xref: /linux/fs/afs/inode.c (revision b134d687dd9369bb2407a23c6ecc9e0a15d8bb20) 
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 static noinline void dump_vnode(struct afs_vnode *vnode, struct afs_vnode *parent_vnode)
33 {
34 	static unsigned long once_only;
35 
36 	pr_warn("kAFS: AFS vnode with undefined type %u\n",
37 		vnode->status.type);
38 	pr_warn("kAFS: A=%d m=%o s=%llx v=%llx\n",
39 		vnode->status.abort_code,
40 		vnode->status.mode,
41 		vnode->status.size,
42 		vnode->status.data_version);
43 	pr_warn("kAFS: vnode %llx:%llx:%x\n",
44 		vnode->fid.vid,
45 		vnode->fid.vnode,
46 		vnode->fid.unique);
47 	if (parent_vnode)
48 		pr_warn("kAFS: dir %llx:%llx:%x\n",
49 			parent_vnode->fid.vid,
50 			parent_vnode->fid.vnode,
51 			parent_vnode->fid.unique);
52 
53 	if (!test_and_set_bit(0, &once_only))
54 		dump_stack();
55 }
56 
57 /*
58  * Initialise an inode from the vnode status.
59  */
60 static int afs_inode_init_from_status(struct afs_vnode *vnode, struct key *key,
61 				      struct afs_vnode *parent_vnode)
62 {
63 	struct inode *inode = AFS_VNODE_TO_I(vnode);
64 
65 	_debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
66 	       vnode->status.type,
67 	       vnode->status.nlink,
68 	       (unsigned long long) vnode->status.size,
69 	       vnode->status.data_version,
70 	       vnode->status.mode);
71 
72 	read_seqlock_excl(&vnode->cb_lock);
73 
74 	afs_update_inode_from_status(vnode, &vnode->status, NULL,
75 				     AFS_VNODE_NOT_YET_SET);
76 
77 	switch (vnode->status.type) {
78 	case AFS_FTYPE_FILE:
79 		inode->i_mode	= S_IFREG | vnode->status.mode;
80 		inode->i_op	= &afs_file_inode_operations;
81 		inode->i_fop	= &afs_file_operations;
82 		inode->i_mapping->a_ops	= &afs_fs_aops;
83 		break;
84 	case AFS_FTYPE_DIR:
85 		inode->i_mode	= S_IFDIR | vnode->status.mode;
86 		inode->i_op	= &afs_dir_inode_operations;
87 		inode->i_fop	= &afs_dir_file_operations;
88 		inode->i_mapping->a_ops	= &afs_dir_aops;
89 		break;
90 	case AFS_FTYPE_SYMLINK:
91 		/* Symlinks with a mode of 0644 are actually mountpoints. */
92 		if ((vnode->status.mode & 0777) == 0644) {
93 			inode->i_flags |= S_AUTOMOUNT;
94 
95 			set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
96 
97 			inode->i_mode	= S_IFDIR | 0555;
98 			inode->i_op	= &afs_mntpt_inode_operations;
99 			inode->i_fop	= &afs_mntpt_file_operations;
100 			inode->i_mapping->a_ops	= &afs_fs_aops;
101 		} else {
102 			inode->i_mode	= S_IFLNK | vnode->status.mode;
103 			inode->i_op	= &afs_symlink_inode_operations;
104 			inode->i_mapping->a_ops	= &afs_fs_aops;
105 		}
106 		inode_nohighmem(inode);
107 		break;
108 	default:
109 		dump_vnode(vnode, parent_vnode);
110 		read_sequnlock_excl(&vnode->cb_lock);
111 		return afs_protocol_error(NULL, -EBADMSG, afs_eproto_file_type);
112 	}
113 
114 	inode->i_blocks		= 0;
115 	vnode->invalid_before	= vnode->status.data_version;
116 
117 	read_sequnlock_excl(&vnode->cb_lock);
118 	return 0;
119 }
120 
121 /*
122  * Fetch file status from the volume.
