1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/xattr.h>
13 #include <linux/posix_acl.h>
14 #include <linux/random.h>
15 #include <linux/sort.h>
18 #include "mds_client.h"
20 #include <linux/ceph/decode.h>
23 * Ceph inode operations
25 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
26 * setattr, etc.), xattr helpers, and helpers for assimilating
27 * metadata returned by the MDS into our cache.
29 * Also define helpers for doing asynchronous writeback, invalidation,
30 * and truncation for the benefit of those who can't afford to block
31 * (typically because they are in the message handler path).
34 static const struct inode_operations ceph_symlink_iops;
36 static void ceph_inode_work(struct work_struct *work);
39 * find or create an inode, given the ceph ino number
41 static int ceph_set_ino_cb(struct inode *inode, void *data)
43 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
44 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
51 ino_t t = ceph_vino_to_ino(vino);
53 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
55 return ERR_PTR(-ENOMEM);
56 if (inode->i_state & I_NEW) {
57 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58 inode, ceph_vinop(inode), (u64)inode->i_ino);
59 unlock_new_inode(inode);
62 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
68 * get/constuct snapdir inode for a given directory
70 struct inode *ceph_get_snapdir(struct inode *parent)
72 struct ceph_vino vino = {
73 .ino = ceph_ino(parent),
76 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
77 struct ceph_inode_info *ci = ceph_inode(inode);
79 BUG_ON(!S_ISDIR(parent->i_mode));
82 inode->i_mode = parent->i_mode;
83 inode->i_uid = parent->i_uid;
84 inode->i_gid = parent->i_gid;
85 inode->i_op = &ceph_snapdir_iops;
86 inode->i_fop = &ceph_snapdir_fops;
87 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
92 const struct inode_operations ceph_file_iops = {
93 .permission = ceph_permission,
94 .setattr = ceph_setattr,
95 .getattr = ceph_getattr,
96 .listxattr = ceph_listxattr,
97 .get_acl = ceph_get_acl,
98 .set_acl = ceph_set_acl,
103 * We use a 'frag tree' to keep track of the MDS's directory fragments
104 * for a given inode (usually there is just a single fragment). We
105 * need to know when a child frag is delegated to a new MDS, or when
106 * it is flagged as replicated, so we can direct our requests
111 * find/create a frag in the tree
113 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
117 struct rb_node *parent = NULL;
118 struct ceph_inode_frag *frag;
121 p = &ci->i_fragtree.rb_node;
124 frag = rb_entry(parent, struct ceph_inode_frag, node);
125 c = ceph_frag_compare(f, frag->frag);
134 frag = kmalloc(sizeof(*frag), GFP_NOFS);
136 return ERR_PTR(-ENOMEM);
143 rb_link_node(&frag->node, parent, p);
144 rb_insert_color(&frag->node, &ci->i_fragtree);
146 dout("get_or_create_frag added %llx.%llx frag %x\n",
147 ceph_vinop(&ci->vfs_inode), f);
152 * find a specific frag @f
154 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
156 struct rb_node *n = ci->i_fragtree.rb_node;
159 struct ceph_inode_frag *frag =
160 rb_entry(n, struct ceph_inode_frag, node);
161 int c = ceph_frag_compare(f, frag->frag);
173 * Choose frag containing the given value @v. If @pfrag is
174 * specified, copy the frag delegation info to the caller if
177 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
178 struct ceph_inode_frag *pfrag, int *found)
180 u32 t = ceph_frag_make(0, 0);
181 struct ceph_inode_frag *frag;
189 WARN_ON(!ceph_frag_contains_value(t, v));
190 frag = __ceph_find_frag(ci, t);
192 break; /* t is a leaf */
193 if (frag->split_by == 0) {
195 memcpy(pfrag, frag, sizeof(*pfrag));
202 nway = 1 << frag->split_by;
203 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
204 frag->split_by, nway);
205 for (i = 0; i < nway; i++) {
206 n = ceph_frag_make_child(t, frag->split_by, i);
207 if (ceph_frag_contains_value(n, v)) {
214 dout("choose_frag(%x) = %x\n", v, t);
219 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
220 struct ceph_inode_frag *pfrag, int *found)
223 mutex_lock(&ci->i_fragtree_mutex);
224 ret = __ceph_choose_frag(ci, v, pfrag, found);
225 mutex_unlock(&ci->i_fragtree_mutex);
230 * Process dirfrag (delegation) info from the mds. Include leaf
231 * fragment in tree ONLY if ndist > 0. Otherwise, only
232 * branches/splits are included in i_fragtree)
234 static int ceph_fill_dirfrag(struct inode *inode,
235 struct ceph_mds_reply_dirfrag *dirinfo)
237 struct ceph_inode_info *ci = ceph_inode(inode);
238 struct ceph_inode_frag *frag;
239 u32 id = le32_to_cpu(dirinfo->frag);
240 int mds = le32_to_cpu(dirinfo->auth);
241 int ndist = le32_to_cpu(dirinfo->ndist);
246 spin_lock(&ci->i_ceph_lock);
248 diri_auth = ci->i_auth_cap->mds;
249 spin_unlock(&ci->i_ceph_lock);
251 if (mds == -1) /* CDIR_AUTH_PARENT */
254 mutex_lock(&ci->i_fragtree_mutex);
255 if (ndist == 0 && mds == diri_auth) {
256 /* no delegation info needed. */
257 frag = __ceph_find_frag(ci, id);
260 if (frag->split_by == 0) {
261 /* tree leaf, remove */
262 dout("fill_dirfrag removed %llx.%llx frag %x"
263 " (no ref)\n", ceph_vinop(inode), id);
264 rb_erase(&frag->node, &ci->i_fragtree);
267 /* tree branch, keep and clear */
268 dout("fill_dirfrag cleared %llx.%llx frag %x"
269 " referral\n", ceph_vinop(inode), id);
277 /* find/add this frag to store mds delegation info */
278 frag = __get_or_create_frag(ci, id);
280 /* this is not the end of the world; we can continue
281 with bad/inaccurate delegation info */
282 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
283 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
289 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
290 for (i = 0; i < frag->ndist; i++)
291 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
292 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
293 ceph_vinop(inode), frag->frag, frag->ndist);
296 mutex_unlock(&ci->i_fragtree_mutex);
300 static int frag_tree_split_cmp(const void *l, const void *r)
302 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
303 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
304 return ceph_frag_compare(le32_to_cpu(ls->frag),
305 le32_to_cpu(rs->frag));
308 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
311 return f == ceph_frag_make(0, 0);
312 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
314 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
317 static int ceph_fill_fragtree(struct inode *inode,
318 struct ceph_frag_tree_head *fragtree,
319 struct ceph_mds_reply_dirfrag *dirinfo)
321 struct ceph_inode_info *ci = ceph_inode(inode);
322 struct ceph_inode_frag *frag, *prev_frag = NULL;
323 struct rb_node *rb_node;
324 unsigned i, split_by, nsplits;
328 mutex_lock(&ci->i_fragtree_mutex);
329 nsplits = le32_to_cpu(fragtree->nsplits);
330 if (nsplits != ci->i_fragtree_nsplits) {
332 } else if (nsplits) {
333 i = prandom_u32() % nsplits;
334 id = le32_to_cpu(fragtree->splits[i].frag);
335 if (!__ceph_find_frag(ci, id))
337 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
338 rb_node = rb_first(&ci->i_fragtree);
339 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
340 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
343 if (!update && dirinfo) {
344 id = le32_to_cpu(dirinfo->frag);
345 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
352 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
353 frag_tree_split_cmp, NULL);
356 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
357 rb_node = rb_first(&ci->i_fragtree);
358 for (i = 0; i < nsplits; i++) {
359 id = le32_to_cpu(fragtree->splits[i].frag);
360 split_by = le32_to_cpu(fragtree->splits[i].by);
361 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
362 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
363 "frag %x split by %d\n", ceph_vinop(inode),
364 i, nsplits, id, split_by);
369 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
370 if (ceph_frag_compare(frag->frag, id) >= 0) {
371 if (frag->frag != id)
374 rb_node = rb_next(rb_node);
377 rb_node = rb_next(rb_node);
378 /* delete stale split/leaf node */
379 if (frag->split_by > 0 ||
380 !is_frag_child(frag->frag, prev_frag)) {
381 rb_erase(&frag->node, &ci->i_fragtree);
382 if (frag->split_by > 0)
383 ci->i_fragtree_nsplits--;
389 frag = __get_or_create_frag(ci, id);
393 if (frag->split_by == 0)
394 ci->i_fragtree_nsplits++;
395 frag->split_by = split_by;
396 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
400 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
401 rb_node = rb_next(rb_node);
402 /* delete stale split/leaf node */
403 if (frag->split_by > 0 ||
404 !is_frag_child(frag->frag, prev_frag)) {
405 rb_erase(&frag->node, &ci->i_fragtree);
406 if (frag->split_by > 0)
407 ci->i_fragtree_nsplits--;
412 mutex_unlock(&ci->i_fragtree_mutex);
417 * initialize a newly allocated inode.
