1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_btree.h"
15#include "xfs_ialloc.h"
16#include "xfs_ialloc_btree.h"
17#include "xfs_iwalk.h"
18#include "xfs_itable.h"
19#include "xfs_error.h"
20#include "xfs_icache.h"
21#include "xfs_health.h"
22#include "xfs_trans.h"
23
24/*
25 * Bulk Stat
26 * =========
27 *
28 * Use the inode walking functions to fill out struct xfs_bulkstat for every
29 * allocated inode, then pass the stat information to some externally provided
30 * iteration function.
31 */
32
33struct xfs_bstat_chunk {
34	bulkstat_one_fmt_pf	formatter;
35	struct xfs_ibulk	*breq;
36	struct xfs_bulkstat	*buf;
37};
38
39/*
40 * Fill out the bulkstat info for a single inode and report it somewhere.
41 *
42 * bc->breq->lastino is effectively the inode cursor as we walk through the
43 * filesystem.  Therefore, we update it any time we need to move the cursor
44 * forward, regardless of whether or not we're sending any bstat information
45 * back to userspace.  If the inode is internal metadata or, has been freed
46 * out from under us, we just simply keep going.
47 *
48 * However, if any other type of error happens we want to stop right where we
49 * are so that userspace will call back with exact number of the bad inode and
50 * we can send back an error code.
51 *
52 * Note that if the formatter tells us there's no space left in the buffer we
53 * move the cursor forward and abort the walk.
54 */
55STATIC int
56xfs_bulkstat_one_int(
57	struct xfs_mount	*mp,
58	struct mnt_idmap	*idmap,
59	struct xfs_trans	*tp,
60	xfs_ino_t		ino,
61	struct xfs_bstat_chunk	*bc)
62{
63	struct user_namespace	*sb_userns = mp->m_super->s_user_ns;
64	struct xfs_inode	*ip;		/* incore inode pointer */
65	struct inode		*inode;
66	struct xfs_bulkstat	*buf = bc->buf;
67	xfs_extnum_t		nextents;
68	int			error = -EINVAL;
69	vfsuid_t		vfsuid;
70	vfsgid_t		vfsgid;
71
72	if (xfs_internal_inum(mp, ino))
73		goto out_advance;
74
75	error = xfs_iget(mp, tp, ino,
76			 (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED),
77			 XFS_ILOCK_SHARED, &ip);
78	if (error == -ENOENT || error == -EINVAL)
79		goto out_advance;
80	if (error)
81		goto out;
82
83	/* Reload the incore unlinked list to avoid failure in inodegc. */
84	if (xfs_inode_unlinked_incomplete(ip)) {
85		error = xfs_inode_reload_unlinked_bucket(tp, ip);
86		if (error) {
87			xfs_iunlock(ip, XFS_ILOCK_SHARED);
88			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
89			xfs_irele(ip);
90			return error;
91		}
92	}
93
94	ASSERT(ip != NULL);
95	ASSERT(ip->i_imap.im_blkno != 0);
96	inode = VFS_I(ip);
97	vfsuid = i_uid_into_vfsuid(idmap, inode);
98	vfsgid = i_gid_into_vfsgid(idmap, inode);
99
100	/* xfs_iget returns the following without needing
101	 * further change.