123  */
124 int afs_fetch_status(struct afs_vnode *vnode, struct key *key, bool new_inode)
125 {
126 	struct afs_fs_cursor fc;
127 	int ret;
128 
129 	_enter("%s,{%llx:%llu.%u,S=%lx}",
130 	       vnode->volume->name,
131 	       vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
132 	       vnode->flags);
133 
134 	ret = -ERESTARTSYS;
135 	if (afs_begin_vnode_operation(&fc, vnode, key)) {
136 		while (afs_select_fileserver(&fc)) {
137 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
138 			afs_fs_fetch_file_status(&fc, NULL, new_inode);
139 		}
140 
141 		afs_check_for_remote_deletion(&fc, fc.vnode);
142 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
143 		ret = afs_end_vnode_operation(&fc);
144 	}
145 
146 	_leave(" = %d", ret);
147 	return ret;
148 }
149 
150 /*
151  * iget5() comparator
152  */
153 int afs_iget5_test(struct inode *inode, void *opaque)
154 {
155 	struct afs_iget_data *data = opaque;
156 	struct afs_vnode *vnode = AFS_FS_I(inode);
157 
158 	return memcmp(&vnode->fid, &data->fid, sizeof(data->fid)) == 0;
159 }
160 
161 /*
162  * iget5() comparator for inode created by autocell operations
163  *
164  * These pseudo inodes don't match anything.
165  */
166 static int afs_iget5_pseudo_dir_test(struct inode *inode, void *opaque)
167 {
168 	return 0;
169 }
170 
171 /*
172  * iget5() inode initialiser
173  */
174 static int afs_iget5_set(struct inode *inode, void *opaque)
175 {
176 	struct afs_iget_data *data = opaque;
177 	struct afs_vnode *vnode = AFS_FS_I(inode);
178 
179 	vnode->fid = data->fid;
180 	vnode->volume = data->volume;
181 
182 	/* YFS supports 96-bit vnode IDs, but Linux only supports
183 	 * 64-bit inode numbers.
184 	 */
185 	inode->i_ino = data->fid.vnode;
186 	inode->i_generation = data->fid.unique;
187 	return 0;
188 }
189 
190 /*
191  * Create an inode for a dynamic root directory or an autocell dynamic
192  * automount dir.
193  */
194 struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
195 {
196 	struct afs_iget_data data;
197 	struct afs_super_info *as;
198 	struct afs_vnode *vnode;
199 	struct inode *inode;
200 	static atomic_t afs_autocell_ino;
201 
202 	_enter("");
203 
204 	as = sb->s_fs_info;
205 	if (as->volume) {
206 		data.volume = as->volume;
207 		data.fid.vid = as->volume->vid;
208 	}
209 	if (root) {
210 		data.fid.vnode = 1;
211 		data.fid.unique = 1;
212 	} else {
213 		data.fid.vnode = atomic_inc_return(&afs_autocell_ino);
214 		data.fid.unique = 0;
215 	}
216 
217 	inode = iget5_locked(sb, data.fid.vnode,
218 			     afs_iget5_pseudo_dir_test, afs_iget5_set,
219 			     &data);
220 	if (!inode) {
221 		_leave(" = -ENOMEM");
222 		return ERR_PTR(-ENOMEM);
223 	}
224 
225 	_debug("GOT INODE %p { ino=%lu, vl=%llx, vn=%llx, u=%x }",
226 	       inode, inode->i_ino, data.fid.vid, data.fid.vnode,
227 	       data.fid.unique);
228 
229 	vnode = AFS_FS_I(inode);
230 
231 	/* there shouldn't be an existing inode */
232 	BUG_ON(!(inode->i_state & I_NEW));
233 
234 	inode->i_size		= 0;
235 	inode->i_mode		= S_IFDIR | S_IRUGO | S_IXUGO;
236 	if (root) {
237 		inode->i_op	= &afs_dynroot_inode_operations;
238 		inode->i_fop	= &afs_dynroot_file_operations;
239 	} else {
240 		inode->i_op	= &afs_autocell_inode_operations;
241 	}
242 	set_nlink(inode, 2);
243 	inode->i_uid		= GLOBAL_ROOT_UID;
244 	inode->i_gid		= GLOBAL_ROOT_GID;
245 	inode->i_ctime = inode->i_atime = inode->i_mtime = current_time(inode);
246 	inode->i_blocks		= 0;
247 	inode_set_iversion_raw(inode, 0);
248 	inode->i_generation	= 0;
249 
250 	set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
251 	if (!root) {
252 		set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
253 		inode->i_flags |= S_AUTOMOUNT;
254 	}
255 
256 	inode->i_flags |= S_NOATIME;
257 	unlock_new_inode(inode);
258 	_leave(" = %p", inode);
259 	return inode;
260 }
261 
262 /*
263  * Get a cache cookie for an inode.
