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"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
31 #include "xfs_dir2_priv.h"
32 #include "xfs_error.h"
33 #include "xfs_trace.h"
35 #include "xfs_trans.h"
38 * Directory file type support functions
40 static unsigned char xfs_dir3_filetype_table[] = {
41 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
42 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
50 if (!xfs_sb_version_hasftype(&mp->m_sb))
53 if (filetype >= XFS_DIR3_FT_MAX)
56 return xfs_dir3_filetype_table[filetype];
59 * @mode, if set, indicates that the type field needs to be set up.
60 * This uses the transformation from file mode to DT_* as defined in linux/fs.h
61 * for file type specification. This will be propagated into the directory
62 * structure if appropriate for the given operation and filesystem config.
64 const unsigned char xfs_mode_to_ftype[S_IFMT >> S_SHIFT] = {
65 [0] = XFS_DIR3_FT_UNKNOWN,
66 [S_IFREG >> S_SHIFT] = XFS_DIR3_FT_REG_FILE,
67 [S_IFDIR >> S_SHIFT] = XFS_DIR3_FT_DIR,
68 [S_IFCHR >> S_SHIFT] = XFS_DIR3_FT_CHRDEV,
69 [S_IFBLK >> S_SHIFT] = XFS_DIR3_FT_BLKDEV,
70 [S_IFIFO >> S_SHIFT] = XFS_DIR3_FT_FIFO,
71 [S_IFSOCK >> S_SHIFT] = XFS_DIR3_FT_SOCK,
72 [S_IFLNK >> S_SHIFT] = XFS_DIR3_FT_SYMLINK,
77 struct xfs_da_args *args,
78 struct dir_context *ctx)
80 int i; /* shortform entry number */
81 struct xfs_inode *dp = args->dp; /* incore directory inode */
82 xfs_dir2_dataptr_t off; /* current entry's offset */
83 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
84 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
85 xfs_dir2_dataptr_t dot_offset;
86 xfs_dir2_dataptr_t dotdot_offset;
88 struct xfs_da_geometry *geo = args->geo;
90 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
92 * Give up if the directory is way too short.
94 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
95 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
99 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
100 ASSERT(dp->i_df.if_u1.if_data != NULL);
102 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
104 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
107 * If the block number in the offset is out of range, we're done.
109 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
113 * Precalculate offsets for . and .. as we will always need them.
115 * XXX(hch): the second argument is sometimes 0 and sometimes
118 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
119 dp->d_ops->data_dot_offset);
120 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
121 dp->d_ops->data_dotdot_offset);
124 * Put . entry unless we're starting past it.
126 if (ctx->pos <= dot_offset) {
127 ctx->pos = dot_offset & 0x7fffffff;
128 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
133 * Put .. entry unless we're starting past it.
135 if (ctx->pos <= dotdot_offset) {
136 ino = dp->d_ops->sf_get_parent_ino(sfp);
137 ctx->pos = dotdot_offset & 0x7fffffff;
138 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
143 * Loop while there are more entries and put'ing works.
145 sfep = xfs_dir2_sf_firstentry(sfp);
146 for (i = 0; i < sfp->count; i++) {
149 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
150 xfs_dir2_sf_get_offset(sfep));
152 if (ctx->pos > off) {
153 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
157 ino = dp->d_ops->sf_get_ino(sfp, sfep);
158 filetype = dp->d_ops->sf_get_ftype(sfep);
159 ctx->pos = off & 0x7fffffff;
160 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
161 xfs_dir3_get_dtype(dp->i_mount, filetype)))
163 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
166 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
172 * Readdir for block directories.
175 xfs_dir2_block_getdents(
176 struct xfs_da_args *args,
177 struct dir_context *ctx)
179 struct xfs_inode *dp = args->dp; /* incore directory inode */
180 xfs_dir2_data_hdr_t *hdr; /* block header */
181 struct xfs_buf *bp; /* buffer for block */
182 xfs_dir2_block_tail_t *btp; /* block tail */
183 xfs_dir2_data_entry_t *dep; /* block data entry */
184 xfs_dir2_data_unused_t *dup; /* block unused entry */
185 char *endptr; /* end of the data entries */
186 int error; /* error return value */
187 char *ptr; /* current data entry */
188 int wantoff; /* starting block offset */
190 struct xfs_da_geometry *geo = args->geo;
193 * If the block number in the offset is out of range, we're done.