419 struct inode *ceph_alloc_inode(struct super_block *sb)
421 struct ceph_inode_info *ci;
424 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
428 dout("alloc_inode %p\n", &ci->vfs_inode);
430 spin_lock_init(&ci->i_ceph_lock);
433 ci->i_inline_version = 0;
434 ci->i_time_warp_seq = 0;
435 ci->i_ceph_flags = 0;
436 atomic64_set(&ci->i_ordered_count, 1);
437 atomic64_set(&ci->i_release_count, 1);
438 atomic64_set(&ci->i_complete_seq[0], 0);
439 atomic64_set(&ci->i_complete_seq[1], 0);
440 ci->i_symlink = NULL;
445 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
446 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
448 ci->i_fragtree = RB_ROOT;
449 mutex_init(&ci->i_fragtree_mutex);
451 ci->i_xattrs.blob = NULL;
452 ci->i_xattrs.prealloc_blob = NULL;
453 ci->i_xattrs.dirty = false;
454 ci->i_xattrs.index = RB_ROOT;
455 ci->i_xattrs.count = 0;
456 ci->i_xattrs.names_size = 0;
457 ci->i_xattrs.vals_size = 0;
458 ci->i_xattrs.version = 0;
459 ci->i_xattrs.index_version = 0;
461 ci->i_caps = RB_ROOT;
462 ci->i_auth_cap = NULL;
463 ci->i_dirty_caps = 0;
464 ci->i_flushing_caps = 0;
465 INIT_LIST_HEAD(&ci->i_dirty_item);
466 INIT_LIST_HEAD(&ci->i_flushing_item);
467 ci->i_prealloc_cap_flush = NULL;
468 INIT_LIST_HEAD(&ci->i_cap_flush_list);
469 init_waitqueue_head(&ci->i_cap_wq);
470 ci->i_hold_caps_min = 0;
471 ci->i_hold_caps_max = 0;
472 INIT_LIST_HEAD(&ci->i_cap_delay_list);
473 INIT_LIST_HEAD(&ci->i_cap_snaps);
474 ci->i_head_snapc = NULL;
477 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
478 ci->i_nr_by_mode[i] = 0;
480 mutex_init(&ci->i_truncate_mutex);
481 ci->i_truncate_seq = 0;
482 ci->i_truncate_size = 0;
483 ci->i_truncate_pending = 0;
486 ci->i_reported_size = 0;
487 ci->i_wanted_max_size = 0;
488 ci->i_requested_max_size = 0;
492 ci->i_rdcache_ref = 0;
495 ci->i_wrbuffer_ref = 0;
496 ci->i_wrbuffer_ref_head = 0;
497 atomic_set(&ci->i_filelock_ref, 0);
498 atomic_set(&ci->i_shared_gen, 1);
499 ci->i_rdcache_gen = 0;
500 ci->i_rdcache_revoking = 0;
502 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
503 INIT_LIST_HEAD(&ci->i_unsafe_iops);
504 spin_lock_init(&ci->i_unsafe_lock);
506 ci->i_snap_realm = NULL;
507 INIT_LIST_HEAD(&ci->i_snap_realm_item);
508 INIT_LIST_HEAD(&ci->i_snap_flush_item);
510 INIT_WORK(&ci->i_work, ceph_inode_work);
513 ceph_fscache_inode_init(ci);
515 return &ci->vfs_inode;
518 void ceph_free_inode(struct inode *inode)
520 struct ceph_inode_info *ci = ceph_inode(inode);
522 kfree(ci->i_symlink);
523 kmem_cache_free(ceph_inode_cachep, ci);
526 void ceph_destroy_inode(struct inode *inode)
528 struct ceph_inode_info *ci = ceph_inode(inode);
529 struct ceph_inode_frag *frag;
532 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
534 ceph_fscache_unregister_inode_cookie(ci);
536 __ceph_remove_caps(ci);
538 if (__ceph_has_any_quota(ci))
539 ceph_adjust_quota_realms_count(inode, false);
542 * we may still have a snap_realm reference if there are stray
543 * caps in i_snap_caps.
545 if (ci->i_snap_realm) {
546 struct ceph_mds_client *mdsc =
547 ceph_inode_to_client(inode)->mdsc;
548 if (ceph_snap(inode) == CEPH_NOSNAP) {
549 struct ceph_snap_realm *realm = ci->i_snap_realm;
550 dout(" dropping residual ref to snap realm %p\n",
552 spin_lock(&realm->inodes_with_caps_lock);
553 list_del_init(&ci->i_snap_realm_item);
554 ci->i_snap_realm = NULL;
555 if (realm->ino == ci->i_vino.ino)
557 spin_unlock(&realm->inodes_with_caps_lock);
558 ceph_put_snap_realm(mdsc, realm);
560 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
561 ci->i_snap_realm = NULL;
565 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
566 frag = rb_entry(n, struct ceph_inode_frag, node);
567 rb_erase(n, &ci->i_fragtree);
570 ci->i_fragtree_nsplits = 0;
572 __ceph_destroy_xattrs(ci);
573 if (ci->i_xattrs.blob)
574 ceph_buffer_put(ci->i_xattrs.blob);
575 if (ci->i_xattrs.prealloc_blob)
576 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
578 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
581 int ceph_drop_inode(struct inode *inode)
584 * Positve dentry and corresponding inode are always accompanied
585 * in MDS reply. So no need to keep inode in the cache after
586 * dropping all its aliases.
591 static inline blkcnt_t calc_inode_blocks(u64 size)
593 return (size + (1<<9) - 1) >> 9;
597 * Helpers to fill in size, ctime, mtime, and atime. We have to be
598 * careful because either the client or MDS may have more up to date
599 * info, depending on which capabilities are held, and whether
600 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
601 * and size are monotonically increasing, except when utimes() or
602 * truncate() increments the corresponding _seq values.)
