Linux Audio

Check our new training course

Linux debugging, profiling, tracing and performance analysis training

Apr 14-17, 2025
Register
Loading...
Note: File does not exist in v3.1.
  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			return -EFSCORRUPTED;
123		if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
124			    xfs_dir3_get_dtype(mp, filetype)))
125			return 0;
126		sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
127	}
128
129	ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
130								0x7fffffff;
131	return 0;
132}
133
134/*
135 * Readdir for block directories.
136 */
137STATIC int
138xfs_dir2_block_getdents(
139	struct xfs_da_args	*args,
140	struct dir_context	*ctx,
141	unsigned int		*lock_mode)
142{
143	struct xfs_inode	*dp = args->dp;	/* incore directory inode */
144	struct xfs_buf		*bp;		/* buffer for block */
145	int			error;		/* error return value */
146	int			wantoff;	/* starting block offset */
147	xfs_off_t		cook;
148	struct xfs_da_geometry	*geo = args->geo;
149	unsigned int		offset, next_offset;
150	unsigned int		end;
151
152	/*
153	 * If the block number in the offset is out of range, we're done.
154	 */
155	if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
156		return 0;
157
158	error = xfs_dir3_block_read(args->trans, dp, &bp);
159	if (error)
160		return error;
161
162	xfs_iunlock(dp, *lock_mode);
163	*lock_mode = 0;
164
165	/*
166	 * Extract the byte offset we start at from the seek pointer.
167	 * We'll skip entries before this.
168	 */
169	wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
170	xfs_dir3_data_check(dp, bp);
171
172	/*
173	 * Loop over the data portion of the block.
174	 * Each object is a real entry (dep) or an unused one (dup).
175	 */
176	end = xfs_dir3_data_end_offset(geo, bp->b_addr);
177	for (offset = geo->data_entry_offset;
178	     offset < end;
179	     offset = next_offset) {
180		struct xfs_dir2_data_unused	*dup = bp->b_addr + offset;
181		struct xfs_dir2_data_entry	*dep = bp->b_addr + offset;
182		uint8_t filetype;
183
184		/*
185		 * Unused, skip it.
186		 */
187		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
188			next_offset = offset + be16_to_cpu(dup->length);
189			continue;
190		}
191
192		/*
193		 * Bump pointer for the next iteration.
194		 */
195		next_offset = offset +
196			xfs_dir2_data_entsize(dp->i_mount, dep->namelen);
197
198		/*
199		 * The entry is before the desired starting point, skip it.
200		 */
201		if (offset < wantoff)
202			continue;
203
204		cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, offset);
205
206		ctx->pos = cook & 0x7fffffff;
207		filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
208		/*
209		 * If it didn't fit, set the final offset to here & return.
210		 */
211		if (XFS_IS_CORRUPT(dp->i_mount,
212				   !xfs_dir2_namecheck(dep->name,
213						       dep->namelen))) {
214			error = -EFSCORRUPTED;
215			goto out_rele;
216		}
217		if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
218			    be64_to_cpu(dep->inumber),
219			    xfs_dir3_get_dtype(dp->i_mount, filetype)))
220			goto out_rele;
221	}
222
223	/*
224	 * Reached the end of the block.
225	 * Set the offset to a non-existent block 1 and return.
226	 */
227	ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
228								0x7fffffff;
229out_rele:
230	xfs_trans_brelse(args->trans, bp);
231	return error;
232}
233
234/*
235 * Read a directory block and initiate readahead for blocks beyond that.
236 * We maintain a sliding readahead window of the remaining space in the
237 * buffer rounded up to the nearest block.
238 */
239STATIC int
240xfs_dir2_leaf_readbuf(
241	struct xfs_da_args	*args,
242	size_t			bufsize,
243	xfs_dir2_off_t		*cur_off,
244	xfs_dablk_t		*ra_blk,
245	struct xfs_buf		**bpp)
246{
247	struct xfs_inode	*dp = args->dp;
248	struct xfs_buf		*bp = NULL;
249	struct xfs_da_geometry	*geo = args->geo;
250	struct xfs_ifork	*ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);
251	struct xfs_bmbt_irec	map;
252	struct blk_plug		plug;
253	xfs_dir2_off_t		new_off;
254	xfs_dablk_t		next_ra;
255	xfs_dablk_t		map_off;
256	xfs_dablk_t		last_da;
257	struct xfs_iext_cursor	icur;
258	int			ra_want;
259	int			error = 0;
260
261	error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
262	if (error)
263		goto out;
264
265	/*
266	 * Look for mapped directory blocks at or above the current offset.