195 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
198 error = xfs_dir3_block_read(NULL, dp, &bp);
203 * Extract the byte offset we start at from the seek pointer.
204 * We'll skip entries before this.
206 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
208 xfs_dir3_data_check(dp, bp);
210 * Set up values for the loop.
212 btp = xfs_dir2_block_tail_p(geo, hdr);
213 ptr = (char *)dp->d_ops->data_entry_p(hdr);
214 endptr = (char *)xfs_dir2_block_leaf_p(btp);
217 * Loop over the data portion of the block.
218 * Each object is a real entry (dep) or an unused one (dup).
220 while (ptr < endptr) {
223 dup = (xfs_dir2_data_unused_t *)ptr;
227 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
228 ptr += be16_to_cpu(dup->length);
232 dep = (xfs_dir2_data_entry_t *)ptr;
235 * Bump pointer for the next iteration.
237 ptr += dp->d_ops->data_entsize(dep->namelen);
239 * The entry is before the desired starting point, skip it.
241 if ((char *)dep - (char *)hdr < wantoff)
244 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
245 (char *)dep - (char *)hdr);
247 ctx->pos = cook & 0x7fffffff;
248 filetype = dp->d_ops->data_get_ftype(dep);
250 * If it didn't fit, set the final offset to here & return.
252 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
253 be64_to_cpu(dep->inumber),
254 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
255 xfs_trans_brelse(NULL, bp);
261 * Reached the end of the block.
262 * Set the offset to a non-existent block 1 and return.
264 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
266 xfs_trans_brelse(NULL, bp);
270 struct xfs_dir2_leaf_map_info {
271 xfs_extlen_t map_blocks; /* number of fsbs in map */
272 xfs_dablk_t map_off; /* last mapped file offset */
273 int map_size; /* total entries in *map */
274 int map_valid; /* valid entries in *map */
275 int nmap; /* mappings to ask xfs_bmapi */
276 xfs_dir2_db_t curdb; /* db for current block */
277 int ra_current; /* number of read-ahead blks */
278 int ra_index; /* *map index for read-ahead */
279 int ra_offset; /* map entry offset for ra */
280 int ra_want; /* readahead count wanted */
281 struct xfs_bmbt_irec map[]; /* map vector for blocks */
285 xfs_dir2_leaf_readbuf(
286 struct xfs_da_args *args,
288 struct xfs_dir2_leaf_map_info *mip,
289 xfs_dir2_off_t *curoff,
290 struct xfs_buf **bpp)
292 struct xfs_inode *dp = args->dp;
293 struct xfs_buf *bp = *bpp;
294 struct xfs_bmbt_irec *map = mip->map;
295 struct blk_plug plug;
300 struct xfs_da_geometry *geo = args->geo;
303 * If we have a buffer, we need to release it and
304 * take it out of the mapping.
308 xfs_trans_brelse(NULL, bp);
310 mip->map_blocks -= geo->fsbcount;
312 * Loop to get rid of the extents for the
315 for (i = geo->fsbcount; i > 0; ) {
316 j = min_t(int, map->br_blockcount, i);
317 map->br_blockcount -= j;
318 map->br_startblock += j;
319 map->br_startoff += j;
321 * If mapping is done, pitch it from
324 if (!map->br_blockcount && --mip->map_valid)
325 memmove(&map[0], &map[1],
326 sizeof(map[0]) * mip->map_valid);
332 * Recalculate the readahead blocks wanted.
334 mip->ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog)) - 1;
335 ASSERT(mip->ra_want >= 0);
338 * If we don't have as many as we want, and we haven't
339 * run out of data blocks, get some more mappings.
