2 * Copyright (c) 2002 Red Hat, Inc. All rights reserved.
4 * This software may be freely redistributed under the terms of the
5 * GNU General Public License.
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.
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/init.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>
28 static const struct inode_operations afs_symlink_inode_operations = {
29 .get_link = page_get_link,
30 .listxattr = afs_listxattr,
33 static noinline void dump_vnode(struct afs_vnode *vnode, struct afs_vnode *parent_vnode)
35 static unsigned long once_only;
37 pr_warn("kAFS: AFS vnode with undefined type %u\n",
39 pr_warn("kAFS: A=%d m=%o s=%llx v=%llx\n",
40 vnode->status.abort_code,
43 vnode->status.data_version);
44 pr_warn("kAFS: vnode %llx:%llx:%x\n",
49 pr_warn("kAFS: dir %llx:%llx:%x\n",
50 parent_vnode->fid.vid,
51 parent_vnode->fid.vnode,
52 parent_vnode->fid.unique);
54 if (!test_and_set_bit(0, &once_only))
59 * Initialise an inode from the vnode status.
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)
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);
71 _debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
74 (unsigned long long) status->size,
78 write_seqlock(&vnode->cb_lock);
80 vnode->status = *status;
82 t = status->mtime_client;
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);
90 switch (status->type) {
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;
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;
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;
108 set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
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;
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;
119 inode_nohighmem(inode);
122 dump_vnode(vnode, parent_vnode);
123 write_sequnlock(&vnode->cb_lock);
124 return afs_protocol_error(NULL, -EBADMSG, afs_eproto_file_type);
128 * Estimate 512 bytes blocks used, rounded up to nearest 1K
129 * for consistency with other AFS clients.
131 inode->i_blocks = ((i_size_read(inode) + 1023) >> 10) << 1;
132 i_size_write(&vnode->vfs_inode, status->size);
134 vnode->invalid_before = status->data_version;
135 inode_set_iversion_raw(&vnode->vfs_inode, status->data_version);
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();
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));
146 rcu_assign_pointer(vnode->cb_interest, afs_get_cb_interest(cbi));
149 set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
152 write_sequnlock(&vnode->cb_lock);
153 afs_put_cb_interest(afs_v2net(vnode), old_cbi);
158 * Update the core inode struct from a returned status record.
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)
165 struct afs_file_status *status = &scb->status;
168 bool data_changed = false;
170 BUG_ON(test_bit(AFS_VNODE_UNSET, &vnode->flags));
172 if (status->type != vnode->status.type) {
173 pr_warning("Vnode %llx:%llx:%x changed type %u to %u\n",
177 status->type, vnode->status.type);
178 afs_protocol_error(NULL, -EBADMSG, afs_eproto_bad_status);
182 if (status->nlink != vnode->status.nlink)
183 set_nlink(&vnode->vfs_inode, status->nlink);
185 if (status->owner != vnode->status.owner)
186 vnode->vfs_inode.i_uid = make_kuid(&init_user_ns, status->owner);
188 if (status->group != vnode->status.group)
189 vnode->vfs_inode.i_gid = make_kgid(&init_user_ns, status->group);
191 if (status->mode != vnode->status.mode) {
192 mode = vnode->vfs_inode.i_mode;
194 mode |= status->mode;
195 WRITE_ONCE(vnode->vfs_inode.i_mode, mode);
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;
203 if (vnode->status.data_version != status->data_version)
206 vnode->status = *status;
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);
220 set_bit(AFS_VNODE_ZAP_DATA, &vnode->flags);
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
227 if (test_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
228 data_changed = false;
232 inode_set_iversion_raw(&vnode->vfs_inode, status->data_version);
233 i_size_write(&vnode->vfs_inode, status->size);
238 * Apply a callback to a vnode.
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)
245 struct afs_cb_interest *old;
246 struct afs_callback *cb = &scb->callback;
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);
256 set_bit(AFS_VNODE_CB_PROMISED, &vnode->flags);
261 * Apply the received status and callback to an inode all in the same critical
262 * section to avoid races with afs_validate().
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)
270 if (fc->ac.error != 0)
273 write_seqlock(&vnode->cb_lock);
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);
282 if (scb->have_status)
283 afs_apply_status(fc, vnode, scb, expected_version);
285 afs_apply_callback(fc, vnode, scb, cb_break);
288 write_sequnlock(&vnode->cb_lock);
290 if (fc->ac.error == 0 && scb->have_status)
291 afs_cache_permit(vnode, fc->key, cb_break, scb);
295 * Fetch file status from the volume.
