xref: /linux/fs/bcachefs/fs-ioctl.c (revision 300a90b2cb5d442879e6398920c49aebbd5c8e40) 
1 // SPDX-License-Identifier: GPL-2.0
2 #ifndef NO_BCACHEFS_FS
3 
4 #include "bcachefs.h"
5 #include "chardev.h"
6 #include "dirent.h"
7 #include "fs.h"
8 #include "fs-common.h"
9 #include "fs-ioctl.h"
10 #include "quota.h"
11 
12 #include <linux/compat.h>
13 #include <linux/fsnotify.h>
14 #include <linux/mount.h>
15 #include <linux/namei.h>
16 #include <linux/security.h>
17 #include <linux/writeback.h>
18 
19 #define FS_IOC_GOINGDOWN	     _IOR('X', 125, __u32)
20 #define FSOP_GOING_FLAGS_DEFAULT	0x0	/* going down */
21 #define FSOP_GOING_FLAGS_LOGFLUSH	0x1	/* flush log but not data */
22 #define FSOP_GOING_FLAGS_NOLOGFLUSH	0x2	/* don't flush log nor data */
23 
24 struct flags_set {
25 	unsigned		mask;
26 	unsigned		flags;
27 
28 	unsigned		projid;
29 
30 	bool			set_projinherit;
31 	bool			projinherit;
32 };
33 
34 static int bch2_inode_flags_set(struct btree_trans *trans,
35 				struct bch_inode_info *inode,
36 				struct bch_inode_unpacked *bi,
37 				void *p)
38 {
39 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
40 	/*
41 	 * We're relying on btree locking here for exclusion with other ioctl
42 	 * calls - use the flags in the btree (@bi), not inode->i_flags:
43 	 */
44 	struct flags_set *s = p;
45 	unsigned newflags = s->flags;
46 	unsigned oldflags = bi->bi_flags & s->mask;
47 
48 	if (((newflags ^ oldflags) & (BCH_INODE_append|BCH_INODE_immutable)) &&
49 	    !capable(CAP_LINUX_IMMUTABLE))
50 		return -EPERM;
51 
52 	if (!S_ISREG(bi->bi_mode) &&
53 	    !S_ISDIR(bi->bi_mode) &&
54 	    (newflags & (BCH_INODE_nodump|BCH_INODE_noatime)) != newflags)
55 		return -EINVAL;
56 
57 	if (s->set_projinherit) {
58 		bi->bi_fields_set &= ~(1 << Inode_opt_project);
59 		bi->bi_fields_set |= ((int) s->projinherit << Inode_opt_project);
60 	}
61 
62 	bi->bi_flags &= ~s->mask;
63 	bi->bi_flags |= newflags;
64 
65 	bi->bi_ctime = timespec_to_bch2_time(c, current_time(&inode->v));
66 	return 0;
67 }
68 
69 static int bch2_ioc_getflags(struct bch_inode_info *inode, int __user *arg)
70 {
71 	unsigned flags = map_flags(bch_flags_to_uflags, inode->ei_inode.bi_flags);
72 
73 	return put_user(flags, arg);
74 }
75 
76 static int bch2_ioc_setflags(struct bch_fs *c,
77 			     struct file *file,
78 			     struct bch_inode_info *inode,
79 			     void __user *arg)
80 {
81 	struct flags_set s = { .mask = map_defined(bch_flags_to_uflags) };
82 	unsigned uflags;
83 	int ret;
84 
85 	if (get_user(uflags, (int __user *) arg))
86 		return -EFAULT;
87 
88 	s.flags = map_flags_rev(bch_flags_to_uflags, uflags);
89 	if (uflags)
90 		return -EOPNOTSUPP;
91 
92 	ret = mnt_want_write_file(file);
93 	if (ret)
94 		return ret;
95 
96 	inode_lock(&inode->v);
97 	if (!inode_owner_or_capable(file_mnt_idmap(file), &inode->v)) {
