xref: /openbmc/linux/fs/afs/inode.c (revision 2eb0f624b709e78ec8e2f4c3412947703db99301)
1 /*
2  * Copyright (c) 2002 Red Hat, Inc. All rights reserved.
3  *
4  * This software may be freely redistributed under the terms of the
5  * GNU General Public License.
6  *
7  * You should have received a copy of the GNU General Public License
8  * along with this program; if not, write to the Free Software
9  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
10  *
11  * Authors: David Woodhouse <dwmw2@infradead.org>
12  *          David Howells <dhowells@redhat.com>
13  *
14  */
15 
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/fs.h>
20 #include <linux/pagemap.h>
21 #include <linux/sched.h>
22 #include <linux/mount.h>
23 #include <linux/namei.h>
24 #include <linux/iversion.h>
25 #include "internal.h"
26 
27 static const struct inode_operations afs_symlink_inode_operations = {
28 	.get_link	= page_get_link,
29 	.listxattr	= afs_listxattr,
30 };
31 
32 /*
33  * map the AFS file status to the inode member variables
34  */
35 static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
36 {
37 	struct inode *inode = AFS_VNODE_TO_I(vnode);
38 	bool changed;
39 
40 	_debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
41 	       vnode->status.type,
42 	       vnode->status.nlink,
43 	       (unsigned long long) vnode->status.size,
44 	       vnode->status.data_version,
45 	       vnode->status.mode);
46 
47 	read_seqlock_excl(&vnode->cb_lock);
48 
49 	switch (vnode->status.type) {
50 	case AFS_FTYPE_FILE:
51 		inode->i_mode	= S_IFREG | vnode->status.mode;
52 		inode->i_op	= &afs_file_inode_operations;
53 		inode->i_fop	= &afs_file_operations;
54 		break;
55 	case AFS_FTYPE_DIR:
56 		inode->i_mode	= S_IFDIR | vnode->status.mode;
57 		inode->i_op	= &afs_dir_inode_operations;
58 		inode->i_fop	= &afs_dir_file_operations;
59 		break;
60 	case AFS_FTYPE_SYMLINK:
61 		/* Symlinks with a mode of 0644 are actually mountpoints. */
62 		if ((vnode->status.mode & 0777) == 0644) {
63 			inode->i_flags |= S_AUTOMOUNT;
64 
65 			set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
66 
67 			inode->i_mode	= S_IFDIR | 0555;
68 			inode->i_op	= &afs_mntpt_inode_operations;
69 			inode->i_fop	= &afs_mntpt_file_operations;
70 		} else {
71 			inode->i_mode	= S_IFLNK | vnode->status.mode;
72 			inode->i_op	= &afs_symlink_inode_operations;
73 		}
74 		inode_nohighmem(inode);
75 		break;
76 	default:
77 		printk("kAFS: AFS vnode with undefined type\n");
78 		read_sequnlock_excl(&vnode->cb_lock);
79 		return -EBADMSG;
80 	}
81 
82 	changed = (vnode->status.size != inode->i_size);
83 
84 	set_nlink(inode, vnode->status.nlink);
85 	inode->i_uid		= vnode->status.owner;
86 	inode->i_gid            = vnode->status.group;
87 	inode->i_size		= vnode->status.size;
88 	inode->i_ctime.tv_sec	= vnode->status.mtime_client;
89 	inode->i_ctime.tv_nsec	= 0;
90 	inode->i_atime		= inode->i_mtime = inode->i_ctime;
91 	inode->i_blocks		= 0;
92 	inode->i_generation	= vnode->fid.unique;
93 	inode_set_iversion_raw(inode, vnode->status.data_version);
94 	inode->i_mapping->a_ops	= &afs_fs_aops;
95 
96 	read_sequnlock_excl(&vnode->cb_lock);
97 
98 #ifdef CONFIG_AFS_FSCACHE
99 	if (changed)
100 		fscache_attr_changed(vnode->cache);
101 #endif
102 	return 0;
103 }
104 
105 /*
106  * Fetch file status from the volume.
