xref: /linux/fs/bcachefs/fs-ioctl.c (revision f694f30e81c4ade358eb8c75273bac1a48f0cb8f) 
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-ioctl.h"
9 #include "namei.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 ((newflags ^ oldflags) & BCH_INODE_casefolded) {
58 #ifdef CONFIG_UNICODE
59 		int ret = 0;
60 		/* Not supported on individual files. */
61 		if (!S_ISDIR(bi->bi_mode))
62 			return -EOPNOTSUPP;
63 
64 		/*
65 		 * Make sure the dir is empty, as otherwise we'd need to
66 		 * rehash everything and update the dirent keys.
67 		 */
68 		ret = bch2_empty_dir_trans(trans, inode_inum(inode));
69 		if (ret < 0)
70 			return ret;
71 
72 		ret = bch2_request_incompat_feature(c,bcachefs_metadata_version_casefolding);
73 		if (ret)
74 			return ret;
75 
76 		bch2_check_set_feature(c, BCH_FEATURE_casefolding);
77 #else
78 		printk(KERN_ERR "Cannot use casefolding on a kernel without CONFIG_UNICODE\n");
79 		return -EOPNOTSUPP;
80 #endif
81 	}
82 
83 	if (s->set_projinherit) {
84 		bi->bi_fields_set &= ~(1 << Inode_opt_project);
85 		bi->bi_fields_set |= ((int) s->projinherit << Inode_opt_project);
86 	}
87 
88 	bi->bi_flags &= ~s->mask;
89 	bi->bi_flags |= newflags;
90 
91 	bi->bi_ctime = timespec_to_bch2_time(c, current_time(&inode->v));
92 	return 0;
93 }
94 
95 static int bch2_ioc_getflags(struct bch_inode_info *inode, int __user *arg)
96 {
97 	unsigned flags = map_flags(bch_flags_to_uflags, inode->ei_inode.bi_flags);
98 
99 	return put_user(flags, arg);
100 }
101 
102 static int bch2_ioc_setflags(struct bch_fs *c,
103 			     struct file *file,
104 			     struct bch_inode_info *inode,
105 			     void __user *arg)
106 {
107 	struct flags_set s = { .mask = map_defined(bch_flags_to_uflags) };
108 	unsigned uflags;
109 	int ret;
110 
111 	if (get_user(uflags, (int __user *) arg))
112 		return -EFAULT;
113 
114 	s.flags = map_flags_rev(bch_flags_to_uflags, uflags);
115 	if (uflags)
116 		return -EOPNOTSUPP;
117 
118 	ret = mnt_want_write_file(file);
119 	if (ret)
120 		return ret;
121 
122 	inode_lock(&inode->v);
123 	if (!inode_owner_or_capable(file_mnt_idmap(file), &inode->v)) {
124 		ret = -EACCES;
125 		goto setflags_out;
126 	}
127 
128 	mutex_lock(&inode->ei_update_lock);
129 	ret   = bch2_subvol_is_ro(c, inode->ei_inum.subvol) ?:
130 		bch2_write_inode(c, inode, bch2_inode_flags_set, &s,
131 			       ATTR_CTIME);
132 	mutex_unlock(&inode->ei_update_lock);
133 
134 setflags_out:
135 	inode_unlock(&inode->v);
136 	mnt_drop_write_file(file);
137 	return ret;
138 }
139 
140 static int bch2_ioc_fsgetxattr(struct bch_inode_info *inode,
141 			       struct fsxattr __user *arg)
142 {
143 	struct fsxattr fa = { 0 };
144 
145 	fa.fsx_xflags = map_flags(bch_flags_to_xflags, inode->ei_inode.bi_flags);
146 
147 	if (inode->ei_inode.bi_fields_set & (1 << Inode_opt_project))
148 		fa.fsx_xflags |= FS_XFLAG_PROJINHERIT;
149 
150 	fa.fsx_projid = inode->ei_qid.q[QTYP_PRJ];
151 
152 	if (copy_to_user(arg, &fa, sizeof(fa)))
153 		return -EFAULT;
154 
155 	return 0;
156 }
157 
158 static int fssetxattr_inode_update_fn(struct btree_trans *trans,
159 				      struct bch_inode_info *inode,
160 				      struct bch_inode_unpacked *bi,
161 				      void *p)
162 {
163 	struct flags_set *s = p;