98 		ret = -EACCES;
99 		goto setflags_out;
100 	}
101 
102 	mutex_lock(&inode->ei_update_lock);
103 	ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
104 		bch2_write_inode(c, inode, bch2_inode_flags_set, &s,
105 			       ATTR_CTIME);
106 	mutex_unlock(&inode->ei_update_lock);
107 
108 setflags_out:
109 	inode_unlock(&inode->v);
110 	mnt_drop_write_file(file);
111 	return ret;
112 }
113 
114 static int bch2_ioc_fsgetxattr(struct bch_inode_info *inode,
115 			       struct fsxattr __user *arg)
116 {
117 	struct fsxattr fa = { 0 };
118 
119 	fa.fsx_xflags = map_flags(bch_flags_to_xflags, inode->ei_inode.bi_flags);
120 
121 	if (inode->ei_inode.bi_fields_set & (1 << Inode_opt_project))
122 		fa.fsx_xflags |= FS_XFLAG_PROJINHERIT;
123 
124 	fa.fsx_projid = inode->ei_qid.q[QTYP_PRJ];
125 
126 	if (copy_to_user(arg, &fa, sizeof(fa)))
127 		return -EFAULT;
128 
129 	return 0;
130 }
131 
132 static int fssetxattr_inode_update_fn(struct btree_trans *trans,
133 				      struct bch_inode_info *inode,
134 				      struct bch_inode_unpacked *bi,
135 				      void *p)
136 {
137 	struct flags_set *s = p;
138 
139 	if (s->projid != bi->bi_project) {
140 		bi->bi_fields_set |= 1U << Inode_opt_project;
141 		bi->bi_project = s->projid;
142 	}
143 
144 	return bch2_inode_flags_set(trans, inode, bi, p);
145 }
146 
147 static int bch2_ioc_fssetxattr(struct bch_fs *c,
148 			       struct file *file,
149 			       struct bch_inode_info *inode,
150 			       struct fsxattr __user *arg)
151 {
152 	struct flags_set s = { .mask = map_defined(bch_flags_to_xflags) };
153 	struct fsxattr fa;
154 	int ret;
155 
156 	if (copy_from_user(&fa, arg, sizeof(fa)))
157 		return -EFAULT;
158 
159 	s.set_projinherit = true;
160 	s.projinherit = (fa.fsx_xflags & FS_XFLAG_PROJINHERIT) != 0;
161 	fa.fsx_xflags &= ~FS_XFLAG_PROJINHERIT;
162 
163 	s.flags = map_flags_rev(bch_flags_to_xflags, fa.fsx_xflags);
164 	if (fa.fsx_xflags)
165 		return -EOPNOTSUPP;
166 
167 	if (fa.fsx_projid >= U32_MAX)
168 		return -EINVAL;
169 
170 	/*
171 	 * inode fields accessible via the xattr interface are stored with a +1
172 	 * bias, so that 0 means unset:
173 	 */
174 	s.projid = fa.fsx_projid + 1;
175 
176 	ret = mnt_want_write_file(file);
177 	if (ret)
178 		return ret;
179 
180 	inode_lock(&inode->v);
181 	if (!inode_owner_or_capable(file_mnt_idmap(file), &inode->v)) {
182 		ret = -EACCES;
183 		goto err;
184 	}
185 
186 	mutex_lock(&inode->ei_update_lock);
187 	ret   = bch2_subvol_is_ro(c, inode->ei_subvol) ?:
188 		bch2_set_projid(c, inode, fa.fsx_projid) ?:
189 		bch2_write_inode(c, inode, fssetxattr_inode_update_fn, &s,
190 			       ATTR_CTIME);
191 	mutex_unlock(&inode->ei_update_lock);
192 err:
193 	inode_unlock(&inode->v);
194 	mnt_drop_write_file(file);
195 	return ret;
196 }
197 
198 static int bch2_reinherit_attrs_fn(struct btree_trans *trans,
