xref: /openbmc/linux/fs/afs/inode.c (revision 4fc4dca8)
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 #include "afs_fs.h"
27 
28 static const struct inode_operations afs_symlink_inode_operations = {
29 	.get_link	= page_get_link,
30 	.listxattr	= afs_listxattr,
31 };
32 
33 static noinline void dump_vnode(struct afs_vnode *vnode, struct afs_vnode *parent_vnode)
34 {
35 	static unsigned long once_only;
36 
37 	pr_warn("kAFS: AFS vnode with undefined type %u\n",
38 		vnode->status.type);
39 	pr_warn("kAFS: A=%d m=%o s=%llx v=%llx\n",
40 		vnode->status.abort_code,
41 		vnode->status.mode,
42 		vnode->status.size,
43 		vnode->status.data_version);
44 	pr_warn("kAFS: vnode %llx:%llx:%x\n",
45 		vnode->fid.vid,
46 		vnode->fid.vnode,
47 		vnode->fid.unique);
48 	if (parent_vnode)
49 		pr_warn("kAFS: dir %llx:%llx:%x\n",
50 			parent_vnode->fid.vid,
51 			parent_vnode->fid.vnode,
52 			parent_vnode->fid.unique);
53 
54 	if (!test_and_set_bit(0, &once_only))
55 		dump_stack();
56 }
57 
58 /*
59  * Initialise an inode from the vnode status.
60  */
61 static int afs_inode_init_from_status(struct afs_vnode *vnode, struct key *key,
62 				      struct afs_cb_interest *cbi,
63 				      struct afs_vnode *parent_vnode,
64 				      struct afs_status_cb *scb)
65 {
66 	struct afs_cb_interest *old_cbi = NULL;
67 	struct afs_file_status *status = &scb->status;
68 	struct inode *inode = AFS_VNODE_TO_I(vnode);
69 	struct timespec64 t;
70 
71 	_debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
72 	       status->type,
73 	       status->nlink,
74 	       (unsigned long long) status->size,
75 	       status->data_version,
76 	       status->mode);
77 
78 	write_seqlock(&vnode->cb_lock);
79 
80 	vnode->status = *status;
81 
82 	t = status->mtime_client;
83 	inode->i_ctime = t;
84 	inode->i_mtime = t;
85 	inode->i_atime = t;
86 	inode->i_uid = make_kuid(&init_user_ns, status->owner);
87 	inode->i_gid = make_kgid(&init_user_ns, status->group);
88 	set_nlink(&vnode->vfs_inode, status->nlink);
89 
90 	switch (status->type) {
91 	case AFS_FTYPE_FILE:
92 		inode->i_mode	= S_IFREG | status->mode;
93 		inode->i_op	= &afs_file_inode_operations;
94 		inode->i_fop	= &afs_file_operations;
95 		inode->i_mapping->a_ops	= &afs_fs_aops;
96 		break;
97 	case AFS_FTYPE_DIR:
98 		inode->i_mode	= S_IFDIR | status->mode;
99 		inode->i_op	= &afs_dir_inode_operations;
100 		inode->i_fop	= &afs_dir_file_operations;
101 		inode->i_mapping->a_ops	= &afs_dir_aops;
102 		break;
103 	case AFS_FTYPE_SYMLINK:
104 		/* Symlinks with a mode of 0644 are actually mountpoints. */
105 		if ((status->mode & 0777) == 0644) {
106 			inode->i_flags |= S_AUTOMOUNT;
107 
108 			set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
109 
110 			inode->i_mode	= S_IFDIR | 0555;
111 			inode->i_op	= &afs_mntpt_inode_operations;
112 			inode->i_fop	= &afs_mntpt_file_operations;
113 			inode->i_mapping->a_ops	= &afs_fs_aops;
114 		} else {
115 			inode->i_mode	= S_IFLNK | status->mode;
116 			inode->i_op	= &afs_symlink_inode_operations;
117 			inode->i_mapping->a_ops	= &afs_fs_aops;
118 		}
119 		inode_nohighmem(inode);
120 		break;
121 	default:
122 		dump_vnode(vnode, parent_vnode);
123 		write_sequnlock(&vnode->cb_lock);
124 		return afs_protocol_error(NULL, -EBADMSG, afs_eproto_file_type);
125 	}
126 
127 	/*
128 	 * Estimate 512 bytes  blocks used, rounded up to nearest 1K
129 	 * for consistency with other AFS clients.