604 int ceph_fill_file_size(struct inode *inode, int issued,
605 u32 truncate_seq, u64 truncate_size, u64 size)
607 struct ceph_inode_info *ci = ceph_inode(inode);
610 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
611 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
612 dout("size %lld -> %llu\n", inode->i_size, size);
613 if (size > 0 && S_ISDIR(inode->i_mode)) {
614 pr_err("fill_file_size non-zero size for directory\n");
617 i_size_write(inode, size);
618 inode->i_blocks = calc_inode_blocks(size);
619 ci->i_reported_size = size;
620 if (truncate_seq != ci->i_truncate_seq) {
621 dout("truncate_seq %u -> %u\n",
622 ci->i_truncate_seq, truncate_seq);
623 ci->i_truncate_seq = truncate_seq;
625 /* the MDS should have revoked these caps */
626 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
629 CEPH_CAP_FILE_LAZYIO));
631 * If we hold relevant caps, or in the case where we're
632 * not the only client referencing this file and we
633 * don't hold those caps, then we need to check whether
634 * the file is either opened or mmaped
636 if ((issued & (CEPH_CAP_FILE_CACHE|
637 CEPH_CAP_FILE_BUFFER)) ||
638 mapping_mapped(inode->i_mapping) ||
639 __ceph_caps_file_wanted(ci)) {
640 ci->i_truncate_pending++;
645 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
646 ci->i_truncate_size != truncate_size) {
647 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
649 ci->i_truncate_size = truncate_size;
653 ceph_fscache_invalidate(inode);
658 void ceph_fill_file_time(struct inode *inode, int issued,
659 u64 time_warp_seq, struct timespec64 *ctime,
660 struct timespec64 *mtime, struct timespec64 *atime)
662 struct ceph_inode_info *ci = ceph_inode(inode);
665 if (issued & (CEPH_CAP_FILE_EXCL|
667 CEPH_CAP_FILE_BUFFER|
669 CEPH_CAP_XATTR_EXCL)) {
670 if (ci->i_version == 0 ||
671 timespec64_compare(ctime, &inode->i_ctime) > 0) {
672 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
673 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
674 ctime->tv_sec, ctime->tv_nsec);
675 inode->i_ctime = *ctime;
677 if (ci->i_version == 0 ||
678 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
679 /* the MDS did a utimes() */
680 dout("mtime %lld.%09ld -> %lld.%09ld "
682 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
683 mtime->tv_sec, mtime->tv_nsec,
684 ci->i_time_warp_seq, (int)time_warp_seq);
686 inode->i_mtime = *mtime;
687 inode->i_atime = *atime;
688 ci->i_time_warp_seq = time_warp_seq;
689 } else if (time_warp_seq == ci->i_time_warp_seq) {
690 /* nobody did utimes(); take the max */
691 if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
692 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
693 inode->i_mtime.tv_sec,
694 inode->i_mtime.tv_nsec,
695 mtime->tv_sec, mtime->tv_nsec);
696 inode->i_mtime = *mtime;
698 if (timespec64_compare(atime, &inode->i_atime) > 0) {
699 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
700 inode->i_atime.tv_sec,
701 inode->i_atime.tv_nsec,
702 atime->tv_sec, atime->tv_nsec);
703 inode->i_atime = *atime;
705 } else if (issued & CEPH_CAP_FILE_EXCL) {
706 /* we did a utimes(); ignore mds values */
711 /* we have no write|excl caps; whatever the MDS says is true */
712 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
713 inode->i_ctime = *ctime;
714 inode->i_mtime = *mtime;
715 inode->i_atime = *atime;
716 ci->i_time_warp_seq = time_warp_seq;
721 if (warn) /* time_warp_seq shouldn't go backwards */
722 dout("%p mds time_warp_seq %llu < %u\n",
723 inode, time_warp_seq, ci->i_time_warp_seq);
727 * Populate an inode based on info from mds. May be called on new or
730 static int fill_inode(struct inode *inode, struct page *locked_page,
731 struct ceph_mds_reply_info_in *iinfo,
732 struct ceph_mds_reply_dirfrag *dirinfo,
733 struct ceph_mds_session *session,
734 unsigned long ttl_from, int cap_fmode,
735 struct ceph_cap_reservation *caps_reservation)
737 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
738 struct ceph_mds_reply_inode *info = iinfo->in;
739 struct ceph_inode_info *ci = ceph_inode(inode);
740 int issued, new_issued, info_caps;
741 struct timespec64 mtime, atime, ctime;
742 struct ceph_buffer *xattr_blob = NULL;
743 struct ceph_string *pool_ns = NULL;
744 struct ceph_cap *new_cap = NULL;
747 bool queue_trunc = false;
748 bool new_version = false;
749 bool fill_inline = false;
751 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
752 inode, ceph_vinop(inode), le64_to_cpu(info->version),
755 info_caps = le32_to_cpu(info->cap.caps);
757 /* prealloc new cap struct */
758 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP)
759 new_cap = ceph_get_cap(mdsc, caps_reservation);
762 * prealloc xattr data, if it looks like we'll need it. only
763 * if len > 4 (meaning there are actually xattrs; the first 4
764 * bytes are the xattr count).
766 if (iinfo->xattr_len > 4) {
767 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
769 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
773 if (iinfo->pool_ns_len > 0)
774 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
777 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
778 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
780 spin_lock(&ci->i_ceph_lock);
783 * provided version will be odd if inode value is projected,
784 * even if stable. skip the update if we have newer stable
785 * info (ours>=theirs, e.g. due to racing mds replies), unless
786 * we are getting projected (unstable) info (in which case the
787 * version is odd, and we want ours>theirs).
793 if (ci->i_version == 0 ||
794 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
795 le64_to_cpu(info->version) > (ci->i_version & ~1)))
798 __ceph_caps_issued(ci, &issued);
799 issued |= __ceph_caps_dirty(ci);
800 new_issued = ~issued & info_caps;
803 inode->i_rdev = le32_to_cpu(info->rdev);
804 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
806 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
808 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
809 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
810 inode->i_mode = le32_to_cpu(info->mode);
811 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
812 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
813 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
814 from_kuid(&init_user_ns, inode->i_uid),
815 from_kgid(&init_user_ns, inode->i_gid));
816 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
819 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
820 (issued & CEPH_CAP_LINK_EXCL) == 0)
821 set_nlink(inode, le32_to_cpu(info->nlink));
823 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
824 /* be careful with mtime, atime, size */
825 ceph_decode_timespec64(&atime, &info->atime);
826 ceph_decode_timespec64(&mtime, &info->mtime);
827 ceph_decode_timespec64(&ctime, &info->ctime);
828 ceph_fill_file_time(inode, issued,
829 le32_to_cpu(info->time_warp_seq),
830 &ctime, &mtime, &atime);
833 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
834 ci->i_files = le64_to_cpu(info->files);
835 ci->i_subdirs = le64_to_cpu(info->subdirs);
839 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
840 s64 old_pool = ci->i_layout.pool_id;
841 struct ceph_string *old_ns;
843 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
844 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
845 lockdep_is_held(&ci->i_ceph_lock));
846 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
848 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