264  */
265 static void afs_get_inode_cache(struct afs_vnode *vnode)
266 {
267 #ifdef CONFIG_AFS_FSCACHE
268 	struct {
269 		u32 vnode_id;
270 		u32 unique;
271 		u32 vnode_id_ext[2];	/* Allow for a 96-bit key */
272 	} __packed key;
273 	struct afs_vnode_cache_aux aux;
274 
275 	if (vnode->status.type == AFS_FTYPE_DIR) {
276 		vnode->cache = NULL;
277 		return;
278 	}
279 
280 	key.vnode_id		= vnode->fid.vnode;
281 	key.unique		= vnode->fid.unique;
282 	key.vnode_id_ext[0]	= vnode->fid.vnode >> 32;
283 	key.vnode_id_ext[1]	= vnode->fid.vnode_hi;
284 	aux.data_version	= vnode->status.data_version;
285 
286 	vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
287 					      &afs_vnode_cache_index_def,
288 					      &key, sizeof(key),
289 					      &aux, sizeof(aux),
290 					      vnode, vnode->status.size, true);
291 #endif
292 }
293 
294 /*
295  * inode retrieval
296  */
297 struct inode *afs_iget(struct super_block *sb, struct key *key,
298 		       struct afs_fid *fid, struct afs_file_status *status,
299 		       struct afs_callback *cb, struct afs_cb_interest *cbi,
300 		       struct afs_vnode *parent_vnode)
301 {
302 	struct afs_iget_data data = { .fid = *fid };
303 	struct afs_super_info *as;
304 	struct afs_vnode *vnode;
305 	struct inode *inode;
306 	int ret;
307 
308 	_enter(",{%llx:%llu.%u},,", fid->vid, fid->vnode, fid->unique);
309 
310 	as = sb->s_fs_info;
311 	data.volume = as->volume;
312 
313 	inode = iget5_locked(sb, fid->vnode, afs_iget5_test, afs_iget5_set,
314 			     &data);
315 	if (!inode) {
316 		_leave(" = -ENOMEM");
317 		return ERR_PTR(-ENOMEM);
318 	}
319 
320 	_debug("GOT INODE %p { vl=%llx vn=%llx, u=%x }",
321 	       inode, fid->vid, fid->vnode, fid->unique);
322 
323 	vnode = AFS_FS_I(inode);
324 
325 	/* deal with an existing inode */
326 	if (!(inode->i_state & I_NEW)) {
327 		_leave(" = %p", inode);
328 		return inode;
329 	}
330 
331 	if (!status) {
332 		/* it's a remotely extant inode */
333 		ret = afs_fetch_status(vnode, key, true);
334 		if (ret < 0)
335 			goto bad_inode;
336 	} else {
337 		/* it's an inode we just created */
338 		memcpy(&vnode->status, status, sizeof(vnode->status));
339 
340 		if (!cb) {
341 			/* it's a symlink we just created (the fileserver
342 			 * didn't give us a callback) */
343 			vnode->cb_version = 0;
344 			vnode->cb_type = 0;
345 			vnode->cb_expires_at = ktime_get();
346 		} else {
347 			vnode->cb_version = cb->version;
348 			vnode->cb_type = cb->type;
349 			vnode->cb_expires_at = cb->expires_at;
350 			vnode->cb_interest = afs_get_cb_interest(cbi);
351 			set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
352 		}
353 
354 		vnode->cb_expires_at += ktime_get_real_seconds();
355 	}
356 
357 	ret = afs_inode_init_from_status(vnode, key, parent_vnode);
358 	if (ret < 0)
359 		goto bad_inode;
360 
361 	afs_get_inode_cache(vnode);
362 
363 	/* success */
364 	clear_bit(AFS_VNODE_UNSET, &vnode->flags);
365 	inode->i_flags |= S_NOATIME;
366 	unlock_new_inode(inode);
367 	_leave(" = %p [CB { v=%u t=%u }]", inode, vnode->cb_version, vnode->cb_type);
368 	return inode;
369 
370 	/* failure */
371 bad_inode:
372 	iget_failed(inode);
373 	_leave(" = %d [bad]", ret);
374 	return ERR_PTR(ret);
375 }
376 
377 /*
378  * mark the data attached to an inode as obsolete due to a write on the server
379  * - might also want to ditch all the outstanding writes and dirty pages
380  */
381 void afs_zap_data(struct afs_vnode *vnode)
382 {
383 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
384 
385 #ifdef CONFIG_AFS_FSCACHE
386 	fscache_invalidate(vnode->cache);
