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#include "xfs_error.h"
21#include "xfs_health.h"
22
23/*
24 * Directory file type support functions
25 */
26static unsigned char xfs_dir3_filetype_table[] = {
27	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
28	DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
29};
30
31unsigned char
32xfs_dir3_get_dtype(
33	struct xfs_mount	*mp,
34	uint8_t			filetype)
35{
36	if (!xfs_has_ftype(mp))
37		return DT_UNKNOWN;
38
39	if (filetype >= XFS_DIR3_FT_MAX)
40		return DT_UNKNOWN;
41
42	return xfs_dir3_filetype_table[filetype];
43}
44
45STATIC int
46xfs_dir2_sf_getdents(
47	struct xfs_da_args	*args,
48	struct dir_context	*ctx)
49{
50	int			i;		/* shortform entry number */
51	struct xfs_inode	*dp = args->dp;	/* incore directory inode */
52	struct xfs_mount	*mp = dp->i_mount;
53	xfs_dir2_dataptr_t	off;		/* current entry's offset */
54	xfs_dir2_sf_entry_t	*sfep;		/* shortform directory entry */
55	struct xfs_dir2_sf_hdr	*sfp = dp->i_df.if_data;
56	xfs_dir2_dataptr_t	dot_offset;
57	xfs_dir2_dataptr_t	dotdot_offset;
58	xfs_ino_t		ino;
59	struct xfs_da_geometry	*geo = args->geo;
60
61	ASSERT(dp->i_df.if_format == XFS_DINODE_FMT_LOCAL);
62	ASSERT(dp->i_df.if_bytes == dp->i_disk_size);
63	ASSERT(sfp != NULL);
64
65	/*
66	 * If the block number in the offset is out of range, we're done.
67	 */
68	if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
69		return 0;
70
71	/*
72	 * Precalculate offsets for "." and ".." as we will always need them.
73	 * This relies on the fact that directories always start with the
74	 * entries for "." and "..".
75	 */
76	dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
77			geo->data_entry_offset);
78	dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
79			geo->data_entry_offset +
80			xfs_dir2_data_entsize(mp, sizeof(".") - 1));
81
82	/*
83	 * Put . entry unless we're starting past it.
84	 */
85	if (ctx->pos <= dot_offset) {
86		ctx->pos = dot_offset & 0x7fffffff;
87		if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
88			return 0;
89	}
90
91	/*
92	 * Put .. entry unless we're starting past it.
93	 */
94	if (ctx->pos <= dotdot_offset) {
95		ino = xfs_dir2_sf_get_parent_ino(sfp);
96		ctx->pos = dotdot_offset & 0x7fffffff;
97		if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
98			return 0;
99	}
100
101	/*
102	 * Loop while there are more entries and put'ing works.
103	 */
104	sfep = xfs_dir2_sf_firstentry(sfp);
105	for (i = 0; i < sfp->count; i++) {
106		uint8_t filetype;
107
108		off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
109				xfs_dir2_sf_get_offset(sfep));
110
111		if (ctx->pos > off) {
112			sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
113			continue;
114		}
115
116		ino = xfs_dir2_sf_get_ino(mp, sfp, sfep);
117		filetype = xfs_dir2_sf_get_ftype(mp, sfep);
118		ctx->pos = off & 0x7fffffff;
119		if (XFS_IS_CORRUPT(dp->i_mount,
120				   !xfs_dir2_namecheck(sfep->name,
121						       sfep->namelen))) {
122			xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
123			return -EFSCORRUPTED;
124		}
125		if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
126			    xfs_dir3_get_dtype(mp, filetype)))
127			return 0;
128		sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
129	}
130
131	ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
132								0x7fffffff;
133	return 0;
134}
135
136/*
137 * Readdir for block directories.
138 */
139STATIC int
140xfs_dir2_block_getdents(
141	struct xfs_da_args	*args,
142	struct dir_context	*ctx,
143	unsigned int		*lock_mode)
144{
145	struct xfs_inode	*dp = args->dp;	/* incore directory inode */
146	struct xfs_buf		*bp;		/* buffer for block */
147	int			error;		/* error return value */
148	int			wantoff;	/* starting block offset */
149	xfs_off_t		cook;
150	struct xfs_da_geometry	*geo = args->geo;
151	unsigned int		offset, next_offset;
152	unsigned int		end;
153
154	/*
155	 * If the block number in the offset is out of range, we're done.
