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
21/*
22 * Directory file type support functions
23 */
24static unsigned char xfs_dir3_filetype_table[] = {
25 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
26 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
27};
28
29unsigned char
30xfs_dir3_get_dtype(
31 struct xfs_mount *mp,
32 uint8_t filetype)
33{
34 if (!xfs_sb_version_hasftype(&mp->m_sb))
35 return DT_UNKNOWN;
36
37 if (filetype >= XFS_DIR3_FT_MAX)
38 return DT_UNKNOWN;
39
40 return xfs_dir3_filetype_table[filetype];
41}
42
43STATIC int
44xfs_dir2_sf_getdents(
45 struct xfs_da_args *args,
46 struct dir_context *ctx)
47{
48 int i; /* shortform entry number */
49 struct xfs_inode *dp = args->dp; /* incore directory inode */
50 xfs_dir2_dataptr_t off; /* current entry's offset */
51 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
52 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
53 xfs_dir2_dataptr_t dot_offset;
54 xfs_dir2_dataptr_t dotdot_offset;
55 xfs_ino_t ino;
56 struct xfs_da_geometry *geo = args->geo;
57
58 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
59 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
60 ASSERT(dp->i_df.if_u1.if_data != NULL);
61
62 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
63
64 /*
65 * If the block number in the offset is out of range, we're done.
66 */
67 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
68 return 0;
69
70 /*
71 * Precalculate offsets for . and .. as we will always need them.
72 *
73 * XXX(hch): the second argument is sometimes 0 and sometimes
74 * geo->datablk
75 */
76 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
77 dp->d_ops->data_dot_offset);
78 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
79 dp->d_ops->data_dotdot_offset);
80
81 /*
82 * Put . entry unless we're starting past it.
83 */
84 if (ctx->pos <= dot_offset) {
85 ctx->pos = dot_offset & 0x7fffffff;
86 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
87 return 0;
88 }
89
90 /*
91 * Put .. entry unless we're starting past it.
92 */
93 if (ctx->pos <= dotdot_offset) {
94 ino = dp->d_ops->sf_get_parent_ino(sfp);
95 ctx->pos = dotdot_offset & 0x7fffffff;
96 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
97 return 0;
98 }
99
100 /*
101 * Loop while there are more entries and put'ing works.
102 */
103 sfep = xfs_dir2_sf_firstentry(sfp);
104 for (i = 0; i < sfp->count; i++) {
105 uint8_t filetype;
106
107 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
108 xfs_dir2_sf_get_offset(sfep));
109
110 if (ctx->pos > off) {
111 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
112 continue;
113 }
114
115 ino = dp->d_ops->sf_get_ino(sfp, sfep);
116 filetype = dp->d_ops->sf_get_ftype(sfep);
117 ctx->pos = off & 0x7fffffff;
118 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
119 xfs_dir3_get_dtype(dp->i_mount, filetype)))
120 return 0;
121 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
122 }
123
124 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
125 0x7fffffff;
126 return 0;
127}
128
129/*
130 * Readdir for block directories.
131 */
132STATIC int
133xfs_dir2_block_getdents(
134 struct xfs_da_args *args,
135 struct dir_context *ctx)
136{
137 struct xfs_inode *dp = args->dp; /* incore directory inode */
138 xfs_dir2_data_hdr_t *hdr; /* block header */
139 struct xfs_buf *bp; /* buffer for block */
140 xfs_dir2_data_entry_t *dep; /* block data entry */
141 xfs_dir2_data_unused_t *dup; /* block unused entry */
142 char *endptr; /* end of the data entries */
143 int error; /* error return value */
144 char *ptr; /* current data entry */
145 int wantoff; /* starting block offset */
146 xfs_off_t cook;
147 struct xfs_da_geometry *geo = args->geo;
148 int lock_mode;
149
150 /*
151 * If the block number in the offset is out of range, we're done.
152 */
153 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
154 return 0;
155
156 lock_mode = xfs_ilock_data_map_shared(dp);
157 error = xfs_dir3_block_read(args->trans, dp, &bp);
158 xfs_iunlock(dp, lock_mode);
159 if (error)
160 return error;
161
162 /*
163 * Extract the byte offset we start at from the seek pointer.
164 * We'll skip entries before this.
165 */
166 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
167 hdr = bp->b_addr;
168 xfs_dir3_data_check(dp, bp);
169 /*
170 * Set up values for the loop.
171 */
172 ptr = (char *)dp->d_ops->data_entry_p(hdr);
173 endptr = xfs_dir3_data_endp(geo, hdr);
174
175 /*
176 * Loop over the data portion of the block.
177 * Each object is a real entry (dep) or an unused one (dup).
178 */
179 while (ptr < endptr) {
180 uint8_t filetype;
181
182 dup = (xfs_dir2_data_unused_t *)ptr;
183 /*
184 * Unused, skip it.
