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.
11 * Authors: David Woodhouse <dwmw2@infradead.org>
12 * David Howells <dhowells@redhat.com>
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,
32 static noinline void dump_vnode(struct afs_vnode *vnode, struct afs_vnode *parent_vnode)
34 static unsigned long once_only;
36 pr_warn("kAFS: AFS vnode with undefined type %u\n", vnode->status.type);
37 pr_warn("kAFS: A=%d m=%o s=%llx v=%llx\n",
38 vnode->status.abort_code,
41 vnode->status.data_version);
42 pr_warn("kAFS: vnode %llx:%llx:%x\n",
47 pr_warn("kAFS: dir %llx:%llx:%x\n",
48 parent_vnode->fid.vid,
49 parent_vnode->fid.vnode,
50 parent_vnode->fid.unique);
52 if (!test_and_set_bit(0, &once_only))
57 * Set parameters for the netfs library
59 static void afs_set_netfs_context(struct afs_vnode *vnode)
61 netfs_inode_init(&vnode->netfs, &afs_req_ops);
65 * Initialise an inode from the vnode status.
67 static int afs_inode_init_from_status(struct afs_operation *op,
68 struct afs_vnode_param *vp,
69 struct afs_vnode *vnode)
71 struct afs_file_status *status = &vp->scb.status;
72 struct inode *inode = AFS_VNODE_TO_I(vnode);
75 _enter("{%llx:%llu.%u} %s",
76 vp->fid.vid, vp->fid.vnode, vp->fid.unique,
77 op->type ? op->type->name : "???");
79 _debug("FS: ft=%d lk=%d sz=%llu ver=%Lu mod=%hu",
82 (unsigned long long) status->size,
86 write_seqlock(&vnode->cb_lock);
88 vnode->cb_v_check = op->cb_v_break;
89 vnode->status = *status;
91 t = status->mtime_client;
92 inode_set_ctime_to_ts(inode, t);
93 inode_set_mtime_to_ts(inode, t);
94 inode_set_atime_to_ts(inode, t);
95 inode->i_flags |= S_NOATIME;
96 inode->i_uid = make_kuid(&init_user_ns, status->owner);
97 inode->i_gid = make_kgid(&init_user_ns, status->group);
98 set_nlink(&vnode->netfs.inode, status->nlink);
100 switch (status->type) {
102 inode->i_mode = S_IFREG | (status->mode & S_IALLUGO);
103 inode->i_op = &afs_file_inode_operations;
104 inode->i_fop = &afs_file_operations;
105 inode->i_mapping->a_ops = &afs_file_aops;
106 mapping_set_large_folios(inode->i_mapping);
109 inode->i_mode = S_IFDIR | (status->mode & S_IALLUGO);
110 inode->i_op = &afs_dir_inode_operations;
111 inode->i_fop = &afs_dir_file_operations;
112 inode->i_mapping->a_ops = &afs_dir_aops;
113 mapping_set_large_folios(inode->i_mapping);
115 case AFS_FTYPE_SYMLINK:
116 /* Symlinks with a mode of 0644 are actually mountpoints. */
117 if ((status->mode & 0777) == 0644) {
118 inode->i_flags |= S_AUTOMOUNT;
120 set_bit(AFS_VNODE_MOUNTPOINT, &vnode->flags);
122 inode->i_mode = S_IFDIR | 0555;
123 inode->i_op = &afs_mntpt_inode_operations;
124 inode->i_fop = &afs_mntpt_file_operations;
125 inode->i_mapping->a_ops = &afs_symlink_aops;
127 inode->i_mode = S_IFLNK | status->mode;
128 inode->i_op = &afs_symlink_inode_operations;
129 inode->i_mapping->a_ops = &afs_symlink_aops;
131 inode_nohighmem(inode);
134 dump_vnode(vnode, op->file[0].vnode != vnode ? op->file[0].vnode : NULL);
135 write_sequnlock(&vnode->cb_lock);
136 return afs_protocol_error(NULL, afs_eproto_file_type);
139 afs_set_i_size(vnode, status->size);
140 afs_set_netfs_context(vnode);
142 vnode->invalid_before = status->data_version;
143 inode_set_iversion_raw(&vnode->netfs.inode, status->data_version);
145 if (!vp->scb.have_cb) {
146 /* it's a symlink we just created (the fileserver
147 * didn't give us a callback) */
148 atomic64_set(&vnode->cb_expires_at, AFS_NO_CB_PROMISE);
150 vnode->cb_server = op->server;
151 atomic64_set(&vnode->cb_expires_at, vp->scb.callback.expires_at);
154 write_sequnlock(&vnode->cb_lock);
159 * Update the core inode struct from a returned status record.