164 
165 	if (s->projid != bi->bi_project) {
166 		bi->bi_fields_set |= 1U << Inode_opt_project;
167 		bi->bi_project = s->projid;
168 	}
169 
170 	return bch2_inode_flags_set(trans, inode, bi, p);
171 }
172 
173 static int bch2_ioc_fssetxattr(struct bch_fs *c,
174 			       struct file *file,
175 			       struct bch_inode_info *inode,
176 			       struct fsxattr __user *arg)
177 {
178 	struct flags_set s = { .mask = map_defined(bch_flags_to_xflags) };
179 	struct fsxattr fa;
180 	int ret;
181 
182 	if (copy_from_user(&fa, arg, sizeof(fa)))
183 		return -EFAULT;
184 
185 	s.set_projinherit = true;
186 	s.projinherit = (fa.fsx_xflags & FS_XFLAG_PROJINHERIT) != 0;
187 	fa.fsx_xflags &= ~FS_XFLAG_PROJINHERIT;
188 
189 	s.flags = map_flags_rev(bch_flags_to_xflags, fa.fsx_xflags);
190 	if (fa.fsx_xflags)
191 		return -EOPNOTSUPP;
192 
193 	if (fa.fsx_projid >= U32_MAX)
194 		return -EINVAL;
195 
196 	/*
197 	 * inode fields accessible via the xattr interface are stored with a +1
198 	 * bias, so that 0 means unset:
199 	 */
200 	s.projid = fa.fsx_projid + 1;
201 
202 	ret = mnt_want_write_file(file);
203 	if (ret)
204 		return ret;
205 
206 	inode_lock(&inode->v);
207 	if (!inode_owner_or_capable(file_mnt_idmap(file), &inode->v)) {
208 		ret = -EACCES;
209 		goto err;
210 	}
211 
212 	mutex_lock(&inode->ei_update_lock);
213 	ret   = bch2_subvol_is_ro(c, inode->ei_inum.subvol) ?:
214 		bch2_set_projid(c, inode, fa.fsx_projid) ?:
215 		bch2_write_inode(c, inode, fssetxattr_inode_update_fn, &s,
216 			       ATTR_CTIME);
217 	mutex_unlock(&inode->ei_update_lock);
218 err:
219 	inode_unlock(&inode->v);
220 	mnt_drop_write_file(file);
221 	return ret;
222 }
223 
224 static int bch2_reinherit_attrs_fn(struct btree_trans *trans,
225 				   struct bch_inode_info *inode,
226 				   struct bch_inode_unpacked *bi,
227 				   void *p)
228 {
229 	struct bch_inode_info *dir = p;
230 
231 	return !bch2_reinherit_attrs(bi, &dir->ei_inode);
232 }
233 
234 static int bch2_ioc_reinherit_attrs(struct bch_fs *c,
235 				    struct file *file,
236 				    struct bch_inode_info *src,
237 				    const char __user *name)
238 {
239 	struct bch_hash_info hash = bch2_hash_info_init(c, &src->ei_inode);
240 	struct bch_inode_info *dst;
241 	struct inode *vinode = NULL;
242 	char *kname = NULL;
243 	struct qstr qstr;
244 	int ret = 0;
245 	subvol_inum inum;
246 
247 	kname = kmalloc(BCH_NAME_MAX, GFP_KERNEL);
248 	if (!kname)
249 		return -ENOMEM;
250 
251 	ret = strncpy_from_user(kname, name, BCH_NAME_MAX);
252 	if (unlikely(ret < 0))
253 		goto err1;
254 
255 	qstr.len	= ret;
256 	qstr.name	= kname;
257 
258 	ret = bch2_dirent_lookup(c, inode_inum(src), &hash, &qstr, &inum);
259 	if (ret)
260 		goto err1;
261 
262 	vinode = bch2_vfs_inode_get(c, inum);
263 	ret = PTR_ERR_OR_ZERO(vinode);
264 	if (ret)
265 		goto err1;
266 
267 	dst = to_bch_ei(vinode);
268 
269 	ret = mnt_want_write_file(file);
270 	if (ret)
271 		goto err2;
272 
273 	bch2_lock_inodes(INODE_UPDATE_LOCK, src, dst);
274 
275 	if (inode_attr_changing(src, dst, Inode_opt_project)) {
276 		ret = bch2_fs_quota_transfer(c, dst,
277 					     src->ei_qid,
278 					     1 << QTYP_PRJ,
279 					     KEY_TYPE_QUOTA_PREALLOC);
280 		if (ret)
281 			goto err3;