102	 */
103	buf->bs_projectid = ip->i_projid;
104	buf->bs_ino = ino;
105	buf->bs_uid = from_kuid(sb_userns, vfsuid_into_kuid(vfsuid));
106	buf->bs_gid = from_kgid(sb_userns, vfsgid_into_kgid(vfsgid));
107	buf->bs_size = ip->i_disk_size;
108
109	buf->bs_nlink = inode->i_nlink;
110	buf->bs_atime = inode_get_atime_sec(inode);
111	buf->bs_atime_nsec = inode_get_atime_nsec(inode);
112	buf->bs_mtime = inode_get_mtime_sec(inode);
113	buf->bs_mtime_nsec = inode_get_mtime_nsec(inode);
114	buf->bs_ctime = inode_get_ctime_sec(inode);
115	buf->bs_ctime_nsec = inode_get_ctime_nsec(inode);
116	buf->bs_gen = inode->i_generation;
117	buf->bs_mode = inode->i_mode;
118
119	buf->bs_xflags = xfs_ip2xflags(ip);
120	buf->bs_extsize_blks = ip->i_extsize;
121
122	nextents = xfs_ifork_nextents(&ip->i_df);
123	if (!(bc->breq->flags & XFS_IBULK_NREXT64))
124		buf->bs_extents = min(nextents, XFS_MAX_EXTCNT_DATA_FORK_SMALL);
125	else
126		buf->bs_extents64 = nextents;
127
128	xfs_bulkstat_health(ip, buf);
129	buf->bs_aextents = xfs_ifork_nextents(&ip->i_af);
130	buf->bs_forkoff = xfs_inode_fork_boff(ip);
131	buf->bs_version = XFS_BULKSTAT_VERSION_V5;
132
133	if (xfs_has_v3inodes(mp)) {
134		buf->bs_btime = ip->i_crtime.tv_sec;
135		buf->bs_btime_nsec = ip->i_crtime.tv_nsec;
136		if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
137			buf->bs_cowextsize_blks = ip->i_cowextsize;
138	}
139
140	switch (ip->i_df.if_format) {
141	case XFS_DINODE_FMT_DEV:
142		buf->bs_rdev = sysv_encode_dev(inode->i_rdev);
143		buf->bs_blksize = BLKDEV_IOSIZE;
144		buf->bs_blocks = 0;
145		break;
146	case XFS_DINODE_FMT_LOCAL:
147		buf->bs_rdev = 0;
148		buf->bs_blksize = mp->m_sb.sb_blocksize;
149		buf->bs_blocks = 0;
150		break;
151	case XFS_DINODE_FMT_EXTENTS:
152	case XFS_DINODE_FMT_BTREE:
153		buf->bs_rdev = 0;
154		buf->bs_blksize = mp->m_sb.sb_blocksize;
155		buf->bs_blocks = ip->i_nblocks + ip->i_delayed_blks;
156		break;
157	}
158	xfs_iunlock(ip, XFS_ILOCK_SHARED);
159	xfs_irele(ip);
160
161	error = bc->formatter(bc->breq, buf);
162	if (error == -ECANCELED)
163		goto out_advance;
164	if (error)
165		goto out;
166
167out_advance:
168	/*
169	 * Advance the cursor to the inode that comes after the one we just
170	 * looked at.  We want the caller to move along if the bulkstat
171	 * information was copied successfully; if we tried to grab the inode
172	 * but it's no longer allocated; or if it's internal metadata.
173	 */
174	bc->breq->startino = ino + 1;
175out:
176	return error;
177}
178
179/* Bulkstat a single inode. */
180int
181xfs_bulkstat_one(
182	struct xfs_ibulk	*breq,
183	bulkstat_one_fmt_pf	formatter)
184{
185	struct xfs_bstat_chunk	bc = {
186		.formatter	= formatter,
187		.breq		= breq,
188	};
189	struct xfs_trans	*tp;
190	int			error;
191
192	if (breq->idmap != &nop_mnt_idmap) {
193		xfs_warn_ratelimited(breq->mp,
194			"bulkstat not supported inside of idmapped mounts.");
195		return -EINVAL;
196	}
197
198	ASSERT(breq->icount == 1);
199
200	bc.buf = kzalloc(sizeof(struct xfs_bulkstat),
201			GFP_KERNEL | __GFP_RETRY_MAYFAIL);
202	if (!bc.buf)
203		return -ENOMEM;
204
205	/*
206	 * Grab an empty transaction so that we can use its recursive buffer
207	 * locking abilities to detect cycles in the inobt without deadlocking.
208	 */
209	error = xfs_trans_alloc_empty(breq->mp, &tp);
210	if (error)
211		goto out;
212
213	error = xfs_bulkstat_one_int(breq->mp, breq->idmap, tp,
214			breq->startino, &bc);
215	xfs_trans_cancel(tp);
216out:
217	kfree(bc.buf);
218
219	/*
220	 * If we reported one inode to userspace then we abort because we hit
221	 * the end of the buffer.  Don't leak that back to userspace.
222	 */
223	if (error == -ECANCELED)
224		error = 0;
225
226	return error;
227}
228
229static int
230xfs_bulkstat_iwalk(
231	struct xfs_mount	*mp,
232	struct xfs_trans	*tp,
233	xfs_ino_t		ino,
234	void			*data)
235{
236	struct xfs_bstat_chunk	*bc = data;
237	int			error;
238
239	error = xfs_bulkstat_one_int(mp, bc->breq->idmap, tp, ino, data);
240	/* bulkstat just skips over missing inodes */
241	if (error == -ENOENT || error == -EINVAL)
242		return 0;
243	return error;
244}
245
246/*
247 * Check the incoming lastino parameter.