161 static void afs_apply_status(struct afs_operation *op,
162 struct afs_vnode_param *vp)
164 struct afs_file_status *status = &vp->scb.status;
165 struct afs_vnode *vnode = vp->vnode;
166 struct inode *inode = &vnode->netfs.inode;
169 bool data_changed = false;
170 bool change_size = vp->set_size;
172 _enter("{%llx:%llu.%u} %s",
173 vp->fid.vid, vp->fid.vnode, vp->fid.unique,
174 op->type ? op->type->name : "???");
176 BUG_ON(test_bit(AFS_VNODE_UNSET, &vnode->flags));
178 if (status->type != vnode->status.type) {
179 pr_warn("Vnode %llx:%llx:%x changed type %u to %u\n",
183 status->type, vnode->status.type);
184 afs_protocol_error(NULL, afs_eproto_bad_status);
188 if (status->nlink != vnode->status.nlink)
189 set_nlink(inode, status->nlink);
191 if (status->owner != vnode->status.owner)
192 inode->i_uid = make_kuid(&init_user_ns, status->owner);
194 if (status->group != vnode->status.group)
195 inode->i_gid = make_kgid(&init_user_ns, status->group);
197 if (status->mode != vnode->status.mode) {
198 mode = inode->i_mode;
200 mode |= status->mode & S_IALLUGO;
201 WRITE_ONCE(inode->i_mode, mode);
204 t = status->mtime_client;
205 inode_set_mtime_to_ts(inode, t);
206 if (vp->update_ctime)
207 inode_set_ctime_to_ts(inode, op->ctime);
209 if (vnode->status.data_version != status->data_version)
212 vnode->status = *status;
214 if (vp->dv_before + vp->dv_delta != status->data_version) {
215 if (vnode->cb_ro_snapshot == atomic_read(&vnode->volume->cb_ro_snapshot) &&
216 atomic64_read(&vnode->cb_expires_at) != AFS_NO_CB_PROMISE)
217 pr_warn("kAFS: vnode modified {%llx:%llu} %llx->%llx %s (op=%x)\n",
218 vnode->fid.vid, vnode->fid.vnode,
219 (unsigned long long)vp->dv_before + vp->dv_delta,
220 (unsigned long long)status->data_version,
221 op->type ? op->type->name : "???",
224 vnode->invalid_before = status->data_version;
225 if (vnode->status.type == AFS_FTYPE_DIR) {
226 if (test_and_clear_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
227 afs_stat_v(vnode, n_inval);
229 set_bit(AFS_VNODE_ZAP_DATA, &vnode->flags);
233 } else if (vnode->status.type == AFS_FTYPE_DIR) {
234 /* Expected directory change is handled elsewhere so
235 * that we can locally edit the directory and save on a
238 if (test_bit(AFS_VNODE_DIR_VALID, &vnode->flags))
239 data_changed = false;
244 inode_set_iversion_raw(inode, status->data_version);
246 /* Only update the size if the data version jumped. If the
247 * file is being modified locally, then we might have our own
248 * idea of what the size should be that's not the same as
249 * what's on the server.
251 vnode->netfs.remote_i_size = status->size;
253 afs_set_i_size(vnode, status->size);
254 inode_set_ctime_to_ts(inode, t);
255 inode_set_atime_to_ts(inode, t);
261 * Apply a callback to a vnode.