107  */
108 int afs_fetch_status(struct afs_vnode *vnode, struct key *key)
109 {
110 	struct afs_fs_cursor fc;
111 	int ret;
112 
113 	_enter("%s,{%x:%u.%u,S=%lx}",
114 	       vnode->volume->name,
115 	       vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
116 	       vnode->flags);
117 
118 	ret = -ERESTARTSYS;
119 	if (afs_begin_vnode_operation(&fc, vnode, key)) {
120 		while (afs_select_fileserver(&fc)) {
121 			fc.cb_break = vnode->cb_break + vnode->cb_s_break;
122 			afs_fs_fetch_file_status(&fc, NULL);
123 		}
124 
125 		afs_check_for_remote_deletion(&fc, fc.vnode);
126 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
127 		ret = afs_end_vnode_operation(&fc);
128 	}
129 
130 	_leave(" = %d", ret);
131 	return ret;
132 }
133 
134 /*
135  * iget5() comparator
136  */
137 int afs_iget5_test(struct inode *inode, void *opaque)
138 {
139 	struct afs_iget_data *data = opaque;
140 
141 	return inode->i_ino == data->fid.vnode &&
142 		inode->i_generation == data->fid.unique;
143 }
144 
145 /*
146  * iget5() comparator for inode created by autocell operations
147  *
148  * These pseudo inodes don't match anything.
149  */
150 static int afs_iget5_pseudo_dir_test(struct inode *inode, void *opaque)
151 {
152 	return 0;
153 }
154 
155 /*
156  * iget5() inode initialiser
157  */
158 static int afs_iget5_set(struct inode *inode, void *opaque)
159 {
160 	struct afs_iget_data *data = opaque;
161 	struct afs_vnode *vnode = AFS_FS_I(inode);
162 
163 	inode->i_ino = data->fid.vnode;
164 	inode->i_generation = data->fid.unique;
165 	vnode->fid = data->fid;
166 	vnode->volume = data->volume;
167 
168 	return 0;
169 }
170 
171 /*
172  * Create an inode for a dynamic root directory or an autocell dynamic
173  * automount dir.
174  */
175 struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
176 {
177 	struct afs_iget_data data;
178 	struct afs_super_info *as;
179 	struct afs_vnode *vnode;
180 	struct inode *inode;
181 	static atomic_t afs_autocell_ino;
182 
183 	_enter("");
184 
185 	as = sb->s_fs_info;
186 	if (as->volume) {
187 		data.volume = as->volume;
188 		data.fid.vid = as->volume->vid;
189 	}
190 	if (root) {
191 		data.fid.vnode = 1;
192 		data.fid.unique = 1;
193 	} else {
194 		data.fid.vnode = atomic_inc_return(&afs_autocell_ino);
195 		data.fid.unique = 0;
196 	}
197 
198 	inode = iget5_locked(sb, data.fid.vnode,
199 			     afs_iget5_pseudo_dir_test, afs_iget5_set,
200 			     &data);
201 	if (!inode) {
202 		_leave(" = -ENOMEM");
203 		return ERR_PTR(-ENOMEM);
204 	}
205 
206 	_debug("GOT INODE %p { ino=%lu, vl=%x, vn=%x, u=%x }",
207 	       inode, inode->i_ino, data.fid.vid, data.fid.vnode,
208 	       data.fid.unique);
209 
210 	vnode = AFS_FS_I(inode);
211 
212 	/* there shouldn't be an existing inode */
213 	BUG_ON(!(inode->i_state & I_NEW));
214 
215 	inode->i_size		= 0;
216 	inode->i_mode		= S_IFDIR | S_IRUGO | S_IXUGO;
217 	if (root) {
218 		inode->i_op	= &afs_dynroot_inode_operations;
219 		inode->i_fop	= &afs_dynroot_file_operations;
220 	} else {
221 		inode->i_op	= &afs_autocell_inode_operations;
222 	}
223 	set_nlink(inode, 2);
224 	inode->i_uid		= GLOBAL_ROOT_UID;
225 	inode->i_gid		= GLOBAL_ROOT_GID;
226 	inode->i_ctime.tv_sec	= get_seconds();
227 	inode->i_ctime.tv_nsec	= 0;
228 	inode->i_atime		= inode->i_mtime = inode->i_ctime;
229 	inode->i_blocks		= 0;
230 	inode_set_iversion_raw(inode, 0);
231 	inode->i_generation	= 0;
232 
233 	set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
234 	if (!root) {
235 		set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
236 		inode->i_flags |= S_AUTOMOUNT;
237 	}
238 
239 	inode->i_flags |= S_NOATIME;
240 	unlock_new_inode(inode);
241 	_leave(" = %p", inode);
242 	return inode;
243 }
244 
245 /*
246  * Get a cache cookie for an inode.
