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