mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / xfs / xfs_dir2_readdir.c
blobba2638d370315cad9ed9a5256bc6867c0f44230d
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
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.
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.
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
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"
37 * Directory file type support functions
39 static 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,
44 static unsigned char
45 xfs_dir3_get_dtype(
46 struct xfs_mount *mp,
47 uint8_t filetype)
49 if (!xfs_sb_version_hasftype(&mp->m_sb))
50 return DT_UNKNOWN;
52 if (filetype >= XFS_DIR3_FT_MAX)
53 return DT_UNKNOWN;
55 return xfs_dir3_filetype_table[filetype];
58 STATIC int
59 xfs_dir2_sf_getdents(
60 struct xfs_da_args *args,
61 struct dir_context *ctx)
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;
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);
77 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
80 * If the block number in the offset is out of range, we're done.
82 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
83 return 0;
86 * Precalculate offsets for . and .. as we will always need them.
88 * XXX(hch): the second argument is sometimes 0 and sometimes
89 * geo->datablk
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);
97 * Put . entry unless we're starting past it.
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;
106 * Put .. entry unless we're starting past it.
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;
116 * Loop while there are more entries and put'ing works.
118 sfep = xfs_dir2_sf_firstentry(sfp);
119 for (i = 0; i < sfp->count; i++) {
120 uint8_t filetype;
122 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
123 xfs_dir2_sf_get_offset(sfep));
125 if (ctx->pos > off) {
126 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
127 continue;
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);
139 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
140 0x7fffffff;
141 return 0;
145 * Readdir for block directories.
147 STATIC int
148 xfs_dir2_block_getdents(
149 struct xfs_da_args *args,
150 struct dir_context *ctx)
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_block_tail_t *btp; /* block tail */
156 xfs_dir2_data_entry_t *dep; /* block data entry */
157 xfs_dir2_data_unused_t *dup; /* block unused entry */
158 char *endptr; /* end of the data entries */
159 int error; /* error return value */
160 char *ptr; /* current data entry */
161 int wantoff; /* starting block offset */
162 xfs_off_t cook;
163 struct xfs_da_geometry *geo = args->geo;
164 int lock_mode;
167 * If the block number in the offset is out of range, we're done.
169 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
170 return 0;
172 lock_mode = xfs_ilock_data_map_shared(dp);
173 error = xfs_dir3_block_read(args->trans, dp, &bp);
174 xfs_iunlock(dp, lock_mode);
175 if (error)
176 return error;
179 * Extract the byte offset we start at from the seek pointer.
180 * We'll skip entries before this.
182 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
183 hdr = bp->b_addr;
184 xfs_dir3_data_check(dp, bp);
186 * Set up values for the loop.
188 btp = xfs_dir2_block_tail_p(geo, hdr);
189 ptr = (char *)dp->d_ops->data_entry_p(hdr);
190 endptr = (char *)xfs_dir2_block_leaf_p(btp);
193 * Loop over the data portion of the block.
194 * Each object is a real entry (dep) or an unused one (dup).
196 while (ptr < endptr) {
197 uint8_t filetype;
199 dup = (xfs_dir2_data_unused_t *)ptr;
201 * Unused, skip it.
203 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
204 ptr += be16_to_cpu(dup->length);
205 continue;
208 dep = (xfs_dir2_data_entry_t *)ptr;
211 * Bump pointer for the next iteration.
213 ptr += dp->d_ops->data_entsize(dep->namelen);
215 * The entry is before the desired starting point, skip it.
217 if ((char *)dep - (char *)hdr < wantoff)
218 continue;
220 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
221 (char *)dep - (char *)hdr);
223 ctx->pos = cook & 0x7fffffff;
224 filetype = dp->d_ops->data_get_ftype(dep);
226 * If it didn't fit, set the final offset to here & return.
228 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
229 be64_to_cpu(dep->inumber),
230 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
231 xfs_trans_brelse(args->trans, bp);
232 return 0;
237 * Reached the end of the block.
238 * Set the offset to a non-existent block 1 and return.
240 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
241 0x7fffffff;
242 xfs_trans_brelse(args->trans, bp);
243 return 0;
247 * Read a directory block and initiate readahead for blocks beyond that.