130 	 */
131 	inode->i_blocks		= ((i_size_read(inode) + 1023) >> 10) << 1;
132 	i_size_write(&vnode->vfs_inode, status->size);
133 
134 	vnode->invalid_before	= status->data_version;
135 	inode_set_iversion_raw(&vnode->vfs_inode, status->data_version);
136 
137 	if (!scb->have_cb) {
138 		/* it's a symlink we just created (the fileserver
139 		 * didn't give us a callback) */
140 		vnode->cb_expires_at = ktime_get_real_seconds();
141 	} else {
142 		vnode->cb_expires_at = scb->callback.expires_at;
143 		old_cbi = rcu_dereference_protected(vnode->cb_interest,
144 						    lockdep_is_held(&vnode->cb_lock.lock));
145 		if (cbi != old_cbi)
146 			rcu_assign_pointer(vnode->cb_interest, afs_get_cb_interest(cbi));
147 		else
148 			old_cbi = NULL;
149 		set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
150 	}
151 
152 	write_sequnlock(&vnode->cb_lock);
153 	afs_put_cb_interest(afs_v2net(vnode), old_cbi);
154 	return 0;
155 }
156 
157 /*
158  * Update the core inode struct from a returned status record.
159  */
160 static void afs_apply_status(struct afs_fs_cursor *fc,
161 			     struct afs_vnode *vnode,
162 			     struct afs_status_cb *scb,
163 			     const afs_dataversion_t *expected_version)
164 {
165 	struct afs_file_status *status = &scb->status;
166 	struct timespec64 t;
167 	umode_t mode;
168 	bool data_changed = false;
169 
170 	BUG_ON(test_bit(AFS_VNODE_UNSET, &vnode->flags));
171 
172 	if (status->type != vnode->status.type) {
173 		pr_warning("Vnode %llx:%llx:%x changed type %u to %u\n",
174 			   vnode->fid.vid,
175 			   vnode->fid.vnode,
176 			   vnode->fid.unique,
177 			   status->type, vnode->status.type);
178 		afs_protocol_error(NULL, -EBADMSG, afs_eproto_bad_status);
179 		return;
180 	}
181 
182 	if (status->nlink != vnode->status.nlink)
183 		set_nlink(&vnode->vfs_inode, status->nlink);
184 
185 	if (status->owner != vnode->status.owner)
186 		vnode->vfs_inode.i_uid = make_kuid(&init_user_ns, status->owner);
187 
188 	if (status->group != vnode->status.group)
189 		vnode->vfs_inode.i_gid = make_kgid(&init_user_ns, status->group);
190 
191 	if (status->mode != vnode->status.mode) {
192 		mode = vnode->vfs_inode.i_mode;
193 		mode &= ~S_IALLUGO;
194 		mode |= status->mode;
195 		WRITE_ONCE(vnode->vfs_inode.i_mode, mode);
196 	}
197 
198 	t = status->mtime_client;
199 	vnode->vfs_inode.i_ctime = t;
200 	vnode->vfs_inode.i_mtime = t;
201 	vnode->vfs_inode.i_atime = t;
202 
203 	if (vnode->status.data_version != status->data_version)
204 		data_changed = true;
205 
206 	vnode->status = *status;
207 
208 	if (expected_version &&
209 	    *expected_version != status->data_version) {
210 		kdebug("vnode modified %llx on {%llx:%llu} [exp %llx] %s",
211 		       (unsigned long long) status->data_version,
212 		       vnode->fid.vid, vnode->fid.vnode,
213 		       (unsigned long long) *expected_version,
214 		       fc->type ? fc->type->name : "???");
215 		vnode->invalid_before = status->data_version;
216 		if (vnode->status.type == AFS_FTYPE_DIR) {
217 			if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
218 				afs_stat_v(vnode, n_inval);
219 		} else {
220 			set_bit(AFS_VNODE_ZAP_DATA, &vnode->flags);
221 		}
222 	} else if (vnode->status.type == AFS_FTYPE_DIR) {
223 		/* Expected directory change is handled elsewhere so
224 		 * that we can locally edit the directory and save on a
225 		 * download.
226 		 */
227 		if (test_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
228 			data_changed = false;
229 	}
230 
231 	if (data_changed) {
232 		inode_set_iversion_raw(&vnode->vfs_inode, status->data_version);
233 		i_size_write(&vnode->vfs_inode, status->size);
234 	}
235 }
236 
237 /*
238  * Apply a callback to a vnode.