247  */
248 static void afs_get_inode_cache(struct afs_vnode *vnode)
249 {
250 #ifdef CONFIG_AFS_FSCACHE
251 	struct {
252 		u32 vnode_id;
253 		u32 unique;
254 		u32 vnode_id_ext[2];	/* Allow for a 96-bit key */
255 	} __packed key;
256 	struct afs_vnode_cache_aux aux;
257 
258 	key.vnode_id		= vnode->fid.vnode;
259 	key.unique		= vnode->fid.unique;
260 	key.vnode_id_ext[0]	= 0;
261 	key.vnode_id_ext[1]	= 0;
262 	aux.data_version	= vnode->status.data_version;
263 
264 	vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
265 					      &afs_vnode_cache_index_def,
266 					      &key, sizeof(key),
267 					      &aux, sizeof(aux),
268 					      vnode, vnode->status.size, true);
269 #endif
270 }
271 
272 /*
273  * inode retrieval
274  */
275 struct inode *afs_iget(struct super_block *sb, struct key *key,
276 		       struct afs_fid *fid, struct afs_file_status *status,
277 		       struct afs_callback *cb, struct afs_cb_interest *cbi)
278 {
279 	struct afs_iget_data data = { .fid = *fid };
280 	struct afs_super_info *as;
281 	struct afs_vnode *vnode;
282 	struct inode *inode;
283 	int ret;
284 
285 	_enter(",{%x:%u.%u},,", fid->vid, fid->vnode, fid->unique);
286 
287 	as = sb->s_fs_info;
288 	data.volume = as->volume;
289 
290 	inode = iget5_locked(sb, fid->vnode, afs_iget5_test, afs_iget5_set,
291 			     &data);
292 	if (!inode) {
293 		_leave(" = -ENOMEM");
294 		return ERR_PTR(-ENOMEM);
295 	}
296 
297 	_debug("GOT INODE %p { vl=%x vn=%x, u=%x }",
298 	       inode, fid->vid, fid->vnode, fid->unique);
299 
300 	vnode = AFS_FS_I(inode);
301 
302 	/* deal with an existing inode */
303 	if (!(inode->i_state & I_NEW)) {
304 		_leave(" = %p", inode);
305 		return inode;
306 	}
307 
308 	if (!status) {
309 		/* it's a remotely extant inode */
310 		ret = afs_fetch_status(vnode, key);
311 		if (ret < 0)
312 			goto bad_inode;
313 	} else {
314 		/* it's an inode we just created */
315 		memcpy(&vnode->status, status, sizeof(vnode->status));
316 
317 		if (!cb) {
318 			/* it's a symlink we just created (the fileserver
319 			 * didn't give us a callback) */
320 			vnode->cb_version = 0;
321 			vnode->cb_type = 0;
322 			vnode->cb_expires_at = 0;
323 		} else {
324 			vnode->cb_version = cb->version;
325 			vnode->cb_type = cb->type;
326 			vnode->cb_expires_at = cb->expiry;
327 			vnode->cb_interest = afs_get_cb_interest(cbi);
328 			set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
329 		}
330 
331 		vnode->cb_expires_at += ktime_get_real_seconds();
332 	}
333 
334 	/* set up caching before mapping the status, as map-status reads the
335 	 * first page of symlinks to see if they're really mountpoints */
336 	inode->i_size = vnode->status.size;
337 	afs_get_inode_cache(vnode);
338 
339 	ret = afs_inode_map_status(vnode, key);
340 	if (ret < 0)
341 		goto bad_inode;
342 
343 	/* success */
344 	clear_bit(AFS_VNODE_UNSET, &vnode->flags);
345 	inode->i_flags |= S_NOATIME;
346 	unlock_new_inode(inode);
347 	_leave(" = %p [CB { v=%u t=%u }]", inode, vnode->cb_version, vnode->cb_type);
348 	return inode;
349 
350 	/* failure */
351 bad_inode:
352 #ifdef CONFIG_AFS_FSCACHE
353 	fscache_relinquish_cookie(vnode->cache, NULL, ret == -ENOENT);
354 	vnode->cache = NULL;
355 #endif
356 	iget_failed(inode);
357 	_leave(" = %d [bad]", ret);
358 	return ERR_PTR(ret);
359 }
360 
361 /*
362  * mark the data attached to an inode as obsolete due to a write on the server
363  * - might also want to ditch all the outstanding writes and dirty pages