282 	}
283 
284 	ret = bch2_write_inode(c, dst, bch2_reinherit_attrs_fn, src, 0);
285 err3:
286 	bch2_unlock_inodes(INODE_UPDATE_LOCK, src, dst);
287 
288 	/* return true if we did work */
289 	if (ret >= 0)
290 		ret = !ret;
291 
292 	mnt_drop_write_file(file);
293 err2:
294 	iput(vinode);
295 err1:
296 	kfree(kname);
297 
298 	return ret;
299 }
300 
301 static int bch2_ioc_getversion(struct bch_inode_info *inode, u32 __user *arg)
302 {
303 	return put_user(inode->v.i_generation, arg);
304 }
305 
306 static int bch2_ioc_getlabel(struct bch_fs *c, char __user *user_label)
307 {
308 	int ret;
309 	size_t len;
310 	char label[BCH_SB_LABEL_SIZE];
311 
312 	BUILD_BUG_ON(BCH_SB_LABEL_SIZE >= FSLABEL_MAX);
313 
314 	mutex_lock(&c->sb_lock);
315 	memcpy(label, c->disk_sb.sb->label, BCH_SB_LABEL_SIZE);
316 	mutex_unlock(&c->sb_lock);
317 
318 	len = strnlen(label, BCH_SB_LABEL_SIZE);
319 	if (len == BCH_SB_LABEL_SIZE) {
320 		bch_warn(c,
321 			"label is too long, return the first %zu bytes",
322 			--len);
323 	}
324 
325 	ret = copy_to_user(user_label, label, len);
326 
327 	return ret ? -EFAULT : 0;
328 }
329 
330 static int bch2_ioc_setlabel(struct bch_fs *c,
331 			     struct file *file,
332 			     struct bch_inode_info *inode,
333 			     const char __user *user_label)
334 {
335 	int ret;
336 	char label[BCH_SB_LABEL_SIZE];
337 
338 	if (!capable(CAP_SYS_ADMIN))
339 		return -EPERM;
340 
341 	if (copy_from_user(label, user_label, sizeof(label)))
342 		return -EFAULT;
343 
344 	if (strnlen(label, BCH_SB_LABEL_SIZE) == BCH_SB_LABEL_SIZE) {
345 		bch_err(c,
346 			"unable to set label with more than %d bytes",
347 			BCH_SB_LABEL_SIZE - 1);
348 		return -EINVAL;
349 	}
350 
351 	ret = mnt_want_write_file(file);
352 	if (ret)
353 		return ret;
354 
355 	mutex_lock(&c->sb_lock);
356 	strscpy(c->disk_sb.sb->label, label, BCH_SB_LABEL_SIZE);
357 	ret = bch2_write_super(c);
358 	mutex_unlock(&c->sb_lock);
359 
360 	mnt_drop_write_file(file);
361 	return ret;
362 }
363 
364 static int bch2_ioc_goingdown(struct bch_fs *c, u32 __user *arg)
365 {
366 	u32 flags;
367 	int ret = 0;
368 
369 	if (!capable(CAP_SYS_ADMIN))
370 		return -EPERM;
371 
372 	if (get_user(flags, arg))
373 		return -EFAULT;
374 
375 	bch_notice(c, "shutdown by ioctl type %u", flags);
376 
377 	switch (flags) {
378 	case FSOP_GOING_FLAGS_DEFAULT:
379 		ret = bdev_freeze(c->vfs_sb->s_bdev);
380 		if (ret)
381 			break;
382 		bch2_journal_flush(&c->journal);
383 		bch2_fs_emergency_read_only(c);
384 		bdev_thaw(c->vfs_sb->s_bdev);
385 		break;
386 	case FSOP_GOING_FLAGS_LOGFLUSH:
387 		bch2_journal_flush(&c->journal);
388 		fallthrough;
389 	case FSOP_GOING_FLAGS_NOLOGFLUSH:
390 		bch2_fs_emergency_read_only(c);
391 		break;
392 	default:
393 		ret = -EINVAL;
394 		break;
395 	}
396 
397 	return ret;
398 }
399 
400 static long bch2_ioctl_subvolume_create(struct bch_fs *c, struct file *filp,
401 					struct bch_ioctl_subvolume arg)
402 {
403 	struct inode *dir;
404 	struct bch_inode_info *inode;
405 	struct user_namespace *s_user_ns;
406 	struct dentry *dst_dentry;
407 	struct path src_path, dst_path;
408 	int how = LOOKUP_FOLLOW;
409 	int error;
410 	subvol_inum snapshot_src = { 0 };
411 	unsigned lookup_flags = 0;