199 				   struct bch_inode_info *inode,
200 				   struct bch_inode_unpacked *bi,
201 				   void *p)
202 {
203 	struct bch_inode_info *dir = p;
204 
205 	return !bch2_reinherit_attrs(bi, &dir->ei_inode);
206 }
207 
208 static int bch2_ioc_reinherit_attrs(struct bch_fs *c,
209 				    struct file *file,
210 				    struct bch_inode_info *src,
211 				    const char __user *name)
212 {
213 	struct bch_hash_info hash = bch2_hash_info_init(c, &src->ei_inode);
214 	struct bch_inode_info *dst;
215 	struct inode *vinode = NULL;
216 	char *kname = NULL;
217 	struct qstr qstr;
218 	int ret = 0;
219 	subvol_inum inum;
220 
221 	kname = kmalloc(BCH_NAME_MAX + 1, GFP_KERNEL);
222 	if (!kname)
223 		return -ENOMEM;
224 
225 	ret = strncpy_from_user(kname, name, BCH_NAME_MAX);
226 	if (unlikely(ret < 0))
227 		goto err1;
228 
229 	qstr.len	= ret;
230 	qstr.name	= kname;
231 
232 	ret = bch2_dirent_lookup(c, inode_inum(src), &hash, &qstr, &inum);
233 	if (ret)
234 		goto err1;
235 
236 	vinode = bch2_vfs_inode_get(c, inum);
237 	ret = PTR_ERR_OR_ZERO(vinode);
238 	if (ret)
239 		goto err1;
240 
241 	dst = to_bch_ei(vinode);
242 
243 	ret = mnt_want_write_file(file);
244 	if (ret)
245 		goto err2;
246 
247 	bch2_lock_inodes(INODE_UPDATE_LOCK, src, dst);
248 
249 	if (inode_attr_changing(src, dst, Inode_opt_project)) {
250 		ret = bch2_fs_quota_transfer(c, dst,
251 					     src->ei_qid,
252 					     1 << QTYP_PRJ,
253 					     KEY_TYPE_QUOTA_PREALLOC);
254 		if (ret)
255 			goto err3;
256 	}
257 
258 	ret = bch2_write_inode(c, dst, bch2_reinherit_attrs_fn, src, 0);
259 err3:
260 	bch2_unlock_inodes(INODE_UPDATE_LOCK, src, dst);
261 
262 	/* return true if we did work */
263 	if (ret >= 0)
264 		ret = !ret;
265 
266 	mnt_drop_write_file(file);
267 err2:
268 	iput(vinode);
269 err1:
270 	kfree(kname);
271 
272 	return ret;
273 }
274 
275 static int bch2_ioc_getversion(struct bch_inode_info *inode, u32 __user *arg)
276 {
277 	return put_user(inode->v.i_generation, arg);
278 }
279 
280 static int bch2_ioc_getlabel(struct bch_fs *c, char __user *user_label)
281 {
282 	int ret;
283 	size_t len;
284 	char label[BCH_SB_LABEL_SIZE];
285 
286 	BUILD_BUG_ON(BCH_SB_LABEL_SIZE >= FSLABEL_MAX);
287 
288 	mutex_lock(&c->sb_lock);
289 	memcpy(label, c->disk_sb.sb->label, BCH_SB_LABEL_SIZE);
290 	mutex_unlock(&c->sb_lock);
291 
292 	len = strnlen(label, BCH_SB_LABEL_SIZE);
293 	if (len == BCH_SB_LABEL_SIZE) {
294 		bch_warn(c,
295 			"label is too long, return the first %zu bytes",
296 			--len);
297 	}
298 
299 	ret = copy_to_user(user_label, label, len);
300 
301 	return ret ? -EFAULT : 0;
302 }
303 
304 static int bch2_ioc_setlabel(struct bch_fs *c,
305 			     struct file *file,
306 			     struct bch_inode_info *inode,
307 			     const char __user *user_label)
308 {
309 	int ret;
310 	char label[BCH_SB_LABEL_SIZE];
