xref: /openbmc/linux/fs/afs/inode.c (revision 2c6467d2) 
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  * Initialise an inode from the vnode status.
34  */
35 static int afs_inode_init_from_status(struct afs_vnode *vnode, struct key *key)
36 {
37 	struct inode *inode = AFS_VNODE_TO_I(vnode);
38 
39 	_debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
40 	       vnode->status.type,
41 	       vnode->status.nlink,
42 	       (unsigned long long) vnode->status.size,
43 	       vnode->status.data_version,
44 	       vnode->status.mode);
45 
46 	read_seqlock_excl(&vnode->cb_lock);
47 
48 	afs_update_inode_from_status(vnode, &vnode->status, NULL,
49 				     AFS_VNODE_NOT_YET_SET);
50 
51 	switch (vnode->status.type) {
52 	case AFS_FTYPE_FILE:
53 		inode->i_mode	= S_IFREG | vnode->status.mode;
54 		inode->i_op	= &afs_file_inode_operations;
55 		inode->i_fop	= &afs_file_operations;
56 		inode->i_mapping->a_ops	= &afs_fs_aops;
57 		break;
58 	case AFS_FTYPE_DIR:
59 		inode->i_mode	= S_IFDIR | vnode->status.mode;
60 		inode->i_op	= &afs_dir_inode_operations;
61 		inode->i_fop	= &afs_dir_file_operations;
62 		inode->i_mapping->a_ops	= &afs_dir_aops;
63 		break;
64 	case AFS_FTYPE_SYMLINK:
65 		/* Symlinks with a mode of 0644 are actually mountpoints. */
66 		if ((vnode->status.mode & 0777) == 0644) {
67 			inode->i_flags |= S_AUTOMOUNT;
68 
69 			set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
70 
71 			inode->i_mode	= S_IFDIR | 0555;
72 			inode->i_op	= &afs_mntpt_inode_operations;
73 			inode->i_fop	= &afs_mntpt_file_operations;
74 			inode->i_mapping->a_ops	= &afs_fs_aops;
75 		} else {
76 			inode->i_mode	= S_IFLNK | vnode->status.mode;
77 			inode->i_op	= &afs_symlink_inode_operations;
78 			inode->i_mapping->a_ops	= &afs_fs_aops;
79 		}
80 		inode_nohighmem(inode);
81 		break;
82 	default:
83 		printk("kAFS: AFS vnode with undefined type\n");
84 		read_sequnlock_excl(&vnode->cb_lock);
85 		return afs_protocol_error(NULL, -EBADMSG, afs_eproto_file_type);
86 	}
87 
88 	inode->i_blocks		= 0;
89 	vnode->invalid_before	= vnode->status.data_version;
90 
91 	read_sequnlock_excl(&vnode->cb_lock);
92 	return 0;
93 }
94 
95 /*
96  * Fetch file status from the volume.
97  */
98 int afs_fetch_status(struct afs_vnode *vnode, struct key *key, bool new_inode)
99 {
100 	struct afs_fs_cursor fc;
101 	int ret;
102 
103 	_enter("%s,{%llx:%llu.%u,S=%lx}",
104 	       vnode->volume->name,
105 	       vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
106 	       vnode->flags);
107 
108 	ret = -ERESTARTSYS;
109 	if (afs_begin_vnode_operation(&fc, vnode, key)) {
110 		while (afs_select_fileserver(&fc)) {
111 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
112 			afs_fs_fetch_file_status(&fc, NULL, new_inode);
113 		}
114 
115 		afs_check_for_remote_deletion(&fc, fc.vnode);
116 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
117 		ret = afs_end_vnode_operation(&fc);
118 	}
119 
120 	_leave(" = %d", ret);
121 	return ret;
122 }
123 
124 /*
125  * iget5() comparator
126  */
127 int afs_iget5_test(struct inode *inode, void *opaque)
128 {
129 	struct afs_iget_data *data = opaque;
130 	struct afs_vnode *vnode = AFS_FS_I(inode);
131 
132 	return memcmp(&vnode->fid, &data->fid, sizeof(data->fid)) == 0;
133 }
134 
135 /*
136  * iget5() comparator for inode created by autocell operations
137  *
138  * These pseudo inodes don't match anything.
