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_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
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{
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;
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{
349 struct xfs_inode *dp = args->dp;
350 struct xfs_mount *mp = dp->i_mount;
351 struct xfs_buf *bp = NULL; /* data block buffer */
352 xfs_dir2_data_entry_t *dep; /* data entry */
353 xfs_dir2_data_unused_t *dup; /* unused entry */
354 struct xfs_da_geometry *geo = args->geo;
355 xfs_dablk_t rablk = 0; /* current readahead block */
356 xfs_dir2_off_t curoff; /* current overall offset */
357 int length; /* temporary length value */
358 int byteoff; /* offset in current block */
359 int lock_mode;
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 lock_mode = xfs_ilock_data_map_shared(dp);
394 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
395 &rablk, &bp);
396 xfs_iunlock(dp, lock_mode);
397 if (error || !bp)
398 break;
399
400 xfs_dir3_data_check(dp, bp);
401 /*
402 * Find our position in the block.
403 */
404 offset = geo->data_entry_offset;
405 byteoff = xfs_dir2_byte_to_off(geo, curoff);
406 /*
407 * Skip past the header.
408 */
409 if (byteoff == 0)
410 curoff += geo->data_entry_offset;
411 /*
412 * Skip past entries until we reach our offset.
413 */
414 else {
415 while (offset < byteoff) {
416 dup = bp->b_addr + offset;
417
418 if (be16_to_cpu(dup->freetag)
419 == XFS_DIR2_DATA_FREE_TAG) {
420
421 length = be16_to_cpu(dup->length);
422 offset += length;
423 continue;
424 }
425 dep = bp->b_addr + offset;
426 length = xfs_dir2_data_entsize(mp,
427 dep->namelen);
428 offset += length;
429 }
430 /*
431 * Now set our real offset.
432 */
433 curoff =
434 xfs_dir2_db_off_to_byte(geo,
435 xfs_dir2_byte_to_db(geo, curoff),
436 offset);
437 if (offset >= geo->blksize)
438 continue;
439 }
440 }
441
442 /*
443 * We have a pointer to an entry. Is it a live one?
444 */
445 dup = bp->b_addr + offset;
446
447 /*
448 * No, it's unused, skip over it.
449 */
450 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
451 length = be16_to_cpu(dup->length);
452 offset += length;
453 curoff += length;
454 continue;
455 }
456
457 dep = bp->b_addr + offset;
458 length = xfs_dir2_data_entsize(mp, dep->namelen);
459 filetype = xfs_dir2_data_get_ftype(mp, dep);
460
461 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
462 if (XFS_IS_CORRUPT(dp->i_mount,
463 !xfs_dir2_namecheck(dep->name,
464 dep->namelen))) {
465 error = -EFSCORRUPTED;
466 break;
467 }
468 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
469 be64_to_cpu(dep->inumber),
470 xfs_dir3_get_dtype(dp->i_mount, filetype)))
471 break;
472
473 /*
474 * Advance to next entry in the block.
475 */
476 offset += length;
477 curoff += length;
478 /* bufsize may have just been a guess; don't go negative */
479 bufsize = bufsize > length ? bufsize - length : 0;
480 }
481
482 /*
483 * All done. Set output offset value to current offset.
484 */
485 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
486 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
487 else
488 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
489 if (bp)
490 xfs_trans_brelse(args->trans, bp);
491 return error;
492}
493
494/*
495 * Read a directory.
496 *
497 * If supplied, the transaction collects locked dir buffers to avoid
498 * nested buffer deadlocks. This function does not dirty the
499 * transaction. The caller should ensure that the inode is locked
500 * before calling this function.