849 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
853 queue_trunc = ceph_fill_file_size(inode, issued,
854 le32_to_cpu(info->truncate_seq),
855 le64_to_cpu(info->truncate_size),
856 le64_to_cpu(info->size));
857 /* only update max_size on auth cap */
858 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
859 ci->i_max_size != le64_to_cpu(info->max_size)) {
860 dout("max_size %lld -> %llu\n", ci->i_max_size,
861 le64_to_cpu(info->max_size));
862 ci->i_max_size = le64_to_cpu(info->max_size);
866 /* layout and rstat are not tracked by capability, update them if
867 * the inode info is from auth mds */
868 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
869 if (S_ISDIR(inode->i_mode)) {
870 ci->i_dir_layout = iinfo->dir_layout;
871 ci->i_rbytes = le64_to_cpu(info->rbytes);
872 ci->i_rfiles = le64_to_cpu(info->rfiles);
873 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
874 ci->i_dir_pin = iinfo->dir_pin;
875 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
880 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
881 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
882 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
883 if (ci->i_xattrs.blob)
884 ceph_buffer_put(ci->i_xattrs.blob);
885 ci->i_xattrs.blob = xattr_blob;
887 memcpy(ci->i_xattrs.blob->vec.iov_base,
888 iinfo->xattr_data, iinfo->xattr_len);
889 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
890 ceph_forget_all_cached_acls(inode);
894 /* finally update i_version */
895 if (le64_to_cpu(info->version) > ci->i_version)
896 ci->i_version = le64_to_cpu(info->version);
898 inode->i_mapping->a_ops = &ceph_aops;
900 switch (inode->i_mode & S_IFMT) {
905 inode->i_blkbits = PAGE_SHIFT;
906 init_special_inode(inode, inode->i_mode, inode->i_rdev);
907 inode->i_op = &ceph_file_iops;
910 inode->i_op = &ceph_file_iops;
911 inode->i_fop = &ceph_file_fops;
914 inode->i_op = &ceph_symlink_iops;
915 if (!ci->i_symlink) {
916 u32 symlen = iinfo->symlink_len;
919 spin_unlock(&ci->i_ceph_lock);
921 if (symlen != i_size_read(inode)) {
922 pr_err("fill_inode %llx.%llx BAD symlink "
923 "size %lld\n", ceph_vinop(inode),
925 i_size_write(inode, symlen);
926 inode->i_blocks = calc_inode_blocks(symlen);
930 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
934 spin_lock(&ci->i_ceph_lock);
938 kfree(sym); /* lost a race */
940 inode->i_link = ci->i_symlink;
943 inode->i_op = &ceph_dir_iops;
944 inode->i_fop = &ceph_dir_fops;
947 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
948 ceph_vinop(inode), inode->i_mode);
951 /* were we issued a capability? */
953 if (ceph_snap(inode) == CEPH_NOSNAP) {
954 ceph_add_cap(inode, session,
955 le64_to_cpu(info->cap.cap_id),
956 cap_fmode, info_caps,
957 le32_to_cpu(info->cap.wanted),
958 le32_to_cpu(info->cap.seq),
959 le32_to_cpu(info->cap.mseq),
960 le64_to_cpu(info->cap.realm),
961 info->cap.flags, &new_cap);
963 /* set dir completion flag? */
964 if (S_ISDIR(inode->i_mode) &&
965 ci->i_files == 0 && ci->i_subdirs == 0 &&
966 (info_caps & CEPH_CAP_FILE_SHARED) &&
967 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
968 !__ceph_dir_is_complete(ci)) {
969 dout(" marking %p complete (empty)\n", inode);
970 i_size_write(inode, 0);
971 __ceph_dir_set_complete(ci,
972 atomic64_read(&ci->i_release_count),
973 atomic64_read(&ci->i_ordered_count));
978 dout(" %p got snap_caps %s\n", inode,
979 ceph_cap_string(info_caps));
980 ci->i_snap_caps |= info_caps;
982 __ceph_get_fmode(ci, cap_fmode);
984 } else if (cap_fmode >= 0) {
985 pr_warn("mds issued no caps on %llx.%llx\n",
987 __ceph_get_fmode(ci, cap_fmode);
990 if (iinfo->inline_version > 0 &&
991 iinfo->inline_version >= ci->i_inline_version) {
992 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
993 ci->i_inline_version = iinfo->inline_version;
994 if (ci->i_inline_version != CEPH_INLINE_NONE &&
995 (locked_page || (info_caps & cache_caps)))
999 spin_unlock(&ci->i_ceph_lock);
1002 ceph_fill_inline_data(inode, locked_page,
1003 iinfo->inline_data, iinfo->inline_len);
1006 wake_up_all(&ci->i_cap_wq);
1008 /* queue truncate if we saw i_size decrease */
1010 ceph_queue_vmtruncate(inode);
1012 /* populate frag tree */
1013 if (S_ISDIR(inode->i_mode))
1014 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1016 /* update delegation info? */
1018 ceph_fill_dirfrag(inode, dirinfo);
1023 ceph_put_cap(mdsc, new_cap);
1025 ceph_buffer_put(xattr_blob);
1026 ceph_put_string(pool_ns);
1031 * caller should hold session s_mutex and dentry->d_lock.
1033 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1034 struct ceph_mds_reply_lease *lease,
1035 struct ceph_mds_session *session,
1036 unsigned long from_time,
1037 struct ceph_mds_session **old_lease_session)
1039 struct ceph_dentry_info *di = ceph_dentry(dentry);
1040 long unsigned duration = le32_to_cpu(lease->duration_ms);
1041 long unsigned ttl = from_time + (duration * HZ) / 1000;
1042 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1044 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1045 dentry, duration, ttl);
1047 /* only track leases on regular dentries */
1048 if (ceph_snap(dir) != CEPH_NOSNAP)
1051 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1052 if (duration == 0) {
1053 __ceph_dentry_dir_lease_touch(di);
1057 if (di->lease_gen == session->s_cap_gen &&
1058 time_before(ttl, di->time))
1059 return; /* we already have a newer lease. */
1061 if (di->lease_session && di->lease_session != session) {
1062 *old_lease_session = di->lease_session;
1063 di->lease_session = NULL;
1066 if (!di->lease_session)
1067 di->lease_session = ceph_get_mds_session(session);
1068 di->lease_gen = session->s_cap_gen;
1069 di->lease_seq = le32_to_cpu(lease->seq);
1070 di->lease_renew_after = half_ttl;
1071 di->lease_renew_from = 0;
1074 __ceph_dentry_lease_touch(di);
1077 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1078 struct ceph_mds_reply_lease *lease,
1079 struct ceph_mds_session *session,
1080 unsigned long from_time)
1082 struct ceph_mds_session *old_lease_session = NULL;
1083 spin_lock(&dentry->d_lock);
1084 __update_dentry_lease(dir, dentry, lease, session, from_time,
1085 &old_lease_session);
1086 spin_unlock(&dentry->d_lock);
1087 if (old_lease_session)
1088 ceph_put_mds_session(old_lease_session);
1092 * update dentry lease without having parent inode locked
1094 static void update_dentry_lease_careful(struct dentry *dentry,
1095 struct ceph_mds_reply_lease *lease,
1096 struct ceph_mds_session *session,
1097 unsigned long from_time,
1098 char *dname, u32 dname_len,
1099 struct ceph_vino *pdvino,
1100 struct ceph_vino *ptvino)
1104 struct ceph_mds_session *old_lease_session = NULL;
1106 spin_lock(&dentry->d_lock);
1107 /* make sure dentry's name matches target */
1108 if (dentry->d_name.len != dname_len ||
1109 memcmp(dentry->d_name.name, dname, dname_len))
1112 dir = d_inode(dentry->d_parent);
1113 /* make sure parent matches dvino */
1114 if (!ceph_ino_compare(dir, pdvino))
1117 /* make sure dentry's inode matches target. NULL ptvino means that
1118 * we expect a negative dentry */
1120 if (d_really_is_negative(dentry))
1122 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1125 if (d_really_is_positive(dentry))
1129 __update_dentry_lease(dir, dentry, lease, session,
1130 from_time, &old_lease_session);
1132 spin_unlock(&dentry->d_lock);
1133 if (old_lease_session)
1134 ceph_put_mds_session(old_lease_session);
1138 * splice a dentry to an inode.
1139 * caller must hold directory i_mutex for this to be safe.