248 * We maintain a sliding readahead window of the remaining space in the
249 * buffer rounded up to the nearest block.
251 STATIC int
252 xfs_dir2_leaf_readbuf(
253 struct xfs_da_args *args,
254 size_t bufsize,
255 xfs_dir2_off_t *cur_off,
256 xfs_dablk_t *ra_blk,
257 struct xfs_buf **bpp)
259 struct xfs_inode *dp = args->dp;
260 struct xfs_buf *bp = NULL;
261 struct xfs_da_geometry *geo = args->geo;
262 struct xfs_ifork *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
263 struct xfs_bmbt_irec map;
264 struct blk_plug plug;
265 xfs_dir2_off_t new_off;
266 xfs_dablk_t next_ra;
267 xfs_dablk_t map_off;
268 xfs_dablk_t last_da;
269 xfs_extnum_t idx;
270 int ra_want;
271 int error = 0;
273 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
274 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
275 if (error)
276 goto out;
280 * Look for mapped directory blocks at or above the current offset.
281 * Truncate down to the nearest directory block to start the scanning
282 * operation.
284 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
285 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
286 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &idx, &map))
287 goto out;
288 if (map.br_startoff >= last_da)
289 goto out;
290 xfs_trim_extent(&map, map_off, last_da - map_off);
292 /* Read the directory block of that first mapping. */
293 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
294 if (new_off > *cur_off)
295 *cur_off = new_off;
296 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, -1, &bp);
297 if (error)
298 goto out;
301 * Start readahead for the next bufsize's worth of dir data blocks.
302 * We may have already issued readahead for some of that range;
303 * ra_blk tracks the last block we tried to read(ahead).
305 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
306 if (*ra_blk >= last_da)
307 goto out;
308 else if (*ra_blk == 0)
309 *ra_blk = map.br_startoff;
310 next_ra = map.br_startoff + geo->fsbcount;
311 if (next_ra >= last_da)
312 goto out_no_ra;
313 if (map.br_blockcount < geo->fsbcount &&
314 !xfs_iext_get_extent(ifp, ++idx, &map))
315 goto out_no_ra;
316 if (map.br_startoff >= last_da)
317 goto out_no_ra;
318 xfs_trim_extent(&map, next_ra, last_da - next_ra);
320 /* Start ra for each dir (not fs) block that has a mapping. */
321 blk_start_plug(&plug);
322 while (ra_want > 0) {
323 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
324 while (ra_want > 0 &&
325 next_ra < map.br_startoff + map.br_blockcount) {
326 if (next_ra >= last_da) {
327 *ra_blk = last_da;
328 break;
330 if (next_ra > *ra_blk) {
331 xfs_dir3_data_readahead(dp, next_ra, -2);
332 *ra_blk = next_ra;
334 ra_want -= geo->fsbcount;
335 next_ra += geo->fsbcount;
337 if (!xfs_iext_get_extent(ifp, ++idx, &map)) {
338 *ra_blk = last_da;
339 break;
342 blk_finish_plug(&plug);
344 out:
345 *bpp = bp;
346 return error;
347 out_no_ra:
348 *ra_blk = last_da;
349 goto out;
353 * Getdents (readdir) for leaf and node directories.
354 * This reads the data blocks only, so is the same for both forms.
356 STATIC int
357 xfs_dir2_leaf_getdents(
358 struct xfs_da_args *args,
359 struct dir_context *ctx,
360 size_t bufsize)
362 struct xfs_inode *dp = args->dp;
363 struct xfs_buf *bp = NULL; /* data block buffer */
364 xfs_dir2_data_hdr_t *hdr; /* data block header */
365 xfs_dir2_data_entry_t *dep; /* data entry */
366 xfs_dir2_data_unused_t *dup; /* unused entry */
367 char *ptr = NULL; /* pointer to current data */
368 struct xfs_da_geometry *geo = args->geo;
369 xfs_dablk_t rablk = 0; /* current readahead block */
370 xfs_dir2_off_t curoff; /* current overall offset */
371 int length; /* temporary length value */
372 int byteoff; /* offset in current block */
373 int lock_mode;
374 int error = 0; /* error return value */
377 * If the offset is at or past the largest allowed value,
378 * give up right away.