139  */
140 static int afs_iget5_pseudo_dir_test(struct inode *inode, void *opaque)
141 {
142 	return 0;
143 }
144 
145 /*
146  * iget5() inode initialiser
147  */
148 static int afs_iget5_set(struct inode *inode, void *opaque)
149 {
150 	struct afs_iget_data *data = opaque;
151 	struct afs_vnode *vnode = AFS_FS_I(inode);
152 
153 	vnode->fid = data->fid;
154 	vnode->volume = data->volume;
155 
156 	/* YFS supports 96-bit vnode IDs, but Linux only supports
157 	 * 64-bit inode numbers.
158 	 */
159 	inode->i_ino = data->fid.vnode;
160 	inode->i_generation = data->fid.unique;
161 	return 0;
162 }
163 
164 /*
165  * Create an inode for a dynamic root directory or an autocell dynamic
166  * automount dir.
167  */
168 struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
169 {
170 	struct afs_iget_data data;
171 	struct afs_super_info *as;
172 	struct afs_vnode *vnode;
173 	struct inode *inode;
174 	static atomic_t afs_autocell_ino;
175 
176 	_enter("");
177 
178 	as = sb->s_fs_info;
179 	if (as->volume) {
180 		data.volume = as->volume;
181 		data.fid.vid = as->volume->vid;
182 	}
183 	if (root) {
184 		data.fid.vnode = 1;
185 		data.fid.unique = 1;
186 	} else {
187 		data.fid.vnode = atomic_inc_return(&afs_autocell_ino);
188 		data.fid.unique = 0;
189 	}
190 
191 	inode = iget5_locked(sb, data.fid.vnode,
192 			     afs_iget5_pseudo_dir_test, afs_iget5_set,
193 			     &data);
194 	if (!inode) {
195 		_leave(" = -ENOMEM");
196 		return ERR_PTR(-ENOMEM);
197 	}
198 
199 	_debug("GOT INODE %p { ino=%lu, vl=%llx, vn=%llx, u=%x }",
200 	       inode, inode->i_ino, data.fid.vid, data.fid.vnode,
201 	       data.fid.unique);
202 
203 	vnode = AFS_FS_I(inode);
204 
205 	/* there shouldn't be an existing inode */
206 	BUG_ON(!(inode->i_state & I_NEW));
207 
208 	inode->i_size		= 0;
209 	inode->i_mode		= S_IFDIR | S_IRUGO | S_IXUGO;
210 	if (root) {
211 		inode->i_op	= &afs_dynroot_inode_operations;
212 		inode->i_fop	= &afs_dynroot_file_operations;
213 	} else {
214 		inode->i_op	= &afs_autocell_inode_operations;
215 	}
216 	set_nlink(inode, 2);
217 	inode->i_uid		= GLOBAL_ROOT_UID;
218 	inode->i_gid		= GLOBAL_ROOT_GID;
219 	inode->i_ctime.tv_sec	= get_seconds();
220 	inode->i_ctime.tv_nsec	= 0;
221 	inode->i_atime		= inode->i_mtime = inode->i_ctime;
222 	inode->i_blocks		= 0;
223 	inode_set_iversion_raw(inode, 0);
224 	inode->i_generation	= 0;
225 
226 	set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
227 	if (!root) {
228 		set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
229 		inode->i_flags |= S_AUTOMOUNT;
230 	}
231 
232 	inode->i_flags |= S_NOATIME;
233 	unlock_new_inode(inode);
234 	_leave(" = %p", inode);
235 	return inode;
236 }
237 
238 /*
239  * Get a cache cookie for an inode.