239  */
240 static void afs_apply_callback(struct afs_fs_cursor *fc,
241 			       struct afs_vnode *vnode,
242 			       struct afs_status_cb *scb,
243 			       unsigned int cb_break)
244 {
245 	struct afs_cb_interest *old;
246 	struct afs_callback *cb = &scb->callback;
247 
248 	if (!afs_cb_is_broken(cb_break, vnode, fc->cbi)) {
249 		vnode->cb_expires_at	= cb->expires_at;
250 		old = rcu_dereference_protected(vnode->cb_interest,
251 						lockdep_is_held(&vnode->cb_lock.lock));
252 		if (old != fc->cbi) {
253 			rcu_assign_pointer(vnode->cb_interest, afs_get_cb_interest(fc->cbi));
254 			afs_put_cb_interest(afs_v2net(vnode), old);
255 		}
256 		set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
257 	}
258 }
259 
260 /*
261  * Apply the received status and callback to an inode all in the same critical
262  * section to avoid races with afs_validate().
263  */
264 void afs_vnode_commit_status(struct afs_fs_cursor *fc,
265 			     struct afs_vnode *vnode,
266 			     unsigned int cb_break,
267 			     const afs_dataversion_t *expected_version,
268 			     struct afs_status_cb *scb)
269 {
270 	if (fc->ac.error != 0)
271 		return;
272 
273 	write_seqlock(&vnode->cb_lock);
274 
275 	if (scb->have_error) {
276 		if (scb->status.abort_code == VNOVNODE) {
277 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
278 			clear_nlink(&vnode->vfs_inode);
279 			__afs_break_callback(vnode);
280 		}
281 	} else {
282 		if (scb->have_status)
283 			afs_apply_status(fc, vnode, scb, expected_version);
284 		if (scb->have_cb)
285 			afs_apply_callback(fc, vnode, scb, cb_break);
286 	}
287 
288 	write_sequnlock(&vnode->cb_lock);
289 
290 	if (fc->ac.error == 0 && scb->have_status)
291 		afs_cache_permit(vnode, fc->key, cb_break, scb);
292 }
293 
294 /*
295  * Fetch file status from the volume.
296  */
297 int afs_fetch_status(struct afs_vnode *vnode, struct key *key, bool is_new,
298 		     afs_access_t *_caller_access)
299 {
300 	struct afs_status_cb *scb;
301 	struct afs_fs_cursor fc;
302 	int ret;
303 
304 	_enter("%s,{%llx:%llu.%u,S=%lx}",
305 	       vnode->volume->name,
306 	       vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
307 	       vnode->flags);
308 
309 	scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL);
310 	if (!scb)
311 		return -ENOMEM;
312 
313 	ret = -ERESTARTSYS;
314 	if (afs_begin_vnode_operation(&fc, vnode, key, true)) {
315 		afs_dataversion_t data_version = vnode->status.data_version;
316 
317 		while (afs_select_fileserver(&fc)) {
318 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
319 			afs_fs_fetch_file_status(&fc, scb, NULL);
320 		}
321 
322 		if (fc.error) {
323 			/* Do nothing. */
324 		} else if (is_new) {
325 			ret = afs_inode_init_from_status(vnode, key, fc.cbi,
326 							 NULL, scb);
327 			fc.error = ret;
328 			if (ret == 0)
329 				afs_cache_permit(vnode, key, fc.cb_break, scb);
330 		} else {
331 			afs_vnode_commit_status(&fc, vnode, fc.cb_break,
332 						&data_version, scb);
333 		}
334 		afs_check_for_remote_deletion(&fc, vnode);
335 		ret = afs_end_vnode_operation(&fc);
336 	}
337 
338 	if (ret == 0 && _caller_access)
339 		*_caller_access = scb->status.caller_access;
340 	kfree(scb);
341 	_leave(" = %d", ret);
342 	return ret;
343 }
344 
345 /*
346  * iget5() comparator
347  */
348 int afs_iget5_test(struct inode *inode, void *opaque)
349 {
350 	struct afs_iget_data *iget_data = opaque;
351 	struct afs_vnode *vnode = AFS_FS_I(inode);
352 
353 	return memcmp(&vnode->fid, &iget_data->fid, sizeof(iget_data->fid)) == 0;
354 }
355 
356 /*
357  * iget5() comparator for inode created by autocell operations
358  *
359  * These pseudo inodes don't match anything.