267	 * Truncate down to the nearest directory block to start the scanning
268	 * operation.
269	 */
270	last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
271	map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
272	if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
273		goto out;
274	if (map.br_startoff >= last_da)
275		goto out;
276	xfs_trim_extent(&map, map_off, last_da - map_off);
277
278	/* Read the directory block of that first mapping. */
279	new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
280	if (new_off > *cur_off)
281		*cur_off = new_off;
282	error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
283	if (error)
284		goto out;
285
286	/*
287	 * Start readahead for the next bufsize's worth of dir data blocks.
288	 * We may have already issued readahead for some of that range;
289	 * ra_blk tracks the last block we tried to read(ahead).
290	 */
291	ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
292	if (*ra_blk >= last_da)
293		goto out;
294	else if (*ra_blk == 0)
295		*ra_blk = map.br_startoff;
296	next_ra = map.br_startoff + geo->fsbcount;
297	if (next_ra >= last_da)
298		goto out_no_ra;
299	if (map.br_blockcount < geo->fsbcount &&
300	    !xfs_iext_next_extent(ifp, &icur, &map))
301		goto out_no_ra;
302	if (map.br_startoff >= last_da)
303		goto out_no_ra;
304	xfs_trim_extent(&map, next_ra, last_da - next_ra);
305
306	/* Start ra for each dir (not fs) block that has a mapping. */
307	blk_start_plug(&plug);
308	while (ra_want > 0) {
309		next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
310		while (ra_want > 0 &&
311		       next_ra < map.br_startoff + map.br_blockcount) {
312			if (next_ra >= last_da) {
313				*ra_blk = last_da;
314				break;
315			}
316			if (next_ra > *ra_blk) {
317				xfs_dir3_data_readahead(dp, next_ra,
318							XFS_DABUF_MAP_HOLE_OK);
319				*ra_blk = next_ra;
320			}
321			ra_want -= geo->fsbcount;
322			next_ra += geo->fsbcount;
323		}
324		if (!xfs_iext_next_extent(ifp, &icur, &map)) {
325			*ra_blk = last_da;
326			break;
327		}
328	}
329	blk_finish_plug(&plug);
330
331out:
332	*bpp = bp;
333	return error;
334out_no_ra:
335	*ra_blk = last_da;
336	goto out;
337}
338
339/*
340 * Getdents (readdir) for leaf and node directories.
341 * This reads the data blocks only, so is the same for both forms.
342 */
343STATIC int
344xfs_dir2_leaf_getdents(
345	struct xfs_da_args	*args,
346	struct dir_context	*ctx,
347	size_t			bufsize,
348	unsigned int		*lock_mode)
349{
350	struct xfs_inode	*dp = args->dp;
351	struct xfs_mount	*mp = dp->i_mount;
352	struct xfs_buf		*bp = NULL;	/* data block buffer */
353	xfs_dir2_data_entry_t	*dep;		/* data entry */
354	xfs_dir2_data_unused_t	*dup;		/* unused entry */
355	struct xfs_da_geometry	*geo = args->geo;
356	xfs_dablk_t		rablk = 0;	/* current readahead block */
357	xfs_dir2_off_t		curoff;		/* current overall offset */
358	int			length;		/* temporary length value */
359	int			byteoff;	/* offset in current block */
360	unsigned int		offset = 0;
361	int			error = 0;	/* error return value */
362
363	/*
364	 * If the offset is at or past the largest allowed value,
365	 * give up right away.
366	 */
367	if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
368		return 0;
369
370	/*
371	 * Inside the loop we keep the main offset value as a byte offset
372	 * in the directory file.
373	 */
374	curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
375
376	/*
377	 * Loop over directory entries until we reach the end offset.
378	 * Get more blocks and readahead as necessary.
379	 */
380	while (curoff < XFS_DIR2_LEAF_OFFSET) {
381		uint8_t filetype;
382
383		/*
384		 * If we have no buffer, or we're off the end of the
385		 * current buffer, need to get another one.
386		 */
387		if (!bp || offset >= geo->blksize) {
388			if (bp) {
389				xfs_trans_brelse(args->trans, bp);
390				bp = NULL;
391			}
392
393			if (*lock_mode == 0)
394				*lock_mode = xfs_ilock_data_map_shared(dp);
395			error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
396					&rablk, &bp);
397			if (error || !bp)
398				break;
399
400			xfs_iunlock(dp, *lock_mode);
401			*lock_mode = 0;
402
403			xfs_dir3_data_check(dp, bp);
404			/*
405			 * Find our position in the block.