240  */
241 static void afs_get_inode_cache(struct afs_vnode *vnode)
242 {
243 #ifdef CONFIG_AFS_FSCACHE
244 	struct {
245 		u32 vnode_id;
246 		u32 unique;
247 		u32 vnode_id_ext[2];	/* Allow for a 96-bit key */
248 	} __packed key;
249 	struct afs_vnode_cache_aux aux;
250 
251 	if (vnode->status.type == AFS_FTYPE_DIR) {
252 		vnode->cache = NULL;
253 		return;
254 	}
255 
256 	key.vnode_id		= vnode->fid.vnode;
257 	key.unique		= vnode->fid.unique;
258 	key.vnode_id_ext[0]	= vnode->fid.vnode >> 32;
259 	key.vnode_id_ext[1]	= vnode->fid.vnode_hi;
260 	aux.data_version	= vnode->status.data_version;
261 
262 	vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
263 					      &afs_vnode_cache_index_def,
264 					      &key, sizeof(key),
265 					      &aux, sizeof(aux),
266 					      vnode, vnode->status.size, true);
267 #endif
268 }
269 
270 /*
271  * inode retrieval
272  */
273 struct inode *afs_iget(struct super_block *sb, struct key *key,
274 		       struct afs_fid *fid, struct afs_file_status *status,
275 		       struct afs_callback *cb, struct afs_cb_interest *cbi)
276 {
277 	struct afs_iget_data data = { .fid = *fid };
278 	struct afs_super_info *as;
279 	struct afs_vnode *vnode;
280 	struct inode *inode;
281 	int ret;
282 
283 	_enter(",{%llx:%llu.%u},,", fid->vid, fid->vnode, fid->unique);
284 
285 	as = sb->s_fs_info;
286 	data.volume = as->volume;
287 
288 	inode = iget5_locked(sb, fid->vnode, afs_iget5_test, afs_iget5_set,
289 			     &data);
290 	if (!inode) {
291 		_leave(" = -ENOMEM");
292 		return ERR_PTR(-ENOMEM);
293 	}
294 
295 	_debug("GOT INODE %p { vl=%llx vn=%llx, u=%x }",
296 	       inode, fid->vid, fid->vnode, fid->unique);
297 
298 	vnode = AFS_FS_I(inode);
299 
300 	/* deal with an existing inode */
301 	if (!(inode->i_state & I_NEW)) {
302 		_leave(" = %p", inode);
303 		return inode;
304 	}
305 
306 	if (!status) {
307 		/* it's a remotely extant inode */
308 		ret = afs_fetch_status(vnode, key, true);
309 		if (ret < 0)
310 			goto bad_inode;
311 	} else {
312 		/* it's an inode we just created */
313 		memcpy(&vnode->status, status, sizeof(vnode->status));
314 
315 		if (!cb) {
316 			/* it's a symlink we just created (the fileserver
317 			 * didn't give us a callback) */
318 			vnode->cb_version = 0;
319 			vnode->cb_type = 0;
320 			vnode->cb_expires_at = ktime_get();
321 		} else {
322 			vnode->cb_version = cb->version;
323 			vnode->cb_type = cb->type;
324 			vnode->cb_expires_at = cb->expires_at;
325 			vnode->cb_interest = afs_get_cb_interest(cbi);
326 			set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
327 		}
328 
329 		vnode->cb_expires_at += ktime_get_real_seconds();
330 	}
331 
332 	ret = afs_inode_init_from_status(vnode, key);
333 	if (ret < 0)
334 		goto bad_inode;
335 
336 	afs_get_inode_cache(vnode);
337 
338 	/* success */
339 	clear_bit(AFS_VNODE_UNSET, &vnode->flags);
340 	inode->i_flags |= S_NOATIME;
341 	unlock_new_inode(inode);
342 	_leave(" = %p [CB { v=%u t=%u }]", inode, vnode->cb_version, vnode->cb_type);
343 	return inode;
344 
345 	/* failure */
346 bad_inode:
347 	iget_failed(inode);
348 	_leave(" = %d [bad]", ret);
349 	return ERR_PTR(ret);
350 }
351 
352 /*
353  * mark the data attached to an inode as obsolete due to a write on the server
354  * - might also want to ditch all the outstanding writes and dirty pages
355  */
356 void afs_zap_data(struct afs_vnode *vnode)
357 {
358 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
359 
360 #ifdef CONFIG_AFS_FSCACHE
361 	fscache_invalidate(vnode->cache);
362 #endif
363 
