xfs_itable.c

  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
 39static inline bool
 40want_metadir_file(
 41	struct xfs_inode	*ip,
 42	struct xfs_ibulk	*breq)
 43{
 44	return xfs_is_metadir_inode(ip) && (breq->flags & XFS_IBULK_METADIR);
 45}
 46
 47/*
 48 * Fill out the bulkstat info for a single inode and report it somewhere.
 49 *
 50 * bc->breq->lastino is effectively the inode cursor as we walk through the
 51 * filesystem.  Therefore, we update it any time we need to move the cursor
 52 * forward, regardless of whether or not we're sending any bstat information
 53 * back to userspace.  If the inode is internal metadata or, has been freed
 54 * out from under us, we just simply keep going.
 55 *
 56 * However, if any other type of error happens we want to stop right where we
 57 * are so that userspace will call back with exact number of the bad inode and
 58 * we can send back an error code.
 59 *
 60 * Note that if the formatter tells us there's no space left in the buffer we
 61 * move the cursor forward and abort the walk.
 62 */
 63STATIC int
 64xfs_bulkstat_one_int(
 65	struct xfs_mount	*mp,
 66	struct mnt_idmap	*idmap,
 67	struct xfs_trans	*tp,
 68	xfs_ino_t		ino,
 69	struct xfs_bstat_chunk	*bc)
 70{
 71	struct user_namespace	*sb_userns = mp->m_super->s_user_ns;
 72	struct xfs_inode	*ip;		/* incore inode pointer */
 73	struct inode		*inode;
 74	struct xfs_bulkstat	*buf = bc->buf;
 75	xfs_extnum_t		nextents;
 76	int			error = -EINVAL;
 77	vfsuid_t		vfsuid;
 78	vfsgid_t		vfsgid;
 79
 80	error = xfs_iget(mp, tp, ino,
 81			 (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED),
 82			 XFS_ILOCK_SHARED, &ip);
 83	if (error == -ENOENT || error == -EINVAL)
 84		goto out_advance;
 85	if (error)
 86		goto out;
 87
 88	/* Reload the incore unlinked list to avoid failure in inodegc. */
 89	if (xfs_inode_unlinked_incomplete(ip)) {
 90		error = xfs_inode_reload_unlinked_bucket(tp, ip);
 91		if (error) {
 92			xfs_iunlock(ip, XFS_ILOCK_SHARED);
 93			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 94			xfs_irele(ip);
 95			return error;
 96		}
 97	}
 98
 99	ASSERT(ip != NULL);
100	ASSERT(ip->i_imap.im_blkno != 0);
101	inode = VFS_I(ip);
102	vfsuid = i_uid_into_vfsuid(idmap, inode);
103	vfsgid = i_gid_into_vfsgid(idmap, inode);
104
105	/*
106	 * If caller wants files from the metadata directories, push out the
107	 * bare minimum information for enabling scrub.
108	 */
109	if (want_metadir_file(ip, bc->breq)) {
110		memset(buf, 0, sizeof(*buf));
111		buf->bs_ino = ino;
112		buf->bs_gen = inode->i_generation;
113		buf->bs_mode = inode->i_mode & S_IFMT;
114		xfs_bulkstat_health(ip, buf);
115		buf->bs_version = XFS_BULKSTAT_VERSION_V5;
116		xfs_iunlock(ip, XFS_ILOCK_SHARED);
117		xfs_irele(ip);
118
119		error = bc->formatter(bc->breq, buf);
120		if (!error || error == -ECANCELED)
121			goto out_advance;
122		goto out;
123	}
124
125	/* If this is a private inode, don't leak its details to userspace. */
126	if (IS_PRIVATE(inode) || xfs_is_sb_inum(mp, ino)) {
127		xfs_iunlock(ip, XFS_ILOCK_SHARED);
128		xfs_irele(ip);
129		error = -EINVAL;
130		goto out_advance;
131	}
132
133	/* xfs_iget returns the following without needing
134	 * further change.