263 static void afs_apply_callback(struct afs_operation *op,
264 struct afs_vnode_param *vp)
266 struct afs_callback *cb = &vp->scb.callback;
267 struct afs_vnode *vnode = vp->vnode;
269 if (!afs_cb_is_broken(vp->cb_break_before, vnode)) {
270 if (op->volume->type == AFSVL_RWVOL)
271 vnode->cb_server = op->server;
272 atomic64_set(&vnode->cb_expires_at, cb->expires_at);
277 * Apply the received status and callback to an inode all in the same critical
278 * section to avoid races with afs_validate().
280 void afs_vnode_commit_status(struct afs_operation *op, struct afs_vnode_param *vp)
282 struct afs_vnode *vnode = vp->vnode;
286 write_seqlock(&vnode->cb_lock);
288 if (vp->scb.have_error) {
289 /* A YFS server will return this from RemoveFile2 and AFS and
290 * YFS will return this from InlineBulkStatus.
292 if (vp->scb.status.abort_code == VNOVNODE) {
293 set_bit(AFS_VNODE_DELETED, &vnode->flags);
294 clear_nlink(&vnode->netfs.inode);
295 __afs_break_callback(vnode, afs_cb_break_for_deleted);
296 op->flags &= ~AFS_OPERATION_DIR_CONFLICT;
298 } else if (vp->scb.have_status) {
299 if (vp->speculative &&
300 (test_bit(AFS_VNODE_MODIFYING, &vnode->flags) ||
301 vp->dv_before != vnode->status.data_version))
302 /* Ignore the result of a speculative bulk status fetch
303 * if it splits around a modification op, thereby
304 * appearing to regress the data version.
307 afs_apply_status(op, vp);
309 afs_apply_callback(op, vp);
310 } else if (vp->op_unlinked && !(op->flags & AFS_OPERATION_DIR_CONFLICT)) {
311 drop_nlink(&vnode->netfs.inode);
312 if (vnode->netfs.inode.i_nlink == 0) {
313 set_bit(AFS_VNODE_DELETED, &vnode->flags);
314 __afs_break_callback(vnode, afs_cb_break_for_deleted);
319 write_sequnlock(&vnode->cb_lock);
321 if (vp->scb.have_status)
322 afs_cache_permit(vnode, op->key, vp->cb_break_before, &vp->scb);
325 static void afs_fetch_status_success(struct afs_operation *op)
327 struct afs_vnode_param *vp = &op->file[op->fetch_status.which];
328 struct afs_vnode *vnode = vp->vnode;
331 if (vnode->netfs.inode.i_state & I_NEW) {
332 ret = afs_inode_init_from_status(op, vp, vnode);
333 afs_op_set_error(op, ret);
335 afs_cache_permit(vnode, op->key, vp->cb_break_before, &vp->scb);
337 afs_vnode_commit_status(op, vp);
341 const struct afs_operation_ops afs_fetch_status_operation = {
342 .issue_afs_rpc = afs_fs_fetch_status,
343 .issue_yfs_rpc = yfs_fs_fetch_status,
344 .success = afs_fetch_status_success,
345 .aborted = afs_check_for_remote_deletion,
349 * Fetch file status from the volume.
351 int afs_fetch_status(struct afs_vnode *vnode, struct key *key, bool is_new,
352 afs_access_t *_caller_access)
354 struct afs_operation *op;
356 _enter("%s,{%llx:%llu.%u,S=%lx}",
358 vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique,
361 op = afs_alloc_operation(key, vnode->volume);
365 afs_op_set_vnode(op, 0, vnode);
368 op->ops = &afs_fetch_status_operation;
369 afs_begin_vnode_operation(op);
370 afs_wait_for_operation(op);
373 *_caller_access = op->file[0].scb.status.caller_access;
374 return afs_put_operation(op);
378 * ilookup() comparator
380 int afs_ilookup5_test_by_fid(struct inode *inode, void *opaque)
382 struct afs_vnode *vnode = AFS_FS_I(inode);
383 struct afs_fid *fid = opaque;
385 return (fid->vnode == vnode->fid.vnode &&
386 fid->vnode_hi == vnode->fid.vnode_hi &&
387 fid->unique == vnode->fid.unique);
393 static int afs_iget5_test(struct inode *inode, void *opaque)
395 struct afs_vnode_param *vp = opaque;
396 //struct afs_vnode *vnode = AFS_FS_I(inode);
398 return afs_ilookup5_test_by_fid(inode, &vp->fid);
402 * iget5() inode initialiser
404 static int afs_iget5_set(struct inode *inode, void *opaque)
406 struct afs_vnode_param *vp = opaque;
407 struct afs_super_info *as = AFS_FS_S(inode->i_sb);
408 struct afs_vnode *vnode = AFS_FS_I(inode);
410 vnode->volume = as->volume;
411 vnode->fid = vp->fid;
413 /* YFS supports 96-bit vnode IDs, but Linux only supports
414 * 64-bit inode numbers.
