xref: /openbmc/linux/fs/xfs/xfs_itable.c (revision be2b81b5)
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 
33 struct 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  */
55 STATIC int
56 xfs_bulkstat_one_int(
57 	struct xfs_mount	*mp,
58 	struct user_namespace	*mnt_userns,
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 	ASSERT(ip != NULL);
84 	ASSERT(ip->i_imap.im_blkno != 0);
85 	inode = VFS_I(ip);
86 	vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
87 	vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
88 
89 	/* xfs_iget returns the following without needing
90 	 * further change.
91 	 */
92 	buf->bs_projectid = ip->i_projid;
93 	buf->bs_ino = ino;
94 	buf->bs_uid = from_kuid(sb_userns, vfsuid_into_kuid(vfsuid));
95 	buf->bs_gid = from_kgid(sb_userns, vfsgid_into_kgid(vfsgid));
96 	buf->bs_size = ip->i_disk_size;
97 
98 	buf->bs_nlink = inode->i_nlink;
99 	buf->bs_atime = inode->i_atime.tv_sec;
100 	buf->bs_atime_nsec = inode->i_atime.tv_nsec;
101 	buf->bs_mtime = inode->i_mtime.tv_sec;
102 	buf->bs_mtime_nsec = inode->i_mtime.tv_nsec;
103 	buf->bs_ctime = inode->i_ctime.tv_sec;
104 	buf->bs_ctime_nsec = inode->i_ctime.tv_nsec;
105 	buf->bs_gen = inode->i_generation;
106 	buf->bs_mode = inode->i_mode;
107 
108 	buf->bs_xflags = xfs_ip2xflags(ip);
109 	buf->bs_extsize_blks = ip->i_extsize;
110 
111 	nextents = xfs_ifork_nextents(&ip->i_df);
112 	if (!(bc->breq->flags & XFS_IBULK_NREXT64))
113 		buf->bs_extents = min(nextents, XFS_MAX_EXTCNT_DATA_FORK_SMALL);
114 	else
115 		buf->bs_extents64 = nextents;
116 
117 	xfs_bulkstat_health(ip, buf);
118 	buf->bs_aextents = xfs_ifork_nextents(&ip->i_af);
119 	buf->bs_forkoff = xfs_inode_fork_boff(ip);
120 	buf->bs_version = XFS_BULKSTAT_VERSION_V5;
121 
122 	if (xfs_has_v3inodes(mp)) {
123 		buf->bs_btime = ip->i_crtime.tv_sec;
124 		buf->bs_btime_nsec = ip->i_crtime.tv_nsec;
125 		if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
126 			buf->bs_cowextsize_blks = ip->i_cowextsize;
127 	}
128 
129 	switch (ip->i_df.if_format) {
130 	case XFS_DINODE_FMT_DEV:
131 		buf->bs_rdev = sysv_encode_dev(inode->i_rdev);
132 		buf->bs_blksize = BLKDEV_IOSIZE;
133 		buf->bs_blocks = 0;
134 		break;
135 	case XFS_DINODE_FMT_LOCAL:
136 		buf->bs_rdev = 0;
137 		buf->bs_blksize = mp->m_sb.sb_blocksize;
138 		buf->bs_blocks = 0;
139 		break;
140 	case XFS_DINODE_FMT_EXTENTS:
141 	case XFS_DINODE_FMT_BTREE:
142 		buf->bs_rdev = 0;
143 		buf->bs_blksize = mp->m_sb.sb_blocksize;
144 		buf->bs_blocks = ip->i_nblocks + ip->i_delayed_blks;
145 		break;
146 	}
147 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
148 	xfs_irele(ip);
149 
150 	error = bc->formatter(bc->breq, buf);
151 	if (error == -ECANCELED)
152 		goto out_advance;
153 	if (error)
154 		goto out;
155 
156 out_advance:
157 	/*
158 	 * Advance the cursor to the inode that comes after the one we just
159 	 * looked at.  We want the caller to move along if the bulkstat
160 	 * information was copied successfully; if we tried to grab the inode
161 	 * but it's no longer allocated; or if it's internal metadata.