311 
312 	if (!capable(CAP_SYS_ADMIN))
313 		return -EPERM;
314 
315 	if (copy_from_user(label, user_label, sizeof(label)))
316 		return -EFAULT;
317 
318 	if (strnlen(label, BCH_SB_LABEL_SIZE) == BCH_SB_LABEL_SIZE) {
319 		bch_err(c,
320 			"unable to set label with more than %d bytes",
321 			BCH_SB_LABEL_SIZE - 1);
322 		return -EINVAL;
323 	}
324 
325 	ret = mnt_want_write_file(file);
326 	if (ret)
327 		return ret;
328 
329 	mutex_lock(&c->sb_lock);
330 	strscpy(c->disk_sb.sb->label, label, BCH_SB_LABEL_SIZE);
331 	mutex_unlock(&c->sb_lock);
332 
333 	ret = bch2_write_super(c);
334 
335 	mnt_drop_write_file(file);
336 	return ret;
337 }
338 
339 static int bch2_ioc_goingdown(struct bch_fs *c, u32 __user *arg)
340 {
341 	u32 flags;
342 	int ret = 0;
343 
344 	if (!capable(CAP_SYS_ADMIN))
345 		return -EPERM;
346 
347 	if (get_user(flags, arg))
348 		return -EFAULT;
349 
350 	bch_notice(c, "shutdown by ioctl type %u", flags);
351 
352 	switch (flags) {
353 	case FSOP_GOING_FLAGS_DEFAULT:
354 		ret = bdev_freeze(c->vfs_sb->s_bdev);
355 		if (ret)
356 			break;
357 		bch2_journal_flush(&c->journal);
358 		bch2_fs_emergency_read_only(c);
359 		bdev_thaw(c->vfs_sb->s_bdev);
360 		break;
361 	case FSOP_GOING_FLAGS_LOGFLUSH:
362 		bch2_journal_flush(&c->journal);
363 		fallthrough;
364 	case FSOP_GOING_FLAGS_NOLOGFLUSH:
365 		bch2_fs_emergency_read_only(c);
366 		break;
367 	default:
368 		ret = -EINVAL;
369 		break;
370 	}
371 
372 	return ret;
373 }
374 
375 static long bch2_ioctl_subvolume_create(struct bch_fs *c, struct file *filp,
376 					struct bch_ioctl_subvolume arg)
377 {
378 	struct inode *dir;
379 	struct bch_inode_info *inode;
380 	struct user_namespace *s_user_ns;
381 	struct dentry *dst_dentry;
382 	struct path src_path, dst_path;
383 	int how = LOOKUP_FOLLOW;
384 	int error;
385 	subvol_inum snapshot_src = { 0 };
386 	unsigned lookup_flags = 0;
387 	unsigned create_flags = BCH_CREATE_SUBVOL;
388 
389 	if (arg.flags & ~(BCH_SUBVOL_SNAPSHOT_CREATE|
390 			  BCH_SUBVOL_SNAPSHOT_RO))
391 		return -EINVAL;
392 
393 	if (!(arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE) &&
394 	    (arg.src_ptr ||
395 	     (arg.flags & BCH_SUBVOL_SNAPSHOT_RO)))
396 		return -EINVAL;
397 
398 	if (arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE)
399 		create_flags |= BCH_CREATE_SNAPSHOT;
400 
401 	if (arg.flags & BCH_SUBVOL_SNAPSHOT_RO)
402 		create_flags |= BCH_CREATE_SNAPSHOT_RO;
403 
404 	if (arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE) {
405 		/* sync_inodes_sb enforce s_umount is locked */
406 		down_read(&c->vfs_sb->s_umount);
407 		sync_inodes_sb(c->vfs_sb);
408 		up_read(&c->vfs_sb->s_umount);
409 	}
410 retry:
411 	if (arg.src_ptr) {
412 		error = user_path_at(arg.dirfd,
413 				(const char __user *)(unsigned long)arg.src_ptr,
414 				how, &src_path);