416 inode->i_ino = vnode->fid.vnode;
417 inode->i_generation = vnode->fid.unique;
422 * Get a cache cookie for an inode.
424 static void afs_get_inode_cache(struct afs_vnode *vnode)
426 #ifdef CONFIG_AFS_FSCACHE
430 __be32 vnode_id_ext[2]; /* Allow for a 96-bit key */
432 struct afs_vnode_cache_aux aux;
434 if (vnode->status.type != AFS_FTYPE_FILE) {
435 vnode->netfs.cache = NULL;
439 key.vnode_id = htonl(vnode->fid.vnode);
440 key.unique = htonl(vnode->fid.unique);
441 key.vnode_id_ext[0] = htonl(vnode->fid.vnode >> 32);
442 key.vnode_id_ext[1] = htonl(vnode->fid.vnode_hi);
443 afs_set_cache_aux(vnode, &aux);
445 afs_vnode_set_cache(vnode,
446 fscache_acquire_cookie(
447 vnode->volume->cache,
448 vnode->status.type == AFS_FTYPE_FILE ?
449 0 : FSCACHE_ADV_SINGLE_CHUNK,
452 i_size_read(&vnode->netfs.inode)));
459 struct inode *afs_iget(struct afs_operation *op, struct afs_vnode_param *vp)
461 struct afs_vnode_param *dvp = &op->file[0];
462 struct super_block *sb = dvp->vnode->netfs.inode.i_sb;
463 struct afs_vnode *vnode;
467 _enter(",{%llx:%llu.%u},,", vp->fid.vid, vp->fid.vnode, vp->fid.unique);
469 inode = iget5_locked(sb, vp->fid.vnode, afs_iget5_test, afs_iget5_set, vp);
471 _leave(" = -ENOMEM");
472 return ERR_PTR(-ENOMEM);
475 vnode = AFS_FS_I(inode);
477 _debug("GOT INODE %p { vl=%llx vn=%llx, u=%x }",
478 inode, vnode->fid.vid, vnode->fid.vnode, vnode->fid.unique);
480 /* deal with an existing inode */
481 if (!(inode->i_state & I_NEW)) {
482 _leave(" = %p", inode);
486 ret = afs_inode_init_from_status(op, vp, vnode);
490 afs_get_inode_cache(vnode);
493 clear_bit(AFS_VNODE_UNSET, &vnode->flags);
494 unlock_new_inode(inode);
495 _leave(" = %p", inode);
501 _leave(" = %d [bad]", ret);
505 static int afs_iget5_set_root(struct inode *inode, void *opaque)
507 struct afs_super_info *as = AFS_FS_S(inode->i_sb);
508 struct afs_vnode *vnode = AFS_FS_I(inode);
510 vnode->volume = as->volume;
511 vnode->fid.vid = as->volume->vid,
512 vnode->fid.vnode = 1;
513 vnode->fid.unique = 1;
515 inode->i_generation = 1;
520 * Set up the root inode for a volume. This is always vnode 1, unique 1 within
523 struct inode *afs_root_iget(struct super_block *sb, struct key *key)
525 struct afs_super_info *as = AFS_FS_S(sb);
526 struct afs_operation *op;
527 struct afs_vnode *vnode;
531 _enter(",{%llx},,", as->volume->vid);
533 inode = iget5_locked(sb, 1, NULL, afs_iget5_set_root, NULL);
535 _leave(" = -ENOMEM");
536 return ERR_PTR(-ENOMEM);
539 _debug("GOT ROOT INODE %p { vl=%llx }", inode, as->volume->vid);
541 BUG_ON(!(inode->i_state & I_NEW));
543 vnode = AFS_FS_I(inode);
544 vnode->cb_v_check = atomic_read(&as->volume->cb_v_break),
545 afs_set_netfs_context(vnode);
547 op = afs_alloc_operation(key, as->volume);
553 afs_op_set_vnode(op, 0, vnode);
556 op->ops = &afs_fetch_status_operation;