162 	 */
163 	bc->breq->startino = ino + 1;
164 out:
165 	return error;
166 }
167 
168 /* Bulkstat a single inode. */
169 int
170 xfs_bulkstat_one(
171 	struct xfs_ibulk	*breq,
172 	bulkstat_one_fmt_pf	formatter)
173 {
174 	struct xfs_bstat_chunk	bc = {
175 		.formatter	= formatter,
176 		.breq		= breq,
177 	};
178 	struct xfs_trans	*tp;
179 	int			error;
180 
181 	if (breq->mnt_userns != &init_user_ns) {
182 		xfs_warn_ratelimited(breq->mp,
183 			"bulkstat not supported inside of idmapped mounts.");
184 		return -EINVAL;
185 	}
186 
187 	ASSERT(breq->icount == 1);
188 
189 	bc.buf = kmem_zalloc(sizeof(struct xfs_bulkstat),
190 			KM_MAYFAIL);
191 	if (!bc.buf)
192 		return -ENOMEM;
193 
194 	/*
195 	 * Grab an empty transaction so that we can use its recursive buffer
196 	 * locking abilities to detect cycles in the inobt without deadlocking.
197 	 */
198 	error = xfs_trans_alloc_empty(breq->mp, &tp);
199 	if (error)
200 		goto out;
201 
202 	error = xfs_bulkstat_one_int(breq->mp, breq->mnt_userns, tp,
203 			breq->startino, &bc);
204 	xfs_trans_cancel(tp);
205 out:
206 	kmem_free(bc.buf);
207 
208 	/*
209 	 * If we reported one inode to userspace then we abort because we hit
210 	 * the end of the buffer.  Don't leak that back to userspace.
211 	 */
212 	if (error == -ECANCELED)
213 		error = 0;
214 
215 	return error;
216 }
217 
218 static int
219 xfs_bulkstat_iwalk(
220 	struct xfs_mount	*mp,
221 	struct xfs_trans	*tp,
222 	xfs_ino_t		ino,
223 	void			*data)
224 {
225 	struct xfs_bstat_chunk	*bc = data;
226 	int			error;
227 
228 	error = xfs_bulkstat_one_int(mp, bc->breq->mnt_userns, tp, ino, data);
229 	/* bulkstat just skips over missing inodes */
230 	if (error == -ENOENT || error == -EINVAL)
231 		return 0;
232 	return error;
233 }
234 
235 /*
236  * Check the incoming lastino parameter.
237  *
238  * We allow any inode value that could map to physical space inside the
239  * filesystem because if there are no inodes there, bulkstat moves on to the
240  * next chunk.  In other words, the magic agino value of zero takes us to the
241  * first chunk in the AG, and an agino value past the end of the AG takes us to
242  * the first chunk in the next AG.
243  *
244  * Therefore we can end early if the requested inode is beyond the end of the
245  * filesystem or doesn't map properly.
246  */
247 static inline bool
248 xfs_bulkstat_already_done(
249 	struct xfs_mount	*mp,
250 	xfs_ino_t		startino)
251 {
252 	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, startino);
253 	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, startino);
254 
255 	return agno >= mp->m_sb.sb_agcount ||
256 	       startino != XFS_AGINO_TO_INO(mp, agno, agino);
257 }
258 
259 /* Return stat information in bulk (by-inode) for the filesystem. */
260 int
261 xfs_bulkstat(
262 	struct xfs_ibulk	*breq,
263 	bulkstat_one_fmt_pf	formatter)
264 {
265 	struct xfs_bstat_chunk	bc = {
266 		.formatter	= formatter,
267 		.breq		= breq,
268 	};
269 	struct xfs_trans	*tp;
270 	unsigned int		iwalk_flags = 0;
271 	int			error;
272 
273 	if (breq->mnt_userns != &init_user_ns) {
274 		xfs_warn_ratelimited(breq->mp,
275 			"bulkstat not supported inside of idmapped mounts.");
276 		return -EINVAL;
277 	}
278 	if (xfs_bulkstat_already_done(breq->mp, breq->startino))
279 		return 0;
280 
281 	bc.buf = kmem_zalloc(sizeof(struct xfs_bulkstat),
282 			KM_MAYFAIL);
283 	if (!bc.buf)
284 		return -ENOMEM;
285 
286 	/*
287 	 * Grab an empty transaction so that we can use its recursive buffer
288 	 * locking abilities to detect cycles in the inobt without deadlocking.