387 #endif
388 
389 	/* nuke all the non-dirty pages that aren't locked, mapped or being
390 	 * written back in a regular file and completely discard the pages in a
391 	 * directory or symlink */
392 	if (S_ISREG(vnode->vfs_inode.i_mode))
393 		invalidate_remote_inode(&vnode->vfs_inode);
394 	else
395 		invalidate_inode_pages2(vnode->vfs_inode.i_mapping);
396 }
397 
398 /*
399  * validate a vnode/inode
400  * - there are several things we need to check
401  *   - parent dir data changes (rm, rmdir, rename, mkdir, create, link,
402  *     symlink)
403  *   - parent dir metadata changed (security changes)
404  *   - dentry data changed (write, truncate)
405  *   - dentry metadata changed (security changes)
406  */
407 int afs_validate(struct afs_vnode *vnode, struct key *key)
408 {
409 	time64_t now = ktime_get_real_seconds();
410 	bool valid;
411 	int ret;
412 
413 	_enter("{v={%llx:%llu} fl=%lx},%x",
414 	       vnode->fid.vid, vnode->fid.vnode, vnode->flags,
415 	       key_serial(key));
416 
417 	/* Quickly check the callback state.  Ideally, we'd use read_seqbegin
418 	 * here, but we have no way to pass the net namespace to the RCU
419 	 * cleanup for the server record.
420 	 */
421 	read_seqlock_excl(&vnode->cb_lock);
422 
423 	if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
424 		if (vnode->cb_s_break != vnode->cb_interest->server->cb_s_break ||
425 		    vnode->cb_v_break != vnode->volume->cb_v_break) {
426 			vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
427 			vnode->cb_v_break = vnode->volume->cb_v_break;
428 			valid = false;
429 		} else if (vnode->status.type == AFS_FTYPE_DIR &&
430 			   (!test_bit(AFS_VNODE_DIR_VALID, &vnode->flags) ||
431 			    vnode->cb_expires_at - 10 <= now)) {
432 			valid = false;
433 		} else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags) ||
434 			   vnode->cb_expires_at - 10 <= now) {
435 			valid = false;
436 		} else {
437 			valid = true;
438 		}
439 	} else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
440 		valid = true;
441 	} else {
442 		vnode->cb_v_break = vnode->volume->cb_v_break;
443 		valid = false;
444 	}
445 
446 	read_sequnlock_excl(&vnode->cb_lock);
447 
448 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
449 		clear_nlink(&vnode->vfs_inode);
450 
451 	if (valid)
452 		goto valid;
453 
454 	down_write(&vnode->validate_lock);
455 
456 	/* if the promise has expired, we need to check the server again to get
457 	 * a new promise - note that if the (parent) directory's metadata was
458 	 * changed then the security may be different and we may no longer have
459 	 * access */
460 	if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
461 		_debug("not promised");
462 		ret = afs_fetch_status(vnode, key, false);
463 		if (ret < 0) {
464 			if (ret == -ENOENT) {
465 				set_bit(AFS_VNODE_DELETED, &vnode->flags);
466 				ret = -ESTALE;
467 			}
468 			goto error_unlock;
469 		}
470 		_debug("new promise [fl=%lx]", vnode->flags);
471 	}
472 
473 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
474 		_debug("file already deleted");
475 		ret = -ESTALE;
476 		goto error_unlock;
477 	}
478 
479 	/* if the vnode's data version number changed then its contents are
480 	 * different */
481 	if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
482 		afs_zap_data(vnode);
483 	up_write(&vnode->validate_lock);
484 valid:
485 	_leave(" = 0");
486 	return 0;
487 
488 error_unlock:
489 	up_write(&vnode->validate_lock);
490 	_leave(" = %d", ret);
491 	return ret;
492 }
493 
494 /*
495  * read the attributes of an inode
496  */