360  */
361 static int afs_iget5_pseudo_dir_test(struct inode *inode, void *opaque)
362 {
363 	return 0;
364 }
365 
366 /*
367  * iget5() inode initialiser
368  */
369 static int afs_iget5_set(struct inode *inode, void *opaque)
370 {
371 	struct afs_iget_data *iget_data = opaque;
372 	struct afs_vnode *vnode = AFS_FS_I(inode);
373 
374 	vnode->fid		= iget_data->fid;
375 	vnode->volume		= iget_data->volume;
376 	vnode->cb_v_break	= iget_data->cb_v_break;
377 	vnode->cb_s_break	= iget_data->cb_s_break;
378 
379 	/* YFS supports 96-bit vnode IDs, but Linux only supports
380 	 * 64-bit inode numbers.
381 	 */
382 	inode->i_ino		= iget_data->fid.vnode;
383 	inode->i_generation	= iget_data->fid.unique;
384 	return 0;
385 }
386 
387 /*
388  * Create an inode for a dynamic root directory or an autocell dynamic
389  * automount dir.
390  */
391 struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
392 {
393 	struct afs_super_info *as;
394 	struct afs_vnode *vnode;
395 	struct inode *inode;
396 	static atomic_t afs_autocell_ino;
397 
398 	struct afs_iget_data iget_data = {
399 		.cb_v_break = 0,
400 		.cb_s_break = 0,
401 	};
402 
403 	_enter("");
404 
405 	as = sb->s_fs_info;
406 	if (as->volume) {
407 		iget_data.volume = as->volume;
408 		iget_data.fid.vid = as->volume->vid;
409 	}
410 	if (root) {
411 		iget_data.fid.vnode = 1;
412 		iget_data.fid.unique = 1;
413 	} else {
414 		iget_data.fid.vnode = atomic_inc_return(&afs_autocell_ino);
415 		iget_data.fid.unique = 0;
416 	}
417 
418 	inode = iget5_locked(sb, iget_data.fid.vnode,
419 			     afs_iget5_pseudo_dir_test, afs_iget5_set,
420 			     &iget_data);
421 	if (!inode) {
422 		_leave(" = -ENOMEM");
423 		return ERR_PTR(-ENOMEM);
424 	}
425 
426 	_debug("GOT INODE %p { ino=%lu, vl=%llx, vn=%llx, u=%x }",
427 	       inode, inode->i_ino, iget_data.fid.vid, iget_data.fid.vnode,
428 	       iget_data.fid.unique);
429 
430 	vnode = AFS_FS_I(inode);
431 
432 	/* there shouldn't be an existing inode */
433 	BUG_ON(!(inode->i_state & I_NEW));
434 
435 	inode->i_size		= 0;
436 	inode->i_mode		= S_IFDIR | S_IRUGO | S_IXUGO;
437 	if (root) {
438 		inode->i_op	= &afs_dynroot_inode_operations;
439 		inode->i_fop	= &afs_dynroot_file_operations;
440 	} else {
441 		inode->i_op	= &afs_autocell_inode_operations;
442 	}
443 	set_nlink(inode, 2);
444 	inode->i_uid		= GLOBAL_ROOT_UID;
445 	inode->i_gid		= GLOBAL_ROOT_GID;
446 	inode->i_ctime = inode->i_atime = inode->i_mtime = current_time(inode);
447 	inode->i_blocks		= 0;
448 	inode_set_iversion_raw(inode, 0);
449 	inode->i_generation	= 0;
450 
451 	set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
452 	if (!root) {
453 		set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
454 		inode->i_flags |= S_AUTOMOUNT;
455 	}
456 
457 	inode->i_flags |= S_NOATIME;
458 	unlock_new_inode(inode);
459 	_leave(" = %p", inode);
460 	return inode;
461 }
462 
463 /*
464  * Get a cache cookie for an inode.