341 if (1 + mip->ra_want > mip->map_blocks &&
342 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
344 * Get more bmaps, fill in after the ones
345 * we already have in the table.
347 mip->nmap = mip->map_size - mip->map_valid;
348 error = xfs_bmapi_read(dp, mip->map_off,
349 xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET) -
351 &map[mip->map_valid], &mip->nmap, 0);
354 * Don't know if we should ignore this or try to return an
355 * error. The trouble with returning errors is that readdir
356 * will just stop without actually passing the error through.
362 * If we got all the mappings we asked for, set the final map
363 * offset based on the last bmap value received. Otherwise,
364 * we've reached the end.
366 if (mip->nmap == mip->map_size - mip->map_valid) {
367 i = mip->map_valid + mip->nmap - 1;
368 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
370 mip->map_off = xfs_dir2_byte_to_da(geo,
371 XFS_DIR2_LEAF_OFFSET);
374 * Look for holes in the mapping, and eliminate them. Count up
377 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
378 if (map[i].br_startblock == HOLESTARTBLOCK) {
380 length = mip->map_valid + mip->nmap - i;
382 memmove(&map[i], &map[i + 1],
383 sizeof(map[i]) * length);
385 mip->map_blocks += map[i].br_blockcount;
389 mip->map_valid += mip->nmap;
393 * No valid mappings, so no more data blocks.
395 if (!mip->map_valid) {
396 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
401 * Read the directory block starting at the first mapping.
403 mip->curdb = xfs_dir2_da_to_db(geo, map->br_startoff);
404 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
405 map->br_blockcount >= geo->fsbcount ?
406 XFS_FSB_TO_DADDR(dp->i_mount, map->br_startblock) :
409 * Should just skip over the data block instead of giving up.
415 * Adjust the current amount of read-ahead: we just read a block that
419 mip->ra_current -= geo->fsbcount;
422 * Do we need more readahead?
424 blk_start_plug(&plug);
425 for (mip->ra_index = mip->ra_offset = i = 0;
426 mip->ra_want > mip->ra_current && i < mip->map_blocks;
427 i += geo->fsbcount) {
428 ASSERT(mip->ra_index < mip->map_valid);
430 * Read-ahead a contiguous directory block.
432 if (i > mip->ra_current &&
433 map[mip->ra_index].br_blockcount >= geo->fsbcount) {
434 xfs_dir3_data_readahead(dp,
435 map[mip->ra_index].br_startoff + mip->ra_offset,
436 XFS_FSB_TO_DADDR(dp->i_mount,
437 map[mip->ra_index].br_startblock +
443 * Read-ahead a non-contiguous directory block. This doesn't
444 * use our mapping, but this is a very rare case.
446 else if (i > mip->ra_current) {
447 xfs_dir3_data_readahead(dp,
448 map[mip->ra_index].br_startoff +
454 * Advance offset through the mapping table.
456 for (j = 0; j < geo->fsbcount; j += length ) {
458 * The rest of this extent but not more than a dir
461 length = min_t(int, geo->fsbcount,
462 map[mip->ra_index].br_blockcount -
464 mip->ra_offset += length;
467 * Advance to the next mapping if this one is used up.
469 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
475 blk_finish_plug(&plug);
483 * Getdents (readdir) for leaf and node directories.
484 * This reads the data blocks only, so is the same for both forms.
487 xfs_dir2_leaf_getdents(
488 struct xfs_da_args *args,
489 struct dir_context *ctx,
492 struct xfs_inode *dp = args->dp;
493 struct xfs_buf *bp = NULL; /* data block buffer */
494 xfs_dir2_data_hdr_t *hdr; /* data block header */
495 xfs_dir2_data_entry_t *dep; /* data entry */
496 xfs_dir2_data_unused_t *dup; /* unused entry */
497 int error = 0; /* error return value */
498 int length; /* temporary length value */
499 int byteoff; /* offset in current block */
500 xfs_dir2_off_t curoff; /* current overall offset */
501 xfs_dir2_off_t newoff; /* new curoff after new blk */
502 char *ptr = NULL; /* pointer to current data */
503 struct xfs_dir2_leaf_map_info *map_info;
504 struct xfs_da_geometry *geo = args->geo;
507 * If the offset is at or past the largest allowed value,
508 * give up right away.