248 *
249 * We allow any inode value that could map to physical space inside the
250 * filesystem because if there are no inodes there, bulkstat moves on to the
251 * next chunk.  In other words, the magic agino value of zero takes us to the
252 * first chunk in the AG, and an agino value past the end of the AG takes us to
253 * the first chunk in the next AG.
254 *
255 * Therefore we can end early if the requested inode is beyond the end of the
256 * filesystem or doesn't map properly.
257 */
258static inline bool
259xfs_bulkstat_already_done(
260	struct xfs_mount	*mp,
261	xfs_ino_t		startino)
262{
263	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, startino);
264	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, startino);
265
266	return agno >= mp->m_sb.sb_agcount ||
267	       startino != XFS_AGINO_TO_INO(mp, agno, agino);
268}
269
270/* Return stat information in bulk (by-inode) for the filesystem. */
271int
272xfs_bulkstat(
273	struct xfs_ibulk	*breq,
274	bulkstat_one_fmt_pf	formatter)
275{
276	struct xfs_bstat_chunk	bc = {
277		.formatter	= formatter,
278		.breq		= breq,
279	};
280	struct xfs_trans	*tp;
281	unsigned int		iwalk_flags = 0;
282	int			error;
283
284	if (breq->idmap != &nop_mnt_idmap) {
285		xfs_warn_ratelimited(breq->mp,
286			"bulkstat not supported inside of idmapped mounts.");
287		return -EINVAL;
288	}
289	if (xfs_bulkstat_already_done(breq->mp, breq->startino))
290		return 0;
291
292	bc.buf = kzalloc(sizeof(struct xfs_bulkstat),
293			GFP_KERNEL | __GFP_RETRY_MAYFAIL);
294	if (!bc.buf)
295		return -ENOMEM;
296
297	/*
298	 * Grab an empty transaction so that we can use its recursive buffer
299	 * locking abilities to detect cycles in the inobt without deadlocking.
300	 */
301	error = xfs_trans_alloc_empty(breq->mp, &tp);
302	if (error)
303		goto out;
304
305	if (breq->flags & XFS_IBULK_SAME_AG)
306		iwalk_flags |= XFS_IWALK_SAME_AG;
307
308	error = xfs_iwalk(breq->mp, tp, breq->startino, iwalk_flags,
309			xfs_bulkstat_iwalk, breq->icount, &bc);
310	xfs_trans_cancel(tp);
311out:
312	kfree(bc.buf);
313
314	/*
315	 * We found some inodes, so clear the error status and return them.
316	 * The lastino pointer will point directly at the inode that triggered
317	 * any error that occurred, so on the next call the error will be
318	 * triggered again and propagated to userspace as there will be no
319	 * formatted inodes in the buffer.
320	 */
321	if (breq->ocount > 0)
322		error = 0;
323
324	return error;
325}
326
327/* Convert bulkstat (v5) to bstat (v1). */
328void
329xfs_bulkstat_to_bstat(
330	struct xfs_mount		*mp,
331	struct xfs_bstat		*bs1,
332	const struct xfs_bulkstat	*bstat)
333{
334	/* memset is needed here because of padding holes in the structure. */
335	memset(bs1, 0, sizeof(struct xfs_bstat));
336	bs1->bs_ino = bstat->bs_ino;
337	bs1->bs_mode = bstat->bs_mode;
338	bs1->bs_nlink = bstat->bs_nlink;
339	bs1->bs_uid = bstat->bs_uid;
340	bs1->bs_gid = bstat->bs_gid;
341	bs1->bs_rdev = bstat->bs_rdev;
342	bs1->bs_blksize = bstat->bs_blksize;
343	bs1->bs_size = bstat->bs_size;
344	bs1->bs_atime.tv_sec = bstat->bs_atime;
345	bs1->bs_mtime.tv_sec = bstat->bs_mtime;
346	bs1->bs_ctime.tv_sec = bstat->bs_ctime;
347	bs1->bs_atime.tv_nsec = bstat->bs_atime_nsec;
348	bs1->bs_mtime.tv_nsec = bstat->bs_mtime_nsec;
349	bs1->bs_ctime.tv_nsec = bstat->bs_ctime_nsec;
350	bs1->bs_blocks = bstat->bs_blocks;
351	bs1->bs_xflags = bstat->bs_xflags;
352	bs1->bs_extsize = XFS_FSB_TO_B(mp, bstat->bs_extsize_blks);
353	bs1->bs_extents = bstat->bs_extents;
354	bs1->bs_gen = bstat->bs_gen;
355	bs1->bs_projid_lo = bstat->bs_projectid & 0xFFFF;
356	bs1->bs_forkoff = bstat->bs_forkoff;
357	bs1->bs_projid_hi = bstat->bs_projectid >> 16;
358	bs1->bs_sick = bstat->bs_sick;
359	bs1->bs_checked = bstat->bs_checked;
360	bs1->bs_cowextsize = XFS_FSB_TO_B(mp, bstat->bs_cowextsize_blks);
361	bs1->bs_dmevmask = 0;
362	bs1->bs_dmstate = 0;
363	bs1->bs_aextents = bstat->bs_aextents;
364}
365
366struct xfs_inumbers_chunk {
367	inumbers_fmt_pf		formatter;
368	struct xfs_ibulk	*breq;
369};
370
371/*
372 * INUMBERS
373 * ========
374 * This is how we export inode btree records to userspace, so that XFS tools
375 * can figure out where inodes are allocated.
