2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
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
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_inode.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.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,
49 if (!xfs_sb_version_hasftype(&mp->m_sb))
52 if (filetype >= XFS_DIR3_FT_MAX)
55 return xfs_dir3_filetype_table[filetype];
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;
71 struct xfs_da_geometry *geo = args->geo;
73 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
75 * Give up if the directory is way too short.
77 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
78 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
82 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
83 ASSERT(dp->i_df.if_u1.if_data != NULL);
85 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
87 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
90 * If the block number in the offset is out of range, we're done.
92 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
96 * Precalculate offsets for . and .. as we will always need them.
98 * XXX(hch): the second argument is sometimes 0 and sometimes
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);
107 * Put . entry unless we're starting past it.
109 if (ctx->pos <= dot_offset) {
110 ctx->pos = dot_offset & 0x7fffffff;
111 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
116 * Put .. entry unless we're starting past it.
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))
126 * Loop while there are more entries and put'ing works.
128 sfep = xfs_dir2_sf_firstentry(sfp);
129 for (i = 0; i < sfp->count; i++) {
132 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
133 xfs_dir2_sf_get_offset(sfep));
135 if (ctx->pos > off) {
136 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
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)))
146 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
149 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
155 * Readdir for block directories.
158 xfs_dir2_block_getdents(
159 struct xfs_da_args *args,
160 struct dir_context *ctx)
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 */
173 struct xfs_da_geometry *geo = args->geo;
177 * If the block number in the offset is out of range, we're done.
179 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
182 lock_mode = xfs_ilock_data_map_shared(dp);
183 error = xfs_dir3_block_read(NULL, dp, &bp);
184 xfs_iunlock(dp, lock_mode);
189 * Extract the byte offset we start at from the seek pointer.
190 * We'll skip entries before this.
192 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
194 xfs_dir3_data_check(dp, bp);
196 * Set up values for the loop.
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);
203 * Loop over the data portion of the block.
204 * Each object is a real entry (dep) or an unused one (dup).
206 while (ptr < endptr) {
209 dup = (xfs_dir2_data_unused_t *)ptr;
213 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
214 ptr += be16_to_cpu(dup->length);
218 dep = (xfs_dir2_data_entry_t *)ptr;
221 * Bump pointer for the next iteration.
223 ptr += dp->d_ops->data_entsize(dep->namelen);
225 * The entry is before the desired starting point, skip it.
227 if ((char *)dep - (char *)hdr < wantoff)
230 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
231 (char *)dep - (char *)hdr);
233 ctx->pos = cook & 0x7fffffff;
234 filetype = dp->d_ops->data_get_ftype(dep);
236 * If it didn't fit, set the final offset to here & return.
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);
247 * Reached the end of the block.
248 * Set the offset to a non-existent block 1 and return.
250 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
252 xfs_trans_brelse(NULL, bp);
256 struct 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 */
271 xfs_dir2_leaf_readbuf(
272 struct xfs_da_args *args,
274 struct xfs_dir2_leaf_map_info *mip,
275 xfs_dir2_off_t *curoff,
276 struct xfs_buf **bpp)
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;
286 struct xfs_da_geometry *geo = args->geo;
289 * If we have a buffer, we need to release it and
290 * take it out of the mapping.
294 xfs_trans_brelse(NULL, bp);
296 mip->map_blocks -= geo->fsbcount;
298 * Loop to get rid of the extents for the
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;
307 * If mapping is done, pitch it from
310 if (!map->br_blockcount && --mip->map_valid)
311 memmove(&map[0], &map[1],
312 sizeof(map[0]) * mip->map_valid);
318 * Recalculate the readahead blocks wanted.
320 mip->ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog)) - 1;
321 ASSERT(mip->ra_want >= 0);
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.
327 if (1 + mip->ra_want > mip->map_blocks &&
328 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
330 * Get more bmaps, fill in after the ones
331 * we already have in the table.
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) -
337 &map[mip->map_valid], &mip->nmap, 0);
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.
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.
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;
356 mip->map_off = xfs_dir2_byte_to_da(geo,
357 XFS_DIR2_LEAF_OFFSET);
360 * Look for holes in the mapping, and eliminate them. Count up
363 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
364 if (map[i].br_startblock == HOLESTARTBLOCK) {
366 length = mip->map_valid + mip->nmap - i;
368 memmove(&map[i], &map[i + 1],
369 sizeof(map[i]) * length);
371 mip->map_blocks += map[i].br_blockcount;
375 mip->map_valid += mip->nmap;
379 * No valid mappings, so no more data blocks.