156	 */
157	if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
158		return 0;
159
160	error = xfs_dir3_block_read(args->trans, dp, &bp);
161	if (error)
162		return error;
163
164	xfs_iunlock(dp, *lock_mode);
165	*lock_mode = 0;
166
167	/*
168	 * Extract the byte offset we start at from the seek pointer.
169	 * We'll skip entries before this.
170	 */
171	wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
172	xfs_dir3_data_check(dp, bp);
173
174	/*
175	 * Loop over the data portion of the block.
176	 * Each object is a real entry (dep) or an unused one (dup).
177	 */
178	end = xfs_dir3_data_end_offset(geo, bp->b_addr);
179	for (offset = geo->data_entry_offset;
180	     offset < end;
181	     offset = next_offset) {
182		struct xfs_dir2_data_unused	*dup = bp->b_addr + offset;
183		struct xfs_dir2_data_entry	*dep = bp->b_addr + offset;
184		uint8_t filetype;
185
186		/*
187		 * Unused, skip it.
188		 */
189		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
190			next_offset = offset + be16_to_cpu(dup->length);
191			continue;
192		}
193
194		/*
195		 * Bump pointer for the next iteration.
196		 */
197		next_offset = offset +
198			xfs_dir2_data_entsize(dp->i_mount, dep->namelen);
199
200		/*
201		 * The entry is before the desired starting point, skip it.
202		 */
203		if (offset < wantoff)
204			continue;
205
206		cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, offset);
207
208		ctx->pos = cook & 0x7fffffff;
209		filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
210		/*
211		 * If it didn't fit, set the final offset to here & return.
212		 */
213		if (XFS_IS_CORRUPT(dp->i_mount,
214				   !xfs_dir2_namecheck(dep->name,
215						       dep->namelen))) {
216			xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
217			error = -EFSCORRUPTED;
218			goto out_rele;
219		}
220		if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
221			    be64_to_cpu(dep->inumber),
222			    xfs_dir3_get_dtype(dp->i_mount, filetype)))
223			goto out_rele;
224	}
225
226	/*
227	 * Reached the end of the block.
228	 * Set the offset to a non-existent block 1 and return.
229	 */
230	ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
231								0x7fffffff;
232out_rele:
233	xfs_trans_brelse(args->trans, bp);
234	return error;
235}
236
237/*
238 * Read a directory block and initiate readahead for blocks beyond that.
239 * We maintain a sliding readahead window of the remaining space in the
240 * buffer rounded up to the nearest block.
241 */
242STATIC int
243xfs_dir2_leaf_readbuf(
244	struct xfs_da_args	*args,
245	size_t			bufsize,
246	xfs_dir2_off_t		*cur_off,
247	xfs_dablk_t		*ra_blk,
248	struct xfs_buf		**bpp)
249{
250	struct xfs_inode	*dp = args->dp;
251	struct xfs_buf		*bp = NULL;
252	struct xfs_da_geometry	*geo = args->geo;
253	struct xfs_ifork	*ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);
254	struct xfs_bmbt_irec	map;
255	struct blk_plug		plug;
256	xfs_dir2_off_t		new_off;
257	xfs_dablk_t		next_ra;
258	xfs_dablk_t		map_off;
259	xfs_dablk_t		last_da;
260	struct xfs_iext_cursor	icur;
261	int			ra_want;
262	int			error = 0;
263
264	error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
265	if (error)
266		goto out;
267
268	/*
269	 * Look for mapped directory blocks at or above the current offset.
270	 * Truncate down to the nearest directory block to start the scanning
271	 * operation.
272	 */
273	last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
274	map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
275	if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
276		goto out;
277	if (map.br_startoff >= last_da)
278		goto out;
279	xfs_trim_extent(&map, map_off, last_da - map_off);
280
281	/* Read the directory block of that first mapping. */
282	new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
283	if (new_off > *cur_off)
284		*cur_off = new_off;
285	error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
286	if (error)
287		goto out;
288
289	/*
290	 * Start readahead for the next bufsize's worth of dir data blocks.