501 */
502int
503xfs_readdir(
504 struct xfs_trans *tp,
505 struct xfs_inode *dp,
506 struct dir_context *ctx,
507 size_t bufsize)
508{
509 struct xfs_da_args args = { NULL };
510 int rval;
511 int v;
512
513 trace_xfs_readdir(dp);
514
515 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
516 return -EIO;
517
518 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
519 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
520
521 args.dp = dp;
522 args.geo = dp->i_mount->m_dir_geo;
523 args.trans = tp;
524
525 if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
526 rval = xfs_dir2_sf_getdents(&args, ctx);
527 else if ((rval = xfs_dir2_isblock(&args, &v)))
528 ;
529 else if (v)
530 rval = xfs_dir2_block_getdents(&args, ctx);
531 else
532 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
533
534 return rval;
535}
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_bit.h"
25#include "xfs_mount.h"
26#include "xfs_da_format.h"
27#include "xfs_da_btree.h"
28#include "xfs_inode.h"
29#include "xfs_dir2.h"
30#include "xfs_dir2_priv.h"
31#include "xfs_error.h"
32#include "xfs_trace.h"
33#include "xfs_bmap.h"
34#include "xfs_trans.h"
35
36/*
37 * Directory file type support functions
38 */
39static unsigned char xfs_dir3_filetype_table[] = {
40 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
41 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
42};
43
44static unsigned char
45xfs_dir3_get_dtype(
46 struct xfs_mount *mp,
47 __uint8_t filetype)
48{
49 if (!xfs_sb_version_hasftype(&mp->m_sb))
50 return DT_UNKNOWN;
51
52 if (filetype >= XFS_DIR3_FT_MAX)
53 return DT_UNKNOWN;
54
55 return xfs_dir3_filetype_table[filetype];
56}
57
58STATIC int
59xfs_dir2_sf_getdents(
60 struct xfs_da_args *args,
61 struct dir_context *ctx)
62{
63 int i; /* shortform entry number */
64 struct xfs_inode *dp = args->dp; /* incore directory inode */
65 xfs_dir2_dataptr_t off; /* current entry's offset */
66 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
67 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
68 xfs_dir2_dataptr_t dot_offset;
69 xfs_dir2_dataptr_t dotdot_offset;
70 xfs_ino_t ino;
71 struct xfs_da_geometry *geo = args->geo;
72
73 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
74 /*
75 * Give up if the directory is way too short.
76 */
77 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
78 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
79 return -EIO;
80 }
81
82 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
83 ASSERT(dp->i_df.if_u1.if_data != NULL);
84
85 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
86
87 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
88
89 /*
90 * If the block number in the offset is out of range, we're done.
91 */
92 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
93 return 0;
94
95 /*
96 * Precalculate offsets for . and .. as we will always need them.
97 *
98 * XXX(hch): the second argument is sometimes 0 and sometimes
99 * geo->datablk
100 */
101 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
102 dp->d_ops->data_dot_offset);
103 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
104 dp->d_ops->data_dotdot_offset);
105
106 /*
107 * Put . entry unless we're starting past it.
108 */
109 if (ctx->pos <= dot_offset) {
110 ctx->pos = dot_offset & 0x7fffffff;
111 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
112 return 0;
113 }
114
115 /*
116 * Put .. entry unless we're starting past it.
117 */
118 if (ctx->pos <= dotdot_offset) {
119 ino = dp->d_ops->sf_get_parent_ino(sfp);
120 ctx->pos = dotdot_offset & 0x7fffffff;
121 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
122 return 0;
123 }
124
125 /*
126 * Loop while there are more entries and put'ing works.
127 */
128 sfep = xfs_dir2_sf_firstentry(sfp);
129 for (i = 0; i < sfp->count; i++) {
130 __uint8_t filetype;
131
132 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
133 xfs_dir2_sf_get_offset(sfep));
134
135 if (ctx->pos > off) {
136 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
137 continue;
138 }
139
140 ino = dp->d_ops->sf_get_ino(sfp, sfep);
141 filetype = dp->d_ops->sf_get_ftype(sfep);
142 ctx->pos = off & 0x7fffffff;
143 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
144 xfs_dir3_get_dtype(dp->i_mount, filetype)))
145 return 0;
146 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
147 }
148
149 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
150 0x7fffffff;
151 return 0;
152}
153
154/*
155 * Readdir for block directories.