510 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
514 * Set up to bmap a number of blocks based on the caller's
515 * buffer size, the directory block size, and the filesystem
518 length = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
519 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
520 (length * sizeof(struct xfs_bmbt_irec)),
522 map_info->map_size = length;
525 * Inside the loop we keep the main offset value as a byte offset
526 * in the directory file.
528 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
531 * Force this conversion through db so we truncate the offset
532 * down to get the start of the data block.
534 map_info->map_off = xfs_dir2_db_to_da(geo,
535 xfs_dir2_byte_to_db(geo, curoff));
538 * Loop over directory entries until we reach the end offset.
539 * Get more blocks and readahead as necessary.
541 while (curoff < XFS_DIR2_LEAF_OFFSET) {
545 * If we have no buffer, or we're off the end of the
546 * current buffer, need to get another one.
548 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
550 error = xfs_dir2_leaf_readbuf(args, bufsize, map_info,
552 if (error || !map_info->map_valid)
556 * Having done a read, we need to set a new offset.
558 newoff = xfs_dir2_db_off_to_byte(geo,
561 * Start of the current block.
566 * Make sure we're in the right block.
568 else if (curoff > newoff)
569 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
572 xfs_dir3_data_check(dp, bp);
574 * Find our position in the block.
576 ptr = (char *)dp->d_ops->data_entry_p(hdr);
577 byteoff = xfs_dir2_byte_to_off(geo, curoff);
579 * Skip past the header.
582 curoff += dp->d_ops->data_entry_offset;
584 * Skip past entries until we reach our offset.
587 while ((char *)ptr - (char *)hdr < byteoff) {
588 dup = (xfs_dir2_data_unused_t *)ptr;
590 if (be16_to_cpu(dup->freetag)
591 == XFS_DIR2_DATA_FREE_TAG) {
593 length = be16_to_cpu(dup->length);
597 dep = (xfs_dir2_data_entry_t *)ptr;
599 dp->d_ops->data_entsize(dep->namelen);
603 * Now set our real offset.
606 xfs_dir2_db_off_to_byte(geo,
607 xfs_dir2_byte_to_db(geo, curoff),
608 (char *)ptr - (char *)hdr);
609 if (ptr >= (char *)hdr + geo->blksize) {
615 * We have a pointer to an entry.
618 dup = (xfs_dir2_data_unused_t *)ptr;
620 * No, it's unused, skip over it.
622 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
623 length = be16_to_cpu(dup->length);
629 dep = (xfs_dir2_data_entry_t *)ptr;
630 length = dp->d_ops->data_entsize(dep->namelen);
631 filetype = dp->d_ops->data_get_ftype(dep);
633 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
634 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
635 be64_to_cpu(dep->inumber),
636 xfs_dir3_get_dtype(dp->i_mount, filetype)))
640 * Advance to next entry in the block.
644 /* bufsize may have just been a guess; don't go negative */
645 bufsize = bufsize > length ? bufsize - length : 0;
649 * All done. Set output offset value to current offset.
651 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
652 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
654 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
657 xfs_trans_brelse(NULL, bp);
666 struct xfs_inode *dp,
667 struct dir_context *ctx,
670 struct xfs_da_args args = { NULL };
675 trace_xfs_readdir(dp);
677 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
680 ASSERT(S_ISDIR(dp->i_d.di_mode));
681 XFS_STATS_INC(xs_dir_getdents);
684 args.geo = dp->i_mount->m_dir_geo;
686 lock_mode = xfs_ilock_data_map_shared(dp);
687 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
688 rval = xfs_dir2_sf_getdents(&args, ctx);
689 else if ((rval = xfs_dir2_isblock(&args, &v)))
692 rval = xfs_dir2_block_getdents(&args, ctx);
694 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
695 xfs_iunlock(dp, lock_mode);