289 	 */
290 	error = xfs_trans_alloc_empty(breq->mp, &tp);
291 	if (error)
292 		goto out;
293 
294 	if (breq->flags & XFS_IBULK_SAME_AG)
295 		iwalk_flags |= XFS_IWALK_SAME_AG;
296 
297 	error = xfs_iwalk(breq->mp, tp, breq->startino, iwalk_flags,
298 			xfs_bulkstat_iwalk, breq->icount, &bc);
299 	xfs_trans_cancel(tp);
300 out:
301 	kmem_free(bc.buf);
302 
303 	/*
304 	 * We found some inodes, so clear the error status and return them.
305 	 * The lastino pointer will point directly at the inode that triggered
306 	 * any error that occurred, so on the next call the error will be
307 	 * triggered again and propagated to userspace as there will be no
308 	 * formatted inodes in the buffer.
309 	 */
310 	if (breq->ocount > 0)
311 		error = 0;
312 
313 	return error;
314 }
315 
316 /* Convert bulkstat (v5) to bstat (v1). */
317 void
318 xfs_bulkstat_to_bstat(
319 	struct xfs_mount		*mp,
320 	struct xfs_bstat		*bs1,
321 	const struct xfs_bulkstat	*bstat)
322 {
323 	/* memset is needed here because of padding holes in the structure. */
324 	memset(bs1, 0, sizeof(struct xfs_bstat));
325 	bs1->bs_ino = bstat->bs_ino;
326 	bs1->bs_mode = bstat->bs_mode;
327 	bs1->bs_nlink = bstat->bs_nlink;
328 	bs1->bs_uid = bstat->bs_uid;
329 	bs1->bs_gid = bstat->bs_gid;
330 	bs1->bs_rdev = bstat->bs_rdev;
331 	bs1->bs_blksize = bstat->bs_blksize;
332 	bs1->bs_size = bstat->bs_size;
333 	bs1->bs_atime.tv_sec = bstat->bs_atime;
334 	bs1->bs_mtime.tv_sec = bstat->bs_mtime;
335 	bs1->bs_ctime.tv_sec = bstat->bs_ctime;
336 	bs1->bs_atime.tv_nsec = bstat->bs_atime_nsec;
337 	bs1->bs_mtime.tv_nsec = bstat->bs_mtime_nsec;
338 	bs1->bs_ctime.tv_nsec = bstat->bs_ctime_nsec;
339 	bs1->bs_blocks = bstat->bs_blocks;
340 	bs1->bs_xflags = bstat->bs_xflags;
341 	bs1->bs_extsize = XFS_FSB_TO_B(mp, bstat->bs_extsize_blks);
342 	bs1->bs_extents = bstat->bs_extents;
343 	bs1->bs_gen = bstat->bs_gen;
344 	bs1->bs_projid_lo = bstat->bs_projectid & 0xFFFF;
345 	bs1->bs_forkoff = bstat->bs_forkoff;
346 	bs1->bs_projid_hi = bstat->bs_projectid >> 16;
347 	bs1->bs_sick = bstat->bs_sick;
348 	bs1->bs_checked = bstat->bs_checked;
349 	bs1->bs_cowextsize = XFS_FSB_TO_B(mp, bstat->bs_cowextsize_blks);
350 	bs1->bs_dmevmask = 0;
351 	bs1->bs_dmstate = 0;
352 	bs1->bs_aextents = bstat->bs_aextents;
353 }
354 
355 struct xfs_inumbers_chunk {
356 	inumbers_fmt_pf		formatter;
357 	struct xfs_ibulk	*breq;
358 };
359 
360 /*
361  * INUMBERS
362  * ========
363  * This is how we export inode btree records to userspace, so that XFS tools
364  * can figure out where inodes are allocated.