156 */
157STATIC int
158xfs_dir2_block_getdents(
159 struct xfs_da_args *args,
160 struct dir_context *ctx)
161{
162 struct xfs_inode *dp = args->dp; /* incore directory inode */
163 xfs_dir2_data_hdr_t *hdr; /* block header */
164 struct xfs_buf *bp; /* buffer for block */
165 xfs_dir2_block_tail_t *btp; /* block tail */
166 xfs_dir2_data_entry_t *dep; /* block data entry */
167 xfs_dir2_data_unused_t *dup; /* block unused entry */
168 char *endptr; /* end of the data entries */
169 int error; /* error return value */
170 char *ptr; /* current data entry */
171 int wantoff; /* starting block offset */
172 xfs_off_t cook;
173 struct xfs_da_geometry *geo = args->geo;
174 int lock_mode;
175
176 /*
177 * If the block number in the offset is out of range, we're done.
178 */
179 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
180 return 0;
181
182 lock_mode = xfs_ilock_data_map_shared(dp);
183 error = xfs_dir3_block_read(NULL, dp, &bp);
184 xfs_iunlock(dp, lock_mode);
185 if (error)
186 return error;
187
188 /*
189 * Extract the byte offset we start at from the seek pointer.
190 * We'll skip entries before this.
191 */
192 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
193 hdr = bp->b_addr;
194 xfs_dir3_data_check(dp, bp);
195 /*
196 * Set up values for the loop.
197 */
198 btp = xfs_dir2_block_tail_p(geo, hdr);
199 ptr = (char *)dp->d_ops->data_entry_p(hdr);
200 endptr = (char *)xfs_dir2_block_leaf_p(btp);
201
202 /*
203 * Loop over the data portion of the block.
204 * Each object is a real entry (dep) or an unused one (dup).
205 */
206 while (ptr < endptr) {
207 __uint8_t filetype;
208
209 dup = (xfs_dir2_data_unused_t *)ptr;
210 /*
211 * Unused, skip it.
212 */
213 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
214 ptr += be16_to_cpu(dup->length);
215 continue;
216 }
217
218 dep = (xfs_dir2_data_entry_t *)ptr;
219
220 /*
221 * Bump pointer for the next iteration.
222 */
223 ptr += dp->d_ops->data_entsize(dep->namelen);
224 /*
225 * The entry is before the desired starting point, skip it.
226 */
227 if ((char *)dep - (char *)hdr < wantoff)
228 continue;
229
230 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
231 (char *)dep - (char *)hdr);
232
233 ctx->pos = cook & 0x7fffffff;
234 filetype = dp->d_ops->data_get_ftype(dep);
235 /*
236 * If it didn't fit, set the final offset to here & return.
237 */
238 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
239 be64_to_cpu(dep->inumber),
240 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
241 xfs_trans_brelse(NULL, bp);
242 return 0;
243 }
244 }
245
246 /*
247 * Reached the end of the block.
248 * Set the offset to a non-existent block 1 and return.
249 */
250 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
251 0x7fffffff;
252 xfs_trans_brelse(NULL, bp);
253 return 0;
254}
255
256struct xfs_dir2_leaf_map_info {
257 xfs_extlen_t map_blocks; /* number of fsbs in map */
258 xfs_dablk_t map_off; /* last mapped file offset */
259 int map_size; /* total entries in *map */
260 int map_valid; /* valid entries in *map */
261 int nmap; /* mappings to ask xfs_bmapi */
262 xfs_dir2_db_t curdb; /* db for current block */
263 int ra_current; /* number of read-ahead blks */
264 int ra_index; /* *map index for read-ahead */
265 int ra_offset; /* map entry offset for ra */
266 int ra_want; /* readahead count wanted */
267 struct xfs_bmbt_irec map[]; /* map vector for blocks */
268};
269
270STATIC int
271xfs_dir2_leaf_readbuf(
272 struct xfs_da_args *args,
273 size_t bufsize,
274 struct xfs_dir2_leaf_map_info *mip,
275 xfs_dir2_off_t *curoff,
276 struct xfs_buf **bpp)
277{
278 struct xfs_inode *dp = args->dp;
279 struct xfs_buf *bp = *bpp;
280 struct xfs_bmbt_irec *map = mip->map;
281 struct blk_plug plug;
282 int error = 0;
283 int length;
284 int i;
285 int j;
286 struct xfs_da_geometry *geo = args->geo;
287
288 /*
289 * If we have a buffer, we need to release it and
290 * take it out of the mapping.