380 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
381 return 0;
384 * Inside the loop we keep the main offset value as a byte offset
385 * in the directory file.
387 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
390 * Loop over directory entries until we reach the end offset.
391 * Get more blocks and readahead as necessary.
393 while (curoff < XFS_DIR2_LEAF_OFFSET) {
394 uint8_t filetype;
397 * If we have no buffer, or we're off the end of the
398 * current buffer, need to get another one.
400 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
401 if (bp) {
402 xfs_trans_brelse(args->trans, bp);
403 bp = NULL;
406 lock_mode = xfs_ilock_data_map_shared(dp);
407 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
408 &rablk, &bp);
409 xfs_iunlock(dp, lock_mode);
410 if (error || !bp)
411 break;
413 hdr = bp->b_addr;
414 xfs_dir3_data_check(dp, bp);
416 * Find our position in the block.
418 ptr = (char *)dp->d_ops->data_entry_p(hdr);
419 byteoff = xfs_dir2_byte_to_off(geo, curoff);
421 * Skip past the header.
423 if (byteoff == 0)
424 curoff += dp->d_ops->data_entry_offset;
426 * Skip past entries until we reach our offset.
428 else {
429 while ((char *)ptr - (char *)hdr < byteoff) {
430 dup = (xfs_dir2_data_unused_t *)ptr;
432 if (be16_to_cpu(dup->freetag)
433 == XFS_DIR2_DATA_FREE_TAG) {
435 length = be16_to_cpu(dup->length);
436 ptr += length;
437 continue;
439 dep = (xfs_dir2_data_entry_t *)ptr;
440 length =
441 dp->d_ops->data_entsize(dep->namelen);
442 ptr += length;
445 * Now set our real offset.
447 curoff =
448 xfs_dir2_db_off_to_byte(geo,
449 xfs_dir2_byte_to_db(geo, curoff),
450 (char *)ptr - (char *)hdr);
451 if (ptr >= (char *)hdr + geo->blksize) {
452 continue;
457 * We have a pointer to an entry.
458 * Is it a live one?
460 dup = (xfs_dir2_data_unused_t *)ptr;
462 * No, it's unused, skip over it.
464 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
465 length = be16_to_cpu(dup->length);
466 ptr += length;
467 curoff += length;
468 continue;
471 dep = (xfs_dir2_data_entry_t *)ptr;
472 length = dp->d_ops->data_entsize(dep->namelen);
473 filetype = dp->d_ops->data_get_ftype(dep);
475 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
476 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
477 be64_to_cpu(dep->inumber),
478 xfs_dir3_get_dtype(dp->i_mount, filetype)))
479 break;
482 * Advance to next entry in the block.
484 ptr += length;
485 curoff += length;
486 /* bufsize may have just been a guess; don't go negative */
487 bufsize = bufsize > length ? bufsize - length : 0;
491 * All done. Set output offset value to current offset.
493 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
494 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
495 else
496 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
497 if (bp)
498 xfs_trans_brelse(args->trans, bp);
499 return error;
503 * Read a directory.
505 * If supplied, the transaction collects locked dir buffers to avoid
506 * nested buffer deadlocks. This function does not dirty the
507 * transaction. The caller should ensure that the inode is locked
508 * before calling this function.
511 xfs_readdir(
512 struct xfs_trans *tp,
513 struct xfs_inode *dp,
514 struct dir_context *ctx,
515 size_t bufsize)
517 struct xfs_da_args args = { NULL };
518 int rval;
519 int v;
521 trace_xfs_readdir(dp);
523 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
524 return -EIO;
526 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
527 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
529 args.dp = dp;
530 args.geo = dp->i_mount->m_dir_geo;
531 args.trans = tp;
533 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
534 rval = xfs_dir2_sf_getdents(&args, ctx);
535 else if ((rval = xfs_dir2_isblock(&args, &v)))
537 else if (v)
538 rval = xfs_dir2_block_getdents(&args, ctx);
539 else
540 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
542 return rval;