1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
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"
16 #include "xfs_dir2_priv.h"
17 #include "xfs_trace.h"
19 #include "xfs_trans.h"
20 #include "xfs_error.h"
23 * Directory file type support functions
25 static 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,
35 if (!xfs_has_ftype(mp))
38 if (filetype >= XFS_DIR3_FT_MAX)
41 return xfs_dir3_filetype_table[filetype];
46 struct xfs_da_args *args,
47 struct dir_context *ctx)
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;
58 struct xfs_da_geometry *geo = args->geo;
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);
64 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
67 * If the block number in the offset is out of range, we're done.
69 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
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 "..".
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));
84 * Put . entry unless we're starting past it.
86 if (ctx->pos <= dot_offset) {
87 ctx->pos = dot_offset & 0x7fffffff;
88 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
93 * Put .. entry unless we're starting past it.
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))
103 * Loop while there are more entries and put'ing works.
105 sfep = xfs_dir2_sf_firstentry(sfp);
106 for (i = 0; i < sfp->count; i++) {
109 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
110 xfs_dir2_sf_get_offset(sfep));
112 if (ctx->pos > off) {
113 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
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,
123 return -EFSCORRUPTED;
124 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
125 xfs_dir3_get_dtype(mp, filetype)))
127 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
130 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
136 * Readdir for block directories.
139 xfs_dir2_block_getdents(
140 struct xfs_da_args *args,
141 struct dir_context *ctx,
142 unsigned int *lock_mode)
144 struct xfs_inode *dp = args->dp; /* incore directory inode */
145 struct xfs_buf *bp; /* buffer for block */
146 int error; /* error return value */
147 int wantoff; /* starting block offset */
149 struct xfs_da_geometry *geo = args->geo;
150 unsigned int offset, next_offset;
154 * If the block number in the offset is out of range, we're done.
156 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
159 error = xfs_dir3_block_read(args->trans, dp, &bp);
163 xfs_iunlock(dp, *lock_mode);
167 * Extract the byte offset we start at from the seek pointer.
168 * We'll skip entries before this.
170 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
171 xfs_dir3_data_check(dp, bp);
174 * Loop over the data portion of the block.
175 * Each object is a real entry (dep) or an unused one (dup).
177 end = xfs_dir3_data_end_offset(geo, bp->b_addr);
178 for (offset = geo->data_entry_offset;
180 offset = next_offset) {
181 struct xfs_dir2_data_unused *dup = bp->b_addr + offset;
182 struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
188 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
189 next_offset = offset + be16_to_cpu(dup->length);
194 * Bump pointer for the next iteration.
196 next_offset = offset +
197 xfs_dir2_data_entsize(dp->i_mount, dep->namelen);
200 * The entry is before the desired starting point, skip it.
202 if (offset < wantoff)
205 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, offset);
207 ctx->pos = cook & 0x7fffffff;
208 filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
210 * If it didn't fit, set the final offset to here & return.
212 if (XFS_IS_CORRUPT(dp->i_mount,
213 !xfs_dir2_namecheck(dep->name,
215 error = -EFSCORRUPTED;
218 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
219 be64_to_cpu(dep->inumber),
220 xfs_dir3_get_dtype(dp->i_mount, filetype)))
225 * Reached the end of the block.
226 * Set the offset to a non-existent block 1 and return.
228 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
231 xfs_trans_brelse(args->trans, bp);
236 * Read a directory block and initiate readahead for blocks beyond that.
237 * We maintain a sliding readahead window of the remaining space in the
238 * buffer rounded up to the nearest block.
241 xfs_dir2_leaf_readbuf(
242 struct xfs_da_args *args,
244 xfs_dir2_off_t *cur_off,
246 struct xfs_buf **bpp)
248 struct xfs_inode *dp = args->dp;
249 struct xfs_buf *bp = NULL;
250 struct xfs_da_geometry *geo = args->geo;
251 struct xfs_ifork *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
252 struct xfs_bmbt_irec map;
253 struct blk_plug plug;
254 xfs_dir2_off_t new_off;
258 struct xfs_iext_cursor icur;
262 error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
267 * Look for mapped directory blocks at or above the current offset.
268 * Truncate down to the nearest directory block to start the scanning
271 last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
272 map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
273 if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
275 if (map.br_startoff >= last_da)
277 xfs_trim_extent(&map, map_off, last_da - map_off);
279 /* Read the directory block of that first mapping. */
280 new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
281 if (new_off > *cur_off)
283 error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
288 * Start readahead for the next bufsize's worth of dir data blocks.
289 * We may have already issued readahead for some of that range;
290 * ra_blk tracks the last block we tried to read(ahead).