465  */
466 static void afs_get_inode_cache(struct afs_vnode *vnode)
467 {
468 #ifdef CONFIG_AFS_FSCACHE
469 	struct {
470 		u32 vnode_id;
471 		u32 unique;
472 		u32 vnode_id_ext[2];	/* Allow for a 96-bit key */
473 	} __packed key;
474 	struct afs_vnode_cache_aux aux;
475 
476 	if (vnode->status.type == AFS_FTYPE_DIR) {
477 		vnode->cache = NULL;
478 		return;
479 	}
480 
481 	key.vnode_id		= vnode->fid.vnode;
482 	key.unique		= vnode->fid.unique;
483 	key.vnode_id_ext[0]	= vnode->fid.vnode >> 32;
484 	key.vnode_id_ext[1]	= vnode->fid.vnode_hi;
485 	aux.data_version	= vnode->status.data_version;
486 
487 	vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
488 					      &afs_vnode_cache_index_def,
489 					      &key, sizeof(key),
490 					      &aux, sizeof(aux),
491 					      vnode, vnode->status.size, true);
492 #endif
493 }
494 
495 /*
496  * inode retrieval
497  */
498 struct inode *afs_iget(struct super_block *sb, struct key *key,
499 		       struct afs_iget_data *iget_data,
500 		       struct afs_status_cb *scb,
501 		       struct afs_cb_interest *cbi,
502 		       struct afs_vnode *parent_vnode)
503 {
504 	struct afs_super_info *as;
505 	struct afs_vnode *vnode;
506 	struct afs_fid *fid = &iget_data->fid;
507 	struct inode *inode;
508 	int ret;
509 
510 	_enter(",{%llx:%llu.%u},,", fid->vid, fid->vnode, fid->unique);
511 
512 	as = sb->s_fs_info;
513 	iget_data->volume = as->volume;
514 
515 	inode = iget5_locked(sb, fid->vnode, afs_iget5_test, afs_iget5_set,
516 			     iget_data);
517 	if (!inode) {
518 		_leave(" = -ENOMEM");
519 		return ERR_PTR(-ENOMEM);
520 	}
521 
522 	_debug("GOT INODE %p { vl=%llx vn=%llx, u=%x }",
523 	       inode, fid->vid, fid->vnode, fid->unique);
524 
525 	vnode = AFS_FS_I(inode);
526 
527 	/* deal with an existing inode */
528 	if (!(inode->i_state & I_NEW)) {
529 		_leave(" = %p", inode);
530 		return inode;
531 	}
532 
533 	if (!scb) {
534 		/* it's a remotely extant inode */
535 		ret = afs_fetch_status(vnode, key, true, NULL);
536 		if (ret < 0)
537 			goto bad_inode;
538 	} else {
539 		ret = afs_inode_init_from_status(vnode, key, cbi, parent_vnode,
540 						 scb);
541 		if (ret < 0)
542 			goto bad_inode;
543 	}
544 
545 	afs_get_inode_cache(vnode);
546 
547 	/* success */
548 	clear_bit(AFS_VNODE_UNSET, &vnode->flags);
549 	inode->i_flags |= S_NOATIME;
550 	unlock_new_inode(inode);
551 	_leave(" = %p", inode);
552 	return inode;
553 
554 	/* failure */
555 bad_inode:
556 	iget_failed(inode);
557 	_leave(" = %d [bad]", ret);
558 	return ERR_PTR(ret);
559 }
560 
561 /*
562  * mark the data attached to an inode as obsolete due to a write on the server
563  * - might also want to ditch all the outstanding writes and dirty pages
564  */
565 void afs_zap_data(struct afs_vnode *vnode)
566 {
567 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
568 
569 #ifdef CONFIG_AFS_FSCACHE
570 	fscache_invalidate(vnode->cache);
571 #endif
572 
573 	/* nuke all the non-dirty pages that aren't locked, mapped or being
574 	 * written back in a regular file and completely discard the pages in a
575 	 * directory or symlink */
576 	if (S_ISREG(vnode->vfs_inode.i_mode))
577 		invalidate_remote_inode(&vnode->vfs_inode);
578 	else
579 		invalidate_inode_pages2(vnode->vfs_inode.i_mapping);
580 }
581 
582 /*
583  * Check the validity of a vnode/inode.