364  */
365 void afs_zap_data(struct afs_vnode *vnode)
366 {
367 	_enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode);
368 
369 #ifdef CONFIG_AFS_FSCACHE
370 	fscache_invalidate(vnode->cache);
371 #endif
372 
373 	/* nuke all the non-dirty pages that aren't locked, mapped or being
374 	 * written back in a regular file and completely discard the pages in a
375 	 * directory or symlink */
376 	if (S_ISREG(vnode->vfs_inode.i_mode))
377 		invalidate_remote_inode(&vnode->vfs_inode);
378 	else
379 		invalidate_inode_pages2(vnode->vfs_inode.i_mapping);
380 }
381 
382 /*
383  * validate a vnode/inode
384  * - there are several things we need to check
385  *   - parent dir data changes (rm, rmdir, rename, mkdir, create, link,
386  *     symlink)
387  *   - parent dir metadata changed (security changes)
388  *   - dentry data changed (write, truncate)
389  *   - dentry metadata changed (security changes)
390  */
391 int afs_validate(struct afs_vnode *vnode, struct key *key)
392 {
393 	time64_t now = ktime_get_real_seconds();
394 	bool valid = false;
395 	int ret;
396 
397 	_enter("{v={%x:%u} fl=%lx},%x",
398 	       vnode->fid.vid, vnode->fid.vnode, vnode->flags,
399 	       key_serial(key));
400 
401 	/* Quickly check the callback state.  Ideally, we'd use read_seqbegin
402 	 * here, but we have no way to pass the net namespace to the RCU
403 	 * cleanup for the server record.
404 	 */
405 	read_seqlock_excl(&vnode->cb_lock);
406 
407 	if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
408 		if (vnode->cb_s_break != vnode->cb_interest->server->cb_s_break) {
409 			vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
410 		} else if (!test_bit(AFS_VNODE_DIR_MODIFIED, &vnode->flags) &&
411 			   !test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags) &&
412 			   vnode->cb_expires_at - 10 > now) {
413 				valid = true;
414 		}
415 	} else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
416 		valid = true;
417 	}
418 
419 	read_sequnlock_excl(&vnode->cb_lock);
420 
421 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
422 		clear_nlink(&vnode->vfs_inode);
423 
424 	if (valid)
425 		goto valid;
426 
427 	mutex_lock(&vnode->validate_lock);
428 
429 	/* if the promise has expired, we need to check the server again to get
430 	 * a new promise - note that if the (parent) directory's metadata was
431 	 * changed then the security may be different and we may no longer have
432 	 * access */
433 	if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
434 		_debug("not promised");
435 		ret = afs_fetch_status(vnode, key);
436 		if (ret < 0) {
437 			if (ret == -ENOENT) {
438 				set_bit(AFS_VNODE_DELETED, &vnode->flags);
439 				ret = -ESTALE;
440 			}
441 			goto error_unlock;
442 		}
443 		_debug("new promise [fl=%lx]", vnode->flags);
444 	}
445 
446 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
447 		_debug("file already deleted");
448 		ret = -ESTALE;
449 		goto error_unlock;
450 	}
451 
452 	/* if the vnode's data version number changed then its contents are
453 	 * different */
454 	if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
455 		afs_zap_data(vnode);
456 
457 	clear_bit(AFS_VNODE_DIR_MODIFIED, &vnode->flags);
458 	mutex_unlock(&vnode->validate_lock);
459 valid:
460 	_leave(" = 0");
461 	return 0;
462 
463 error_unlock:
464 	mutex_unlock(&vnode->validate_lock);
465 	_leave(" = %d", ret);
466 	return ret;
467 }
468 
469 /*
470  * read the attributes of an inode
471  */