185 */
186 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
187 ptr += be16_to_cpu(dup->length);
188 continue;
189 }
190
191 dep = (xfs_dir2_data_entry_t *)ptr;
192
193 /*
194 * Bump pointer for the next iteration.
195 */
196 ptr += dp->d_ops->data_entsize(dep->namelen);
197 /*
198 * The entry is before the desired starting point, skip it.
199 */
200 if ((char *)dep - (char *)hdr < wantoff)
201 continue;
202
203 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
204 (char *)dep - (char *)hdr);
205
206 ctx->pos = cook & 0x7fffffff;
207 filetype = dp->d_ops->data_get_ftype(dep);
208 /*
209 * If it didn't fit, set the final offset to here & return.
210 */
211 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
212 be64_to_cpu(dep->inumber),
213 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
214 xfs_trans_brelse(args->trans, bp);
215 return 0;
216 }
217 }
218
219 /*
220 * Reached the end of the block.
221 * Set the offset to a non-existent block 1 and return.
222 */
223 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
224 0x7fffffff;
225 xfs_trans_brelse(args->trans, bp);
226 return 0;
227}
228
229/*
230 * Read a directory block and initiate readahead for blocks beyond that.
231 * We maintain a sliding readahead window of the remaining space in the
232 * buffer rounded up to the nearest block.
233 */
234STATIC int
235xfs_dir2_leaf_readbuf(
236 struct xfs_da_args *args,
237 size_t bufsize,
238 xfs_dir2_off_t *cur_off,
239 xfs_dablk_t *ra_blk,
240 struct xfs_buf **bpp)
241{
242 struct xfs_inode *dp = args->dp;
243 struct xfs_buf *bp = NULL;
244 struct xfs_da_geometry *geo = args->geo;
245 struct xfs_ifork *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
246 struct xfs_bmbt_irec map;
247 struct blk_plug plug;
248 xfs_dir2_off_t new_off;
249 xfs_dablk_t next_ra;
250 xfs_dablk_t map_off;
251 xfs_dablk_t last_da;
252 struct xfs_iext_cursor icur;
253 int ra_want;
254 int error = 0;
255
256 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
257 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
258 if (error)
259 goto out;
260 }
261
262 /*
263 * Look for mapped directory blocks at or above the current offset.
264 * Truncate down to the nearest directory block to start the scanning
265 * operation.
266 */
267 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
268 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
269 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
270 goto out;
271 if (map.br_startoff >= last_da)
272 goto out;
273 xfs_trim_extent(&map, map_off, last_da - map_off);
274
275 /* Read the directory block of that first mapping. */
276 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
277 if (new_off > *cur_off)
278 *cur_off = new_off;
279 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, -1, &bp);
280 if (error)
281 goto out;
282
283 /*
284 * Start readahead for the next bufsize's worth of dir data blocks.
285 * We may have already issued readahead for some of that range;
286 * ra_blk tracks the last block we tried to read(ahead).
287 */
288 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
289 if (*ra_blk >= last_da)
290 goto out;
291 else if (*ra_blk == 0)
292 *ra_blk = map.br_startoff;
293 next_ra = map.br_startoff + geo->fsbcount;
294 if (next_ra >= last_da)
295 goto out_no_ra;
296 if (map.br_blockcount < geo->fsbcount &&
297 !xfs_iext_next_extent(ifp, &icur, &map))
298 goto out_no_ra;
299 if (map.br_startoff >= last_da)
300 goto out_no_ra;
301 xfs_trim_extent(&map, next_ra, last_da - next_ra);
302
303 /* Start ra for each dir (not fs) block that has a mapping. */
304 blk_start_plug(&plug);
305 while (ra_want > 0) {
306 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
307 while (ra_want > 0 &&
308 next_ra < map.br_startoff + map.br_blockcount) {
309 if (next_ra >= last_da) {
310 *ra_blk = last_da;
311 break;
312 }
313 if (next_ra > *ra_blk) {
314 xfs_dir3_data_readahead(dp, next_ra, -2);
315 *ra_blk = next_ra;
316 }
317 ra_want -= geo->fsbcount;
318 next_ra += geo->fsbcount;
319 }
320 if (!xfs_iext_next_extent(ifp, &icur, &map)) {
321 *ra_blk = last_da;
322 break;
323 }
324 }
325 blk_finish_plug(&plug);
326
327out:
328 *bpp = bp;
329 return error;
330out_no_ra:
331 *ra_blk = last_da;
332 goto out;
333}
334
335/*
336 * Getdents (readdir) for leaf and node directories.
337 * This reads the data blocks only, so is the same for both forms.