584  */
585 bool afs_check_validity(struct afs_vnode *vnode)
586 {
587 	struct afs_cb_interest *cbi;
588 	struct afs_server *server;
589 	struct afs_volume *volume = vnode->volume;
590 	time64_t now = ktime_get_real_seconds();
591 	bool valid, need_clear = false;
592 	unsigned int cb_break, cb_s_break, cb_v_break;
593 	int seq = 0;
594 
595 	do {
596 		read_seqbegin_or_lock(&vnode->cb_lock, &seq);
597 		cb_v_break = READ_ONCE(volume->cb_v_break);
598 		cb_break = vnode->cb_break;
599 
600 		if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
601 			cbi = rcu_dereference(vnode->cb_interest);
602 			server = rcu_dereference(cbi->server);
603 			cb_s_break = READ_ONCE(server->cb_s_break);
604 
605 			if (vnode->cb_s_break != cb_s_break ||
606 			    vnode->cb_v_break != cb_v_break) {
607 				vnode->cb_s_break = cb_s_break;
608 				vnode->cb_v_break = cb_v_break;
609 				need_clear = true;
610 				valid = false;
611 			} else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) {
612 				need_clear = true;
613 				valid = false;
614 			} else if (vnode->cb_expires_at - 10 <= now) {
615 				need_clear = true;
616 				valid = false;
617 			} else {
618 				valid = true;
619 			}
620 		} else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
621 			valid = true;
622 		} else {
623 			vnode->cb_v_break = cb_v_break;
624 			valid = false;
625 		}
626 
627 	} while (need_seqretry(&vnode->cb_lock, seq));
628 
629 	done_seqretry(&vnode->cb_lock, seq);
630 
631 	if (need_clear) {
632 		write_seqlock(&vnode->cb_lock);
633 		if (cb_break == vnode->cb_break)
634 			__afs_break_callback(vnode);
635 		write_sequnlock(&vnode->cb_lock);
636 		valid = false;
637 	}
638 
639 	return valid;
640 }
641 
642 /*
643  * validate a vnode/inode
644  * - there are several things we need to check
645  *   - parent dir data changes (rm, rmdir, rename, mkdir, create, link,
646  *     symlink)
647  *   - parent dir metadata changed (security changes)
648  *   - dentry data changed (write, truncate)
649  *   - dentry metadata changed (security changes)
650  */
651 int afs_validate(struct afs_vnode *vnode, struct key *key)
652 {
653 	bool valid;
654 	int ret;
655 
656 	_enter("{v={%llx:%llu} fl=%lx},%x",
657 	       vnode->fid.vid, vnode->fid.vnode, vnode->flags,
658 	       key_serial(key));
659 
660 	rcu_read_lock();
661 	valid = afs_check_validity(vnode);
662 	rcu_read_unlock();
663 
664 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
665 		clear_nlink(&vnode->vfs_inode);
666 
667 	if (valid)
668 		goto valid;
669 
670 	down_write(&vnode->validate_lock);
671 
672 	/* if the promise has expired, we need to check the server again to get
673 	 * a new promise - note that if the (parent) directory's metadata was
674 	 * changed then the security may be different and we may no longer have
675 	 * access */
676 	if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
677 		_debug("not promised");
678 		ret = afs_fetch_status(vnode, key, false, NULL);
679 		if (ret < 0) {
680 			if (ret == -ENOENT) {
681 				set_bit(AFS_VNODE_DELETED, &vnode->flags);
682 				ret = -ESTALE;
683 			}
684 			goto error_unlock;
685 		}
686 		_debug("new promise [fl=%lx]", vnode->flags);
687 	}
688 
689 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
690 		_debug("file already deleted");
691 		ret = -ESTALE;
692 		goto error_unlock;
693 	}
694 
695 	/* if the vnode's data version number changed then its contents are
696 	 * different */
697 	if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
698 		afs_zap_data(vnode);
699 	up_write(&vnode->validate_lock);
700 valid:
701 	_leave(" = 0");
702 	return 0;
703 
704 error_unlock:
705 	up_write(&vnode->validate_lock);
706 	_leave(" = %d", ret);
707 	return ret;
708 }
709 
710 /*
711  * read the attributes of an inode
712  */
713 int afs_getattr(const struct path *path, struct kstat *stat,
714 		u32 request_mask, unsigned int query_flags)
715 {
716 	struct inode *inode = d_inode(path->dentry);
717 	struct afs_vnode *vnode = AFS_FS_I(inode);