291 */
292
293 if (bp) {
294 xfs_trans_brelse(NULL, bp);
295 bp = NULL;
296 mip->map_blocks -= geo->fsbcount;
297 /*
298 * Loop to get rid of the extents for the
299 * directory block.
300 */
301 for (i = geo->fsbcount; i > 0; ) {
302 j = min_t(int, map->br_blockcount, i);
303 map->br_blockcount -= j;
304 map->br_startblock += j;
305 map->br_startoff += j;
306 /*
307 * If mapping is done, pitch it from
308 * the table.
309 */
310 if (!map->br_blockcount && --mip->map_valid)
311 memmove(&map[0], &map[1],
312 sizeof(map[0]) * mip->map_valid);
313 i -= j;
314 }
315 }
316
317 /*
318 * Recalculate the readahead blocks wanted.
319 */
320 mip->ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog)) - 1;
321 ASSERT(mip->ra_want >= 0);
322
323 /*
324 * If we don't have as many as we want, and we haven't
325 * run out of data blocks, get some more mappings.
326 */
327 if (1 + mip->ra_want > mip->map_blocks &&
328 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
329 /*
330 * Get more bmaps, fill in after the ones
331 * we already have in the table.
332 */
333 mip->nmap = mip->map_size - mip->map_valid;
334 error = xfs_bmapi_read(dp, mip->map_off,
335 xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET) -
336 mip->map_off,
337 &map[mip->map_valid], &mip->nmap, 0);
338
339 /*
340 * Don't know if we should ignore this or try to return an
341 * error. The trouble with returning errors is that readdir
342 * will just stop without actually passing the error through.
343 */
344 if (error)
345 goto out; /* XXX */
346
347 /*
348 * If we got all the mappings we asked for, set the final map
349 * offset based on the last bmap value received. Otherwise,
350 * we've reached the end.
351 */
352 if (mip->nmap == mip->map_size - mip->map_valid) {
353 i = mip->map_valid + mip->nmap - 1;
354 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
355 } else
356 mip->map_off = xfs_dir2_byte_to_da(geo,
357 XFS_DIR2_LEAF_OFFSET);
358
359 /*
360 * Look for holes in the mapping, and eliminate them. Count up
361 * the valid blocks.
362 */
363 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
364 if (map[i].br_startblock == HOLESTARTBLOCK) {
365 mip->nmap--;
366 length = mip->map_valid + mip->nmap - i;
367 if (length)
368 memmove(&map[i], &map[i + 1],
369 sizeof(map[i]) * length);
370 } else {
371 mip->map_blocks += map[i].br_blockcount;
372 i++;
373 }
374 }
375 mip->map_valid += mip->nmap;
376 }
377
378 /*
379 * No valid mappings, so no more data blocks.
380 */
381 if (!mip->map_valid) {
382 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
383 goto out;
384 }
385
386 /*
387 * Read the directory block starting at the first mapping.
388 */
389 mip->curdb = xfs_dir2_da_to_db(geo, map->br_startoff);
390 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
391 map->br_blockcount >= geo->fsbcount ?
392 XFS_FSB_TO_DADDR(dp->i_mount, map->br_startblock) :
393 -1, &bp);
394 /*
395 * Should just skip over the data block instead of giving up.
396 */
397 if (error)
398 goto out; /* XXX */
399
400 /*
401 * Adjust the current amount of read-ahead: we just read a block that
402 * was previously ra.
403 */
404 if (mip->ra_current)
405 mip->ra_current -= geo->fsbcount;
406
407 /*
408 * Do we need more readahead?
409 */
410 blk_start_plug(&plug);
411 for (mip->ra_index = mip->ra_offset = i = 0;
412 mip->ra_want > mip->ra_current && i < mip->map_blocks;
413 i += geo->fsbcount) {
414 ASSERT(mip->ra_index < mip->map_valid);
415 /*
416 * Read-ahead a contiguous directory block.