557 ret = afs_do_sync_operation(op);
561 afs_get_inode_cache(vnode);
563 clear_bit(AFS_VNODE_UNSET, &vnode->flags);
564 unlock_new_inode(inode);
565 _leave(" = %p", inode);
570 _leave(" = %d [bad]", ret);
575 * read the attributes of an inode
577 int afs_getattr(struct mnt_idmap *idmap, const struct path *path,
578 struct kstat *stat, u32 request_mask, unsigned int query_flags)
580 struct inode *inode = d_inode(path->dentry);
581 struct afs_vnode *vnode = AFS_FS_I(inode);
585 _enter("{ ino=%lu v=%u }", inode->i_ino, inode->i_generation);
588 !(query_flags & AT_STATX_DONT_SYNC) &&
589 atomic64_read(&vnode->cb_expires_at) == AFS_NO_CB_PROMISE) {
590 key = afs_request_key(vnode->volume->cell);
593 ret = afs_validate(vnode, key);
600 seq = read_seqbegin(&vnode->cb_lock);
601 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
602 if (test_bit(AFS_VNODE_SILLY_DELETED, &vnode->flags) &&
606 /* Lie about the size of directories. We maintain a locally
607 * edited copy and may make different allocation decisions on
608 * it, but we need to give userspace the server's size.
610 if (S_ISDIR(inode->i_mode))
611 stat->size = vnode->netfs.remote_i_size;
612 } while (read_seqretry(&vnode->cb_lock, seq));
618 * discard an AFS inode
620 int afs_drop_inode(struct inode *inode)
624 if (test_bit(AFS_VNODE_PSEUDODIR, &AFS_FS_I(inode)->flags))
625 return generic_delete_inode(inode);
627 return generic_drop_inode(inode);
633 void afs_evict_inode(struct inode *inode)
635 struct afs_vnode_cache_aux aux;
636 struct afs_vnode *vnode = AFS_FS_I(inode);
638 _enter("{%llx:%llu.%d}",
643 _debug("CLEAR INODE %p", inode);
645 ASSERTCMP(inode->i_ino, ==, vnode->fid.vnode);
647 truncate_inode_pages_final(&inode->i_data);
649 afs_set_cache_aux(vnode, &aux);
650 fscache_clear_inode_writeback(afs_vnode_cache(vnode), inode, &aux);
653 while (!list_empty(&vnode->wb_keys)) {
654 struct afs_wb_key *wbk = list_entry(vnode->wb_keys.next,
655 struct afs_wb_key, vnode_link);
656 list_del(&wbk->vnode_link);
660 fscache_relinquish_cookie(afs_vnode_cache(vnode),
661 test_bit(AFS_VNODE_DELETED, &vnode->flags));
663 afs_prune_wb_keys(vnode);
664 afs_put_permits(rcu_access_pointer(vnode->permit_cache));
665 key_put(vnode->silly_key);
666 vnode->silly_key = NULL;
667 key_put(vnode->lock_key);
668 vnode->lock_key = NULL;
672 static void afs_setattr_success(struct afs_operation *op)
674 struct afs_vnode_param *vp = &op->file[0];
675 struct inode *inode = &vp->vnode->netfs.inode;
676 loff_t old_i_size = i_size_read(inode);
678 op->setattr.old_i_size = old_i_size;
679 afs_vnode_commit_status(op, vp);
680 /* inode->i_size has now been changed. */
682 if (op->setattr.attr->ia_valid & ATTR_SIZE) {
683 loff_t size = op->setattr.attr->ia_size;
684 if (size > old_i_size)
685 pagecache_isize_extended(inode, old_i_size, size);