406			 */
407			offset = geo->data_entry_offset;
408			byteoff = xfs_dir2_byte_to_off(geo, curoff);
409			/*
410			 * Skip past the header.
411			 */
412			if (byteoff == 0)
413				curoff += geo->data_entry_offset;
414			/*
415			 * Skip past entries until we reach our offset.
416			 */
417			else {
418				while (offset < byteoff) {
419					dup = bp->b_addr + offset;
420
421					if (be16_to_cpu(dup->freetag)
422						  == XFS_DIR2_DATA_FREE_TAG) {
423
424						length = be16_to_cpu(dup->length);
425						offset += length;
426						continue;
427					}
428					dep = bp->b_addr + offset;
429					length = xfs_dir2_data_entsize(mp,
430							dep->namelen);
431					offset += length;
432				}
433				/*
434				 * Now set our real offset.
435				 */
436				curoff =
437					xfs_dir2_db_off_to_byte(geo,
438					    xfs_dir2_byte_to_db(geo, curoff),
439					    offset);
440				if (offset >= geo->blksize)
441					continue;
442			}
443		}
444
445		/*
446		 * We have a pointer to an entry.  Is it a live one?
447		 */
448		dup = bp->b_addr + offset;
449
450		/*
451		 * No, it's unused, skip over it.
452		 */
453		if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
454			length = be16_to_cpu(dup->length);
455			offset += length;
456			curoff += length;
457			continue;
458		}
459
460		dep = bp->b_addr + offset;
461		length = xfs_dir2_data_entsize(mp, dep->namelen);
462		filetype = xfs_dir2_data_get_ftype(mp, dep);
463
464		ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
465		if (XFS_IS_CORRUPT(dp->i_mount,
466				   !xfs_dir2_namecheck(dep->name,
467						       dep->namelen))) {
468			error = -EFSCORRUPTED;
469			break;
470		}
471		if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
472			    be64_to_cpu(dep->inumber),
473			    xfs_dir3_get_dtype(dp->i_mount, filetype)))
474			break;
475
476		/*
477		 * Advance to next entry in the block.
478		 */
479		offset += length;
480		curoff += length;
481		/* bufsize may have just been a guess; don't go negative */
482		bufsize = bufsize > length ? bufsize - length : 0;
483	}
484
485	/*
486	 * All done.  Set output offset value to current offset.
487	 */
488	if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
489		ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
490	else
491		ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
492	if (bp)
493		xfs_trans_brelse(args->trans, bp);
494	return error;
495}
496
497/*
498 * Read a directory.
499 *
500 * If supplied, the transaction collects locked dir buffers to avoid
501 * nested buffer deadlocks.  This function does not dirty the
502 * transaction.  The caller must hold the IOLOCK (shared or exclusive)
503 * before calling this function.
504 */
505int
506xfs_readdir(
507	struct xfs_trans	*tp,
508	struct xfs_inode	*dp,
509	struct dir_context	*ctx,
510	size_t			bufsize)
511{
512	struct xfs_da_args	args = { NULL };
513	unsigned int		lock_mode;
514	bool			isblock;
515	int			error;
516
517	trace_xfs_readdir(dp);
518
519	if (xfs_is_shutdown(dp->i_mount))
520		return -EIO;
521	if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
522		return -EIO;
523
524	ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
525	ASSERT(xfs_isilocked(dp, XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
526	XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
527
528	args.dp = dp;
529	args.geo = dp->i_mount->m_dir_geo;
530	args.trans = tp;
531
532	if (dp->i_df.if_format == XFS_DINODE_FMT_LOCAL)
533		return xfs_dir2_sf_getdents(&args, ctx);
534
535	lock_mode = xfs_ilock_data_map_shared(dp);
536	error = xfs_dir2_isblock(&args, &isblock);
537	if (error)
538		goto out_unlock;
539
540	if (isblock) {
541		error = xfs_dir2_block_getdents(&args, ctx, &lock_mode);
542		goto out_unlock;
543	}
544
545	error = xfs_dir2_leaf_getdents(&args, ctx, bufsize, &lock_mode);
546
547out_unlock:
548	if (lock_mode)
549		xfs_iunlock(dp, lock_mode);
550	return error;
551}