417 */
418 if (i > mip->ra_current &&
419 map[mip->ra_index].br_blockcount >= geo->fsbcount) {
420 xfs_dir3_data_readahead(dp,
421 map[mip->ra_index].br_startoff + mip->ra_offset,
422 XFS_FSB_TO_DADDR(dp->i_mount,
423 map[mip->ra_index].br_startblock +
424 mip->ra_offset));
425 mip->ra_current = i;
426 }
427
428 /*
429 * Read-ahead a non-contiguous directory block. This doesn't
430 * use our mapping, but this is a very rare case.
431 */
432 else if (i > mip->ra_current) {
433 xfs_dir3_data_readahead(dp,
434 map[mip->ra_index].br_startoff +
435 mip->ra_offset, -1);
436 mip->ra_current = i;
437 }
438
439 /*
440 * Advance offset through the mapping table.
441 */
442 for (j = 0; j < geo->fsbcount; j += length ) {
443 /*
444 * The rest of this extent but not more than a dir
445 * block.
446 */
447 length = min_t(int, geo->fsbcount,
448 map[mip->ra_index].br_blockcount -
449 mip->ra_offset);
450 mip->ra_offset += length;
451
452 /*
453 * Advance to the next mapping if this one is used up.
454 */
455 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
456 mip->ra_offset = 0;
457 mip->ra_index++;
458 }
459 }
460 }
461 blk_finish_plug(&plug);
462
463out:
464 *bpp = bp;
465 return error;
466}
467
468/*
469 * Getdents (readdir) for leaf and node directories.
470 * This reads the data blocks only, so is the same for both forms.
471 */
472STATIC int
473xfs_dir2_leaf_getdents(
474 struct xfs_da_args *args,
475 struct dir_context *ctx,
476 size_t bufsize)
477{
478 struct xfs_inode *dp = args->dp;
479 struct xfs_buf *bp = NULL; /* data block buffer */
480 xfs_dir2_data_hdr_t *hdr; /* data block header */
481 xfs_dir2_data_entry_t *dep; /* data entry */
482 xfs_dir2_data_unused_t *dup; /* unused entry */
483 int error = 0; /* error return value */
484 int length; /* temporary length value */
485 int byteoff; /* offset in current block */
486 xfs_dir2_off_t curoff; /* current overall offset */
487 xfs_dir2_off_t newoff; /* new curoff after new blk */
488 char *ptr = NULL; /* pointer to current data */
489 struct xfs_dir2_leaf_map_info *map_info;
490 struct xfs_da_geometry *geo = args->geo;
491
492 /*
493 * If the offset is at or past the largest allowed value,
494 * give up right away.
495 */
496 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
497 return 0;
498
499 /*
500 * Set up to bmap a number of blocks based on the caller's
501 * buffer size, the directory block size, and the filesystem
502 * block size.
503 */
504 length = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
505 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
506 (length * sizeof(struct xfs_bmbt_irec)),
507 KM_SLEEP | KM_NOFS);
508 map_info->map_size = length;
509
510 /*
511 * Inside the loop we keep the main offset value as a byte offset
512 * in the directory file.
513 */
514 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
515
516 /*
517 * Force this conversion through db so we truncate the offset
518 * down to get the start of the data block.
519 */
520 map_info->map_off = xfs_dir2_db_to_da(geo,
521 xfs_dir2_byte_to_db(geo, curoff));
522
523 /*
524 * Loop over directory entries until we reach the end offset.
525 * Get more blocks and readahead as necessary.
526 */
527 while (curoff < XFS_DIR2_LEAF_OFFSET) {
528 __uint8_t filetype;
529
530 /*
531 * If we have no buffer, or we're off the end of the
532 * current buffer, need to get another one.
533 */
534 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
535 int lock_mode;
536
537 lock_mode = xfs_ilock_data_map_shared(dp);
538 error = xfs_dir2_leaf_readbuf(args, bufsize, map_info,
539 &curoff, &bp);
540 xfs_iunlock(dp, lock_mode);
541 if (error || !map_info->map_valid)
542 break;
543
544 /*
545 * Having done a read, we need to set a new offset.