1141 static int splice_dentry(struct dentry **pdn, struct inode *in)
1143 struct dentry *dn = *pdn;
1144 struct dentry *realdn;
1146 BUG_ON(d_inode(dn));
1148 if (S_ISDIR(in->i_mode)) {
1149 /* If inode is directory, d_splice_alias() below will remove
1150 * 'realdn' from its origin parent. We need to ensure that
1151 * origin parent's readdir cache will not reference 'realdn'
1153 realdn = d_find_any_alias(in);
1155 struct ceph_dentry_info *di = ceph_dentry(realdn);
1156 spin_lock(&realdn->d_lock);
1158 realdn->d_op->d_prune(realdn);
1161 di->lease_shared_gen = 0;
1164 spin_unlock(&realdn->d_lock);
1169 /* dn must be unhashed */
1170 if (!d_unhashed(dn))
1172 realdn = d_splice_alias(in, dn);
1173 if (IS_ERR(realdn)) {
1174 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1175 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1176 return PTR_ERR(realdn);
1180 dout("dn %p (%d) spliced with %p (%d) "
1181 "inode %p ino %llx.%llx\n",
1183 realdn, d_count(realdn),
1184 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1188 BUG_ON(!ceph_dentry(dn));
1189 dout("dn %p attached to %p ino %llx.%llx\n",
1190 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1196 * Incorporate results into the local cache. This is either just
1197 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1200 * A reply may contain
1201 * a directory inode along with a dentry.
1202 * and/or a target inode
1204 * Called with snap_rwsem (read).
1206 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1208 struct ceph_mds_session *session = req->r_session;
1209 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1210 struct inode *in = NULL;
1211 struct ceph_vino tvino, dvino;
1212 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1215 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1216 rinfo->head->is_dentry, rinfo->head->is_target);
1218 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1219 dout("fill_trace reply is empty!\n");
1220 if (rinfo->head->result == 0 && req->r_parent)
1221 ceph_invalidate_dir_request(req);
1225 if (rinfo->head->is_dentry) {
1226 struct inode *dir = req->r_parent;
1229 err = fill_inode(dir, NULL,
1230 &rinfo->diri, rinfo->dirfrag,
1231 session, req->r_request_started, -1,
1232 &req->r_caps_reservation);
1239 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1240 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1241 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1243 struct dentry *dn, *parent;
1245 BUG_ON(!rinfo->head->is_target);
1246 BUG_ON(req->r_dentry);
1248 parent = d_find_any_alias(dir);
1251 dname.name = rinfo->dname;
1252 dname.len = rinfo->dname_len;
1253 dname.hash = full_name_hash(parent, dname.name, dname.len);
1254 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1255 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1257 dn = d_lookup(parent, &dname);
1258 dout("d_lookup on parent=%p name=%.*s got %p\n",
1259 parent, dname.len, dname.name, dn);
1262 dn = d_alloc(parent, &dname);
1263 dout("d_alloc %p '%.*s' = %p\n", parent,
1264 dname.len, dname.name, dn);
1271 } else if (d_really_is_positive(dn) &&
1272 (ceph_ino(d_inode(dn)) != tvino.ino ||
1273 ceph_snap(d_inode(dn)) != tvino.snap)) {
1274 dout(" dn %p points to wrong inode %p\n",
1276 ceph_dir_clear_ordered(dir);
1287 if (rinfo->head->is_target) {
1288 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1289 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1291 in = ceph_get_inode(sb, tvino);
1296 req->r_target_inode = in;
1298 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1299 session, req->r_request_started,
1300 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1301 rinfo->head->result == 0) ? req->r_fmode : -1,
1302 &req->r_caps_reservation);
1304 pr_err("fill_inode badness %p %llx.%llx\n",
1305 in, ceph_vinop(in));
1311 * ignore null lease/binding on snapdir ENOENT, or else we
1312 * will have trouble splicing in the virtual snapdir later
1314 if (rinfo->head->is_dentry &&
1315 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1316 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1317 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1318 fsc->mount_options->snapdir_name,
1319 req->r_dentry->d_name.len))) {
1321 * lookup link rename : null -> possibly existing inode
1322 * mknod symlink mkdir : null -> new inode
1323 * unlink : linked -> null
1325 struct inode *dir = req->r_parent;
1326 struct dentry *dn = req->r_dentry;
1327 bool have_dir_cap, have_lease;
1331 BUG_ON(d_inode(dn->d_parent) != dir);
1333 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1334 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1336 BUG_ON(ceph_ino(dir) != dvino.ino);
1337 BUG_ON(ceph_snap(dir) != dvino.snap);
1339 /* do we have a lease on the whole dir? */
1341 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1342 CEPH_CAP_FILE_SHARED);
1344 /* do we have a dn lease? */
1345 have_lease = have_dir_cap ||
1346 le32_to_cpu(rinfo->dlease->duration_ms);
1348 dout("fill_trace no dentry lease or dir cap\n");
1351 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1352 struct inode *olddir = req->r_old_dentry_dir;
1355 dout(" src %p '%pd' dst %p '%pd'\n",
1359 dout("fill_trace doing d_move %p -> %p\n",
1360 req->r_old_dentry, dn);
1362 /* d_move screws up sibling dentries' offsets */
1363 ceph_dir_clear_ordered(dir);
1364 ceph_dir_clear_ordered(olddir);
1366 d_move(req->r_old_dentry, dn);
1367 dout(" src %p '%pd' dst %p '%pd'\n",
1372 /* ensure target dentry is invalidated, despite
1373 rehashing bug in vfs_rename_dir */
1374 ceph_invalidate_dentry_lease(dn);
1376 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1377 ceph_dentry(req->r_old_dentry)->offset);
1379 /* swap r_dentry and r_old_dentry in case that
1380 * splice_dentry() gets called later. This is safe
1381 * because no other place will use them */
1382 req->r_dentry = req->r_old_dentry;
1383 req->r_old_dentry = dn;
1388 if (!rinfo->head->is_target) {
1389 dout("fill_trace null dentry\n");
1390 if (d_really_is_positive(dn)) {
1391 dout("d_delete %p\n", dn);
1392 ceph_dir_clear_ordered(dir);
1394 } else if (have_lease) {
1397 update_dentry_lease(dir, dn,
1398 rinfo->dlease, session,
1399 req->r_request_started);
1404 /* attach proper inode */
1405 if (d_really_is_negative(dn)) {
1406 ceph_dir_clear_ordered(dir);
1408 err = splice_dentry(&req->r_dentry, in);
1411 dn = req->r_dentry; /* may have spliced */
1412 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1413 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1414 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1421 update_dentry_lease(dir, dn,
1422 rinfo->dlease, session,
1423 req->r_request_started);
1425 dout(" final dn %p\n", dn);
1426 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1427 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1428 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1429 struct inode *dir = req->r_parent;
1431 /* fill out a snapdir LOOKUPSNAP dentry */
1433 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1434 BUG_ON(!req->r_dentry);
1435 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1436 ceph_dir_clear_ordered(dir);
1438 err = splice_dentry(&req->r_dentry, in);
1441 } else if (rinfo->head->is_dentry && req->r_dentry) {
1442 /* parent inode is not locked, be carefull */
1443 struct ceph_vino *ptvino = NULL;
1444 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1445 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1446 if (rinfo->head->is_target) {
1447 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1448 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1451 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1452 session, req->r_request_started,
1453 rinfo->dname, rinfo->dname_len,
1457 dout("fill_trace done err=%d\n", err);
1462 * Prepopulate our cache with readdir results, leases, etc.
