC int
138xfs_dir2_block_getdents(
139 struct xfs_da_args *args,
140 struct dir_context *ctx,
141 unsigned int *lock_mode)
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 unsigned int offset, next_offset;
150 unsigned int end;
151
152 /*
153 * If the block number in the offset is out of range, we're done.
154 */
155 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
156 return 0;
157
158 error = xfs_dir3_block_read(args->trans, dp, &bp);
159 if (error)
160 return error;
161
162 xfs_iunlock(dp, *lock_mode);
163 *lock_mode = 0;
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;
229out_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 */
239STATIC int
240xfs_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 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
262 if (error)
263 goto out;
264
265 /*
266 * Look for mapped directory blocks at or above the current offset.
267 * Truncate down to the nearest directory block to start the scanning
268 * operation.
269 */
270 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
271 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
272 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
273 goto out;
274 if (map.br_startoff >= last_da)
275 goto out;
276 xfs_trim_extent(&map, map_off, last_da - map_off);
277
278 /* Read the directory block of that first mapping. */
279 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
280 if (new_off > *cur_off)
281 *cur_off = new_off;
282 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
283 if (error)
284 goto out;
285
286 /*
287 * Start readahead for the next bufsize's worth of dir data blocks.
288 * We may have already issued readahead for some of that range;
289 * ra_blk tracks the last block we tried to read(ahead).
290 */
291 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
292 if (*ra_blk >= last_da)
293 goto out;
294 else if (*ra_blk == 0)
295 *ra_blk = map.br_startoff;
296 next_ra = map.br_startoff + geo->fsbcount;
297 if (next_ra >= last_da)
298 goto out_no_ra;
299 if (map.br_blockcount < geo->fsbcount &&
300 !xfs_iext_next_extent(ifp, &icur, &map))
301 goto out_no_ra;
302 if (map.br_startoff >= last_da)
303 goto out_no_ra;
304 xfs_trim_extent(&map, next_ra, last_da - next_ra);
305
306 /* Start ra for each dir (not fs) block that has a mapping. */
307 blk_start_plug(&plug);
308 while (ra_want > 0) {
309 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
310 while (ra_want > 0 &&
311 next_ra < map.br_startoff + map.br_blockcount) {
312 if (next_ra >= last_da) {
313 *ra_blk = last_da;
314 break;
315 }
316 if (next_ra > *ra_blk) {
317 xfs_dir3_data_readahead(dp, next_ra,
318 XFS_DABUF_MAP_HOLE_OK);
319 *ra_blk = next_ra;
320 }
321 ra_want -= geo->fsbcount;
322 next_ra += geo->fsbcount;
323 }
324 if (!xfs_iext_next_extent(ifp, &icur, &map)) {
325 *ra_blk = last_da;
326 break;
327 }
328 }
329 blk_finish_plug(&plug);
330
331out:
332 *bpp = bp;
333 return error;
334out_no_ra:
335 *ra_blk = last_da;
336 goto out;
337}
338
339/*
340 * Getdents (readdir) for leaf and node directories.
341 * This reads the data blocks only, so is the same for both forms.
342 */
343STATIC int
344xfs_dir2_leaf_getdents(
345 struct xfs_da_args *args,
346 struct dir_context *ctx,
347 size_t bufsize,
348 unsigned int *lock_mode)
349{
350 struct xfs_inode *dp = args->dp;
351 struct xfs_mount *mp = dp->i_mount;
352 struct xfs_buf *bp = NULL; /* data block buffer */
353 xfs_dir2_data_entry_t *dep; /* data entry */
354 xfs_dir2_data_unused_t *dup; /* unused entry */
355 struct xfs_da_geometry *geo = args->geo;
356 xfs_dablk_t rablk = 0; /* current readahead block */
357 xfs_dir2_off_t curoff; /* current overall offset */
358 int length; /* temporary length value */
359 int byteoff; /* offset in current block */
360 unsigned int offset = 0;
361 int error = 0; /* error return value */
362
363 /*
364 * If the offset is at or past the largest allowed value,
365 * give up right away.
366 */
367 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
368 return 0;
369
370 /*
371 * Inside the loop we keep the main offset value as a byte offset
372 * in the directory file.
373 */
374 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
375
376 /*
377 * Loop over directory entries until we reach the end offset.
378 * Get more blocks and readahead as necessary.
379 */
380 while (curoff < XFS_DIR2_LEAF_OFFSET) {
381 uint8_t filetype;
382
383 /*
384 * If we have no buffer, or we're off the end of the
385 * current buffer, need to get another one.