365  */
366 
367 /*
368  * Format the inode group structure and report it somewhere.
369  *
370  * Similar to xfs_bulkstat_one_int, lastino is the inode cursor as we walk
371  * through the filesystem so we move it forward unless there was a runtime
372  * error.  If the formatter tells us the buffer is now full we also move the
373  * cursor forward and abort the walk.
374  */
375 STATIC int
376 xfs_inumbers_walk(
377 	struct xfs_mount	*mp,
378 	struct xfs_trans	*tp,
379 	xfs_agnumber_t		agno,
380 	const struct xfs_inobt_rec_incore *irec,
381 	void			*data)
382 {
383 	struct xfs_inumbers	inogrp = {
384 		.xi_startino	= XFS_AGINO_TO_INO(mp, agno, irec->ir_startino),
385 		.xi_alloccount	= irec->ir_count - irec->ir_freecount,
386 		.xi_allocmask	= ~irec->ir_free,
387 		.xi_version	= XFS_INUMBERS_VERSION_V5,
388 	};
389 	struct xfs_inumbers_chunk *ic = data;
390 	int			error;
391 
392 	error = ic->formatter(ic->breq, &inogrp);
393 	if (error && error != -ECANCELED)
394 		return error;
395 
396 	ic->breq->startino = XFS_AGINO_TO_INO(mp, agno, irec->ir_startino) +
397 			XFS_INODES_PER_CHUNK;
398 	return error;
399 }
400 
401 /*
402  * Return inode number table for the filesystem.
403  */
404 int
405 xfs_inumbers(
406 	struct xfs_ibulk	*breq,
407 	inumbers_fmt_pf		formatter)
408 {
409 	struct xfs_inumbers_chunk ic = {
410 		.formatter	= formatter,
411 		.breq		= breq,
412 	};
413 	struct xfs_trans	*tp;
414 	int			error = 0;
415 
416 	if (xfs_bulkstat_already_done(breq->mp, breq->startino))
417 		return 0;
418 
419 	/*
420 	 * Grab an empty transaction so that we can use its recursive buffer
421 	 * locking abilities to detect cycles in the inobt without deadlocking.
422 	 */
423 	error = xfs_trans_alloc_empty(breq->mp, &tp);
424 	if (error)
425 		goto out;
426 
427 	error = xfs_inobt_walk(breq->mp, tp, breq->startino, breq->flags,
428 			xfs_inumbers_walk, breq->icount, &ic);
429 	xfs_trans_cancel(tp);
430 out:
431 
432 	/*
433 	 * We found some inode groups, so clear the error status and return
434 	 * them.  The lastino pointer will point directly at the inode that
435 	 * triggered any error that occurred, so on the next call the error
436 	 * will be triggered again and propagated to userspace as there will be
437 	 * no formatted inode groups in the buffer.
438 	 */
439 	if (breq->ocount > 0)
440 		error = 0;
441 
442 	return error;
443 }
444 
445 /* Convert an inumbers (v5) struct to a inogrp (v1) struct. */
446 void
447 xfs_inumbers_to_inogrp(
448 	struct xfs_inogrp		*ig1,
449 	const struct xfs_inumbers	*ig)
450 {
451 	/* memset is needed here because of padding holes in the structure. */
452 	memset(ig1, 0, sizeof(struct xfs_inogrp));
453 	ig1->xi_startino = ig->xi_startino;
454 	ig1->xi_alloccount = ig->xi_alloccount;
455 	ig1->xi_allocmask = ig->xi_allocmask;
456 }
457