497 int afs_getattr(const struct path *path, struct kstat *stat,
498 		u32 request_mask, unsigned int query_flags)
499 {
500 	struct inode *inode = d_inode(path->dentry);
501 	struct afs_vnode *vnode = AFS_FS_I(inode);
502 	int seq = 0;
503 
504 	_enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
505 
506 	do {
507 		read_seqbegin_or_lock(&vnode->cb_lock, &seq);
508 		generic_fillattr(inode, stat);
509 	} while (need_seqretry(&vnode->cb_lock, seq));
510 
511 	done_seqretry(&vnode->cb_lock, seq);
512 	return 0;
513 }
514 
515 /*
516  * discard an AFS inode
517  */
518 int afs_drop_inode(struct inode *inode)
519 {
520 	_enter("");
521 
522 	if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
523 		return generic_delete_inode(inode);
524 	else
525 		return generic_drop_inode(inode);
526 }
527 
528 /*
529  * clear an AFS inode
530  */
531 void afs_evict_inode(struct inode *inode)
532 {
533 	struct afs_vnode *vnode;
534 
535 	vnode = AFS_FS_I(inode);
536 
537 	_enter("{%llx:%llu.%d}",
538 	       vnode->fid.vid,
539 	       vnode->fid.vnode,
540 	       vnode->fid.unique);
541 
542 	_debug("CLEAR INODE %p", inode);
543 
544 	ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
545 
546 	truncate_inode_pages_final(&inode->i_data);
547 	clear_inode(inode);
548 
549 	if (vnode->cb_interest) {
550 		afs_put_cb_interest(afs_i2net(inode), vnode->cb_interest);
551 		vnode->cb_interest = NULL;
552 	}
553 
554 	while (!list_empty(&vnode->wb_keys)) {
555 		struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
556 						    struct afs_wb_key, vnode_link);
557 		list_del(&wbk->vnode_link);
558 		afs_put_wb_key(wbk);
559 	}
560 
561 #ifdef CONFIG_AFS_FSCACHE
562 	{
563 		struct afs_vnode_cache_aux aux;
564 
565 		aux.data_version = vnode->status.data_version;
566 		fscache_relinquish_cookie(vnode->cache, &aux,
567 					  test_bit(AFS_VNODE_DELETED, &vnode->flags));
568 		vnode->cache = NULL;
569 	}
570 #endif
571 
572 	afs_put_permits(rcu_access_pointer(vnode->permit_cache));
573 	key_put(vnode->silly_key);
574 	vnode->silly_key = NULL;
575 	key_put(vnode->lock_key);
576 	vnode->lock_key = NULL;
577 	_leave("");
578 }
579 
580 /*
581  * set the attributes of an inode
582  */
583 int afs_setattr(struct dentry *dentry, struct iattr *attr)
584 {
585 	struct afs_fs_cursor fc;
586 	struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
587 	struct key *key;
588 	int ret;
589 
590 	_enter("{%llx:%llu},{n=%pd},%x",
591 	       vnode->fid.vid, vnode->fid.vnode, dentry,
592 	       attr->ia_valid);
593 
594 	if (!(attr->ia_valid & (ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
595 				ATTR_MTIME))) {
596 		_leave(" = 0 [unsupported]");
597 		return 0;
598 	}
599 
600 	/* flush any dirty data outstanding on a regular file */
601 	if (S_ISREG(vnode->vfs_inode.i_mode))
602 		filemap_write_and_wait(vnode->vfs_inode.i_mapping);
603 
604 	if (attr->ia_valid & ATTR_FILE) {
605 		key = afs_file_key(attr->ia_file);
606 	} else {
607 		key = afs_request_key(vnode->volume->cell);
608 		if (IS_ERR(key)) {
609 			ret = PTR_ERR(key);
610 			goto error;
611 		}
612 	}
613 
614 	ret = -ERESTARTSYS;
615 	if (afs_begin_vnode_operation(&fc, vnode, key)) {
616 		while (afs_select_fileserver(&fc)) {
617 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
618 			afs_fs_setattr(&fc, attr);
619 		}
620 
621 		afs_check_for_remote_deletion(&fc, fc.vnode);
622 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
623 		ret = afs_end_vnode_operation(&fc);
624 	}
625 
626 	if (!(attr->ia_valid & ATTR_FILE))
627 		key_put(key);
628 
629 error:
630 	_leave(" = %d", ret);
631 	return ret;
632 }
633