412 	unsigned create_flags = BCH_CREATE_SUBVOL;
413 
414 	if (arg.flags & ~(BCH_SUBVOL_SNAPSHOT_CREATE|
415 			  BCH_SUBVOL_SNAPSHOT_RO))
416 		return -EINVAL;
417 
418 	if (!(arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE) &&
419 	    (arg.src_ptr ||
420 	     (arg.flags & BCH_SUBVOL_SNAPSHOT_RO)))
421 		return -EINVAL;
422 
423 	if (arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE)
424 		create_flags |= BCH_CREATE_SNAPSHOT;
425 
426 	if (arg.flags & BCH_SUBVOL_SNAPSHOT_RO)
427 		create_flags |= BCH_CREATE_SNAPSHOT_RO;
428 
429 	if (arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE) {
430 		/* sync_inodes_sb enforce s_umount is locked */
431 		down_read(&c->vfs_sb->s_umount);
432 		sync_inodes_sb(c->vfs_sb);
433 		up_read(&c->vfs_sb->s_umount);
434 	}
435 
436 	if (arg.src_ptr) {
437 		error = user_path_at(arg.dirfd,
438 				(const char __user *)(unsigned long)arg.src_ptr,
439 				how, &src_path);
440 		if (error)
441 			goto err1;
442 
443 		if (src_path.dentry->d_sb->s_fs_info != c) {
444 			path_put(&src_path);
445 			error = -EXDEV;
446 			goto err1;
447 		}
448 
449 		snapshot_src = inode_inum(to_bch_ei(src_path.dentry->d_inode));
450 	}
451 
452 	dst_dentry = user_path_create(arg.dirfd,
453 			(const char __user *)(unsigned long)arg.dst_ptr,
454 			&dst_path, lookup_flags);
455 	error = PTR_ERR_OR_ZERO(dst_dentry);
456 	if (error)
457 		goto err2;
458 
459 	if (dst_dentry->d_sb->s_fs_info != c) {
460 		error = -EXDEV;
461 		goto err3;
462 	}
463 
464 	if (dst_dentry->d_inode) {
465 		error = -BCH_ERR_EEXIST_subvolume_create;
466 		goto err3;
467 	}
468 
469 	dir = dst_path.dentry->d_inode;
470 	if (IS_DEADDIR(dir)) {
471 		error = -BCH_ERR_ENOENT_directory_dead;
472 		goto err3;
473 	}
474 
475 	s_user_ns = dir->i_sb->s_user_ns;
476 	if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
477 	    !kgid_has_mapping(s_user_ns, current_fsgid())) {
478 		error = -EOVERFLOW;
479 		goto err3;
480 	}
481 
482 	error = inode_permission(file_mnt_idmap(filp),
483 				 dir, MAY_WRITE | MAY_EXEC);
484 	if (error)
485 		goto err3;
486 
487 	if (!IS_POSIXACL(dir))
488 		arg.mode &= ~current_umask();
489 
490 	error = security_path_mkdir(&dst_path, dst_dentry, arg.mode);
491 	if (error)
492 		goto err3;
493 
494 	if ((arg.flags & BCH_SUBVOL_SNAPSHOT_CREATE) &&
495 	    !arg.src_ptr)
496 		snapshot_src.subvol = inode_inum(to_bch_ei(dir)).subvol;
497 
498 	down_write(&c->snapshot_create_lock);
499 	inode = __bch2_create(file_mnt_idmap(filp), to_bch_ei(dir),
500 			      dst_dentry, arg.mode|S_IFDIR,
501 			      0, snapshot_src, create_flags);
502 	up_write(&c->snapshot_create_lock);
503 
504 	error = PTR_ERR_OR_ZERO(inode);
505 	if (error)
506 		goto err3;
507 
508 	d_instantiate(dst_dentry, &inode->v);
509 	fsnotify_mkdir(dir, dst_dentry);
510 err3:
511 	done_path_create(&dst_path, dst_dentry);
512 err2:
513 	if (arg.src_ptr)
514 		path_put(&src_path);
515 err1:
516 	return error;
517 }
518 
519 static long bch2_ioctl_subvolume_destroy(struct bch_fs *c, struct file *filp,
520 				struct bch_ioctl_subvolume arg)
521 {
522 	const char __user *name = (void __user *)(unsigned long)arg.dst_ptr;
523 	struct path path;
524 	struct inode *dir;
525 	struct dentry *victim;