364 	/* nuke all the non-dirty pages that aren't locked, mapped or being
365 	 * written back in a regular file and completely discard the pages in a
366 	 * directory or symlink */
367 	if (S_ISREG(vnode->vfs_inode.i_mode))
368 		invalidate_remote_inode(&vnode->vfs_inode);
369 	else
370 		invalidate_inode_pages2(vnode->vfs_inode.i_mapping);
371 }
372 
373 /*
374  * validate a vnode/inode
375  * - there are several things we need to check
376  *   - parent dir data changes (rm, rmdir, rename, mkdir, create, link,
377  *     symlink)
378  *   - parent dir metadata changed (security changes)
379  *   - dentry data changed (write, truncate)
380  *   - dentry metadata changed (security changes)
381  */
382 int afs_validate(struct afs_vnode *vnode, struct key *key)
383 {
384 	time64_t now = ktime_get_real_seconds();
385 	bool valid;
386 	int ret;
387 
388 	_enter("{v={%llx:%llu} fl=%lx},%x",
389 	       vnode->fid.vid, vnode->fid.vnode, vnode->flags,
390 	       key_serial(key));
391 
392 	/* Quickly check the callback state.  Ideally, we'd use read_seqbegin
393 	 * here, but we have no way to pass the net namespace to the RCU
394 	 * cleanup for the server record.
395 	 */
396 	read_seqlock_excl(&vnode->cb_lock);
397 
398 	if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
399 		if (vnode->cb_s_break != vnode->cb_interest->server->cb_s_break ||
400 		    vnode->cb_v_break != vnode->volume->cb_v_break) {
401 			vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
402 			vnode->cb_v_break = vnode->volume->cb_v_break;
403 			valid = false;
404 		} else if (vnode->status.type == AFS_FTYPE_DIR &&
405 			   (!test_bit(AFS_VNODE_DIR_VALID, &vnode->flags) ||
406 			    vnode->cb_expires_at - 10 <= now)) {
407 			valid = false;
408 		} else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags) ||
409 			   vnode->cb_expires_at - 10 <= now) {
410 			valid = false;
411 		} else {
412 			valid = true;
413 		}
414 	} else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
415 		valid = true;
416 	} else {
417 		vnode->cb_s_break = vnode->cb_interest->server->cb_s_break;
418 		vnode->cb_v_break = vnode->volume->cb_v_break;
419 		valid = false;
420 	}
421 
422 	read_sequnlock_excl(&vnode->cb_lock);
423 
424 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
425 		clear_nlink(&vnode->vfs_inode);
426 
427 	if (valid)
428 		goto valid;
429 
430 	down_write(&vnode->validate_lock);
431 
432 	/* if the promise has expired, we need to check the server again to get
433 	 * a new promise - note that if the (parent) directory's metadata was
434 	 * changed then the security may be different and we may no longer have
435 	 * access */
436 	if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
437 		_debug("not promised");
438 		ret = afs_fetch_status(vnode, key, false);
439 		if (ret < 0) {
440 			if (ret == -ENOENT) {
441 				set_bit(AFS_VNODE_DELETED, &vnode->flags);
442 				ret = -ESTALE;
443 			}
444 			goto error_unlock;
445 		}
446 		_debug("new promise [fl=%lx]", vnode->flags);
447 	}
448 
449 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
450 		_debug("file already deleted");
451 		ret = -ESTALE;
452 		goto error_unlock;
453 	}
454 
455 	/* if the vnode's data version number changed then its contents are
456 	 * different */
457 	if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
458 		afs_zap_data(vnode);
459 	up_write(&vnode->validate_lock);
460 valid:
461 	_leave(" = 0");
462 	return 0;
463 
464 error_unlock:
465 	up_write(&vnode->validate_lock);
466 	_leave(" = %d", ret);
467 	return ret;
468 }
469 