415 		if (error)
416 			goto err1;
417 
418 		if (src_path.dentry->d_sb->s_fs_info != c) {
419 			path_put(&src_path);
420 			error = -EXDEV;
421 			goto err1;
422 		}
423 
424 		snapshot_src = inode_inum(to_bch_ei(src_path.dentry->d_inode));
425 	}
426 
427 	dst_dentry = user_path_create(arg.dirfd,
428 			(const char __user *)(unsigned long)arg.dst_ptr,
429 			&dst_path, lookup_flags);
430 	error = PTR_ERR_OR_ZERO(dst_dentry);
431 	if (error)
432 		goto err2;
433 
434 	if (dst_dentry->d_sb->s_fs_info != c) {
435 		error = -EXDEV;
436 		goto err3;
437 	}
438 
439 	if (dst_dentry->d_inode) {
440 		error = -BCH_ERR_EEXIST_subvolume_create;
441 		goto err3;
442 	}
443 
444 	dir = dst_path.dentry->d_inode;
445 	if (IS_DEADDIR(dir)) {
446 		error = -BCH_ERR_ENOENT_directory_dead;
447 		goto err3;
448 	}
449 
450 	s_user_ns = dir->i_sb->s_user_ns;
451 	if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
452 	    !kgid_has_mapping(s_user_ns, current_fsgid())) {
453 		error = -EOVERFLOW;
454 		goto err3;
455 	}
456 
457 	error = inode_permission(file_mnt_idmap(filp),
458 				 dir, MAY_WRITE | MAY_EXEC);
459 	if (error)
460 		goto err3;
461 
462 	if (!IS_POSIXACL(dir))
463 		arg.mode &= ~current_umask();
464 
465 	error = security_path_mkdir(&dst_path, dst_dentry, arg.mode);
466 	if (error)
467 		goto err3;
468 
469 	if ((arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE) &&
470 	    !arg.src_ptr)
471 		snapshot_src.subvol = inode_inum(to_bch_ei(dir)).subvol;
472 
473 	down_write(&c->snapshot_create_lock);
474 	inode = __bch2_create(file_mnt_idmap(filp), to_bch_ei(dir),
475 			      dst_dentry, arg.mode|S_IFDIR,
476 			      0, snapshot_src, create_flags);
477 	up_write(&c->snapshot_create_lock);
478 
479 	error = PTR_ERR_OR_ZERO(inode);
480 	if (error)
481 		goto err3;
482 
483 	d_instantiate(dst_dentry, &inode->v);
484 	fsnotify_mkdir(dir, dst_dentry);
485 err3:
486 	done_path_create(&dst_path, dst_dentry);
487 err2:
488 	if (arg.src_ptr)
489 		path_put(&src_path);
490 
491 	if (retry_estale(error, lookup_flags)) {
492 		lookup_flags |= LOOKUP_REVAL;
493 		goto retry;
494 	}
495 err1:
496 	return error;
497 }
498 
499 static long bch2_ioctl_subvolume_destroy(struct bch_fs *c, struct file *filp,
500 				struct bch_ioctl_subvolume arg)
501 {
502 	const char __user *name = (void __user *)(unsigned long)arg.dst_ptr;
503 	struct path path;
504 	struct inode *dir;
505 	struct dentry *victim;
506 	int ret = 0;
507 
508 	if (arg.flags)
509 		return -EINVAL;
510 
511 	victim = user_path_locked_at(arg.dirfd, name, &path);
512 	if (IS_ERR(victim))
513 		return PTR_ERR(victim);
514 
515 	dir = d_inode(path.dentry);
516 	if (victim->d_sb->s_fs_info != c) {
517 		ret = -EXDEV;
518 		goto err;
519 	}
520 	if (!d_is_positive(victim)) {
521 		ret = -ENOENT;
522 		goto err;
523 	}