297 int afs_fetch_status(struct afs_vnode *vnode, struct key *key, bool is_new,
298 afs_access_t *_caller_access)
300 struct afs_status_cb *scb;
301 struct afs_fs_cursor fc;
304 _enter("%s,{%llx:%llu.%u,S=%lx}",
306 vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
309 scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL);
314 if (afs_begin_vnode_operation(&fc, vnode, key, true)) {
315 afs_dataversion_t data_version = vnode->status.data_version;
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);
325 ret = afs_inode_init_from_status(vnode, key, fc.cbi,
329 afs_cache_permit(vnode, key, fc.cb_break, scb);
331 afs_vnode_commit_status(&fc, vnode, fc.cb_break,
334 afs_check_for_remote_deletion(&fc, vnode);
335 ret = afs_end_vnode_operation(&fc);
338 if (ret == 0 && _caller_access)
339 *_caller_access = scb->status.caller_access;
341 _leave(" = %d", ret);
348 int afs_iget5_test(struct inode *inode, void *opaque)
350 struct afs_iget_data *iget_data = opaque;
351 struct afs_vnode *vnode = AFS_FS_I(inode);
353 return memcmp(&vnode->fid, &iget_data->fid, sizeof(iget_data->fid)) == 0;
357 * iget5() comparator for inode created by autocell operations
359 * These pseudo inodes don't match anything.
361 static int afs_iget5_pseudo_dir_test(struct inode *inode, void *opaque)
367 * iget5() inode initialiser
369 static int afs_iget5_set(struct inode *inode, void *opaque)
371 struct afs_iget_data *iget_data = opaque;
372 struct afs_vnode *vnode = AFS_FS_I(inode);
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;
379 /* YFS supports 96-bit vnode IDs, but Linux only supports
380 * 64-bit inode numbers.
382 inode->i_ino = iget_data->fid.vnode;
383 inode->i_generation = iget_data->fid.unique;
388 * Create an inode for a dynamic root directory or an autocell dynamic
391 struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
393 struct afs_super_info *as;
394 struct afs_vnode *vnode;
396 static atomic_t afs_autocell_ino;
398 struct afs_iget_data iget_data = {
407 iget_data.volume = as->volume;
408 iget_data.fid.vid = as->volume->vid;
411 iget_data.fid.vnode = 1;
412 iget_data.fid.unique = 1;
414 iget_data.fid.vnode = atomic_inc_return(&afs_autocell_ino);
415 iget_data.fid.unique = 0;
418 inode = iget5_locked(sb, iget_data.fid.vnode,
419 afs_iget5_pseudo_dir_test, afs_iget5_set,
422 _leave(" = -ENOMEM");
423 return ERR_PTR(-ENOMEM);
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);
430 vnode = AFS_FS_I(inode);
432 /* there shouldn't be an existing inode */
433 BUG_ON(!(inode->i_state & I_NEW));
436 inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO;
438 inode->i_op = &afs_dynroot_inode_operations;
439 inode->i_fop = &afs_dynroot_file_operations;
441 inode->i_op = &afs_autocell_inode_operations;
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);
448 inode_set_iversion_raw(inode, 0);
449 inode->i_generation = 0;
451 set_bit(AFS_VNODE_PSEUDODIR, &vnode->flags);
453 set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
454 inode->i_flags |= S_AUTOMOUNT;
457 inode->i_flags |= S_NOATIME;
458 unlock_new_inode(inode);
459 _leave(" = %p", inode);
464 * Get a cache cookie for an inode.
466 static void afs_get_inode_cache(struct afs_vnode *vnode)
468 #ifdef CONFIG_AFS_FSCACHE
472 u32 vnode_id_ext[2]; /* Allow for a 96-bit key */
474 struct afs_vnode_cache_aux aux;
476 if (vnode->status.type == AFS_FTYPE_DIR) {
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;
487 vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
488 &afs_vnode_cache_index_def,
491 vnode, vnode->status.size, true);
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)
504 struct afs_super_info *as;
505 struct afs_vnode *vnode;
506 struct afs_fid *fid = &iget_data->fid;
510 _enter(",{%llx:%llu.%u},,", fid->vid, fid->vnode, fid->unique);
513 iget_data->volume = as->volume;
515 inode = iget5_locked(sb, fid->vnode, afs_iget5_test, afs_iget5_set,
518 _leave(" = -ENOMEM");
519 return ERR_PTR(-ENOMEM);
522 _debug("GOT INODE %p { vl=%llx vn=%llx, u=%x }",
523 inode, fid->vid, fid->vnode, fid->unique);
525 vnode = AFS_FS_I(inode);
527 /* deal with an existing inode */
528 if (!(inode->i_state & I_NEW)) {
529 _leave(" = %p", inode);
534 /* it's a remotely extant inode */
535 ret = afs_fetch_status(vnode, key, true, NULL);
539 ret = afs_inode_init_from_status(vnode, key, cbi, parent_vnode,
545 afs_get_inode_cache(vnode);
548 clear_bit(AFS_VNODE_UNSET, &vnode->flags);
549 inode->i_flags |= S_NOATIME;
550 unlock_new_inode(inode);
551 _leave(" = %p", inode);
557 _leave(" = %d [bad]", ret);
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
565 void afs_zap_data(struct afs_vnode *vnode)
567 _enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
569 #ifdef CONFIG_AFS_FSCACHE
570 fscache_invalidate(vnode->cache);
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);
579 invalidate_inode_pages2(vnode->vfs_inode.i_mapping);
583 * Check the validity of a vnode/inode.