472 int afs_getattr(const struct path *path, struct kstat *stat,
473 		u32 request_mask, unsigned int query_flags)
474 {
475 	struct inode *inode = d_inode(path->dentry);
476 	struct afs_vnode *vnode = AFS_FS_I(inode);
477 	int seq = 0;
478 
479 	_enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
480 
481 	do {
482 		read_seqbegin_or_lock(&vnode->cb_lock, &seq);
483 		generic_fillattr(inode, stat);
484 	} while (need_seqretry(&vnode->cb_lock, seq));
485 
486 	done_seqretry(&vnode->cb_lock, seq);
487 	return 0;
488 }
489 
490 /*
491  * discard an AFS inode
492  */
493 int afs_drop_inode(struct inode *inode)
494 {
495 	_enter("");
496 
497 	if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
498 		return generic_delete_inode(inode);
499 	else
500 		return generic_drop_inode(inode);
501 }
502 
503 /*
504  * clear an AFS inode
505  */
506 void afs_evict_inode(struct inode *inode)
507 {
508 	struct afs_vnode *vnode;
509 
510 	vnode = AFS_FS_I(inode);
511 
512 	_enter("{%x:%u.%d}",
513 	       vnode->fid.vid,
514 	       vnode->fid.vnode,
515 	       vnode->fid.unique);
516 
517 	_debug("CLEAR INODE %p", inode);
518 
519 	ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
520 
521 	truncate_inode_pages_final(&inode->i_data);
522 	clear_inode(inode);
523 
524 	if (vnode->cb_interest) {
525 		afs_put_cb_interest(afs_i2net(inode), vnode->cb_interest);
526 		vnode->cb_interest = NULL;
527 	}
528 
529 	while (!list_empty(&vnode->wb_keys)) {
530 		struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
531 						    struct afs_wb_key, vnode_link);
532 		list_del(&wbk->vnode_link);
533 		afs_put_wb_key(wbk);
534 	}
535 
536 #ifdef CONFIG_AFS_FSCACHE
537 	{
538 		struct afs_vnode_cache_aux aux;
539 
540 		aux.data_version = vnode->status.data_version;
541 		fscache_relinquish_cookie(vnode->cache, &aux,
542 					  test_bit(AFS_VNODE_DELETED, &vnode->flags));
543 		vnode->cache = NULL;
544 	}
545 #endif
546 
547 	afs_put_permits(vnode->permit_cache);
548 	_leave("");
549 }
550 
551 /*
552  * set the attributes of an inode
553  */
554 int afs_setattr(struct dentry *dentry, struct iattr *attr)
555 {
556 	struct afs_fs_cursor fc;
557 	struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
558 	struct key *key;
559 	int ret;
560 
561 	_enter("{%x:%u},{n=%pd},%x",
562 	       vnode->fid.vid, vnode->fid.vnode, dentry,
563 	       attr->ia_valid);
564 
565 	if (!(attr->ia_valid & (ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
566 				ATTR_MTIME))) {
567 		_leave(" = 0 [unsupported]");
568 		return 0;
569 	}
570 
571 	/* flush any dirty data outstanding on a regular file */
572 	if (S_ISREG(vnode->vfs_inode.i_mode))
573 		filemap_write_and_wait(vnode->vfs_inode.i_mapping);
574 
575 	if (attr->ia_valid & ATTR_FILE) {
576 		key = afs_file_key(attr->ia_file);
577 	} else {
578 		key = afs_request_key(vnode->volume->cell);
579 		if (IS_ERR(key)) {
580 			ret = PTR_ERR(key);
581 			goto error;
582 		}
583 	}
584 
585 	ret = -ERESTARTSYS;
586 	if (afs_begin_vnode_operation(&fc, vnode, key)) {
587 		while (afs_select_fileserver(&fc)) {
588 			fc.cb_break = vnode->cb_break + vnode->cb_s_break;
589 			afs_fs_setattr(&fc, attr);
590 		}
591 
592 		afs_check_for_remote_deletion(&fc, fc.vnode);
593 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
594 		ret = afs_end_vnode_operation(&fc);
595 	}
596 
597 	if (!(attr->ia_valid & ATTR_FILE))
598 		key_put(key);
599 
600 error:
601 	_leave(" = %d", ret);
602 	return ret;
603 }
604