689 static void afs_setattr_edit_file(struct afs_operation *op)
691 struct afs_vnode_param *vp = &op->file[0];
692 struct inode *inode = &vp->vnode->netfs.inode;
694 if (op->setattr.attr->ia_valid & ATTR_SIZE) {
695 loff_t size = op->setattr.attr->ia_size;
696 loff_t i_size = op->setattr.old_i_size;
699 truncate_pagecache(inode, size);
701 fscache_resize_cookie(afs_vnode_cache(vp->vnode),
702 vp->scb.status.size);
706 static const struct afs_operation_ops afs_setattr_operation = {
707 .issue_afs_rpc = afs_fs_setattr,
708 .issue_yfs_rpc = yfs_fs_setattr,
709 .success = afs_setattr_success,
710 .edit_dir = afs_setattr_edit_file,
714 * set the attributes of an inode
716 int afs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
719 const unsigned int supported =
720 ATTR_SIZE | ATTR_MODE | ATTR_UID | ATTR_GID |
721 ATTR_MTIME | ATTR_MTIME_SET | ATTR_TIMES_SET | ATTR_TOUCH;
722 struct afs_operation *op;
723 struct afs_vnode *vnode = AFS_FS_I(d_inode(dentry));
724 struct inode *inode = &vnode->netfs.inode;
728 _enter("{%llx:%llu},{n=%pd},%x",
729 vnode->fid.vid, vnode->fid.vnode, dentry,
732 if (!(attr->ia_valid & supported)) {
733 _leave(" = 0 [unsupported]");
737 i_size = i_size_read(inode);
738 if (attr->ia_valid & ATTR_SIZE) {
739 if (!S_ISREG(inode->i_mode))
742 ret = inode_newsize_ok(inode, attr->ia_size);
746 if (attr->ia_size == i_size)
747 attr->ia_valid &= ~ATTR_SIZE;
750 fscache_use_cookie(afs_vnode_cache(vnode), true);
752 /* Prevent any new writebacks from starting whilst we do this. */
753 down_write(&vnode->validate_lock);
755 if ((attr->ia_valid & ATTR_SIZE) && S_ISREG(inode->i_mode)) {
756 loff_t size = attr->ia_size;
758 /* Wait for any outstanding writes to the server to complete */
759 loff_t from = min(size, i_size);
760 loff_t to = max(size, i_size);
761 ret = filemap_fdatawait_range(inode->i_mapping, from, to);
765 /* Don't talk to the server if we're just shortening in-memory
766 * writes that haven't gone to the server yet.
768 if (!(attr->ia_valid & (supported & ~ATTR_SIZE & ~ATTR_MTIME)) &&
769 attr->ia_size < i_size &&
770 attr->ia_size > vnode->status.size) {
771 truncate_pagecache(inode, attr->ia_size);
772 fscache_resize_cookie(afs_vnode_cache(vnode),
774 i_size_write(inode, attr->ia_size);
780 op = afs_alloc_operation(((attr->ia_valid & ATTR_FILE) ?
781 afs_file_key(attr->ia_file) : NULL),
788 afs_op_set_vnode(op, 0, vnode);
789 op->setattr.attr = attr;
791 if (attr->ia_valid & ATTR_SIZE) {
792 op->file[0].dv_delta = 1;
793 op->file[0].set_size = true;
795 op->ctime = attr->ia_ctime;
796 op->file[0].update_ctime = 1;
797 op->file[0].modification = true;
799 op->ops = &afs_setattr_operation;
800 ret = afs_do_sync_operation(op);
803 up_write(&vnode->validate_lock);
804 fscache_unuse_cookie(afs_vnode_cache(vnode), NULL, NULL);
805 _leave(" = %d", ret);