135	 */
136	buf->bs_projectid = ip->i_projid;
137	buf->bs_ino = ino;
138	buf->bs_uid = from_kuid(sb_userns, vfsuid_into_kuid(vfsuid));
139	buf->bs_gid = from_kgid(sb_userns, vfsgid_into_kgid(vfsgid));
140	buf->bs_size = ip->i_disk_size;
141
142	buf->bs_nlink = inode->i_nlink;
143	buf->bs_atime = inode_get_atime_sec(inode);
144	buf->bs_atime_nsec = inode_get_atime_nsec(inode);
145	buf->bs_mtime = inode_get_mtime_sec(inode);
146	buf->bs_mtime_nsec = inode_get_mtime_nsec(inode);
147	buf->bs_ctime = inode_get_ctime_sec(inode);
148	buf->bs_ctime_nsec = inode_get_ctime_nsec(inode);
149	buf->bs_gen = inode->i_generation;
150	buf->bs_mode = inode->i_mode;
151
152	buf->bs_xflags = xfs_ip2xflags(ip);
153	buf->bs_extsize_blks = ip->i_extsize;
154
155	nextents = xfs_ifork_nextents(&ip->i_df);
156	if (!(bc->breq->flags & XFS_IBULK_NREXT64))
157		buf->bs_extents = min(nextents, XFS_MAX_EXTCNT_DATA_FORK_SMALL);
158	else
159		buf->bs_extents64 = nextents;
160
161	xfs_bulkstat_health(ip, buf);
162	buf->bs_aextents = xfs_ifork_nextents(&ip->i_af);
163	buf->bs_forkoff = xfs_inode_fork_boff(ip);
164	buf->bs_version = XFS_BULKSTAT_VERSION_V5;
165
166	if (xfs_has_v3inodes(mp)) {
167		buf->bs_btime = ip->i_crtime.tv_sec;
168		buf->bs_btime_nsec = ip->i_crtime.tv_nsec;
169		if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
170			buf->bs_cowextsize_blks = ip->i_cowextsize;
171	}
172
173	switch (ip->i_df.if_format) {
174	case XFS_DINODE_FMT_DEV:
175		buf->bs_rdev = sysv_encode_dev(inode->i_rdev);
176		buf->bs_blksize = BLKDEV_IOSIZE;
177		buf->bs_blocks = 0;
178		break;
179	case XFS_DINODE_FMT_LOCAL:
180		buf->bs_rdev = 0;
181		buf->bs_blksize = mp->m_sb.sb_blocksize;
182		buf->bs_blocks = 0;
183		break;
184	case XFS_DINODE_FMT_EXTENTS:
185	case XFS_DINODE_FMT_BTREE:
186		buf->bs_rdev = 0;
187		buf->bs_blksize = mp->m_sb.sb_blocksize;
188		buf->bs_blocks = ip->i_nblocks + ip->i_delayed_blks;
189		break;
190	}
191	xfs_iunlock(ip, XFS_ILOCK_SHARED);
192	xfs_irele(ip);
193
194	error = bc->formatter(bc->breq, buf);
195	if (error == -ECANCELED)
196		goto out_advance;
197	if (error)
198		goto out;
199
200out_advance:
201	/*
202	 * Advance the cursor to the inode that comes after the one we just
203	 * looked at.  We want the caller to move along if the bulkstat
204	 * information was copied successfully; if we tried to grab the inode
205	 * but it's no longer allocated; or if it's internal metadata.
206	 */
207	bc->breq->startino = ino + 1;
208out:
209	return error;
210}
211
212/* Bulkstat a single inode. */
213int
214xfs_bulkstat_one(
215	struct xfs_ibulk	*breq,
216	bulkstat_one_fmt_pf	formatter)
217{
218	struct xfs_bstat_chunk	bc = {
219		.formatter	= formatter,
220		.breq		= breq,
221	};
222	struct xfs_trans	*tp;
223	int			error;
224
225	if (breq->idmap != &nop_mnt_idmap) {
226		xfs_warn_ratelimited(breq->mp,
227			"bulkstat not supported inside of idmapped mounts.");
228		return -EINVAL;
229	}
230
231	ASSERT(breq->icount == 1);
232
233	bc.buf = kzalloc(sizeof(struct xfs_bulkstat),
234			GFP_KERNEL | __GFP_RETRY_MAYFAIL);
235	if (!bc.buf)
236		return -ENOMEM;
237
238	/*
239	 * Grab an empty transaction so that we can use its recursive buffer
240	 * locking abilities to detect cycles in the inobt without deadlocking.
241	 */
242	tp = xfs_trans_alloc_empty(breq->mp);
243	error = xfs_bulkstat_one_int(breq->mp, breq->idmap, tp,
244			breq->startino, &bc);
245	xfs_trans_cancel(tp);
246	kfree(bc.buf);
247
248	/*
249	 * If we reported one inode to userspace then we abort because we hit
250	 * the end of the buffer.  Don't leak that back to userspace.