526 	int ret = 0;
527 
528 	if (arg.flags)
529 		return -EINVAL;
530 
531 	victim = user_path_locked_at(arg.dirfd, name, &path);
532 	if (IS_ERR(victim))
533 		return PTR_ERR(victim);
534 
535 	dir = d_inode(path.dentry);
536 	if (victim->d_sb->s_fs_info != c) {
537 		ret = -EXDEV;
538 		goto err;
539 	}
540 
541 	ret =   inode_permission(file_mnt_idmap(filp), d_inode(victim), MAY_WRITE) ?:
542 		__bch2_unlink(dir, victim, true);
543 	if (!ret) {
544 		fsnotify_rmdir(dir, victim);
545 		d_invalidate(victim);
546 	}
547 err:
548 	inode_unlock(dir);
549 	dput(victim);
550 	path_put(&path);
551 	return ret;
552 }
553 
554 long bch2_fs_file_ioctl(struct file *file, unsigned cmd, unsigned long arg)
555 {
556 	struct bch_inode_info *inode = file_bch_inode(file);
557 	struct bch_fs *c = inode->v.i_sb->s_fs_info;
558 	long ret;
559 
560 	switch (cmd) {
561 	case FS_IOC_GETFLAGS:
562 		ret = bch2_ioc_getflags(inode, (int __user *) arg);
563 		break;
564 
565 	case FS_IOC_SETFLAGS:
566 		ret = bch2_ioc_setflags(c, file, inode, (int __user *) arg);
567 		break;
568 
569 	case FS_IOC_FSGETXATTR:
570 		ret = bch2_ioc_fsgetxattr(inode, (void __user *) arg);
571 		break;
572 
573 	case FS_IOC_FSSETXATTR:
574 		ret = bch2_ioc_fssetxattr(c, file, inode,
575 					  (void __user *) arg);
576 		break;
577 
578 	case BCHFS_IOC_REINHERIT_ATTRS:
579 		ret = bch2_ioc_reinherit_attrs(c, file, inode,
580 					       (void __user *) arg);
581 		break;
582 
583 	case FS_IOC_GETVERSION:
584 		ret = bch2_ioc_getversion(inode, (u32 __user *) arg);
585 		break;
586 
587 	case FS_IOC_SETVERSION:
588 		ret = -ENOTTY;
589 		break;
590 
591 	case FS_IOC_GETFSLABEL:
592 		ret = bch2_ioc_getlabel(c, (void __user *) arg);
593 		break;
594 
595 	case FS_IOC_SETFSLABEL:
596 		ret = bch2_ioc_setlabel(c, file, inode, (const void __user *) arg);
597 		break;
598 
599 	case FS_IOC_GOINGDOWN:
600 		ret = bch2_ioc_goingdown(c, (u32 __user *) arg);
601 		break;
602 
603 	case BCH_IOCTL_SUBVOLUME_CREATE: {
604 		struct bch_ioctl_subvolume i;
605 
606 		ret = copy_from_user(&i, (void __user *) arg, sizeof(i))
607 			? -EFAULT
608 			: bch2_ioctl_subvolume_create(c, file, i);
609 		break;
610 	}
611 
612 	case BCH_IOCTL_SUBVOLUME_DESTROY: {
613 		struct bch_ioctl_subvolume i;
614 
615 		ret = copy_from_user(&i, (void __user *) arg, sizeof(i))
616 			? -EFAULT
617 			: bch2_ioctl_subvolume_destroy(c, file, i);
618 		break;
619 	}
620 
621 	default:
622 		ret = bch2_fs_ioctl(c, cmd, (void __user *) arg);
623 		break;
624 	}
625 
626 	return bch2_err_class(ret);
627 }
628 
629 #ifdef CONFIG_COMPAT
630 long bch2_compat_fs_ioctl(struct file *file, unsigned cmd, unsigned long arg)
631 {
632 	/* These are just misnamed, they actually get/put from/to user an int */
633 	switch (cmd) {
634 	case FS_IOC32_GETFLAGS:
635 		cmd = FS_IOC_GETFLAGS;
636 		break;
637 	case FS_IOC32_SETFLAGS:
638 		cmd = FS_IOC_SETFLAGS;
639 		break;
640 	case FS_IOC32_GETVERSION:
641 		cmd = FS_IOC_GETVERSION;
642 		break;
643 	case FS_IOC_GETFSLABEL:
644 	case FS_IOC_SETFSLABEL:
645 		break;
646 	default:
647 		return -ENOIOCTLCMD;
648 	}
649 	return bch2_fs_file_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
650 }
651 #endif
652 
653 #endif /* NO_BCACHEFS_FS */
654