291	 * We may have already issued readahead for some of that range;
292	 * ra_blk tracks the last block we tried to read(ahead).
293	 */
294	ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
295	if (*ra_blk >= last_da)
296		goto out;
297	else if (*ra_blk == 0)
298		*ra_blk = map.br_startoff;
299	next_ra = map.br_startoff + geo->fsbcount;
300	if (next_ra >= last_da)
301		goto out_no_ra;
302	if (map.br_blockcount < geo->fsbcount &&
303	    !xfs_iext_next_extent(ifp, &icur, &map))
304		goto out_no_ra;
305	if (map.br_startoff >= last_da)
306		goto out_no_ra;
307	xfs_trim_extent(&map, next_ra, last_da - next_ra);
308
309	/* Start ra for each dir (not fs) block that has a mapping. */
310	blk_start_plug(&plug);
311	while (ra_want > 0) {
312		next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
313		while (ra_want > 0 &&
314		       next_ra < map.br_startoff + map.br_blockcount) {
315			if (next_ra >= last_da) {
316				*ra_blk = last_da;
317				break;
318			}
319			if (next_ra > *ra_blk) {
320				xfs_dir3_data_readahead(dp, next_ra,
321							XFS_DABUF_MAP_HOLE_OK);
322				*ra_blk = next_ra;
323			}
324			ra_want -= geo->fsbcount;
325			next_ra += geo->fsbcount;
326		}
327		if (!xfs_iext_next_extent(ifp, &icur, &map)) {
328			*ra_blk = last_da;
329			break;
330		}
331	}
332	blk_finish_plug(&plug);
333
334out:
335	*bpp = bp;
336	return error;
337out_no_ra:
338	*ra_blk = last_da;
339	goto out;
340}
341
342/*
343 * Getdents (readdir) for leaf and node directories.
344 * This reads the data blocks only, so is the same for both forms.
345 */
346STATIC int
347xfs_dir2_leaf_getdents(
348	struct xfs_da_args	*args,
349	struct dir_context	*ctx,
350	size_t			bufsize,
351	unsigned int		*lock_mode)
352{
353	struct xfs_inode	*dp = args->dp;
354	struct xfs_mount	*mp = dp->i_mount;
355	struct xfs_buf		*bp = NULL;	/* data block buffer */
356	xfs_dir2_data_entry_t	*dep;		/* data entry */
357	xfs_dir2_data_unused_t	*dup;		/* unused entry */
358	struct xfs_da_geometry	*geo = args->geo;
359	xfs_dablk_t		rablk = 0;	/* current readahead block */
360	xfs_dir2_off_t		curoff;		/* current overall offset */
361	int			length;		/* temporary length value */
362	int			byteoff;	/* offset in current block */
363	unsigned int		offset = 0;
364	int			error = 0;	/* error return value */
365
366	/*
367	 * If the offset is at or past the largest allowed value,
368	 * give up right away.
369	 */
370	if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
371		return 0;
372
373	/*
374	 * Inside the loop we keep the main offset value as a byte offset
375	 * in the directory file.
376	 */
377	curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
378
379	/*
380	 * Loop over directory entries until we reach the end offset.
381	 * Get more blocks and readahead as necessary.
382	 */
383	while (curoff < XFS_DIR2_LEAF_OFFSET) {
384		uint8_t filetype;
385
386		/*
387		 * If we have no buffer, or we're off the end of the
388		 * current buffer, need to get another one.
389		 */
390		if (!bp || offset >= geo->blksize) {
391			if (bp) {
392				xfs_trans_brelse(args->trans, bp);
393				bp = NULL;
394			}
395
396			if (*lock_mode == 0)
397				*lock_mode = xfs_ilock_data_map_shared(dp);
398			error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
399					&rablk, &bp);
400			if (error || !bp)
401				break;
402
403			xfs_iunlock(dp, *lock_mode);
404			*lock_mode = 0;
405
406			xfs_dir3_data_check(dp, bp);
407			/*
408			 * Find our position in the block.