470 /*
471  * read the attributes of an inode
472  */
473 int afs_getattr(const struct path *path, struct kstat *stat,
474 		u32 request_mask, unsigned int query_flags)
475 {
476 	struct inode *inode = d_inode(path->dentry);
477 	struct afs_vnode *vnode = AFS_FS_I(inode);
478 	int seq = 0;
479 
480 	_enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
481 
482 	do {
483 		read_seqbegin_or_lock(&vnode->cb_lock, &seq);
484 		generic_fillattr(inode, stat);
485 	} while (need_seqretry(&vnode->cb_lock, seq));
486 
487 	done_seqretry(&vnode->cb_lock, seq);
488 	return 0;
489 }
490 
491 /*
492  * discard an AFS inode
493  */
494 int afs_drop_inode(struct inode *inode)
495 {
496 	_enter("");
497 
498 	if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
499 		return generic_delete_inode(inode);
500 	else
501 		return generic_drop_inode(inode);
502 }
503 
504 /*
505  * clear an AFS inode
506  */
507 void afs_evict_inode(struct inode *inode)
508 {
509 	struct afs_vnode *vnode;
510 
511 	vnode = AFS_FS_I(inode);
512 
513 	_enter("{%llx:%llu.%d}",
514 	       vnode->fid.vid,
515 	       vnode->fid.vnode,
516 	       vnode->fid.unique);
517 
518 	_debug("CLEAR INODE %p", inode);
519 
520 	ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
521 
522 	truncate_inode_pages_final(&inode->i_data);
523 	clear_inode(inode);
524 
525 	if (vnode->cb_interest) {
526 		afs_put_cb_interest(afs_i2net(inode), vnode->cb_interest);
527 		vnode->cb_interest = NULL;
528 	}
529 
530 	while (!list_empty(&vnode->wb_keys)) {
531 		struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
532 						    struct afs_wb_key, vnode_link);
533 		list_del(&wbk->vnode_link);
534 		afs_put_wb_key(wbk);
535 	}
536 
537 #ifdef CONFIG_AFS_FSCACHE
538 	{
539 		struct afs_vnode_cache_aux aux;
540 
541 		aux.data_version = vnode->status.data_version;
542 		fscache_relinquish_cookie(vnode->cache, &aux,
543 					  test_bit(AFS_VNODE_DELETED, &vnode->flags));
544 		vnode->cache = NULL;
545 	}
546 #endif
547 
548 	afs_put_permits(rcu_access_pointer(vnode->permit_cache));
549 	_leave("");
550 }
551 
552 /*
553  * set the attributes of an inode
554  */
555 int afs_setattr(struct dentry *dentry, struct iattr *attr)
556 {
557 	struct afs_fs_cursor fc;
558 	struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
559 	struct key *key;
560 	int ret;
561 
562 	_enter("{%llx:%llu},{n=%pd},%x",
563 	       vnode->fid.vid, vnode->fid.vnode, dentry,
564 	       attr->ia_valid);
565 
566 	if (!(attr->ia_valid & (ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
567 				ATTR_MTIME))) {
568 		_leave(" = 0 [unsupported]");
569 		return 0;
570 	}
571 
572 	/* flush any dirty data outstanding on a regular file */
573 	if (S_ISREG(vnode->vfs_inode.i_mode))
574 		filemap_write_and_wait(vnode->vfs_inode.i_mapping);
575 
576 	if (attr->ia_valid & ATTR_FILE) {
577 		key = afs_file_key(attr->ia_file);
578 	} else {
579 		key = afs_request_key(vnode->volume->cell);
580 		if (IS_ERR(key)) {
581 			ret = PTR_ERR(key);
582 			goto error;
583 		}
584 	}
585 
586 	ret = -ERESTARTSYS;
587 	if (afs_begin_vnode_operation(&fc, vnode, key)) {
588 		while (afs_select_fileserver(&fc)) {
589 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
590 			afs_fs_setattr(&fc, attr);
591 		}
592 
593 		afs_check_for_remote_deletion(&fc, fc.vnode);
594 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
595 		ret = afs_end_vnode_operation(&fc);
596 	}
597 
598 	if (!(attr->ia_valid & ATTR_FILE))
599 		key_put(key);
600 
601 error:
602 	_leave(" = %d", ret);
603 	return ret;
604 }
605