524 	ret = __bch2_unlink(dir, victim, true);
525 	if (!ret) {
526 		fsnotify_rmdir(dir, victim);
527 		d_delete(victim);
528 	}
529 err:
530 	inode_unlock(dir);
531 	dput(victim);
532 	path_put(&path);
533 	return ret;
534 }
535 
536 long bch2_fs_file_ioctl(struct file *file, unsigned cmd, unsigned long arg)
537 {
538 	struct bch_inode_info *inode = file_bch_inode(file);
539 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
540 	long ret;
541 
542 	switch (cmd) {
543 	case FS_IOC_GETFLAGS:
544 		ret = bch2_ioc_getflags(inode, (int __user *) arg);
545 		break;
546 
547 	case FS_IOC_SETFLAGS:
548 		ret = bch2_ioc_setflags(c, file, inode, (int __user *) arg);
549 		break;
550 
551 	case FS_IOC_FSGETXATTR:
552 		ret = bch2_ioc_fsgetxattr(inode, (void __user *) arg);
553 		break;
554 
555 	case FS_IOC_FSSETXATTR:
556 		ret = bch2_ioc_fssetxattr(c, file, inode,
557 					  (void __user *) arg);
558 		break;
559 
560 	case BCHFS_IOC_REINHERIT_ATTRS:
561 		ret = bch2_ioc_reinherit_attrs(c, file, inode,
562 					       (void __user *) arg);
563 		break;
564 
565 	case FS_IOC_GETVERSION:
566 		ret = bch2_ioc_getversion(inode, (u32 __user *) arg);
567 		break;
568 
569 	case FS_IOC_SETVERSION:
570 		ret = -ENOTTY;
571 		break;
572 
573 	case FS_IOC_GETFSLABEL:
574 		ret = bch2_ioc_getlabel(c, (void __user *) arg);
575 		break;
576 
577 	case FS_IOC_SETFSLABEL:
578 		ret = bch2_ioc_setlabel(c, file, inode, (const void __user *) arg);
579 		break;
580 
581 	case FS_IOC_GOINGDOWN:
582 		ret = bch2_ioc_goingdown(c, (u32 __user *) arg);
583 		break;
584 
585 	case BCH_IOCTL_SUBVOLUME_CREATE: {
586 		struct bch_ioctl_subvolume i;
587 
588 		ret = copy_from_user(&i, (void __user *) arg, sizeof(i))
589 			? -EFAULT
590 			: bch2_ioctl_subvolume_create(c, file, i);
591 		break;
592 	}
593 
594 	case BCH_IOCTL_SUBVOLUME_DESTROY: {
595 		struct bch_ioctl_subvolume i;
596 
597 		ret = copy_from_user(&i, (void __user *) arg, sizeof(i))
598 			? -EFAULT
599 			: bch2_ioctl_subvolume_destroy(c, file, i);
600 		break;
601 	}
602 
603 	default:
604 		ret = bch2_fs_ioctl(c, cmd, (void __user *) arg);
605 		break;
606 	}
607 
608 	return bch2_err_class(ret);
609 }
610 
611 #ifdef CONFIG_COMPAT
612 long bch2_compat_fs_ioctl(struct file *file, unsigned cmd, unsigned long arg)
613 {
614 	/* These are just misnamed, they actually get/put from/to user an int */
615 	switch (cmd) {
616 	case FS_IOC32_GETFLAGS:
617 		cmd = FS_IOC_GETFLAGS;
618 		break;
619 	case FS_IOC32_SETFLAGS:
620 		cmd = FS_IOC_SETFLAGS;
621 		break;
622 	case FS_IOC32_GETVERSION:
623 		cmd = FS_IOC_GETVERSION;
624 		break;
625 	case FS_IOC_GETFSLABEL:
626 	case FS_IOC_SETFSLABEL:
627 		break;
628 	default:
629 		return -ENOIOCTLCMD;
630 	}
631 	return bch2_fs_file_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
632 }
633 #endif
634 
635 #endif /* NO_BCACHEFS_FS */
636