292 ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
293 if (*ra_blk >= last_da)
295 else if (*ra_blk == 0)
296 *ra_blk = map.br_startoff;
297 next_ra = map.br_startoff + geo->fsbcount;
298 if (next_ra >= last_da)
300 if (map.br_blockcount < geo->fsbcount &&
301 !xfs_iext_next_extent(ifp, &icur, &map))
303 if (map.br_startoff >= last_da)
305 xfs_trim_extent(&map, next_ra, last_da - next_ra);
307 /* Start ra for each dir (not fs) block that has a mapping. */
308 blk_start_plug(&plug);
309 while (ra_want > 0) {
310 next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
311 while (ra_want > 0 &&
312 next_ra < map.br_startoff + map.br_blockcount) {
313 if (next_ra >= last_da) {
317 if (next_ra > *ra_blk) {
318 xfs_dir3_data_readahead(dp, next_ra,
319 XFS_DABUF_MAP_HOLE_OK);
322 ra_want -= geo->fsbcount;
323 next_ra += geo->fsbcount;
325 if (!xfs_iext_next_extent(ifp, &icur, &map)) {
330 blk_finish_plug(&plug);
341 * Getdents (readdir) for leaf and node directories.
342 * This reads the data blocks only, so is the same for both forms.
345 xfs_dir2_leaf_getdents(
346 struct xfs_da_args *args,
347 struct dir_context *ctx,
349 unsigned int *lock_mode)
351 struct xfs_inode *dp = args->dp;
352 struct xfs_mount *mp = dp->i_mount;
353 struct xfs_buf *bp = NULL; /* data block buffer */
354 xfs_dir2_data_entry_t *dep; /* data entry */
355 xfs_dir2_data_unused_t *dup; /* unused entry */
356 struct xfs_da_geometry *geo = args->geo;
357 xfs_dablk_t rablk = 0; /* current readahead block */
358 xfs_dir2_off_t curoff; /* current overall offset */
359 int length; /* temporary length value */
360 int byteoff; /* offset in current block */
361 unsigned int offset = 0;
362 int error = 0; /* error return value */
365 * If the offset is at or past the largest allowed value,
366 * give up right away.
368 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
372 * Inside the loop we keep the main offset value as a byte offset
373 * in the directory file.
375 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
378 * Loop over directory entries until we reach the end offset.
379 * Get more blocks and readahead as necessary.
381 while (curoff < XFS_DIR2_LEAF_OFFSET) {
385 * If we have no buffer, or we're off the end of the
386 * current buffer, need to get another one.
388 if (!bp || offset >= geo->blksize) {
390 xfs_trans_brelse(args->trans, bp);
395 *lock_mode = xfs_ilock_data_map_shared(dp);
396 error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
401 xfs_iunlock(dp, *lock_mode);
404 xfs_dir3_data_check(dp, bp);
406 * Find our position in the block.
408 offset = geo->data_entry_offset;
409 byteoff = xfs_dir2_byte_to_off(geo, curoff);
411 * Skip past the header.
414 curoff += geo->data_entry_offset;
416 * Skip past entries until we reach our offset.
419 while (offset < byteoff) {
420 dup = bp->b_addr + offset;
422 if (be16_to_cpu(dup->freetag)
423 == XFS_DIR2_DATA_FREE_TAG) {
425 length = be16_to_cpu(dup->length);
429 dep = bp->b_addr + offset;
430 length = xfs_dir2_data_entsize(mp,
435 * Now set our real offset.
438 xfs_dir2_db_off_to_byte(geo,
439 xfs_dir2_byte_to_db(geo, curoff),
441 if (offset >= geo->blksize)
447 * We have a pointer to an entry. Is it a live one?
449 dup = bp->b_addr + offset;
452 * No, it's unused, skip over it.
454 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
455 length = be16_to_cpu(dup->length);
461 dep = bp->b_addr + offset;
462 length = xfs_dir2_data_entsize(mp, dep->namelen);
463 filetype = xfs_dir2_data_get_ftype(mp, dep);
465 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
466 if (XFS_IS_CORRUPT(dp->i_mount,
467 !xfs_dir2_namecheck(dep->name,
469 error = -EFSCORRUPTED;
472 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
473 be64_to_cpu(dep->inumber),
474 xfs_dir3_get_dtype(dp->i_mount, filetype)))
478 * Advance to next entry in the block.
482 /* bufsize may have just been a guess; don't go negative */
483 bufsize = bufsize > length ? bufsize - length : 0;
487 * All done. Set output offset value to current offset.
489 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
490 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
492 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
494 xfs_trans_brelse(args->trans, bp);
501 * If supplied, the transaction collects locked dir buffers to avoid
502 * nested buffer deadlocks. This function does not dirty the
503 * transaction. The caller must hold the IOLOCK (shared or exclusive)
504 * before calling this function.
508 struct xfs_trans *tp,
509 struct xfs_inode *dp,
510 struct dir_context *ctx,
513 struct xfs_da_args args = { NULL };
514 unsigned int lock_mode;
518 trace_xfs_readdir(dp);
520 if (xfs_is_shutdown(dp->i_mount))
523 ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
524 ASSERT(xfs_isilocked(dp, XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
525 XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
528 args.geo = dp->i_mount->m_dir_geo;
531 if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
532 return xfs_dir2_sf_getdents(&args, ctx);
534 lock_mode = xfs_ilock_data_map_shared(dp);
535 error = xfs_dir2_isblock(&args, &isblock);
540 error = xfs_dir2_block_getdents(&args, ctx, &lock_mode);
544 error = xfs_dir2_leaf_getdents(&args, ctx, bufsize, &lock_mode);
548 xfs_iunlock(dp, lock_mode);