546 */
547 newoff = xfs_dir2_db_off_to_byte(geo,
548 map_info->curdb, 0);
549 /*
550 * Start of the current block.
551 */
552 if (curoff < newoff)
553 curoff = newoff;
554 /*
555 * Make sure we're in the right block.
556 */
557 else if (curoff > newoff)
558 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
559 map_info->curdb);
560 hdr = bp->b_addr;
561 xfs_dir3_data_check(dp, bp);
562 /*
563 * Find our position in the block.
564 */
565 ptr = (char *)dp->d_ops->data_entry_p(hdr);
566 byteoff = xfs_dir2_byte_to_off(geo, curoff);
567 /*
568 * Skip past the header.
569 */
570 if (byteoff == 0)
571 curoff += dp->d_ops->data_entry_offset;
572 /*
573 * Skip past entries until we reach our offset.
574 */
575 else {
576 while ((char *)ptr - (char *)hdr < byteoff) {
577 dup = (xfs_dir2_data_unused_t *)ptr;
578
579 if (be16_to_cpu(dup->freetag)
580 == XFS_DIR2_DATA_FREE_TAG) {
581
582 length = be16_to_cpu(dup->length);
583 ptr += length;
584 continue;
585 }
586 dep = (xfs_dir2_data_entry_t *)ptr;
587 length =
588 dp->d_ops->data_entsize(dep->namelen);
589 ptr += length;
590 }
591 /*
592 * Now set our real offset.
593 */
594 curoff =
595 xfs_dir2_db_off_to_byte(geo,
596 xfs_dir2_byte_to_db(geo, curoff),
597 (char *)ptr - (char *)hdr);
598 if (ptr >= (char *)hdr + geo->blksize) {
599 continue;
600 }
601 }
602 }
603 /*
604 * We have a pointer to an entry.
605 * Is it a live one?
606 */
607 dup = (xfs_dir2_data_unused_t *)ptr;
608 /*
609 * No, it's unused, skip over it.
610 */
611 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
612 length = be16_to_cpu(dup->length);
613 ptr += length;
614 curoff += length;
615 continue;
616 }
617
618 dep = (xfs_dir2_data_entry_t *)ptr;
619 length = dp->d_ops->data_entsize(dep->namelen);
620 filetype = dp->d_ops->data_get_ftype(dep);
621
622 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
623 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
624 be64_to_cpu(dep->inumber),
625 xfs_dir3_get_dtype(dp->i_mount, filetype)))
626 break;
627
628 /*
629 * Advance to next entry in the block.
630 */
631 ptr += length;
632 curoff += length;
633 /* bufsize may have just been a guess; don't go negative */
634 bufsize = bufsize > length ? bufsize - length : 0;
635 }
636
637 /*
638 * All done. Set output offset value to current offset.
639 */
640 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
641 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
642 else
643 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
644 kmem_free(map_info);
645 if (bp)
646 xfs_trans_brelse(NULL, bp);
647 return error;
648}
649
650/*
651 * Read a directory.
652 */
653int
654xfs_readdir(
655 struct xfs_inode *dp,
656 struct dir_context *ctx,
657 size_t bufsize)
658{
659 struct xfs_da_args args = { NULL };
660 int rval;
661 int v;
662
663 trace_xfs_readdir(dp);
664
665 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
666 return -EIO;
667
668 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
669 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
670
671 args.dp = dp;
672 args.geo = dp->i_mount->m_dir_geo;
673
674 xfs_ilock(dp, XFS_IOLOCK_SHARED);
675 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
676 rval = xfs_dir2_sf_getdents(&args, ctx);
677 else if ((rval = xfs_dir2_isblock(&args, &v)))
678 ;
679 else if (v)
680 rval = xfs_dir2_block_getdents(&args, ctx);
681 else
682 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
683 xfs_iunlock(dp, XFS_IOLOCK_SHARED);
684
685 return rval;
686}