251	 */
252	if (error == -ECANCELED)
253		error = 0;
254
255	return error;
256}
257
258static int
259xfs_bulkstat_iwalk(
260	struct xfs_mount	*mp,
261	struct xfs_trans	*tp,
262	xfs_ino_t		ino,
263	void			*data)
264{
265	struct xfs_bstat_chunk	*bc = data;
266	int			error;
267
268	error = xfs_bulkstat_one_int(mp, bc->breq->idmap, tp, ino, data);
269	/* bulkstat just skips over missing inodes */
270	if (error == -ENOENT || error == -EINVAL)
271		return 0;
272	return error;
273}
274
275/*
276 * Check the incoming lastino parameter.
277 *
278 * We allow any inode value that could map to physical space inside the
279 * filesystem because if there are no inodes there, bulkstat moves on to the
280 * next chunk.  In other words, the magic agino value of zero takes us to the
281 * first chunk in the AG, and an agino value past the end of the AG takes us to
282 * the first chunk in the next AG.
283 *
284 * Therefore we can end early if the requested inode is beyond the end of the
285 * filesystem or doesn't map properly.
286 */
287static inline bool
288xfs_bulkstat_already_done(
289	struct xfs_mount	*mp,
290	xfs_ino_t		startino)
291{
292	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, startino);
293	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, startino);
294
295	return agno >= mp->m_sb.sb_agcount ||
296	       startino != XFS_AGINO_TO_INO(mp, agno, agino);
297}
298
299/* Return stat information in bulk (by-inode) for the filesystem. */
300int
301xfs_bulkstat(
302	struct xfs_ibulk	*breq,
303	bulkstat_one_fmt_pf	formatter)
304{
305	struct xfs_bstat_chunk	bc = {
306		.formatter	= formatter,
307		.breq		= breq,
308	};
309	struct xfs_trans	*tp;
310	int			error;
311
312	if (breq->idmap != &nop_mnt_idmap) {
313		xfs_warn_ratelimited(breq->mp,
314			"bulkstat not supported inside of idmapped mounts.");
315		return -EINVAL;
316	}
317	if (xfs_bulkstat_already_done(breq->mp, breq->startino))
318		return 0;
319
320	bc.buf = kzalloc(sizeof(struct xfs_bulkstat),
321			GFP_KERNEL | __GFP_RETRY_MAYFAIL);
322	if (!bc.buf)
323		return -ENOMEM;
324
325	/*
326	 * Grab an empty transaction so that we can use its recursive buffer
327	 * locking abilities to detect cycles in the inobt without deadlocking.
328	 */
329	tp = xfs_trans_alloc_empty(breq->mp);
330	error = xfs_iwalk(breq->mp, tp, breq->startino, breq->iwalk_flags,
331			xfs_bulkstat_iwalk, breq->icount, &bc);
332	xfs_trans_cancel(tp);
333	kfree(bc.buf);
334
335	/*
336	 * We found some inodes, so clear the error status and return them.
337	 * The lastino pointer will point directly at the inode that triggered
338	 * any error that occurred, so on the next call the error will be
339	 * triggered again and propagated to userspace as there will be no
340	 * formatted inodes in the buffer.
341	 */
342	if (breq->ocount > 0)
343		error = 0;
344
345	return error;
346}
347
348/* Convert bulkstat (v5) to bstat (v1). */
349void
350xfs_bulkstat_to_bstat(
351	struct xfs_mount		*mp,
352	struct xfs_bstat		*bs1,
353	const struct xfs_bulkstat	*bstat)
354{
355	/* memset is needed here because of padding holes in the structure. */
356	memset(bs1, 0, sizeof(struct xfs_bstat));
357	bs1->bs_ino = bstat->bs_ino;
358	bs1->bs_mode = bstat->bs_mode;
359	bs1->bs_nlink = bstat->bs_nlink;
360	bs1->bs_uid = bstat->bs_uid;
361	bs1->bs_gid = bstat->bs_gid;
362	bs1->bs_rdev = bstat->bs_rdev;
363	bs1->bs_blksize = bstat->bs_blksize;
364	bs1->bs_size = bstat->bs_size;
365	bs1->bs_atime.tv_sec = bstat->bs_atime;
366	bs1->bs_mtime.tv_sec = bstat->bs_mtime;
367	bs1->bs_ctime.tv_sec = bstat->bs_ctime;
368	bs1->bs_atime.tv_nsec = bstat->bs_atime_nsec;
369	bs1->bs_mtime.tv_nsec = bstat->bs_mtime_nsec;
370	bs1->bs_ctime.tv_nsec = bstat->bs_ctime_nsec;
371	bs1->bs_blocks = bstat->bs_blocks;
372	bs1->bs_xflags = bstat->bs_xflags;
373	bs1->bs_extsize = XFS_FSB_TO_B(mp, bstat->bs_extsize_blks);
374	bs1->bs_extents = bstat->bs_extents;
375	bs1->bs_gen = bstat->bs_gen;
376	bs1->bs_projid_lo = bstat->bs_projectid & 0xFFFF;
377	bs1->bs_forkoff = bstat->bs_forkoff;
378	bs1->bs_projid_hi = bstat->bs_projectid >> 16;
379	bs1->bs_sick = bstat->bs_sick;
380	bs1->bs_checked = bstat->bs_checked;
381	bs1->bs_cowextsize = XFS_FSB_TO_B(mp, bstat->bs_cowextsize_blks);
382	bs1->bs_dmevmask = 0;
383	bs1->bs_dmstate = 0;
384	bs1->bs_aextents = bstat->bs_aextents;
385}
386
387struct xfs_inumbers_chunk {
388	inumbers_fmt_pf		formatter;
389	struct xfs_ibulk	*breq;
390};
391
392/*
393 * INUMBERS
394 * ========
395 * This is how we export inode btree records to userspace, so that XFS tools
396 * can figure out where inodes are allocated.