381 if (!mip->map_valid) {
382 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
387 * Read the directory block starting at the first mapping.
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) :
395 * Should just skip over the data block instead of giving up.
401 * Adjust the current amount of read-ahead: we just read a block that
405 mip->ra_current -= geo->fsbcount;
408 * Do we need more readahead?
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);
416 * Read-ahead a contiguous directory block.
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 +
429 * Read-ahead a non-contiguous directory block. This doesn't
430 * use our mapping, but this is a very rare case.
432 else if (i > mip->ra_current) {
433 xfs_dir3_data_readahead(dp,
434 map[mip->ra_index].br_startoff +
440 * Advance offset through the mapping table.
442 for (j = 0; j < geo->fsbcount; j += length ) {
444 * The rest of this extent but not more than a dir
447 length = min_t(int, geo->fsbcount,
448 map[mip->ra_index].br_blockcount -
450 mip->ra_offset += length;
453 * Advance to the next mapping if this one is used up.
455 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
461 blk_finish_plug(&plug);
469 * Getdents (readdir) for leaf and node directories.
470 * This reads the data blocks only, so is the same for both forms.
473 xfs_dir2_leaf_getdents(
474 struct xfs_da_args *args,
475 struct dir_context *ctx,
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;
493 * If the offset is at or past the largest allowed value,
494 * give up right away.
496 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
500 * Set up to bmap a number of blocks based on the caller's
501 * buffer size, the directory block size, and the filesystem
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)),
508 map_info->map_size = length;
511 * Inside the loop we keep the main offset value as a byte offset
512 * in the directory file.
514 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
517 * Force this conversion through db so we truncate the offset
518 * down to get the start of the data block.
520 map_info->map_off = xfs_dir2_db_to_da(geo,
521 xfs_dir2_byte_to_db(geo, curoff));
524 * Loop over directory entries until we reach the end offset.
525 * Get more blocks and readahead as necessary.
527 while (curoff < XFS_DIR2_LEAF_OFFSET) {
531 * If we have no buffer, or we're off the end of the
532 * current buffer, need to get another one.
534 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
537 lock_mode = xfs_ilock_data_map_shared(dp);
538 error = xfs_dir2_leaf_readbuf(args, bufsize, map_info,
540 xfs_iunlock(dp, lock_mode);
541 if (error || !map_info->map_valid)
545 * Having done a read, we need to set a new offset.
547 newoff = xfs_dir2_db_off_to_byte(geo,
550 * Start of the current block.
555 * Make sure we're in the right block.
557 else if (curoff > newoff)
558 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
561 xfs_dir3_data_check(dp, bp);
563 * Find our position in the block.
565 ptr = (char *)dp->d_ops->data_entry_p(hdr);
566 byteoff = xfs_dir2_byte_to_off(geo, curoff);
568 * Skip past the header.
571 curoff += dp->d_ops->data_entry_offset;
573 * Skip past entries until we reach our offset.
576 while ((char *)ptr - (char *)hdr < byteoff) {
577 dup = (xfs_dir2_data_unused_t *)ptr;
579 if (be16_to_cpu(dup->freetag)
580 == XFS_DIR2_DATA_FREE_TAG) {
582 length = be16_to_cpu(dup->length);
586 dep = (xfs_dir2_data_entry_t *)ptr;
588 dp->d_ops->data_entsize(dep->namelen);
592 * Now set our real offset.
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) {
604 * We have a pointer to an entry.
607 dup = (xfs_dir2_data_unused_t *)ptr;
609 * No, it's unused, skip over it.
611 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
612 length = be16_to_cpu(dup->length);
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);
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)))
629 * Advance to next entry in the block.
633 /* bufsize may have just been a guess; don't go negative */
634 bufsize = bufsize > length ? bufsize - length : 0;
638 * All done. Set output offset value to current offset.
640 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
641 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
643 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
646 xfs_trans_brelse(NULL, bp);
655 struct xfs_inode *dp,
656 struct dir_context *ctx,
659 struct xfs_da_args args = { NULL };
663 trace_xfs_readdir(dp);
665 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
668 ASSERT(S_ISDIR(dp->i_d.di_mode));
669 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
672 args.geo = dp->i_mount->m_dir_geo;
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)))
680 rval = xfs_dir2_block_getdents(&args, ctx);
682 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
683 xfs_iunlock(dp, XFS_IOLOCK_SHARED);