409			 */
410			offset = geo->data_entry_offset;
411			byteoff = xfs_dir2_byte_to_off(geo, curoff);
412			/*
413			 * Skip past the header.
414			 */
415			if (byteoff == 0)
416				curoff += geo->data_entry_offset;
417			/*
418			 * Skip past entries until we reach our offset.
419			 */
420			else {
421				while (offset < byteoff) {
422					dup = bp->b_addr + offset;
423
424					if (be16_to_cpu(dup->freetag)
425						  == XFS_DIR2_DATA_FREE_TAG) {
426
427						length = be16_to_cpu(dup->length);
428						offset += length;
429						continue;
430					}
431					dep = bp->b_addr + offset;
432					length = xfs_dir2_data_entsize(mp,
433							dep->namelen);
434					offset += length;
435				}
436				/*
437				 * Now set our real offset.
438				 */
439				curoff =
440					xfs_dir2_db_off_to_byte(geo,
441					    xfs_dir2_byte_to_db(geo, curoff),
442					    offset);
443				if (offset >= geo->blksize)
444					continue;
445			}
446		}
447
448		/*
449		 * We have a pointer to an entry.  Is it a live one?
450		 */
451		dup = bp->b_addr + offset;
452
453		/*
454		 * No, it's unused, skip over it.
455		 */
456		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
457			length = be16_to_cpu(dup->length);
458			offset += length;
459			curoff += length;
460			continue;
461		}
462
463		dep = bp->b_addr + offset;
464		length = xfs_dir2_data_entsize(mp, dep->namelen);
465		filetype = xfs_dir2_data_get_ftype(mp, dep);
466
467		ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
468		if (XFS_IS_CORRUPT(dp->i_mount,
469				   !xfs_dir2_namecheck(dep->name,
470						       dep->namelen))) {
471			xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
472			error = -EFSCORRUPTED;
473			break;
474		}
475		if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
476			    be64_to_cpu(dep->inumber),
477			    xfs_dir3_get_dtype(dp->i_mount, filetype)))
478			break;
479
480		/*
481		 * Advance to next entry in the block.
482		 */
483		offset += length;
484		curoff += length;
485		/* bufsize may have just been a guess; don't go negative */
486		bufsize = bufsize > length ? bufsize - length : 0;
487	}
488
489	/*
490	 * All done.  Set output offset value to current offset.
491	 */
492	if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
493		ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
494	else
495		ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
496	if (bp)
497		xfs_trans_brelse(args->trans, bp);
498	return error;
499}
500
501/*
502 * Read a directory.
503 *
504 * If supplied, the transaction collects locked dir buffers to avoid
505 * nested buffer deadlocks.  This function does not dirty the
506 * transaction.  The caller must hold the IOLOCK (shared or exclusive)
507 * before calling this function.
508 */
509int
510xfs_readdir(
511	struct xfs_trans	*tp,
512	struct xfs_inode	*dp,
513	struct dir_context	*ctx,
514	size_t			bufsize)
515{
516	struct xfs_da_args	args = { NULL };
517	unsigned int		lock_mode;
518	bool			isblock;
519	int			error;
520
521	trace_xfs_readdir(dp);
522
523	if (xfs_is_shutdown(dp->i_mount))
524		return -EIO;
525	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
526		return -EIO;
527
528	ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
529	xfs_assert_ilocked(dp, XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL);
530	XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
531
532	args.dp = dp;
533	args.geo = dp->i_mount->m_dir_geo;
534	args.trans = tp;
535
536	if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
537		return xfs_dir2_sf_getdents(&args, ctx);
538
539	lock_mode = xfs_ilock_data_map_shared(dp);
540	error = xfs_dir2_isblock(&args, &isblock);
541	if (error)
542		goto out_unlock;
543
544	if (isblock) {
545		error = xfs_dir2_block_getdents(&args, ctx, &lock_mode);
546		goto out_unlock;
547	}
548
549	error = xfs_dir2_leaf_getdents(&args, ctx, bufsize, &lock_mode);
550
551out_unlock:
552	if (lock_mode)
553		xfs_iunlock(dp, lock_mode);
554	return error;
555}
556