1464 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1465 struct ceph_mds_session *session)
1467 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1470 for (i = 0; i < rinfo->dir_nr; i++) {
1471 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1472 struct ceph_vino vino;
1476 vino.ino = le64_to_cpu(rde->inode.in->ino);
1477 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1479 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1482 dout("new_inode badness got %d\n", err);
1485 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1486 req->r_request_started, -1,
1487 &req->r_caps_reservation);
1489 pr_err("fill_inode badness on %p got %d\n", in, rc);
1492 /* avoid calling iput_final() in mds dispatch threads */
1493 ceph_async_iput(in);
1499 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1503 put_page(ctl->page);
1508 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1509 struct ceph_readdir_cache_control *ctl,
1510 struct ceph_mds_request *req)
1512 struct ceph_inode_info *ci = ceph_inode(dir);
1513 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1514 unsigned idx = ctl->index % nsize;
1515 pgoff_t pgoff = ctl->index / nsize;
1517 if (!ctl->page || pgoff != page_index(ctl->page)) {
1518 ceph_readdir_cache_release(ctl);
1520 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1522 ctl->page = find_lock_page(&dir->i_data, pgoff);
1525 return idx == 0 ? -ENOMEM : 0;
1527 /* reading/filling the cache are serialized by
1528 * i_mutex, no need to use page lock */
1529 unlock_page(ctl->page);
1530 ctl->dentries = kmap(ctl->page);
1532 memset(ctl->dentries, 0, PAGE_SIZE);
1535 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1536 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1537 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1538 ctl->dentries[idx] = dn;
1541 dout("disable readdir cache\n");
1547 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1548 struct ceph_mds_session *session)
1550 struct dentry *parent = req->r_dentry;
1551 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1552 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1556 int err = 0, skipped = 0, ret, i;
1557 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1558 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1561 struct ceph_readdir_cache_control cache_ctl = {};
1563 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1564 return readdir_prepopulate_inodes_only(req, session);
1566 if (rinfo->hash_order) {
1568 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1570 strlen(req->r_path2));
1571 last_hash = ceph_frag_value(last_hash);
1572 } else if (rinfo->offset_hash) {
1573 /* mds understands offset_hash */
1574 WARN_ON_ONCE(req->r_readdir_offset != 2);
1575 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1579 if (rinfo->dir_dir &&
1580 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1581 dout("readdir_prepopulate got new frag %x -> %x\n",
1582 frag, le32_to_cpu(rinfo->dir_dir->frag));
1583 frag = le32_to_cpu(rinfo->dir_dir->frag);
1584 if (!rinfo->hash_order)
1585 req->r_readdir_offset = 2;
1588 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1589 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1590 rinfo->dir_nr, parent);
1592 dout("readdir_prepopulate %d items under dn %p\n",
1593 rinfo->dir_nr, parent);
1595 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1597 if (ceph_frag_is_leftmost(frag) &&
1598 req->r_readdir_offset == 2 &&
1599 !(rinfo->hash_order && last_hash)) {
1600 /* note dir version at start of readdir so we can
1601 * tell if any dentries get dropped */
1602 req->r_dir_release_cnt =
1603 atomic64_read(&ci->i_release_count);
1604 req->r_dir_ordered_cnt =
1605 atomic64_read(&ci->i_ordered_count);
1606 req->r_readdir_cache_idx = 0;
1610 cache_ctl.index = req->r_readdir_cache_idx;
1611 fpos_offset = req->r_readdir_offset;
1613 /* FIXME: release caps/leases if error occurs */
1614 for (i = 0; i < rinfo->dir_nr; i++) {
1615 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1616 struct ceph_vino tvino;
1618 dname.name = rde->name;
1619 dname.len = rde->name_len;
1620 dname.hash = full_name_hash(parent, dname.name, dname.len);
1622 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1623 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1625 if (rinfo->hash_order) {
1626 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1627 rde->name, rde->name_len);
1628 hash = ceph_frag_value(hash);
1629 if (hash != last_hash)
1632 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1634 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1638 dn = d_lookup(parent, &dname);
1639 dout("d_lookup on parent=%p name=%.*s got %p\n",
1640 parent, dname.len, dname.name, dn);
1643 dn = d_alloc(parent, &dname);
1644 dout("d_alloc %p '%.*s' = %p\n", parent,
1645 dname.len, dname.name, dn);
1647 dout("d_alloc badness\n");
1651 } else if (d_really_is_positive(dn) &&
1652 (ceph_ino(d_inode(dn)) != tvino.ino ||
1653 ceph_snap(d_inode(dn)) != tvino.snap)) {
1654 struct ceph_dentry_info *di = ceph_dentry(dn);
1655 dout(" dn %p points to wrong inode %p\n",
1658 spin_lock(&dn->d_lock);
1659 if (di->offset > 0 &&
1660 di->lease_shared_gen ==
1661 atomic_read(&ci->i_shared_gen)) {
1662 __ceph_dir_clear_ordered(ci);
1665 spin_unlock(&dn->d_lock);
1673 if (d_really_is_positive(dn)) {
1676 in = ceph_get_inode(parent->d_sb, tvino);
1678 dout("new_inode badness\n");
1686 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1687 req->r_request_started, -1,
1688 &req->r_caps_reservation);
1690 pr_err("fill_inode badness on %p\n", in);
1691 if (d_really_is_negative(dn)) {
1692 /* avoid calling iput_final() in mds
1693 * dispatch threads */
1694 ceph_async_iput(in);
1701 if (d_really_is_negative(dn)) {
1702 if (ceph_security_xattr_deadlock(in)) {
1703 dout(" skip splicing dn %p to inode %p"
1704 " (security xattr deadlock)\n", dn, in);
1705 ceph_async_iput(in);
1710 err = splice_dentry(&dn, in);
1715 ceph_dentry(dn)->offset = rde->offset;
1717 update_dentry_lease(d_inode(parent), dn,
1718 rde->lease, req->r_session,
1719 req->r_request_started);
1721 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1722 ret = fill_readdir_cache(d_inode(parent), dn,
1731 if (err == 0 && skipped == 0) {
1732 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1733 req->r_readdir_cache_idx = cache_ctl.index;
1735 ceph_readdir_cache_release(&cache_ctl);
1736 dout("readdir_prepopulate done\n");
1740 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1742 struct ceph_inode_info *ci = ceph_inode(inode);
1745 spin_lock(&ci->i_ceph_lock);
1746 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1747 i_size_write(inode, size);
1748 inode->i_blocks = calc_inode_blocks(size);
1750 ret = __ceph_should_report_size(ci);
1752 spin_unlock(&ci->i_ceph_lock);
1757 * Put reference to inode, but avoid calling iput_final() in current thread.
1758 * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1761 void ceph_async_iput(struct inode *inode)
1766 if (atomic_add_unless(&inode->i_count, -1, 1))
1768 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1769 &ceph_inode(inode)->i_work))
1771 /* queue work failed, i_count must be at least 2 */
1776 * Write back inode data in a worker thread. (This can't be done
1777 * in the message handler context.)