718 	int seq = 0;
719 
720 	_enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
721 
722 	do {
723 		read_seqbegin_or_lock(&vnode->cb_lock, &seq);
724 		generic_fillattr(inode, stat);
725 	} while (need_seqretry(&vnode->cb_lock, seq));
726 
727 	done_seqretry(&vnode->cb_lock, seq);
728 	return 0;
729 }
730 
731 /*
732  * discard an AFS inode
733  */
734 int afs_drop_inode(struct inode *inode)
735 {
736 	_enter("");
737 
738 	if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
739 		return generic_delete_inode(inode);
740 	else
741 		return generic_drop_inode(inode);
742 }
743 
744 /*
745  * clear an AFS inode
746  */
747 void afs_evict_inode(struct inode *inode)
748 {
749 	struct afs_cb_interest *cbi;
750 	struct afs_vnode *vnode;
751 
752 	vnode = AFS_FS_I(inode);
753 
754 	_enter("{%llx:%llu.%d}",
755 	       vnode->fid.vid,
756 	       vnode->fid.vnode,
757 	       vnode->fid.unique);
758 
759 	_debug("CLEAR INODE %p", inode);
760 
761 	ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
762 
763 	truncate_inode_pages_final(&inode->i_data);
764 	clear_inode(inode);
765 
766 	write_seqlock(&vnode->cb_lock);
767 	cbi = rcu_dereference_protected(vnode->cb_interest,
768 					lockdep_is_held(&vnode->cb_lock.lock));
769 	if (cbi) {
770 		afs_put_cb_interest(afs_i2net(inode), cbi);
771 		rcu_assign_pointer(vnode->cb_interest, NULL);
772 	}
773 	write_sequnlock(&vnode->cb_lock);
774 
775 	while (!list_empty(&vnode->wb_keys)) {
776 		struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
777 						    struct afs_wb_key, vnode_link);
778 		list_del(&wbk->vnode_link);
779 		afs_put_wb_key(wbk);
780 	}
781 
782 #ifdef CONFIG_AFS_FSCACHE
783 	{
784 		struct afs_vnode_cache_aux aux;
785 
786 		aux.data_version = vnode->status.data_version;
787 		fscache_relinquish_cookie(vnode->cache, &aux,
788 					  test_bit(AFS_VNODE_DELETED, &vnode->flags));
789 		vnode->cache = NULL;
790 	}
791 #endif
792 
793 	afs_prune_wb_keys(vnode);
794 	afs_put_permits(rcu_access_pointer(vnode->permit_cache));
795 	key_put(vnode->silly_key);
796 	vnode->silly_key = NULL;
797 	key_put(vnode->lock_key);
798 	vnode->lock_key = NULL;
799 	_leave("");
800 }
801 
802 /*
803  * set the attributes of an inode
804  */
805 int afs_setattr(struct dentry *dentry, struct iattr *attr)
806 {
807 	struct afs_fs_cursor fc;
808 	struct afs_status_cb *scb;
809 	struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
810 	struct key *key;
811 	int ret = -ENOMEM;
812 
813 	_enter("{%llx:%llu},{n=%pd},%x",
814 	       vnode->fid.vid, vnode->fid.vnode, dentry,
815 	       attr->ia_valid);
816 
817 	if (!(attr->ia_valid & (ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
818 				ATTR_MTIME))) {
819 		_leave(" = 0 [unsupported]");
820 		return 0;
821 	}
822 
823 	scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL);
824 	if (!scb)
825 		goto error;
826 
827 	/* flush any dirty data outstanding on a regular file */
828 	if (S_ISREG(vnode->vfs_inode.i_mode))
829 		filemap_write_and_wait(vnode->vfs_inode.i_mapping);
830 
831 	if (attr->ia_valid & ATTR_FILE) {
832 		key = afs_file_key(attr->ia_file);
833 	} else {
834 		key = afs_request_key(vnode->volume->cell);
835 		if (IS_ERR(key)) {
836 			ret = PTR_ERR(key);
837 			goto error_scb;
838 		}
839 	}
840 
841 	ret = -ERESTARTSYS;
842 	if (afs_begin_vnode_operation(&fc, vnode, key, false)) {
843 		afs_dataversion_t data_version = vnode->status.data_version;
844 
845 		if (attr->ia_valid & ATTR_SIZE)
846 			data_version++;
847 
848 		while (afs_select_fileserver(&fc)) {
849 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
850 			afs_fs_setattr(&fc, attr, scb);
851 		}
852 
853 		afs_check_for_remote_deletion(&fc, vnode);
854 		afs_vnode_commit_status(&fc, vnode, fc.cb_break,
855 					&data_version, scb);
856 		ret = afs_end_vnode_operation(&fc);
857 	}
858 
859 	if (!(attr->ia_valid & ATTR_FILE))
860 		key_put(key);
861 
862 error_scb:
863 	kfree(scb);
864 error:
865 	_leave(" = %d", ret);
866 	return ret;
867 }
868