338 */
339STATIC int
340xfs_dir2_leaf_getdents(
341 struct xfs_da_args *args,
342 struct dir_context *ctx,
343 size_t bufsize)
344{
345 struct xfs_inode *dp = args->dp;
346 struct xfs_buf *bp = NULL; /* data block buffer */
347 xfs_dir2_data_hdr_t *hdr; /* data block header */
348 xfs_dir2_data_entry_t *dep; /* data entry */
349 xfs_dir2_data_unused_t *dup; /* unused entry */
350 char *ptr = NULL; /* pointer to current data */
351 struct xfs_da_geometry *geo = args->geo;
352 xfs_dablk_t rablk = 0; /* current readahead block */
353 xfs_dir2_off_t curoff; /* current overall offset */
354 int length; /* temporary length value */
355 int byteoff; /* offset in current block */
356 int lock_mode;
357 int error = 0; /* error return value */
358
359 /*
360 * If the offset is at or past the largest allowed value,
361 * give up right away.
362 */
363 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
364 return 0;
365
366 /*
367 * Inside the loop we keep the main offset value as a byte offset
368 * in the directory file.
369 */
370 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
371
372 /*
373 * Loop over directory entries until we reach the end offset.
374 * Get more blocks and readahead as necessary.
375 */
376 while (curoff < XFS_DIR2_LEAF_OFFSET) {
377 uint8_t filetype;
378
379 /*
380 * If we have no buffer, or we're off the end of the
381 * current buffer, need to get another one.
382 */
383 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
384 if (bp) {
385 xfs_trans_brelse(args->trans, bp);
386 bp = NULL;
387 }
388
389 lock_mode = xfs_ilock_data_map_shared(dp);
390 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
391 &rablk, &bp);
392 xfs_iunlock(dp, lock_mode);
393 if (error || !bp)
394 break;
395
396 hdr = bp->b_addr;
397 xfs_dir3_data_check(dp, bp);
398 /*
399 * Find our position in the block.
400 */
401 ptr = (char *)dp->d_ops->data_entry_p(hdr);
402 byteoff = xfs_dir2_byte_to_off(geo, curoff);
403 /*
404 * Skip past the header.
405 */
406 if (byteoff == 0)
407 curoff += dp->d_ops->data_entry_offset;
408 /*
409 * Skip past entries until we reach our offset.
410 */
411 else {
412 while ((char *)ptr - (char *)hdr < byteoff) {
413 dup = (xfs_dir2_data_unused_t *)ptr;
414
415 if (be16_to_cpu(dup->freetag)
416 == XFS_DIR2_DATA_FREE_TAG) {
417
418 length = be16_to_cpu(dup->length);
419 ptr += length;
420 continue;
421 }
422 dep = (xfs_dir2_data_entry_t *)ptr;
423 length =
424 dp->d_ops->data_entsize(dep->namelen);
425 ptr += length;
426 }
427 /*
428 * Now set our real offset.
429 */
430 curoff =
431 xfs_dir2_db_off_to_byte(geo,
432 xfs_dir2_byte_to_db(geo, curoff),
433 (char *)ptr - (char *)hdr);
434 if (ptr >= (char *)hdr + geo->blksize) {
435 continue;
436 }
437 }
438 }
439 /*
440 * We have a pointer to an entry.
441 * Is it a live one?
442 */
443 dup = (xfs_dir2_data_unused_t *)ptr;
444 /*
445 * No, it's unused, skip over it.
446 */
447 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
448 length = be16_to_cpu(dup->length);
449 ptr += length;
450 curoff += length;
451 continue;
452 }
453
454 dep = (xfs_dir2_data_entry_t *)ptr;
455 length = dp->d_ops->data_entsize(dep->namelen);
456 filetype = dp->d_ops->data_get_ftype(dep);
457
458 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
459 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
460 be64_to_cpu(dep->inumber),
461 xfs_dir3_get_dtype(dp->i_mount, filetype)))
462 break;
463
464 /*
465 * Advance to next entry in the block.
466 */
467 ptr += length;
468 curoff += length;
469 /* bufsize may have just been a guess; don't go negative */
470 bufsize = bufsize > length ? bufsize - length : 0;
471 }
472
473 /*
474 * All done. Set output offset value to current offset.
475 */
476 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
477 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
478 else
479 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
480 if (bp)
481 xfs_trans_brelse(args->trans, bp);
482 return error;
483}
484
485/*
486 * Read a directory.
487 *
488 * If supplied, the transaction collects locked dir buffers to avoid
489 * nested buffer deadlocks. This function does not dirty the
490 * transaction. The caller should ensure that the inode is locked
491 * before calling this function.
492 */
493int
494xfs_readdir(
495 struct xfs_trans *tp,
496 struct xfs_inode *dp,
497 struct dir_context *ctx,
498 size_t bufsize)
499{
500 struct xfs_da_args args = { NULL };
501 int rval;
502 int v;
503
504 trace_xfs_readdir(dp);
505
506 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
507 return -EIO;
508
509 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
510 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
511
512 args.dp = dp;
513 args.geo = dp->i_mount->m_dir_geo;
514 args.trans = tp;
515
516 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
517 rval = xfs_dir2_sf_getdents(&args, ctx);
518 else if ((rval = xfs_dir2_isblock(&args, &v)))
519 ;
520 else if (v)
521 rval = xfs_dir2_block_getdents(&args, ctx);
522 else
523 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
524
525 return rval;
526}