386 */
387 if (!bp || offset >= geo->blksize) {
388 if (bp) {
389 xfs_trans_brelse(args->trans, bp);
390 bp = NULL;
391 }
392
393 if (*lock_mode == 0)
394 *lock_mode = xfs_ilock_data_map_shared(dp);
395 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
396 &rablk, &bp);
397 if (error || !bp)
398 break;
399
400 xfs_iunlock(dp, *lock_mode);
401 *lock_mode = 0;
402
403 xfs_dir3_data_check(dp, bp);
404 /*
405 * Find our position in the block.
406 */
407 offset = geo->data_entry_offset;
408 byteoff = xfs_dir2_byte_to_off(geo, curoff);
409 /*
410 * Skip past the header.
411 */
412 if (byteoff == 0)
413 curoff += geo->data_entry_offset;
414 /*
415 * Skip past entries until we reach our offset.
416 */
417 else {
418 while (offset < byteoff) {
419 dup = bp->b_addr + offset;
420
421 if (be16_to_cpu(dup->freetag)
422 == XFS_DIR2_DATA_FREE_TAG) {
423
424 length = be16_to_cpu(dup->length);
425 offset += length;
426 continue;
427 }
428 dep = bp->b_addr + offset;
429 length = xfs_dir2_data_entsize(mp,
430 dep->namelen);
431 offset += length;
432 }
433 /*
434 * Now set our real offset.
435 */
436 curoff =
437 xfs_dir2_db_off_to_byte(geo,
438 xfs_dir2_byte_to_db(geo, curoff),
439 offset);
440 if (offset >= geo->blksize)
441 continue;
442 }
443 }
444
445 /*
446 * We have a pointer to an entry. Is it a live one?
447 */
448 dup = bp->b_addr + offset;
449
450 /*
451 * No, it's unused, skip over it.
452 */
453 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
454 length = be16_to_cpu(dup->length);
455 offset += length;
456 curoff += length;
457 continue;
458 }
459
460 dep = bp->b_addr + offset;
461 length = xfs_dir2_data_entsize(mp, dep->namelen);
462 filetype = xfs_dir2_data_get_ftype(mp, dep);
463
464 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
465 if (XFS_IS_CORRUPT(dp->i_mount,
466 !xfs_dir2_namecheck(dep->name,
467 dep->namelen))) {
468 error = -EFSCORRUPTED;
469 break;
470 }
471 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
472 be64_to_cpu(dep->inumber),
473 xfs_dir3_get_dtype(dp->i_mount, filetype)))
474 break;
475
476 /*
477 * Advance to next entry in the block.
478 */
479 offset += length;
480 curoff += length;
481 /* bufsize may have just been a guess; don't go negative */
482 bufsize = bufsize > length ? bufsize - length : 0;
483 }
484
485 /*
486 * All done. Set output offset value to current offset.
487 */
488 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
489 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
490 else
491 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
492 if (bp)
493 xfs_trans_brelse(args->trans, bp);
494 return error;
495}
496
497/*
498 * Read a directory.
499 *
500 * If supplied, the transaction collects locked dir buffers to avoid
501 * nested buffer deadlocks. This function does not dirty the
502 * transaction. The caller must hold the IOLOCK (shared or exclusive)
503 * before calling this function.
504 */
505int
506xfs_readdir(
507 struct xfs_trans *tp,
508 struct xfs_inode *dp,
509 struct dir_context *ctx,
510 size_t bufsize)
511{
512 struct xfs_da_args args = { NULL };
513 unsigned int lock_mode;
514 bool isblock;
515 int error;
516
517 trace_xfs_readdir(dp);
518
519 if (xfs_is_shutdown(dp->i_mount))
520 return -EIO;
521 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
522 return -EIO;
523
524 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
525 ASSERT(xfs_isilocked(dp, XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
526 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
527
528 args.dp = dp;
529 args.geo = dp->i_mount->m_dir_geo;
530 args.trans = tp;
531
532 if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
533 return xfs_dir2_sf_getdents(&args, ctx);
534
535 lock_mode = xfs_ilock_data_map_shared(dp);
536 error = xfs_dir2_isblock(&args, &isblock);
537 if (error)
538 goto out_unlock;
539
540 if (isblock) {
541 error = xfs_dir2_block_getdents(&args, ctx, &lock_mode);
542 goto out_unlock;
543 }
544
545 error = xfs_dir2_leaf_getdents(&args, ctx, bufsize, &lock_mode);
546
547out_unlock:
548 if (lock_mode)
549 xfs_iunlock(dp, lock_mode);
550 return error;
551}