1779 void ceph_queue_writeback(struct inode *inode)
1781 struct ceph_inode_info *ci = ceph_inode(inode);
1782 set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1785 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1787 dout("ceph_queue_writeback %p\n", inode);
1789 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1790 inode, ci->i_work_mask);
1796 * queue an async invalidation
1798 void ceph_queue_invalidate(struct inode *inode)
1800 struct ceph_inode_info *ci = ceph_inode(inode);
1801 set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1804 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1805 &ceph_inode(inode)->i_work)) {
1806 dout("ceph_queue_invalidate %p\n", inode);
1808 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1809 inode, ci->i_work_mask);
1815 * Queue an async vmtruncate. If we fail to queue work, we will handle
1816 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1818 void ceph_queue_vmtruncate(struct inode *inode)
1820 struct ceph_inode_info *ci = ceph_inode(inode);
1821 set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1824 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1826 dout("ceph_queue_vmtruncate %p\n", inode);
1828 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1829 inode, ci->i_work_mask);
1834 static void ceph_do_invalidate_pages(struct inode *inode)
1836 struct ceph_inode_info *ci = ceph_inode(inode);
1837 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1841 mutex_lock(&ci->i_truncate_mutex);
1843 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1844 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1845 inode, ceph_ino(inode));
1846 mapping_set_error(inode->i_mapping, -EIO);
1847 truncate_pagecache(inode, 0);
1848 mutex_unlock(&ci->i_truncate_mutex);
1852 spin_lock(&ci->i_ceph_lock);
1853 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1854 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1855 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1856 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1858 spin_unlock(&ci->i_ceph_lock);
1859 mutex_unlock(&ci->i_truncate_mutex);
1862 orig_gen = ci->i_rdcache_gen;
1863 spin_unlock(&ci->i_ceph_lock);
1865 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1866 pr_err("invalidate_pages %p fails\n", inode);
1869 spin_lock(&ci->i_ceph_lock);
1870 if (orig_gen == ci->i_rdcache_gen &&
1871 orig_gen == ci->i_rdcache_revoking) {
1872 dout("invalidate_pages %p gen %d successful\n", inode,
1874 ci->i_rdcache_revoking--;
1877 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1878 inode, orig_gen, ci->i_rdcache_gen,
1879 ci->i_rdcache_revoking);
1880 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1883 spin_unlock(&ci->i_ceph_lock);
1884 mutex_unlock(&ci->i_truncate_mutex);
1887 ceph_check_caps(ci, 0, NULL);
1891 * Make sure any pending truncation is applied before doing anything
1892 * that may depend on it.
1894 void __ceph_do_pending_vmtruncate(struct inode *inode)
1896 struct ceph_inode_info *ci = ceph_inode(inode);
1898 int wrbuffer_refs, finish = 0;
1900 mutex_lock(&ci->i_truncate_mutex);
1902 spin_lock(&ci->i_ceph_lock);
1903 if (ci->i_truncate_pending == 0) {
1904 dout("__do_pending_vmtruncate %p none pending\n", inode);
1905 spin_unlock(&ci->i_ceph_lock);
1906 mutex_unlock(&ci->i_truncate_mutex);
1911 * make sure any dirty snapped pages are flushed before we
1912 * possibly truncate them.. so write AND block!
1914 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1915 spin_unlock(&ci->i_ceph_lock);
1916 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1918 filemap_write_and_wait_range(&inode->i_data, 0,
1919 inode->i_sb->s_maxbytes);
1923 /* there should be no reader or writer */
1924 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1926 to = ci->i_truncate_size;
1927 wrbuffer_refs = ci->i_wrbuffer_ref;
1928 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1929 ci->i_truncate_pending, to);
1930 spin_unlock(&ci->i_ceph_lock);
1932 truncate_pagecache(inode, to);
1934 spin_lock(&ci->i_ceph_lock);
1935 if (to == ci->i_truncate_size) {
1936 ci->i_truncate_pending = 0;
1939 spin_unlock(&ci->i_ceph_lock);
1943 mutex_unlock(&ci->i_truncate_mutex);
1945 if (wrbuffer_refs == 0)
1946 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1948 wake_up_all(&ci->i_cap_wq);
1951 static void ceph_inode_work(struct work_struct *work)
1953 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1955 struct inode *inode = &ci->vfs_inode;
1957 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1958 dout("writeback %p\n", inode);
1959 filemap_fdatawrite(&inode->i_data);
1961 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1962 ceph_do_invalidate_pages(inode);
1964 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1965 __ceph_do_pending_vmtruncate(inode);
1973 static const struct inode_operations ceph_symlink_iops = {
1974 .get_link = simple_get_link,
1975 .setattr = ceph_setattr,
1976 .getattr = ceph_getattr,
1977 .listxattr = ceph_listxattr,
1980 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1982 struct ceph_inode_info *ci = ceph_inode(inode);
1983 const unsigned int ia_valid = attr->ia_valid;
1984 struct ceph_mds_request *req;
1985 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1986 struct ceph_cap_flush *prealloc_cf;
1988 int release = 0, dirtied = 0;
1991 int inode_dirty_flags = 0;
1992 bool lock_snap_rwsem = false;
1994 prealloc_cf = ceph_alloc_cap_flush();
1998 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2001 ceph_free_cap_flush(prealloc_cf);
2002 return PTR_ERR(req);
2005 spin_lock(&ci->i_ceph_lock);
2006 issued = __ceph_caps_issued(ci, NULL);
2008 if (!ci->i_head_snapc &&
2009 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2010 lock_snap_rwsem = true;
2011 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2012 spin_unlock(&ci->i_ceph_lock);
2013 down_read(&mdsc->snap_rwsem);
2014 spin_lock(&ci->i_ceph_lock);
2015 issued = __ceph_caps_issued(ci, NULL);
2019 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2021 if (ia_valid & ATTR_UID) {
2022 dout("setattr %p uid %d -> %d\n", inode,
2023 from_kuid(&init_user_ns, inode->i_uid),
2024 from_kuid(&init_user_ns, attr->ia_uid));
2025 if (issued & CEPH_CAP_AUTH_EXCL) {
2026 inode->i_uid = attr->ia_uid;
2027 dirtied |= CEPH_CAP_AUTH_EXCL;
2028 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2029 !uid_eq(attr->ia_uid, inode->i_uid)) {
2030 req->r_args.setattr.uid = cpu_to_le32(
2031 from_kuid(&init_user_ns, attr->ia_uid));
2032 mask |= CEPH_SETATTR_UID;
2033 release |= CEPH_CAP_AUTH_SHARED;
2036 if (ia_valid & ATTR_GID) {
2037 dout("setattr %p gid %d -> %d\n", inode,
2038 from_kgid(&init_user_ns, inode->i_gid),
2039 from_kgid(&init_user_ns, attr->ia_gid));
2040 if (issued & CEPH_CAP_AUTH_EXCL) {
2041 inode->i_gid = attr->ia_gid;
2042 dirtied |= CEPH_CAP_AUTH_EXCL;
2043 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2044 !gid_eq(attr->ia_gid, inode->i_gid)) {
2045 req->r_args.setattr.gid = cpu_to_le32(
2046 from_kgid(&init_user_ns, attr->ia_gid));
2047 mask |= CEPH_SETATTR_GID;
2048 release |= CEPH_CAP_AUTH_SHARED;
2051 if (ia_valid & ATTR_MODE) {
2052 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2054 if (issued & CEPH_CAP_AUTH_EXCL) {
2055 inode->i_mode = attr->ia_mode;
2056 dirtied |= CEPH_CAP_AUTH_EXCL;
2057 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2058 attr->ia_mode != inode->i_mode) {
2059 inode->i_mode = attr->ia_mode;
2060 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2061 mask |= CEPH_SETATTR_MODE;
2062 release |= CEPH_CAP_AUTH_SHARED;
2066 if (ia_valid & ATTR_ATIME) {
2067 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2068 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2069 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2070 if (issued & CEPH_CAP_FILE_EXCL) {
2071 ci->i_time_warp_seq++;