585 bool afs_check_validity(struct afs_vnode *vnode)
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;
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;
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);
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;
611 } else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) {
614 } else if (vnode->cb_expires_at - 10 <= now) {
620 } else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
623 vnode->cb_v_break = cb_v_break;
627 } while (need_seqretry(&vnode->cb_lock, seq));
629 done_seqretry(&vnode->cb_lock, seq);
632 write_seqlock(&vnode->cb_lock);
633 if (cb_break == vnode->cb_break)
634 __afs_break_callback(vnode);
635 write_sequnlock(&vnode->cb_lock);
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,
647 * - parent dir metadata changed (security changes)
648 * - dentry data changed (write, truncate)
649 * - dentry metadata changed (security changes)
651 int afs_validate(struct afs_vnode *vnode, struct key *key)
656 _enter("{v={%llx:%llu} fl=%lx},%x",
657 vnode->fid.vid, vnode->fid.vnode, vnode->flags,
661 valid = afs_check_validity(vnode);
664 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
665 clear_nlink(&vnode->vfs_inode);
670 down_write(&vnode->validate_lock);
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
676 if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
677 _debug("not promised");
678 ret = afs_fetch_status(vnode, key, false, NULL);
680 if (ret == -ENOENT) {
681 set_bit(AFS_VNODE_DELETED, &vnode->flags);
686 _debug("new promise [fl=%lx]", vnode->flags);
689 if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
690 _debug("file already deleted");
695 /* if the vnode's data version number changed then its contents are
697 if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
699 up_write(&vnode->validate_lock);
705 up_write(&vnode->validate_lock);
706 _leave(" = %d", ret);
711 * read the attributes of an inode
713 int afs_getattr(const struct path *path, struct kstat *stat,
714 u32 request_mask, unsigned int query_flags)
716 struct inode *inode = d_inode(path->dentry);
717 struct afs_vnode *vnode = AFS_FS_I(inode);
720 _enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
723 read_seqbegin_or_lock(&vnode->cb_lock, &seq);
724 generic_fillattr(inode, stat);
725 } while (need_seqretry(&vnode->cb_lock, seq));
727 done_seqretry(&vnode->cb_lock, seq);
732 * discard an AFS inode
734 int afs_drop_inode(struct inode *inode)
738 if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
739 return generic_delete_inode(inode);
741 return generic_drop_inode(inode);
747 void afs_evict_inode(struct inode *inode)
749 struct afs_cb_interest *cbi;
750 struct afs_vnode *vnode;
752 vnode = AFS_FS_I(inode);
754 _enter("{%llx:%llu.%d}",
759 _debug("CLEAR INODE %p", inode);
761 ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
763 truncate_inode_pages_final(&inode->i_data);
766 write_seqlock(&vnode->cb_lock);
767 cbi = rcu_dereference_protected(vnode->cb_interest,
768 lockdep_is_held(&vnode->cb_lock.lock));
770 afs_put_cb_interest(afs_i2net(inode), cbi);
771 rcu_assign_pointer(vnode->cb_interest, NULL);
773 write_sequnlock(&vnode->cb_lock);
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);
782 #ifdef CONFIG_AFS_FSCACHE
784 struct afs_vnode_cache_aux aux;
786 aux.data_version = vnode->status.data_version;
787 fscache_relinquish_cookie(vnode->cache, &aux,
788 test_bit(AFS_VNODE_DELETED, &vnode->flags));
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;
803 * set the attributes of an inode
805 int afs_setattr(struct dentry *dentry, struct iattr *attr)
807 struct afs_fs_cursor fc;
808 struct afs_status_cb *scb;
809 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
813 _enter("{%llx:%llu},{n=%pd},%x",
814 vnode->fid.vid, vnode->fid.vnode, dentry,
817 if (!(attr->ia_valid & (ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
819 _leave(" = 0 [unsupported]");
823 scb = kzalloc(sizeof(struct afs_status_cb), GFP_KERNEL);
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);
831 if (attr->ia_valid & ATTR_FILE) {
832 key = afs_file_key(attr->ia_file);
834 key = afs_request_key(vnode->volume->cell);
842 if (afs_begin_vnode_operation(&fc, vnode, key, false)) {
843 afs_dataversion_t data_version = vnode->status.data_version;
845 if (attr->ia_valid & ATTR_SIZE)
848 while (afs_select_fileserver(&fc)) {
849 fc.cb_break = afs_calc_vnode_cb_break(vnode);
850 afs_fs_setattr(&fc, attr, scb);
853 afs_check_for_remote_deletion(&fc, vnode);
854 afs_vnode_commit_status(&fc, vnode, fc.cb_break,
856 ret = afs_end_vnode_operation(&fc);
859 if (!(attr->ia_valid & ATTR_FILE))
865 _leave(" = %d", ret);