376 */
377
378/*
379 * Format the inode group structure and report it somewhere.
380 *
381 * Similar to xfs_bulkstat_one_int, lastino is the inode cursor as we walk
382 * through the filesystem so we move it forward unless there was a runtime
383 * error.  If the formatter tells us the buffer is now full we also move the
384 * cursor forward and abort the walk.
385 */
386STATIC int
387xfs_inumbers_walk(
388	struct xfs_mount	*mp,
389	struct xfs_trans	*tp,
390	xfs_agnumber_t		agno,
391	const struct xfs_inobt_rec_incore *irec,
392	void			*data)
393{
394	struct xfs_inumbers	inogrp = {
395		.xi_startino	= XFS_AGINO_TO_INO(mp, agno, irec->ir_startino),
396		.xi_alloccount	= irec->ir_count - irec->ir_freecount,
397		.xi_allocmask	= ~irec->ir_free,
398		.xi_version	= XFS_INUMBERS_VERSION_V5,
399	};
400	struct xfs_inumbers_chunk *ic = data;
401	int			error;
402
403	error = ic->formatter(ic->breq, &inogrp);
404	if (error && error != -ECANCELED)
405		return error;
406
407	ic->breq->startino = XFS_AGINO_TO_INO(mp, agno, irec->ir_startino) +
408			XFS_INODES_PER_CHUNK;
409	return error;
410}
411
412/*
413 * Return inode number table for the filesystem.
414 */
415int
416xfs_inumbers(
417	struct xfs_ibulk	*breq,
418	inumbers_fmt_pf		formatter)
419{
420	struct xfs_inumbers_chunk ic = {
421		.formatter	= formatter,
422		.breq		= breq,
423	};
424	struct xfs_trans	*tp;
425	int			error = 0;
426
427	if (xfs_bulkstat_already_done(breq->mp, breq->startino))
428		return 0;
429
430	/*
431	 * Grab an empty transaction so that we can use its recursive buffer
432	 * locking abilities to detect cycles in the inobt without deadlocking.
433	 */
434	error = xfs_trans_alloc_empty(breq->mp, &tp);
435	if (error)
436		goto out;
437
438	error = xfs_inobt_walk(breq->mp, tp, breq->startino, breq->flags,
439			xfs_inumbers_walk, breq->icount, &ic);
440	xfs_trans_cancel(tp);
441out:
442
443	/*
444	 * We found some inode groups, so clear the error status and return
445	 * them.  The lastino pointer will point directly at the inode that
446	 * triggered any error that occurred, so on the next call the error
447	 * will be triggered again and propagated to userspace as there will be
448	 * no formatted inode groups in the buffer.
449	 */
450	if (breq->ocount > 0)
451		error = 0;
452
453	return error;
454}
455
456/* Convert an inumbers (v5) struct to a inogrp (v1) struct. */
457void
458xfs_inumbers_to_inogrp(
459	struct xfs_inogrp		*ig1,
460	const struct xfs_inumbers	*ig)
461{
462	/* memset is needed here because of padding holes in the structure. */
463	memset(ig1, 0, sizeof(struct xfs_inogrp));
464	ig1->xi_startino = ig->xi_startino;
465	ig1->xi_alloccount = ig->xi_alloccount;
466	ig1->xi_allocmask = ig->xi_allocmask;
467}
468