2072 inode->i_atime = attr->ia_atime;
2073 dirtied |= CEPH_CAP_FILE_EXCL;
2074 } else if ((issued & CEPH_CAP_FILE_WR) &&
2075 timespec64_compare(&inode->i_atime,
2076 &attr->ia_atime) < 0) {
2077 inode->i_atime = attr->ia_atime;
2078 dirtied |= CEPH_CAP_FILE_WR;
2079 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2080 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2081 ceph_encode_timespec64(&req->r_args.setattr.atime,
2083 mask |= CEPH_SETATTR_ATIME;
2084 release |= CEPH_CAP_FILE_SHARED |
2085 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2088 if (ia_valid & ATTR_MTIME) {
2089 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2090 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2091 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2092 if (issued & CEPH_CAP_FILE_EXCL) {
2093 ci->i_time_warp_seq++;
2094 inode->i_mtime = attr->ia_mtime;
2095 dirtied |= CEPH_CAP_FILE_EXCL;
2096 } else if ((issued & CEPH_CAP_FILE_WR) &&
2097 timespec64_compare(&inode->i_mtime,
2098 &attr->ia_mtime) < 0) {
2099 inode->i_mtime = attr->ia_mtime;
2100 dirtied |= CEPH_CAP_FILE_WR;
2101 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2102 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2103 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2105 mask |= CEPH_SETATTR_MTIME;
2106 release |= CEPH_CAP_FILE_SHARED |
2107 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2110 if (ia_valid & ATTR_SIZE) {
2111 dout("setattr %p size %lld -> %lld\n", inode,
2112 inode->i_size, attr->ia_size);
2113 if ((issued & CEPH_CAP_FILE_EXCL) &&
2114 attr->ia_size > inode->i_size) {
2115 i_size_write(inode, attr->ia_size);
2116 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2117 ci->i_reported_size = attr->ia_size;
2118 dirtied |= CEPH_CAP_FILE_EXCL;
2119 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2120 attr->ia_size != inode->i_size) {
2121 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2122 req->r_args.setattr.old_size =
2123 cpu_to_le64(inode->i_size);
2124 mask |= CEPH_SETATTR_SIZE;
2125 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2126 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2130 /* these do nothing */
2131 if (ia_valid & ATTR_CTIME) {
2132 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2133 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2134 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2135 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2136 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2137 only ? "ctime only" : "ignored");
2140 * if kernel wants to dirty ctime but nothing else,
2141 * we need to choose a cap to dirty under, or do
2142 * a almost-no-op setattr
2144 if (issued & CEPH_CAP_AUTH_EXCL)
2145 dirtied |= CEPH_CAP_AUTH_EXCL;
2146 else if (issued & CEPH_CAP_FILE_EXCL)
2147 dirtied |= CEPH_CAP_FILE_EXCL;
2148 else if (issued & CEPH_CAP_XATTR_EXCL)
2149 dirtied |= CEPH_CAP_XATTR_EXCL;
2151 mask |= CEPH_SETATTR_CTIME;
2154 if (ia_valid & ATTR_FILE)
2155 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2158 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2160 inode->i_ctime = attr->ia_ctime;
2164 spin_unlock(&ci->i_ceph_lock);
2165 if (lock_snap_rwsem)
2166 up_read(&mdsc->snap_rwsem);
2168 if (inode_dirty_flags)
2169 __mark_inode_dirty(inode, inode_dirty_flags);
2173 req->r_inode = inode;
2175 req->r_inode_drop = release;
2176 req->r_args.setattr.mask = cpu_to_le32(mask);
2177 req->r_num_caps = 1;
2178 req->r_stamp = attr->ia_ctime;
2179 err = ceph_mdsc_do_request(mdsc, NULL, req);
2181 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2182 ceph_cap_string(dirtied), mask);
2184 ceph_mdsc_put_request(req);
2185 ceph_free_cap_flush(prealloc_cf);
2187 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2188 __ceph_do_pending_vmtruncate(inode);
2196 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2198 struct inode *inode = d_inode(dentry);
2199 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2202 if (ceph_snap(inode) != CEPH_NOSNAP)
2205 err = setattr_prepare(dentry, attr);
2209 if ((attr->ia_valid & ATTR_SIZE) &&
2210 attr->ia_size > max(inode->i_size, fsc->max_file_size))
2213 if ((attr->ia_valid & ATTR_SIZE) &&
2214 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2217 err = __ceph_setattr(inode, attr);
2219 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2220 err = posix_acl_chmod(inode, attr->ia_mode);
2226 * Verify that we have a lease on the given mask. If not,
2227 * do a getattr against an mds.
2229 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2230 int mask, bool force)
2232 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2233 struct ceph_mds_client *mdsc = fsc->mdsc;
2234 struct ceph_mds_request *req;
2238 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2239 dout("do_getattr inode %p SNAPDIR\n", inode);
2243 dout("do_getattr inode %p mask %s mode 0%o\n",
2244 inode, ceph_cap_string(mask), inode->i_mode);
2245 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2248 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2249 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2251 return PTR_ERR(req);
2252 req->r_inode = inode;
2254 req->r_num_caps = 1;
2255 req->r_args.getattr.mask = cpu_to_le32(mask);
2256 req->r_locked_page = locked_page;
2257 err = ceph_mdsc_do_request(mdsc, NULL, req);
2258 if (locked_page && err == 0) {
2259 u64 inline_version = req->r_reply_info.targeti.inline_version;
2260 if (inline_version == 0) {
2261 /* the reply is supposed to contain inline data */
2263 } else if (inline_version == CEPH_INLINE_NONE) {
2266 err = req->r_reply_info.targeti.inline_len;
2269 ceph_mdsc_put_request(req);
2270 dout("do_getattr result=%d\n", err);
2276 * Check inode permissions. We verify we have a valid value for
2277 * the AUTH cap, then call the generic handler.
2279 int ceph_permission(struct inode *inode, int mask)
2283 if (mask & MAY_NOT_BLOCK)
2286 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2289 err = generic_permission(inode, mask);
2293 /* Craft a mask of needed caps given a set of requested statx attrs. */
2294 static int statx_to_caps(u32 want)
2298 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME))
2299 mask |= CEPH_CAP_AUTH_SHARED;
2301 if (want & (STATX_NLINK|STATX_CTIME))
2302 mask |= CEPH_CAP_LINK_SHARED;
2304 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2306 mask |= CEPH_CAP_FILE_SHARED;
2308 if (want & (STATX_CTIME))
2309 mask |= CEPH_CAP_XATTR_SHARED;
2315 * Get all the attributes. If we have sufficient caps for the requested attrs,
2316 * then we can avoid talking to the MDS at all.
2318 int ceph_getattr(const struct path *path, struct kstat *stat,
2319 u32 request_mask, unsigned int flags)
2321 struct inode *inode = d_inode(path->dentry);
2322 struct ceph_inode_info *ci = ceph_inode(inode);
2325 /* Skip the getattr altogether if we're asked not to sync */
2326 if (!(flags & AT_STATX_DONT_SYNC)) {
2327 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2328 flags & AT_STATX_FORCE_SYNC);
2333 generic_fillattr(inode, stat);
2334 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2335 if (ceph_snap(inode) == CEPH_NOSNAP)
2336 stat->dev = inode->i_sb->s_dev;
2338 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2340 if (S_ISDIR(inode->i_mode)) {
2341 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2343 stat->size = ci->i_rbytes;
2345 stat->size = ci->i_files + ci->i_subdirs;
2347 stat->blksize = 65536;
2349 * Some applications rely on the number of st_nlink
2350 * value on directories to be either 0 (if unlinked)
2351 * or 2 + number of subdirectories.
2353 if (stat->nlink == 1)
2354 /* '.' + '..' + subdirs */
2355 stat->nlink = 1 + 1 + ci->i_subdirs;
2358 /* Mask off any higher bits (e.g. btime) until we have support */
2359 stat->result_mask = request_mask & STATX_BASIC_STATS;