397 */
398
399/*
400 * Format the inode group structure and report it somewhere.
401 *
402 * Similar to xfs_bulkstat_one_int, lastino is the inode cursor as we walk
403 * through the filesystem so we move it forward unless there was a runtime
404 * error.  If the formatter tells us the buffer is now full we also move the
405 * cursor forward and abort the walk.
406 */
407STATIC int
408xfs_inumbers_walk(
409	struct xfs_mount	*mp,
410	struct xfs_trans	*tp,
411	xfs_agnumber_t		agno,
412	const struct xfs_inobt_rec_incore *irec,
413	void			*data)
414{
415	struct xfs_inumbers	inogrp = {
416		.xi_startino	= XFS_AGINO_TO_INO(mp, agno, irec->ir_startino),
417		.xi_alloccount	= irec->ir_count - irec->ir_freecount,
418		.xi_allocmask	= ~irec->ir_free,
419		.xi_version	= XFS_INUMBERS_VERSION_V5,
420	};
421	struct xfs_inumbers_chunk *ic = data;
422	int			error;
423
424	error = ic->formatter(ic->breq, &inogrp);
425	if (error && error != -ECANCELED)
426		return error;
427
428	ic->breq->startino = XFS_AGINO_TO_INO(mp, agno, irec->ir_startino) +
429			XFS_INODES_PER_CHUNK;
430	return error;
431}
432
433/*
434 * Return inode number table for the filesystem.
435 */
436int
437xfs_inumbers(
438	struct xfs_ibulk	*breq,
439	inumbers_fmt_pf		formatter)
440{
441	struct xfs_inumbers_chunk ic = {
442		.formatter	= formatter,
443		.breq		= breq,
444	};
445	struct xfs_trans	*tp;
446	int			error = 0;
447
448	if (xfs_bulkstat_already_done(breq->mp, breq->startino))
449		return 0;
450
451	/*
452	 * Grab an empty transaction so that we can use its recursive buffer
453	 * locking abilities to detect cycles in the inobt without deadlocking.
454	 */
455	tp = xfs_trans_alloc_empty(breq->mp);
456	error = xfs_inobt_walk(breq->mp, tp, breq->startino, breq->iwalk_flags,
457			xfs_inumbers_walk, breq->icount, &ic);
458	xfs_trans_cancel(tp);
459
460	/*
461	 * We found some inode groups, so clear the error status and return
462	 * them.  The lastino pointer will point directly at the inode that
463	 * triggered any error that occurred, so on the next call the error
464	 * will be triggered again and propagated to userspace as there will be
465	 * no formatted inode groups in the buffer.
466	 */
467	if (breq->ocount > 0)
468		error = 0;
469
470	return error;
471}
472
473/* Convert an inumbers (v5) struct to a inogrp (v1) struct. */
474void
475xfs_inumbers_to_inogrp(
476	struct xfs_inogrp		*ig1,
477	const struct xfs_inumbers	*ig)
478{
479	/* memset is needed here because of padding holes in the structure. */
480	memset(ig1, 0, sizeof(struct xfs_inogrp));
481	ig1->xi_startino = ig->xi_startino;
482	ig1->xi_alloccount = ig->xi_alloccount;
483	ig1->xi_allocmask = ig->xi_allocmask;
484}